Attribute Prediction Using Attribute Combinations
A method and system are presented in which statistical predictions are generated to indicate whether an individual has or will acquire an attribute designated in a query. The predictions are generated based on a first set of attributes associated with the individual and a second set of attribute combinations and statistical results that indicate the strength of association of each attribute combination with the query attribute.
Latest EXPANSE NETWORKS, INC. Patents:
This application claims priority to U.S. Provisional Application Ser. No. 60895236, which was filed on Mar. 16, 2007, and which is incorporated herein by reference in its entirety.
BRIEF DESCRIPTION OF THE DRAWINGSThe following detailed description will be better understood when read in conjunction with the appended drawings, in which there is shown one or more of the multiple embodiments of the present invention. It should be understood, however, that the various embodiments are not limited to the precise arrangements and instrumentalities shown in the drawings.
Described herein are methods, computer systems, databases and software for identifying combinations of attributes associated with individuals that co-occur with key attributes, such as specific disorders, behaviors and traits. Also described are databases as well as database systems for creating and accessing databases describing those attributes and for performing analyses based on those attributes. The methods, computer systems and software are useful for identifying intricate combinations of attributes that predispose human beings toward having or developing specific disorders, behaviors and traits of interest, determining the level of predisposition of an individual towards such attributes, and revealing which attribute associations can be added or eliminated to effectively modify what may have been hereto believed to be destiny. The methods, computer systems and software are also applicable for tissues and non-human organisms, as well as for identifying combinations of attributes that correlate with or cause behaviors and outcomes in complex non-living systems including molecules, electrical and mechanical systems and various devices and apparatus whose functionality is dependent on a multitude of attributes.
Previous methods have been largely unsuccessful in determining the complex combinations of attributes that predispose individuals to most disorders, behaviors and traits. The level of resolution afforded by the data typically used is too low, the number and types of attributes considered is too limited, and the sensitivity to detect low frequency, high complexity combinations is lacking. The desirability of being able to determine the complex combinations of attributes that predispose an individual to physical or behavioral disorders has clear implications for improving individualized diagnoses, choosing the most effective therapeutic regimens, making beneficial lifestyle changes that prevent disease and promote health, and reducing associated health care expenditures. It is also desirable to determine those combinations of attributes that promote certain behaviors and traits such as success in sports, music, school, leadership, career and relationships.
Advances in technology within the field of genetics, provides the ability to achieve maximum resolution of the entire genome. Discovery and characterization of epigenetic modifications—reversible chemical modifications of DNA and structural modification of chromatin that dramatically alter gene expression—has provided an additional level of information that may be altered due to environmental conditions, life experiences and aging. Along with a collection of diverse nongenetic attributes including physical, behavioral, situational and historical attributes associated with an organism, the present invention provides the ability to utilize the above information to enable prediction of the predisposition of an organism toward developing a specific attribute of interest provided in a query.
There are approximately 25,000 genes in the human genome. Of these, approximately 1,000 of these genes are involved in monogenic disorders, which are disorders whose sole cause is due to the properties of a single gene. This collection of disorders represents less than two percent of all human disorders. The remaining 98 percent of human disorders, termed complex disorders, are caused by multiple genetic influences or a combination of multiple genetic and non-genetic influences, still yet to be determined due to their resistance to current methods of discovery.
Previous methods using genetic information have suffered from either a lack of high resolution information, very limited coverage of total genomic information, or both. Genetic markers such as single nucleotide polymorphisms (SNPs) do not provide a complete picture of a gene's nucleotide sequence or the total genetic variability of the individual. The SNPs typically used occur at a frequency of at least 5% in the population. However, the majority of genetic variation that exists in the population occurs at frequencies below 1%. Furthermore, SNPs are spaced hundreds of nucleotides apart and do not account for genetic variation that occurs in the genetic sequence lying between, which is vastly more sequence than the single nucleotide position represented by an SNP. SNPs are typically located within gene coding regions and do not allow consideration of 98% of the 3 billion base pairs of genetic code in the human genome that does not encode gene sequences. Other markers such as STS, gene locus markers and chromosome loci markers also provide very low resolution and incomplete coverage of the genome. Complete and partial sequencing of an individual's genome provides the ability to incorporate that detailed information into the analysis of factors contributing toward expressed attributes.
Genomic influence on traits is now known to involve more than just the DNA nucleotide sequence of the genome. Regulation of expression of the genome can be influenced significantly by epigenetic modification of the genomic DNA and chromatin (3-dimensional genomic DNA with bound proteins). Termed the epigenome, this additional level of information can make genes in an individual's genome behave as if they were absent. Epigenetic modification can dramatically affect the expression of approximately at least 6% of all genes.
Epigenetic modification silences the activity of gene regulatory regions required to permit gene expression. Genes can undergo epigenetic silencing as a result of methylation of cytosines occurring in CpG dinucleotide motifs, and to a lesser extent by deacetylation of chromatin-associated histone proteins which inhibit gene expression by creating 3-dimensional conformational changes in chromatin. Assays such as bisulfite sequencing, differential methyl hybridization using microarrays, methylation sensitive polymerase chain reaction, and mass spectrometry enable the detection of cytosine nucleotide methylation while chromosome immunoprecipitation (CHIP) can be used to detect histone acetylation states of chromatin.
In one embodiment, epigenetic attributes are incorporated in the present invention to provide certain functionality. First, major mental disorders such as schizophrenia and bipolar mood disorder are thought to be caused by or at least greatly influenced by epigenetic imprinting of genes. Second, all epigenetic modification characterized to date is reversible in nature, allowing for the potential therapeutic manipulation of the epigenome to alter the course and occurrence of disease and certain behaviors. Third, because epigenetic modification of the genome occurs in response to experiences and stimuli encountered during prenatal and postnatal life, epigenetic data can help fill gaps resulting from unobtainable personal data, and reinforce or even substitute for unreliable self-reported data such as life experiences and environmental exposures.
In addition to genetic and epigenetic attributes, numerous other attributes likely influence the development of traits and disorders. The remaining attributes can be classified as either physical, behavioral, situational or historical.
In the present invention the term ‘attributes’ rather than the term ‘factors’ is used since many of the entities are characteristics associated with an individual that may have no influence on the vast majority of their traits, behaviors and disorders. As such, there may be many instances during execution of the methods described herein when a particular attribute does not act as a factor in determining predisposition. Nonetheless, every attribute remains a potentially important characteristic of the individual and may contribute to predisposition toward some other attribute or subset of attributes queried during subsequent or future implementation of the methods described herein. An individual possesses many associated attributes which may be collectively referred to as an attribute profile associated with that individual. In one embodiment, an attribute profile can be considered as being comprised of the attributes that are present (i.e., occur) in that profile, as well as being comprised of the various combinations (i.e., combinations and subcombinations) of those attributes. The attribute profile of an individual is preferably provided to embodiments of the present invention as a dataset record whose association with the individual can be indicated by a unique identifier contained in the dataset record. An actual attribute of an individual can be represented by an attribute descriptor in attribute profiles, records, datasets, and databases. Herein, both actual attributes and attribute descriptors may be referred to simply as attributes. In one embodiment, statistical relationships and associations between attribute descriptors are a direct result of relationships and associations between actual attributes of an individual. In the present disclosure, the term ‘individual’ can refer to an individual group, person, organism, organ, tissue, cell, virus, molecule, thing, entity or state, wherein a state includes but is not limited to a state-of-being, an operational state or a status. Individuals, attribute profiles and attributes can be real and/or measurable, or they may be hypothetical and/or not directly observable.
In one embodiment the present invention can be used to discover combinations of attributes regardless of number or type, in a population of any size, that cause predisposition to an attribute of interest. In doing so, this embodiment also has the ability to provide a list of attributes one can add or subtract from an existing profile of attributes in order to respectively increase or decrease the strength of predisposition toward the attribute of interest. The ability to accurately detect predisposing attribute combinations naturally benefits from being supplied with datasets representing large numbers of individuals and having a large number and variety of attributes for each. Nevertheless, the present invention will function properly with a minimal number of individuals and attributes. One embodiment of the present invention can be used to detect not only attributes that have a direct (causal) effect on an attribute of interest, but also those attributes that do not have a direct effect such as instrumental variables (i.e., correlative attributes), which are attributes that correlate with and can be used to predict predisposition for the attribute of interest but are not causal. For simplicity of terminology, both types of attributes are referred to herein as predisposing attributes, or simply attributes, that contribute toward predisposition toward the attribute of interest, regardless of whether the contribution or correlation is direct or indirect.
It is beneficial, but not necessary, in most instances, that the individuals whose data is supplied for the method be representative of the individual or population of individuals for which the predictions are desired. In a preferred embodiment, the attribute categories collectively encompass all potential attributes of an individual. Each attribute of an individual can be appropriately placed in one or more attribute categories of the methods, system and software of the invention. Attributes and the various categories of attributes can be defined as follows:
-
- a) attribute: a quality, trait, characteristic, relationship, property, factor or object associated with or possessed by an individual;
- b) genetic attribute: any genome, genotype, haplotype, chromatin, chromosome, chromosome locus, chromosomal material, deoxyribonucleic acid (DNA), allele, gene, gene cluster, gene locus, gene polymorphism, gene marker, nucleotide, single nucleotide polymorphism (SNP), restriction fragment length polymorphism (RFLP), variable tandem repeat (VTR), genetic marker, sequence marker, sequence tagged site (STS), plasmid, transcription unit, transcription product, ribonucleic acid (RNA), and copy DNA (cDNA), including the nucleotide sequence and encoded amino acid sequence of any of the above;
- c) epigenetic attribute: any feature of the genetic material—all genomic, vector and plasmid DNA, and chromatin—that affects gene expression in a manner that is heritable during somatic cell divisions and sometimes heritable in germline transmission, but that is nonmutational to the DNA sequence and is therefore fundamentally reversible, including but not limited to methylation of DNA nucleotides and acetylation of chromatin-associated histone proteins;
- d) pangenetic attribute: any genetic or epigenetic attribute;
- e) physical attribute: any material quality, trait, characteristic, property or factor of an individual present at the atomic, molecular, cellular, tissue, organ or organism level, excluding genetic and epigenetic attributes;
- f) behavioral attribute: any singular, periodic, or aperiodic response, action or habit of an individual to internal or external stimuli, including but not limited to an action, reflex, emotion or psychological state that is controlled or created by the nervous system on either a conscious or subconscious level;
- g) situational attribute: any object, condition, influence, or milieu that surrounds, impacts or contacts an individual;
- h) historical attribute: any genetic, epigenetic, physical, behavioral or situational attribute that was associated with or possessed by an individual in the past. As such, the historical attribute refers to a past state of the individual and may no longer describe the current state.
The methods, systems, software, and databases described herein apply to and are suitable for use with not only humans, but for other organisms as well. The methods, systems, software and databases may also be used for applications that consider attribute identification, predisposition potential and destiny modification for organs, tissues, individual cells, and viruses in vitro and in vivo. For example, the methods can be applied to behavior modification of individual cells being grown and studied in a laboratory incubator by providing pangenetic attributes of the cells, physical attributes of the cells such as size, shape and surface receptor densities, and situational attributes of the cells such as levels of oxygen and carbon dioxide in the incubator, temperature of the incubator, and levels of glucose and other nutrients in the liquid growth medium. Using these and other attributes, the methods, systems, software and databases can then be used to predict predisposition of the cells for such characteristics as susceptibility to infection by viruses, general growth rate, morphology, and differentiation potential. The methods, systems, software, and databases described herein can also be applied to complex non-living systems to, for example, predict the behavior of molecules or the performance of electrical devices or machinery subject to a large number of variables.
Conversion/formatting engine 220 of
In a preferred mode of comparison between nucleotide sequences, a direct sequence comparison that that requires two or more sequences to be the same at the nucleotide sequence level is performed. To increase efficiency at the cost of loosing information contained in non-gene-coding regions of the genome, a direct sequence comparison between genomic sequences may use only gene coding and gene regulatory sequences since these represent the expressed and expression-controlling portions of the genome, respectively. In embodiments where computing power and time are available, a comparison of the whole genome can be used as opposed to comparison of only the 2% which encodes genes and gene regulatory sequences since the noncoding region of the genome may still have effects on genome expression which influence attribute predisposition.
In one embodiment, comparison engine 222 is permitted some degree of flexibility in comparison of nucleotide sequences, so that the exact identity within protein encoding regions is not always required. For example, when a single nucleotide difference between two sequences is deemed unlikely to result in a functional difference between the two encoded proteins, it is beneficial to make the determination that the two sequences are the same even though they are actually not identical. The reason for allowing non-identical matches being that since the nucleotide difference is functionally silent it should not have a differential effect on attribute predisposition. A number of rules can be provided to comparison engine 222 to guide such decision making. These rules are based on the knowledge of several phenomena. For example, a single nucleotide difference in the 3rd nucleotide position of a codon—termed the wobble position—often does not change the identity of the amino acid encoded by the codon, and therefore may not change the amino acid sequence of the encoded protein. Determination of whether or not a particular nucleotide change in a wobble position alters the encoded protein amino acid sequence is easily made based on published information known to those in the art. Other exemplary types of changes that are unlikely to affect protein function are those that are known to be silent, those that result in conservative amino acid changes particularly in non-enzymatic, non-catalytic, nonantigenic or non-transmembrane domains of the protein, and those that simply alter the location of truncation of a protein within the same domain of one protein relative to another.
Allowing flexibility in sequence matching can increase the number of sequences determined to be identical, but may also reduce the sensitivity of the invention to detect predisposing attributes. There may be sequence changes which are thought to be innocuous or inconsequential based on current scientific knowledge that in actuality are not. For example, nucleotide changes in the wobble codon position that do not change the amino acid sequence may appear to be inconsequential, but may actually impact the stability of the intermediary RNA transcript required for translation of nucleotide sequence into the encoded protein, thus having a significant effect on ultimate levels of expressed protein. Therefore, application of the rules can be left to up the user's discretion or automatically applied when comparing small populations where the low opportunity for exact matches resulting from small sample size increases the probability of obtaining an uninformative result.
In one embodiment, when a particular set of rules fails to provide sufficient detection of predisposing attributes, the rules can be modified in order to provide higher granularity or resolution for the discovery of predisposing attributes. As such, nucleotide changes in the wobble codon position may be examined in certain applications. Similarly, the brand of cigarettes smoked may be a required attribute to discover some predisposing attributes, but not others. By varying the rules, the appropriate level of granularity or resolution can be determined. In one embodiment, the rules are varied on a test population (which can be comprised of both attribute-positive and attribute-negative individuals) in an effort to determine the most appropriate rules for the greater population.
Based on this knowledge, the following additional rules can be applied by comparison engine 222 when comparing two nucleotide sequences:
-
- a) a direct sequence comparison may determine two protein-encoding nucleotide sequences to be the same if they encode the same protein and differ only by one or more nucleotides within the open reading frame that do not alter the amino acid sequence of the encoded protein;
- b) a direct sequence comparison may determine two protein-encoding nucleotide sequences to be the same if they encode the same protein and differ only by one or more nucleotides that result in conservative amino acid substitutions within the amino acid sequence of the encoded protein;
- c) a direct sequence comparison may determine two protein-encoding nucleotide sequences to be the same if they encode the same protein and differ only by one or more nucleotides that result in conservative amino acid substitutions anywhere within the amino acid sequence of the encoded protein except for enzymatic, transmembrane or antigen-recognition domains;
- d) a direct sequence comparison may determine two protein-encoding nucleotide sequences to be the same if they encode the same protein and differ only by one or more nucleotides that result in silent amino acid substitutions;
- e) a direct sequence comparison may determine two nucleotide sequences that do not encode amino acid sequences to be the same if they differ only by the identity of nucleotide mutations occurring at the same position within both sequences;
- f) a direct sequence comparison may determine two protein-encoding nucleotide sequences to be the same if they encode the same protein and differ only by one or more conservative missense mutations within the open reading frame of the encoded protein;
- g) a direct sequence comparison may determine two protein-encoding nucleotide sequences to be the same if they encode the same protein and differ only by one or more conservative missense mutations anywhere within the open reading frame of the encoded protein except for those regions of the open reading frame that encode enzymatic, transmembrane or antigen-recognition domains of the protein;
- h) a direct sequence comparison may determine two protein-encoding nucleotide sequences to be the same if they encode the same protein and differ only by one or more silent mutations;
- i) a direct sequence comparison may determine two protein-encoding nucleotide sequences to be the same if they encode the same protein and differ only by the locations of nonsense mutations within the same domain of the encoded protein;
- j) a direct sequence comparison may determine two protein-encoding nucleotide sequences to be the same if they encode the same protein and differ only by the locations of frameshift mutations within the same respective domain of the encoded protein.
As shown in
The above classes and attributes represent the current condition of the individual. In the event that the individual (e.g. consumer 810) had a diagnosis for an ailment in the past, the same classification methodology can be applied, but with an “h” placed after the attribute number to denote a historical attribute. For example, 510.1.4h can be used to create an attribute to indicate that the individual suffered a stroke in the past, as opposed to 510.1.4 which indicates the individual is currently suffering a stroke or the immediate aftereffects. Using this approach, historical classes and attributes mirroring the current classes and attributes can be created, as illustrated by historical physical health class 510h, historical physical diagnoses class 510.1h, historical basic physical class 520h, historical height class 520.1h, historical detailed physical class 530h, and historical hormone levels class 530.1h. In an alternate embodiment historical classes and historical attributes are not utilized. Rather, time stamping of the diagnoses or event is used. In this approach, an attribute of 510.1.4-05FEB03 would indicate that the individual suffered a stroke on Feb. 5, 2003. Alternate classification schemes and attribute classes/classifications can be used and will be understood by one of skill in the art. In one embodiment, time stamping of attributes is preferred in order to permit accurate determination of those attributes or attribute combinations that are associated with an attribute of interest (i.e., a query attribute or target attribute) in a causative or predictive relationship, or alternatively, those attributes or attribute combinations that are associated with an attribute of interest in a consequential or symptomatic relationship. In one embodiment, only attributes bearing a time stamp that predates the time stamp of the attribute of interest are processed by the methods. In another embodiment, only attributes bearing a time stamp that postdates the time stamp of the attribute of interest are processed by the methods. In another embodiment, both attributes that predate and attributes that postdate an attribute of interest are processed by the methods.
As further shown in
In one embodiment, the classes and indexing illustrated in
In one embodiment, commercial databases such as credit databases, databases containing purchase information (e.g. frequent shopper information) can be used as either the basis for extracting attributes for the classes such as those in financial subclass 630 and historical financial subclass 630h, or for direct mapping of the information in those databases to situational attributes. Similarly, accounting information such as that maintained by the consumer 810 of
Measurement of financial attributes such as those illustrated and described with respect to
As discussed with respect to
Unified Modeling Language (“UML”) can be used to model and/or describe methods and systems and provide the basis for better understanding their functionality and internal operation as well as describing interfaces with external components, systems and people using standardized notation. When used herein, UML diagrams including, but not limited to, use case diagrams, class diagrams and activity diagrams, are meant to serve as an aid in describing the embodiments of the present invention but do not constrain implementation thereof to any particular hardware or software embodiments. Unless otherwise noted, the notation used with respect to the UML diagrams contained herein is consistent with the UML 2.0 specification or variants thereof and is understood by those skilled in the art.
In one embodiment this involves the contribution by consumer 810 of a swab of the inside of the cheek, a blood sample, or contribution of other biological specimen associated with consumer 810 from which genetic and epigenetic data can be obtained. In one embodiment, genetic database administrator 830 causes the genetic sample to be analyzed through a determine genetic and epigenetic attributes use case 850. Consumer 810 or clinician 820 may collect physical attributes through a describe physical attributes use case 842. Similarly, behavioral, situational, and historical attributes are collected from consumer 810 or clinician 820 via describe behavioral attributes use case 844, describe situational attributes use case 846, and describe historical attributes use case 848, respectively. Clinician 820 or consumer 810 can then enter a query attribute through receive query attribute use case 852. Attribute determination system 800 then, based on attributes of large query-attribute positive and query-attribute negative populations, determines which attributes and combinations of attributes, extending across the pangenetic (genetic/epigenetic), physical, behavioral, situational, and historical attribute categories, are statistically related to the query attribute. As previously discussed, and with respect to
Physical, behavioral, situational and historical attribute data may be stored or processed in a manner that allows retention of maximum resolution and accuracy of the data while also allowing flexible comparison of the data so that important shared similarities between individuals are not overlooked. This is taken into account when processing narrow and extreme attribute values, or smaller populations of individuals where the reduced number of individuals makes the occurrence of identical matches of attributes rare. In these and other circumstances, flexible treatment and comparison of attributes can reveal predisposing attributes that are related to or legitimately derive from the original attribute values but have broader scope, lower resolution, and extended or compounded values compared to the original attributes. In one embodiment, attributes and attribute values can be qualitative (categorical) or quantitative (numerical). In a further embodiment, attributes and attribute values can be discrete or continuous.
There are several ways flexible treatment and comparison of attributes can be accomplished. As shown in
In one embodiment, individual attributes may be expanded into one or more sets containing attributes having values, levels or degrees that are above, below, surrounding or including that of the original attribute. In one embodiment, attributes can be used to create new attributes that are broader or narrower in scope than the original attribute. In one embodiment, attributes can be used to compute new attributes that are related to the original attribute. As an example,
With respect to the aforementioned method of collection, inaccuracies can occur, sometimes due to outright misrepresentations of the individual's habits. For example, it is not uncommon for patients to self-report alcohol consumption levels which are significantly below actual levels. Such situations can occur even when a clinician/physician is involved, as the patient reports consumption levels to the clinician/physician that are significantly below their actual consumption levels. Similarly, it is not uncommon for an individual to over-report the amount of exercise they get.
In one embodiment, disparate sources of data including consumption data as derived from purchase records, data from blood and urine tests, and other observed characteristics are used to derive attributes such as those shown in
In one embodiment, the heuristic rules take into account attributes such as household size and self-reported data to assist in the derivation of the desired attribute. For example, if purchase data is used in a heuristic rule, household size and even the number of self-reported smokers in the household, can be used to help determine actual levels of consumption of tobacco by the individual. In one embodiment, household members are tracked individually, and the heuristic rules provide for the ability to approximately assign consumption levels to different people in the household. Details such as individual brand usages or preferences may be used to help assign consumptions within the household. As such, the heuristic rules can be applied to a number of disparate pieces of data to assist in extracting one or more attributes.
The methods, systems, software and databases described herein are able to achieve determination of complex combinations of predisposing attributes not only as a consequence of the resolution and breadth of data used, but also as a consequence of the process methodology used for discovery of predisposing attributes. An attribute may have no effect on expression of another attribute unless it occurs in the proper context, the proper context being co-occurrence with one or more additional predisposing attributes. In combination with one or more additional attributes of the right type and degree, an attribute may be a significant contributor to predisposition of the organism for developing the attribute of interest. This contribution is likely to remain undetected if attributes are evaluated individually. As an example, complex diseases require a specific combination of multiple attributes to promote expression of the disease. The required disease-predisposing attribute combinations will occur in a significant percentage of those that have or develop the disease and will occur at a lower frequency in a group of unaffected individuals.
Although the previous two figures present frequencies of occurrence as percentages, for the methods of the present invention the frequencies of occurrence of attribute combinations are can be stored as ratios for both the query-attribute-positive individuals and the query-attribute-negative individuals. Referring to
The frequencies of occurrence of an attribute or attribute combination, when compared for two or more groups of individuals with respect to a query attribute, are statistical results that can indicate strength of association of the attribute combination with a query attribute. Frequencies of occurrence can also be utilized by statistical computation engine 224 to compute additional statistical results for strength of association of the attribute combinations with the query attribute. The statistical measures used may include, but are not limited to, prevalence, incidence, probability, absolute risk, relative risk, attributable risk, excess risk, odds (a.k.a. likelihood), and odds ratio (a.k.a. likelihood ratio). Absolute risk (a.k.a. probability), relative risk, odds, and odds ratio are the preferred statistical computations for the present invention. Among these, absolute risk and relative risk are the more preferable statistical computations because their values can still be calculated for an attribute combination in instances where the frequency of occurrence of the attribute combination in the query-attribute-negative group is zero. Odds and odds ratio are undefined in instances where the frequency of occurrence of the attribute combination in the query-attribute-negative group is zero, because in that situation their computation requires division by zero which is mathematically undefined. One embodiment of the present invention, when supplied with ample data, is expected to routinely yield frequencies of occurrence of zero in query-attribute-negative groups because of its ability to discover large predisposing attribute combinations that are exclusively associated with the query attribute.
In one embodiment, results for absolute risk and relative risk can be interpreted as follows with respect to an attribute combination predicting association with a query attribute: 1) if absolute risk=1.0, and relative risk=undefined, then the attribute combination is sufficient and necessary to cause association with the query attribute, 2) if absolute risk=1.0, and relative riskundefined, then the attribute combination is sufficient but not necessary to cause association with the query attribute, 3) if absolute risk<1.0, and relative riskundefined, then the attribute combination is neither sufficient nor necessary to cause association with the query attribute, and 4) if absolute risk<1.0, and relative risk=undefined, then the attribute combination is not sufficient but is necessary to cause association with the query attribute. In an alternate embodiment, relative risk=undefined can be interpreted to mean that there are two or more attribute combinations, rather than just one attribute combination, that can cause association with the query attribute. In one embodiment, an absolute risk<1.0 can be interpreted to mean one or more of the following: 1) the association status of one or more attributes, as provided to the methods, is inaccurate or missing (null), 2) not enough attributes have been collected, provided to or processed by the methods, or 3) the resolution afforded by the attributes that have been provided is too narrow or too broad. These interpretations can be used to increase accuracy and utility of the methods for use in many applications including but not limited to attribute combination discovery, attribute prediction, predisposition prediction, predisposition modification and destiny modification.
The statistical results obtained from computing the statistical measures can be subjected to inclusion, elimination, filtering, and evaluation based on meeting one or more statistical requirements. Statistical requirements can include but are not limited to numerical thresholds, statistical minimum or maximum values, and statistical significance/confidence values.
One embodiment of the present invention can be used in many types of statistical analyses including but not limited to Bayesian analyses (e.g., Bayesian probabilities, Bayesian classifiers, Bayesian classification tree analyses, Bayesian networks), linear regression analyses, non-linear regression analyses, multiple linear regression analyses, uniform analyses, Gaussian analyses, hierarchical analyses, recursive partitioning (e.g., classification and regression trees), resampling methods (e.g., bootstrapping, cross-validation, jackknife), Markov methods (e.g., Hidden Markov Models, Regular Markov Models, Markov Blanket algorithms), kernel methods (e.g., Support Vector Machine, Fisher's linear discriminant analysis, principle components analysis, canonical correlation analysis, ridge regression, spectral clustering, matching pursuit, partial least squares), multivariate data analyses including cluster analyses, discriminant analyses and factor analyses, parametric statistical methods (e.g., ANOVA), non-parametric inferential statistical methods (i.e., binomial test, Anderson-Darling test, chi-square test, Cochran's Q, Cohen's kappa, Efron-Petrosian Test, Fisher's exact test, Friedman two-way analysis of variance by ranks, Kendall's tau, Kendall's W, Kolmogorov-Smirnov test, Kruskal-Wallis one-way analysis of variance by ranks, Kuiper's test, Mann-Whitney U or Wilcoxon rank sum test, McNemar's test, median test, Pitman's permutation test, Siegel-Tukey test, Spearman's rank correlation coefficient, Student-Newman-Keuls test, Wald-Wolfowitz runs test, Wilcoxon signed-rank test).
In one embodiment, the methods, databases, software and systems of the present invention can be used to produce data for use in and/or results for the above statistical analyses. In another embodiment, the methods, databases, software and systems of the present invention can be used to independently verify the results produced by the above statistical analyses.
In one embodiment a method is provided which accesses a first dataset containing attributes associated with a set of query-attribute-positive individuals and query-attribute-negative individuals, the attributes being pangenetic, physical, behavioral and situational attributes associated with individuals, and creates a second dataset of attributes associated with a query-attribute-positive individual but not associated with one or more query-attribute-negative individuals. A third dataset can be created containing attributes of the second dataset that are either associated with one or more query-attribute-positive individuals or are not present in any of the query-attribute-negative individuals, along with the frequency of occurrence in the query-attribute-positive individuals and the frequency of occurrence in the query-attribute-negative individuals. A statistical computation can be performed for each attribute combination, based on the frequency of occurrence, the statistical computation result indicating the strength of association, as measured by one or more well known statistical measures, between each attribute combination and the query attribute. The process can be repeated for a number of query attributes, and multiple query-positive individuals can be studied to create a computer-stored and machine-accessible compilation of different attribute combinations that co-occur with the queried attributes. The compilation can be ranked and co-occurring attribute combinations not having a minimum strength of association with the query attribute can be eliminated from the compilation.
Similarly, a system can be developed which contains a subsystem for accessing a query attribute, a second subsystem for accessing a set of databases containing pangenetic, physical, behavioral, and situational attributes associated with a plurality of query-attribute positive, and query-attribute negative individuals, a data processing subsystem for identifying combinations of pangenetic, physical, behavioral, and situational attributes associated with query-attribute positive individuals, but not with query-attribute negative individuals, and a calculating subsystem for determining a set of statistical results that indicates a strength of association between the combinations of pangenetic, physical, behavioral, and situational attributes with the query attribute. The system can also include a communications subsystem for retrieving at least some of pangenetic, physical, behavioral, and situational attributes from at least one external database; a ranking subsystem for ranking the co-occurring attributes according to the strength of the association of each co-occurring attribute with the query attribute; and a storage subsystem for storing the set of statistical results indicating the strength of association between the combinations of pangenetic, physical, behavioral, and situational attributes and the query attribute. The various subsystems can be discrete components, configurations of electronic circuits within other circuits, software modules running on computing platforms including classes of objects and object code, or individual commands or lines of code working in conjunction with one or more Central Processing Units (CPUs). A variety of storage units can be used including but not limited to electronic, magnetic, electromagnetic, optical, opto-magnetic and electro-optical storage.
In one application the method and/or system is used in conjunction with a plurality of databases, such as those that would be maintained by health-insurance providers, employers, or health-care providers, which serve to store the aforementioned attributes. In one embodiment the pangenetic (genetic and epigenetic) data is stored separately from the other attribute data and is accessed by the system/method. In another embodiment the pangenetic data is stored with the other attribute data. A user, such as a clinician, physician or patient, can input a query attribute, and that query attribute can form the basis for determination of the attribute combinations associated with that query attribute. In one embodiment the associations will have been previously stored and are retrieved and displayed to the user, with the highest ranked (most strongly associated) combinations appearing first. In an alternate embodiment the calculation is made at the time the query is entered, and a threshold can be used to determine the number of attribute combinations that are to be displayed.
1st dataset 1322 in the flow chart of
In the example illustrated in
As shown in the flowchart of
In select query-attribute-positive individualN step 1304, individual #1 is selected in this example for comparison of their attributes with those of other individuals. In store attributes step 1306, those attributes of the selected individual #1 that are not associated with a portion (e.g., one or more individuals) of the query-attribute-negative group (or alternatively, a randomly selected subgroup of query-attribute-negative individuals) are stored in 2nd dataset 1324 as potential candidate attributes for contributing to predisposition toward the query attribute. In one embodiment this initial comparison step is used to increase efficiency of the method by eliminating those attributes that are associated with all of the query-attribute-negative individuals. Because such attributes occur with a frequency of 100% in the query-attribute-negative group, they cannot occur at a higher frequency in the query-attribute-positive group and are therefore not candidates for contributing to predisposition toward the query attribute. Therefore, this step ensures that only attributes of the individual that occur with a frequency of less than 100% in the query-attribute-negative group are stored in the 2nd dataset. This step is especially useful for handling genetic attributes since the majority of the approximately three billion nucleotide attributes of the human genome are identically shared among individuals and may be eliminated from further comparison before advancing to subsequent steps.
As mentioned above, this initial comparison to effectively eliminate attributes that are not potential candidates may be performed against a randomly selected subgroup of query-attribute-negative individuals. Using a small subgroup of individuals for the comparison increases efficiency and prevents the need to perform a comparison against the entire query-attribute-negative population which may consist of thousands or even millions of individuals. In one embodiment, such a subgroup preferably consists of at least 20, but as few as 10, randomly selected query-attribute-negative individuals.
For the present example, only those attributes having a status value of 1 for individual #1 and a status value of 0 for one or more query-attribute-negative individuals are stored as potential candidate attributes, but in one embodiment those attributes having a status value of 0 for individual #1 and a status value of 1 for one or more query-attribute-negative individuals (i.e., attributes I, K, Q and W) can also be stored as candidate attributes, and may be referred to as candidate not-attributes of individual #1.
In store statistical results step 1310, the frequencies of occurrence previously stored in 3rd dataset 1326 are used to compute statistical results for the attribute combinations which indicate the strength of association of each attribute combination with the query attribute. As mentioned previously, the statistical computations used may include prevalence, incidence, absolute risk (a.k.a. probability), attributable risk, excess risk, relative risk, odds and odds ratio. In one embodiment, absolute risk, relative risk, odds and odds ratio are the statistical computations performed (see formulas in
For the sake of brevity, only the individual #1 was selected and processed in the method, thereby determining only the predisposing attribute combinations of individual #1 and those individuals of the group that also happen to possess one or more of those attribute combinations. However, one can proceed to exhaustively determine all predisposing attribute combinations in the query-attribute-positive group and build a complete 3rd dataset for the population with respect to query attribute ‘A’. As shown in the flow chart of
In store significantly associated attribute combinations step 1314, 4th dataset 1328 may be created by selecting and storing only those attribute combinations and their associated data from the 3rd dataset that show a minimum statistical association with the query attribute. The minimum statistical association may be a minimum positive, negative, neutral or combined association determined by either the user or the system. This determination can be made based on the statistical results previously stored in 3rd dataset 1326. As an example, the determination can be made based on the results computed for relative risk. Statistically, a relative risk of >1.0 indicates a positive association between the attribute combination and the query attribute, while a relative risk of 1.0 indicates no association, and a relative risk of <1.0 indicates a negative association.
It can be left up to the user or made dependent on the particular application as to which statistical measure and what degree of statistical association is used as the criteria for determining inclusion of attribute combinations in the 4th dataset. In this way, 4th dataset 1328 can be presented in the form of a report which contains only those attribute combinations determined to be predisposing toward the query attribute above a selected threshold of significant association for the individual or population of individuals.
In many applications it will be desirable to determine predisposing attribute combinations for additional query attributes within the same population of individuals. In one embodiment this is accomplished by repeating the entire method for each additional query attribute and either creating new 2nd, 3rd and 4th datasets, or appending the results into the existing datasets with associated identifiers that clearly indicate what data results correspond to which query attributes. In this way, a comprehensive database containing datasets of predisposing attribute combinations for many different query attributes may be created.
In one embodiment of a method for creating an attribute combinations database, attribute profile records of individuals that have nulls for one or more attribute values are not processed by the method or are eliminated from the 1st dataset before initiating the method. In another embodiment, attribute profile records of individuals that have nulls for one or more attribute values are only processed by the method if those attribute values that are nulls are deemed inconsequential for the particular query or application. In another embodiment, a population of individuals having one or more individual attribute profile records containing nulls for one or more attribute values are only processed for those attributes that have values (non-nulls) for every individual of that population.
In one embodiment of a method for creating an attribute combinations database, frequencies of occurrence and statistical results for strength of association of existing attribute combinations in the attribute combinations dataset are updated based on the attribute profile of an individual processed by the method. In another embodiment, frequencies of occurrence and statistical results for strength of association of existing attribute combinations in the attribute combinations dataset are not updated based on the attribute profile of an individual processed by the method. In another embodiment, the processing of an individual by the method can require first comparing the individuals' attribute profile to the preexisting attribute combinations dataset to determine which attribute combinations in the dataset are also present in the individual's attribute profile, and then in a further embodiment, based on the individual's attribute profile, updating the frequencies of occurrence and statistical results for strength of association of those attribute combinations in the dataset that are also present in the individual's attribute profile, without further processing the individual or their attributes by the method.
The 3rd and 4th datasets created by performing the above methods for creation of a database of attribute combinations can be used for additional methods of the invention that enable: 1) identification of predisposing attribute combinations toward a key attribute of interest, 2) predisposition prediction for an individual toward a key attribute of interest, and 3) intelligent individual destiny modification provided as predisposition predictions resulting from the addition or elimination of specific attribute associations.
In one embodiment a method of identifying predisposing attribute combinations is provided which accesses a first dataset containing attribute combinations and statistical computation results that indicate the potential of each attribute combination to co-occur with a query attribute, the attributes being pangenetic, physical, behavioral, and situational attributes. A tabulation can be performed to provide, based on the statistical computation results, those predisposing attribute combinations that are most likely to co-occur with the query attribute, or a rank-ordering of predisposing attribute combinations of the first dataset that co-occur with the query attribute.
Similarly, a system can be developed which contains a subsystem for accessing or receiving a query attribute, a second subsystem for accessing a dataset containing attribute combinations of pangenetic, physical, behavioral and situational attributes that co-occur with one or more query attributes, a communications subsystem for retrieving the attribute combinations from at least one external database, and a data processing subsystem for tabulating the attribute combinations. The various subsystems can be discrete components, configurations of electronic circuits within other circuits, software modules running on computing platforms including classes of objects and object code, or individual commands or lines of code working in conjunction with one or more Central Processing Units (CPUs). A variety of storage units can be used including but not limited to electronic, magnetic, electromagnetic, optical, opto-magnetic and electro-optical storage.
In one application the method and/or system is used in conjunction with one or more databases, such as those that would be maintained by health-insurance providers, employers, or health-care providers, which can serve to store the aforementioned attribute combinations and corresponding statistical results. In one embodiment the attribute combinations are stored in a separate dataset from the statistical results and the correspondence is achieved using identifiers or keys present in (shared across) both datasets. In another embodiment the attribute combinations and corresponding statistical results data is stored with the other attribute data. A user, such as a clinician, physician or patient, can input a query attribute, and that query attribute can form the basis for tabulating attribute combinations associated with that query attribute. In one embodiment the associations will have been previously stored and are retrieved and displayed to the user, with the highest ranked (most strongly associated) combinations appearing first. In an alternate embodiment the tabulation is performed at the time the query attribute is entered, and a threshold can be used to determine the number of attribute combinations that are to be displayed.
As an example, a minimum strength of association requiring relative risk 1.0 may be chosen. Based on this chosen requirement, the tabulated list of attribute combinations shown in
In one embodiment a method for predicting predisposition of an individual for query attributes of interest is provided which accesses a first dataset containing attributes associated with an individual and a second dataset containing attribute combinations and statistical computation results that indicate strength of association of each attribute combination with a query attribute, the attributes being pangenetic, physical, behavioral and situational attributes. A comparison can be performed to determine the largest attribute combination of the second dataset that is also present in the first dataset and that meets a minimum statistical requirement, the result being stored in a third dataset. The process can be repeated for a number of query attributes. A tabulation can be performed to provide a predisposition prediction listing indicating the predisposition of the individual for each of the query attributes. In one embodiment, predisposition can be defined as a statistical result indicating strength of association between an attribute or attribute combination and a query attribute.
Similarly, a system can be developed which contains a subsystem for accessing or receiving a query attribute, a second subsystem for accessing a dataset containing attributes of an individual, a third subsystem for accessing attribute combinations of pangenetic, physical, behavioral, and situational attributes that co-occur with one or more query attributes, a communications subsystem for retrieving the attribute combinations from at least one external database, and a data processing subsystem for comparing and tabulating the attribute combinations. The various subsystems can be discrete components, configurations of electronic circuits within other circuits, software modules running on computing platforms including classes of objects and object code, or individual commands or lines of code working in conjunction with one or more Central Processing Units (CPUs). A variety of storage units can be used including but not limited to electronic, magnetic, electromagnetic, optical, opto-magnetic and electro-optical storage.
In one application the method and/or system is used in conjunction with one or more databases, such as those that would be maintained by health-insurance providers, employers, or health-care providers, which can serve to store the aforementioned attribute combinations and corresponding statistical results. In one embodiment the attribute combinations are stored in a separate dataset from the statistical results and the correspondence is achieved using identifiers or keys present in (shared across) both datasets. In another embodiment the attribute combinations and corresponding statistical results data is stored with the other attribute data. A user, such as a clinician, physician or patient, can input a query attribute, and that query attribute can form the basis for tabulating attribute combinations associated with that query attribute. In one embodiment the associations will have been previously stored and are retrieved and displayed to the user, with the highest ranked (most strongly associated) combinations appearing first. In an alternate embodiment the tabulation is performed at the time the query attribute is entered, and a threshold can be used to determine the number of attribute combinations that are to be displayed.
For this example, query attribute ‘A’ is submitted by a user in a query. In access attributes step 2102 the attributes of an individual whose attribute profile is contained in a 1st dataset 2122 are accessed. A representative 1st dataset for individual #112 is shown in
In one embodiment, the tabulation can be provided in a form suitable for visual output, such as a visual graphic display or printed report. Attribute combinations do not need to be reported in predisposition prediction and can be omitted or masked so as to provide only the query attributes of interest and the individual's predisposition prediction for each. In creating a tabulated report for viewing by a consumer, counselor, agent, physician, patient or consumer, tabulating the statistical predictions can include substituting the terminology ‘absolute risk’ and ‘relative risk’ with the terminology ‘absolute potential’ and ‘relative potential’, since the term ‘risk’ carries negative connotations typically associated with the potential for developing undesirable conditions like diseases. This substitution may be desirable when the present invention is used to predict predisposition for desirable attributes such as specific talents or success in careers and sports. Also, the numerical result of absolute risk is a mathematical probability that can be converted to chance by simply multiplying it by 100%. It may be desirable to make this conversion during tabulation since chance is more universally understood than mathematical probability. Similarly, relative risk can be represented as a multiplier, which may facilitate its interpretation. The resulting tabulated results for this example are shown in
In one embodiment a method for individual destiny modification is provided which accesses a first dataset containing attributes associated with an individual and a second dataset containing attribute combinations and statistical computation results that indicate strength of association of each attribute combination with a query attribute, the attributes being pangenetic, physical, behavioral and situational attributes. A comparison can be performed to identify the largest attribute combination of the second dataset that consists of attributes of the first dataset. Then, attribute combinations of the second dataset that either contain that identified attribute combination or consist of attributes from that identified attribute combination can be stored in a third dataset. The content of the third dataset can be transmitted as a tabulation of attribute combinations and corresponding statistical results which indicate strengths of association of each attribute combination with the query attribute, thereby providing predisposition potentials for the individual toward the query attribute given possession of those attribute combinations. In one embodiment destiny can be defined as statistical predisposition toward having or acquiring one or more specific attributes.
Similarly, a system can be developed which contains a subsystem for accessing or receiving a query attribute, a second subsystem for accessing a dataset containing attributes of an individual, a third subsystem for accessing attribute combinations of pangenetic, physical, behavioral, and situational attributes that co-occur with one or more query attributes, a communications subsystem for retrieving the attribute combinations from at least one external database, and a data processing subsystem for comparing and tabulating the attribute combinations. The various subsystems can be discrete components, configurations of electronic circuits within other circuits, software modules running on computing platforms including classes of objects and object code, or individual commands or lines of code working in conjunction with one or more Central Processing Units (CPUs). A variety of storage units can be used including but not limited to electronic, magnetic, electromagnetic, optical, opto-magnetic, and electro-optical storage.
In one application the method and/or system is used in conjunction with one or more databases, such as those that would be maintained by health-insurance providers, employers, or health-care providers, which can serve to store the aforementioned attribute combinations and corresponding statistical results. In one embodiment the attribute combinations are stored in a separate dataset from the statistical results and the correspondence is achieved using identifiers or keys present in (shared across) both datasets. In another embodiment the attribute combinations and corresponding statistical results data is stored with the other attribute data. A user, such as a clinician, physician or patient, can input a query attribute, and that query attribute can form the basis for tabulating attribute combinations associated with that query attribute. In one embodiment the associations will have been previously stored and are retrieved and displayed to the user, with the highest ranked (most strongly associated) combinations appearing first. In an alternate embodiment the tabulation is performed at the time the query attribute is entered, and a threshold can be used to determine the number of attribute combinations that are to be displayed.
In one embodiment the method for individual destiny modification is used to identify and report attributes that the individual may modify to increase or decrease their chances of having a particular attribute or outcome. In one embodiment, the tabulation of attribute combinations produced by the method of destiny modification is filtered to eliminate those attribute combinations that contain one or more attributes that are not modifiable. In an alternate embodiment, modifiable attributes are prioritized for modification in order to enable efficient destiny (i.e., predisposition) modification. In one embodiment, non-historical attributes are considered modifiable while historical attributes are considered not modifiable. In another embodiment, non-historical behavioral attributes are considered to be the most easily or readily modifiable attributes. In another embodiment, non-historical situational attributes are considered to be the most easily or readily modifiable attributes. In another embodiment, non-historical physical attributes are considered to be the most easily or readily modifiable attributes. In another embodiment, non-historical pangenetic attributes are considered to be the most easily or readily modifiable attributes. In one embodiment, the modifiable attributes are ranked or otherwise presented in a manner that indicates which are the most easily or readily modifiable, which may include creating categories or classes of modifiable attributes, or alternatively, reporting attributes organized according to the attribute categories of the invention.
In biological organisms and systems, age and sex type are two somewhat unique and powerful attributes that influence the expression of many other attributes. For example, age is a primary factor associated with: predicting onset and progression of age-associated diseases in humans and animals; acquiring training and life experiences that lead to success in career, sports and music; and determining life-style choices. Similarly, biological sex type is correlated with profound differences in expression of physical, behavioral and situational attributes. The inclusion of accurate data for the age and sex of individuals is very important for acquiring accurate and valid results from the methods of the present invention. In one embodiment, specific values of age and sex that aggregate with a query attribute can be determined by the methods of the present invention, just as for other attributes, to either co-occur or not co-occur in attribute combinations that are associated with a query attribute. In one embodiment results of the methods can be filtered according to age and/or sex. In other embodiments a population or subpopulation can be selected according to age and/or sex (age-matching and/or sex-matching) and then only that subpopulation subjected to additional processing by methods of the present invention. In another embodiment, an age-matched and/or sex-matched population may be used to form query-attribute-positive and query-attribute-negative groups. In another embodiment, the sex and/or age of an individual is used to select a population of age-matched and/or sex-matched individuals for creation of an attribute combinations database. In another embodiment, the sex and/or age of an individual is used to select a subpopulation of age-matched and/or sex-matched individuals for comparison in methods of identifying predisposing attribute combinations, individual predisposition prediction and individual destiny modification. In another embodiment, summary statistics for age and/or sex are included with the output results of the methods. In another embodiment, summary statistics for age and/or sex are included with the output results of the methods when other attributes are omitted or masked for privacy.
Additional embodiments are envisioned which implement a preselection of individuals processed by methods of the present invention. In one embodiment, preselection is a selection or pooling of one or more populations or subpopulations of individuals from one or more datasets or databases based on particular values of attributes such as income, occupation, disease status, zip code or marital status for example. Preselecting populations and subpopulations based on possession of one or more specific attributes can serve to focus a query on the most representative population, reduce noise by removing irrelevant individuals whose attribute data may contribute to increasing error in the results, and decrease computing time required to execute the methods by reducing the size of the population to be processed. Also, using preselection to define and separate different populations enables comparison of predisposing attribute combinations toward the same query attribute between those populations. For example, if two separate subpopulations are selected—a first population of individuals that earn over $100,000/year and a second population of individuals that earn less that $10,000/year—and each subpopulation is processed separately to identify predisposing attribute combinations for a query attribute of alcoholism, a comparison of the identities, frequencies of occurrence, and strengths of association of predisposing attribute combinations that lead to alcoholism in individuals that earn over $100,000 can be made with those of individuals that earn less than $10,000. In one embodiment, predisposing attribute combinations that are present in one preselected population and absent in a second preselected population are identified. In one embodiment, the frequencies of occurrence and/or statistical strengths of association of predisposing attribute combinations are compared between two or more preselected populations. In one embodiment, only a single preselected population is selected and processed by the methods of the present invention.
Additional embodiments of the methods of the present invention are possible. In one embodiment, two or more mutually exclusive (having no attributes in common) predisposing attribute combinations for a query attribute are identified for a single individual and can be tabulated and presented as output. In one embodiment the query attribute can be an attribute combination, and can be termed a query attribute combination. By submitting a query attribute combination to the methods of the present invention, the ability to identify attribute combinations that predispose toward other attribute combinations is enabled.
In one embodiment of the methods of the present invention, statistical measures for strength of association of attribute combinations are not stored in a dataset containing the attribute combinations, but rather, are calculated at any time (on as-needed basis) from the frequencies of occurrence of the stored attribute combinations. In one embodiment only a portion of the results from a method of the present invention are presented, reported or displayed as output. In one embodiment, the results may be presented as a graphical display or printout including but not limited to a 2-dimensional, 3-dimensional or multi-dimensional axis, pie-chart, flowchart, bar-graph, histogram, cluster chart, dendrogram, tree or pictogram.
Methods for predisposing attributes identification, predisposition prediction and intelligent destiny modification are subject to error and noise. A prominent cause of error and noise in methods is bias in the attribute data or in the distribution of the population from which the data is collected. In one embodiment, bias can manifest as inaccurate frequencies of occurrence and strengths of association between attribute combinations and query attributes, inaccurate lists of attributes determined to co-occur with a query attribute, inaccurate predictions of an individual's predisposition toward query attributes, and inaccurate lists of modifiable attributes for destiny modification. Bias can result from inaccurate data supplied to methods of the present invention, primarily as a consequence of inaccurate reporting and self-reporting of attribute data but also as a consequence of collecting attributes from populations that are biased, skewed or unrepresentative of the individual or population for which predisposition predictions are desired. Error can also result as consequence of faulty attribute data collection such as misdirected or improperly worded questionnaires.
If bias exists and is left unchecked, it can have different effects depending on whether the bias exists with the query attribute, or whether the bias exists in one or more of the co-occurring attributes of an attribute combination. At a minimum, the existence of bias in the attribute data or population distribution may result in slightly inaccurate results for frequency of occurrence of attributes and attribute combinations, and inaccurate statistical strengths of association between attribute combinations and query attributes. When bias is present at higher levels, results for frequency of occurrence and strengths of association can be moderately to highly inaccurate, even producing false positives (Type I Error) and false negatives (Type II Error), where a false positive is the mistaken identification of an attribute association that actually does not exist (or does not exist differentially in one population relative to another) and a false negative is a mistaken unidentification of an attribute association that actually does exist (or exists differentially in one population relative to another).
For the methods described herein, it is possible to minimize error and noise by ensuring that accurate (unbiased) attribute data is provided to the methods and that representative populations of individuals are used as the basis for creating attribute combinations datasets. It is anticipated that some degree of inaccuracy of input data will be present. The following disclosure indicates types of sources of error and noise and ways to identify, avoid and compensate for inaccurate attribute data and unrepresentative populations.
Selection bias is a major source of error and refers to bias that results from using a population of individuals that are not representative of the population for which results and predictions are desired. For example, if a query for attribute combinations that predispose an individual to becoming a professional basketball player is entered against an attributes combination dataset that was created with an over-representation of professional basketball players relative to the general population, then smaller attribute combinations that are associated with both professional basketball players and individuals that are not professional basketball players will receive artificially inflated statistical strengths of association with the query attribute, giving a false impression that one needs fewer predisposing attributes than are actually required to achieve the goal with a high degree of probability. Selection bias is largely under the control of those responsible for collecting attribute profiles for individuals of the population and creating datasets that contain that information. Selecting a misrepresentative set of individuals will obviously result in selection bias as discussed above. Sending questionnaires to a representative set of individuals but failing to receive completed questionnaires from a particular subpopulation, such as a very busy group of business professionals who failed to take time to fill out and return the questionnaire, will also result in selection bias if the returned questionnaires are used to complete a database without ensuring that the set of responses are a balanced and representative set for the population as a whole. Therefore, in one embodiment, administrators of the methods described herein use a variety of techniques to ensure that appropriate and representative populations are used so that selection bias is not present in the attribute profiles and attribute combination datasets used as input data for the methods.
Information bias is the second major class of bias and encompasses error due to inaccuracies in the collected attribute data. The information bias class comprises several subclasses including misclassification bias, interview bias, surveillance bias, surrogate interview bias, recall bias and reporting bias.
Misclassification bias refers to bias resulting from misclassifying an individual as attribute-positive when they are attribute-negative, or vice-versa. To help eliminate this type of bias, it is possible to assign a null for an attribute in circumstances where an accurate value for the attribute cannot be ensured.
Interview bias refers to bias resulting from deriving attributes from questions or means of information collection that are not correctly designed to obtain accurate attribute values. This type of bias is primarily under the control of those administrators that design and administer the various modes of attribute collection, and as such, they can ensure that the means of attribute collection employed are correctly designed and validated for collecting accurate values of the targeted attributes.
Surveillance bias refers to bias that results from more closely or more frequently monitoring one subpopulation of individuals relative to others, thereby resulting in collection of more accurate and/or more complete attribute data for that subpopulation. This is common in cases of individuals suffering from disease, which results in their constant and close monitoring by experienced professionals who may collect more accurate and more complete attribute data about many aspects of the individual, including trivial, routine and common attributes that are not restricted to the medical field. An administrator of the methods described herein can seek to reduce this bias by either excluding attribute information obtained as a consequence of surveillance bias or by ensuring that equivalent attribute information is provided for all members of the representative population used for the methods.
Surrogate interview bias refers to bias that results from obtaining inaccurate attribute information about an individual from a second-hand source such as a friend or relative. For example, when an individual dies, the only source of certain attribute information may be from a parent or spouse of the individual who may have inaccurate perception or memory of certain attributes of the deceased individual. To help avoid this type of bias, it is preferable that a surrogate provider of attribute information be instructed to refrain from providing attribute values for which they are uncertain and instead assign a null for those attributes.
Recall bias refers to enhanced or diminished memory recall of attribute values in one subpopulation of individuals versus another. This again may occur in individuals that are subject to extreme situations such as chronic illness, where the individual is much more conscious and attentive to small details of their life and environment to which others would pay little attention and therefore not recall as accurately. This type of bias results from inaccuracy in self-reporting and can be difficult to detect and control for. Therefore, to minimize this type of bias, it is recommended that attempts to collect self-reported data be made over a period of time in which individuals are aware of attributes that are being collected and may even keep a record or journal for attributes that are subject to significant recall bias. Also, whenever more accurate means than self-reporting can be used to collect attribute values, the more accurate means should be used.
Reporting bias refers to bias resulting from intentional misrepresentation of attribute values. This occurs when individuals underestimate the value for an attribute or underreport or fail to report an attribute they perceive as undesirable or are in denial over, or alternatively, when they overestimate the value for an attribute or overreport or invent possession of an attribute they perceive as desirable. For example, individuals typically knowingly underestimate the quantity of alcohol they drink, but overestimate the amount of time they spend exercising. One approach to encourage accurate self-reporting of attribute values can be to allow the individual to control their attribute profile record and keep their identity masked or anonymous in results output or during use of their data by others, when creating attribute combinations databases for example. If bias can be determined to exist and quantified at least in relative terms, another approach can be to use mathematical compensation/correction of the attribute value reported by the individual by multiplying their reported value by a coefficient or numerical adjustment factor in order to obtain an accurate value. In one embodiment this type of adjustment can be performed at the time the data is collected. In another embodiment this type of adjustment can be performed during conversion and reformatting of data by data conversion/formatting engine 220.
In one embodiment data conversion/formatting engine 220 works toward the removal of biases by the application of rules which assist in the identification of biased (suspect) attributes. In one embodiment the rules cause the insertion of null attributes where the existing attribute is suspect. In an alternate embodiment, rules are applied to identify suspect attributes (e.g. overreporting of exercise, underreporting of alcohol consumption) and corrective factors are applied to those attributes. For example, if it is determined that users self report consumption of alcohol at about ⅓ the actual rate consumed, the rules can, when attributes are suspect, increase the self-reported attribute by a factor of 1.5-3.0 depending on how the attribute is believed to be suspect. In large databases (e.g. health care databases) the size of the database can be used in conjunction with specific investigations (detailed data collection on test groups) to help develop rules to both identify and address biases.
In an alternate embodiment, actual possession of attributes and accurate values for self-reported attributes are determined using a multipronged data collection approach wherein multiple different inquires or means of attribute collection are used to collect a value for an attribute prone to bias. One example of this approach is to employ a questionnaire that asks multiple different questions to acquire the same attribute value. For example, if one wants to collect the attribute value for the number of cigarettes a person smokes each week, a questionnaire can include the following questions which are designed to directly or indirectly acquire this information: “how many cigarettes do you smoke each day?”, “how many packs of cigarettes do you smoke each day?”, “how many packs of cigarettes do you smoke each week?”, “how many packs of cigarettes do purchase each day? each week?”, “how many cartons of cigarettes do you purchase each month?”, “how much money do you spend on cigarettes each day?, each week? each month?”, “how many smoking breaks do you take at work each day?”. Another example is to ask a person to self-report how much time they spend exercising and also collect information from their gym that shows the time they swipe-in and swipe-out with their membership card. In this way, multiple sources of values for an attribute can be obtained and the values compared, cross-validated, deleted, filtered, adjusted, or averaged to help ensure storing accurate values for attributes.
In one embodiment the comparison, cross-validation, deletion, filtering, adjusting and averaging of attribute values can be performed during conversion and reformatting of data by data conversion/formatting engine 220. In one embodiment, multiple values obtained for a single attribute are averaged to obtain a final value for the attribute. In one embodiment, values for an attribute are discarded based on discrepancies between multiple values for an attribute. In one embodiment, one value for an attribute is chosen from among multiple values obtained for the attribute based on a comparison of the multiple values. In an alternate embodiment, reported values that appear out of an acceptable range (e.g. statistical outliers) are discarded and the final attribute value is determined from the remaining reported values.
Although calculation of the following mathematical measures are not performed in the examples presented herein, statistical measures of confidence including but not limited to variance, standard deviation, confidence intervals, coefficients of variation, correlation coefficients, residuals, t values (e.g., student's t test, one- and two-tailed t-distributions), ANOVA, correlation coefficients (e.g., regression coefficient, Pearson product-moment correlation coefficient), standard error and p-values can be computed for the results of methods of the current invention, the computation of which is known to those of skill in the art. In one embodiment, these confidence measures provide a level or degree of confidence in the numerical results of the methods so that the formal, standardized, legal, ethical, business, economic, medical, scientific, or peer-reviewable conclusions and decision-making can be made based on the results. In another embodiment, these measures are computed and compared for frequencies of occurrence of attribute combinations during creation of an attribute combinations database, for example to determine whether the difference between frequencies of occurrence of an attribute combination for the query-attribute-positive and query-attribute-negative groups is statistically significant for the purpose, in a further embodiment, of eliminating those attribute combinations that do not have a statistically significant difference in frequency of occurrence between the two populations. Levels of significance and confidence thresholds can be chosen based on user preference, implementation requirements, or standards of the various industries and fields of application.
Aside from the purposes indicated in the above methods, the present invention can also be used for investigation of attribute interactions forming the basis for predisposition. For example, embodiments of the methods can be used to reveal which attributes have diverse and wide-ranging interactions, which attributes have subtle interactions, which attributes have additive effects and which attributes have multiplicative or exponential synergistic interactions with other attributes.
In one embodiment, synergistic interactions are particularly important because they have multiplicative or exponential effects on predisposition, rather than simple additive effects, and can increase predisposition by many fold, sometimes by as much as 1000 fold. These types of synergistic interactions are common occurrences in biological systems. For example, synergistic interactions routinely occur with drugs introduced into biological systems. Depending on the circumstances, this synergism can lead to beneficial synergistic increases in drug potency or to synergistic adverse drug reactions. Synergism also occurs in opportunistic infections by microbes. Synergism between attributes may also occur in development of physical and behavioral traits. For example, cigarette smoking and asbestos exposure are known to synergize in multiplicative fashion to cause lung cancer. The same is true for smoking combined with uranium radiation exposure. Exposure to bacterial aflatoxin ingested via farm products combined with chronic hepatitis B infection synergistically causes liver cancer. Revealing synergistic interactions can be invaluable for intelligent and efficient targeting of therapies, treatments, training regimens, and lifestyle alterations to either increase or decrease predisposition toward an attribute of interest in the most rapid and efficient manner.
In the ‘before’ situation shown in
In the various embodiments of the present invention, the question as to how the results are to be used can be considered in the application of a particular embodiment of the method of attribute identification. In instances where the goal is to determine how to reduce predisposition toward an undesirable attribute for example, then utilizing one embodiment of the method to determine the identity of predisposing attribute combinations and then proceeding to eliminate an individual's association with those attributes is one way to reduce predisposition toward that attribute. However, one may also attempt to decrease predisposition by applying an embodiment of the method to determine those attribute combinations that are predisposing toward an attribute that is the opposite of the undesirable attribute, and then proceed to introduce association with those attributes to direct predisposition of the individual toward that opposing attribute. In other words, the attributes that predispose toward a key attribute may in many cases not be simple opposite of attributes that predispose to the opposite of the key attribute. Approaching this from both angles may provide additional effectiveness in achieving the goal of how to most effectively modify predisposition toward a key attribute of interest. In one embodiment both approaches are applied simultaneously to increase success in reaching the goal of destiny modification.
Confidentiality of personal attribute data can be a major concern to individuals that submit their data for analysis. Various embodiments of the present invention are envisioned in which the identity of an individual linked directly or indirectly to their data, or masked, or provided by privileged access or express permission, including but not limited to the following embodiments. In one embodiment the identity of individuals are linked to their raw attribute profiles. In one embodiment the identity of individuals are linked directly to their raw attribute profiles. In one embodiment the identity of individuals are linked indirectly to their raw attribute profiles. In one embodiment the identity of individuals are anonymously linked to their raw attribute profiles. In one embodiment the identity of individuals are linked to their raw attribute profiles using a nondescriptive alphanumeric identifier. In one embodiment the identity of individuals are linked to the attribute combinations they possess as stored in one or more datasets of the methods. In one embodiment the linkage of identity is direct. In one embodiment the linkage of identity is indirect. In one embodiment the linkage of identity requires anonymizing the identity of the individual. In one embodiment the linkage of identity requires use of a nondescriptive alphanumeric identifier.
Various embodiments of the present invention are envisioned in which data is made public, or held private, or provided restricted/privileged access granted upon express permission and include but are not limited to the following embodiments. In one embodiment, the identity of attributes and statistical results produced in the output of the methods are provided only to the individual whose attribute profile was accessed for the query. In one embodiment, the identity of attributes and statistical results produced in the output of the methods are provided only to the individual that submitted or authorized the query. In one embodiment, the identity of attributes and statistical results produced in the output of the methods are provided only to the individual consumer that paid for the query. In one embodiment, the identity of attributes and statistical results produced in the output of the methods are provided only to a commercial organization that submitted, authorized or paid for the query. In one embodiment, the identities of attributes in the output results from methods of the present invention are omitted or masked. In one embodiment, the identity of attributes can be omitted, masked or granted privileged access to by others as dictated by the individual whose attribute profile was accessed for the query. In one embodiment, the identity of attributes can be made accessible to a government employee, legal professional, medical professional, or other professional legally bound to secrecy. In one embodiment, the identity of attributes can be omitted, masked or granted privileged access to by others as dictated by a government employee, legal professional, or medical professional. In one embodiment, the identity of attributes can be omitted, masked or granted privileged access to by others as dictated by a commercial organization.
It will be appreciated by one of skill in the art that the present methods, systems, software and databases can be implemented on a number of computing platforms, and that
With respect to
In one embodiment, and as illustrated in
In one embodiment, the datasets of the methods of the present invention may be combined into a single dataset. In another embodiment the datasets may be kept separated. Separate datasets may be stored on a single computing device or distributed across a plurality of devices. Data, datasets, databases, methods and software of the present invention can be embodied on computer-readable media and computer-readable memory devices.
In one embodiment, at least a portion of the attribute data for one or more individuals is obtained from medical records. In one embodiment, at least a portion of the attribute data for one or more individuals is accessed, retrieved or obtained (directly or indirectly) from a centralized medical records database. In one embodiment, at least a portion of the attribute data for one or more individuals is accessed or retrieved from a centralized medical records database over a computer network.
The methods, systems, software and databases described herein have a number of industrial applications pertaining to the identification of attributes and combinations of attributes related to a query attribute, creation of databases including the attributes, combinations of attributes, strength of association with the query attribute, and rankings of strength of association with the query attribute, use of the identified attributes, combinations of attributes, and strength of association of attributes with the query attribute in making a variety of decisions related to lifestyle, lifestyle modification, diagnosis, medical treatment, eventual outcome (e.g. destiny), possibilities for destiny modification, and sensitivity analysis (impact or lack thereof of modification of certain attributes).
In one embodiment the methods, system, software, and databases described herein are used as part of a web based health analysis and diagnostics system in which one or more service providers utilize pangenetic information (attributes) in conjunction with physical, situational, and behavioral, attributes to provide services such as longevity analysis, insurance optimization (determination of recommended policies and amounts), and medication impact analysis. In these scenarios, the methods described herein are applied using appropriate query attributes to determine such parameters as the likelihood that the patient will develop or has a particular disease, or make an inquiry related to likelihood of disease development. In one embodiment, the genetic sample is mailed to an analysis center, where genetic and epigenetic sequencing is performed, and the data stored in an appropriate database. Clinician 820 of
With respect to general analysis of medical conditions, the web based system can be used to evaluate insurance coverage (amounts and types) and provide recommendations for coverage based on the specific illness risks and attributes possessed by the consumer, evaluate the impact (or lack thereof) of lifestyle changes, the impact and effectiveness of medications. Such analyses can also be made in view of predisposition predictions that can indicate probable future development of a disorder, thereby allowing preparations for insurance coverage and therapeutic preventive measures to be taken in advance of the disorder.
As previously discussed, the system can be used for web based strength and weakness identification, by allowing the consumer or clinician to query the system to assess the probability that an individual has a particular strength or weakness. In one embodiment, parents query the system to determine if their child (from which a biological sample was taken) will have particular strengths (e.g. music or sports) and to determine what behavioral attributes should be adopted to maximize the probability of success at that endeavor, assuming there is an identified “natural talent” as suggested by combinations of attributes associated with that endeavor. Various service providers, including genetic and epigenetic profiling entities, can interact with the system over a network (e.g., the internet) and allow the consumer or clinician to interact with the system over a network through a web-based interface to obtain the destiny or attribute information.
In one embodiment a web based goal achievement tool is presented in which the consumer enters one or more goals, and the system returns modifiable attributes which have been identified using the aforementioned analysis tools, indicating how the consumer can best obtain the desired goal(s) given their pangenetic, physical, situational, and behavioral makeup.
In one embodiment, potential relationship/life/marriage partners are located based on the pangenetic, physical, situational, and behavioral attributes of those individuals, as measured against an attribute model of a suitable partner developed for the consumer. The attribute model of the suitable partner can be developed using a number of techniques, including but not limited to, modeling of partner attributes based on attributes of individuals with which the individual has had previous successful relationships, determination of appropriate “complementary” attributes to the consumer based on statistical studies of individuals with similar attributes to the consumer who are in successful relationships and examination of their partner's attributes (determination of appropriate complementary attributes), and an ab initio determination of appropriate partner attributes. Once the attribute model for the most suitable potential partner has been developed, a database containing pangenetic, physical, situational and behavioral attribute data for potential partners for the consumer can be searched for the purpose of partner identification. In an alternate embodiment a consumer indicates persons they believe have suitable partner qualities including physical attraction (based on photos or video segments) as well as attributes described in profiles associated with the persons and their photos. In one embodiment the system uses genetic and epigenetic information associated with those individuals to create a subpopulation of individuals which the consumer believes they are attracted to, and examines a variety of data associated with that subpopulation (e.g., all available attribute data including genetic and epigenetic data) to determine attributes that are indicative of desirability to that consumer. In one embodiment the system uses those attributes to locate more individuals that could be potentially of interest to the consumer and presents those individuals to the consumer as potential partners.
Although the aforementioned methods, systems, software and databases have been described as incorporating and utilizing pangenetic, physical, situational and behavioral data, embodiments not utilizing pangenetic information are possible, with those embodiments being based solely on physical, situational and behavioral data. Such embodiments can be utilized to accomplish the tasks described above with respect to the analysis of biological systems, as well as for the analysis of complex non-living systems which contain a multitude of attributes. As an example, a non-biological application of the methodology and systems described herein would be for the analysis of complex electrical or electrical-mechanical systems in order to identify probable failure mechanisms (e.g. attributes leading to failure) and as such increase reliability through the identification of those failure-associated attributes. Additionally, the aforementioned embodiments are based on the use of information from multiple attribute categories. Embodiments in which attribute information from a single attribute category (pangenetic, behavioral, physical, or situational) can be used in circumstances where attributes from a single category dominate in the development of a condition or outcome.
Embodiments of the present invention can be used for a variety of methods, databases, software and systems including but not limited to: pattern recognition; feature extraction; binary search trees and binary prediction tree modeling; decision trees; neural networks and self-learning systems; belief networks; classification systems; classifier-based systems; clustering algorithms; nondeterministic algorithms (e.g., Monte Carlo methods); deterministic algorithms; scoring systems; decision-making systems; decision-based training systems; complex supervised learning systems; process control systems; chaos analysis systems; interaction, association and correlation mapping systems; relational databases; navigation and autopilot systems; communications systems and interfaces; career management; job placement and hiring; dating services; marriage counseling; relationship evaluation; animal companion compatibility evaluation; living environment evaluation; disease and health management and assessment; genetic assessment and counseling; genetic engineering; genetic linkage studies; genetic screening; genetic drift and evolution discovery; ancestry investigation; criminal investigation; forensics; criminal profiling; psychological profiling; adoption placement and planning; fertility and pregnancy evaluation and planning; family planning; social services; infrastructure planning; species preservation; organism cloning; organism design and evaluation; apparatus design and evaluation; invention design and evaluation; clinical investigation; epidemiological investigation; etiology investigation; diagnosis, prognosis, treatment, prescription and therapy prediction, formulation and delivery; adverse outcome avoidance (i.e. prophylaxis); data mining; bioinformatics; biomarker development; physiological profiling; rational drug design; drug interaction prediction; drug screening; pharmaceutical formulation; molecular modeling; xenobiotic side-effect prediction; microarray analysis; dietary analysis and recommendation; processed foods formulation; census evaluation and planning; population dynamics assessment; ecological and environmental preservation; environmental health; land management; agriculture planning; crisis and disaster prediction, prevention, planning and analysis; pandemic and epidemic prediction, prevention, planning and analysis; weather forecasting; goal formulation and goal achievement assessment; risk assessment; formulating recommendations; asset management; task management; consulting; marketing and advertising; cost analysis; business development; economics forecasting and planning; stock market prediction; lifestyle modification; time management; emergency intervention; operational/failure status evaluation and prediction; system failure analysis; optimization analysis; architectural design; and product appearance, ergonomics, efficiency, efficacy and reliability engineering (i.e., product development).
The embodiments of the present invention may be implemented with any combination of hardware and software. If implemented as a computer-implemented apparatus, the present invention is implemented using means for performing all of the steps and functions described above.
The embodiments of the present invention can be included in an article of manufacture (e.g., one or more computer program products) having, for instance, computer useable media. The media has embodied therein, for instance, computer readable program code means for providing and facilitating the mechanisms of the present invention. The article of manufacture can be included as part of a computer system or sold separately.
While specific embodiments have been described in detail in the foregoing detailed description and illustrated in the accompanying drawings, it will be appreciated by those skilled in the art that various modifications and alternatives to those details could be developed in light of the overall teachings of the disclosure and the broad inventive concepts thereof. It is understood, therefore, that the scope of the present invention is not limited to the particular examples and implementations disclosed herein, but is intended to cover modifications within the spirit and scope thereof as defined by the appended claims and any and all equivalents thereof.
Claims
1. A computer based method for attribute prediction, comprising:
- a) receiving a query attribute;
- b) accessing a first set containing attributes associated with an individual;
- c) accessing a second set containing attribute combinations and statistical results that indicate the strength of association of each of the attribute combinations with the query attribute;
- d) identifying one or more attribute combinations in the second set that consist of attributes from the first set; and
- e) generating, based on the identified attribute combinations and the statistical results, one or more statistical predictions indicating the potential association between the individual and the query attribute.
2. The computer based method of claim 1, further comprising:
- f) storing the statistical predictions.
3. The computer based method of claim 1, further comprising:
- f) associating the statistical predictions with the individual and storing the association.
4. The computer based method of claim 3, wherein storing the association comprises storing a link to at least one of: an identifier of the individual, the first set, and a record of the individual.
5. The computer based method of claim 1, further comprising:
- f) repeating steps (a)-(e) for a succession of query attributes to create a tabulated attribute prediction report, profile or record.
6. The computer based method of claim 1, further comprising:
- f) transmitting the statistical predictions.
7. The computer based method of claim 1, further comprising:
- f) transmitting the statistical predictions to create an attribute prediction report, profile or record for the individual.
8. The computer based method of claim 1, further comprising:
- f) updating the attribute combinations and statistical results of the second set, based on the attribute content of the first set.
9. The computer based method of claim 1, wherein the query attribute comprises a combination of two or more attributes.
10. The computer based method of claim 1, wherein the identity of the individual is masked or anonymized.
11. The computer based method of claim 1, wherein at least one of the statistical predictions is generated based on the statistical results for the largest attribute combination identified in step (d).
12. The computer based method of claim 7, wherein the statistical predictions must meet at least one statistical requirement in order to be transmitted.
13. The computer based method of claim 12, wherein the at least one statistical requirement is selected from the group comprising: a minimum statistical value, a maximum statistical value, and a value of statistical significance.
14. The computer based method of claim 1, further comprising:
- f) eliminating one or more statistical predictions that are based on attribute combinations which lack one or more specified attributes.
15. The computer based method of claim 1, further comprising:
- f) eliminating one or more statistical predictions that are based on attribute combinations which contain one or more specified attributes.
16. A computer based system for attribute prediction, comprising:
- a) a data receiving subsystem for receiving a query attribute;
- b) a data accessing subsystem comprising: i) a first data accessing subsystem for accessing a first set containing attributes associated with an individual; ii) a second data accessing subsystem for accessing a second set containing attribute combinations and statistical results that indicate the strength of association of each of the attribute combinations with the query attribute;
- c) a data processing subsystem comprising: i) a data comparison subsystem for identifying one or more attribute combinations in the second set that consist of attributes from the first set; and ii) a statistical prediction subsystem for generating, based on the identified attribute combinations and the statistical results, one or more statistical predictions indicating the potential association between the individual and the query attribute.
17. The computer based system of claim 16, further comprising:
- d) a data storage subsystem for storing the statistical predictions and for storing an association between the statistical predictions and the individual.
18. The computer based system of claim 16, further comprising:
- d) a communications subsystem for transmitting the statistical predictions.
19. The computer based system of claim 16, wherein the data processing subsystem further comprises:
- iii) a database updating subsystem for updating the attribute combinations and the statistical results of the second set based on the attribute content of the first set.
20. The computer based system of claim 16, wherein the data processing subsystem further comprises:
- iii) a data elimination subsystem for eliminating one or more statistical predictions that are based on attribute combinations which lack one or more specified attributes, or alternatively, which contain one or more specified attributes.
Type: Application
Filed: May 13, 2007
Publication Date: Sep 18, 2008
Applicant: EXPANSE NETWORKS, INC. (Furlong, PA)
Inventors: Andrew Alexander Kenedy (Sugar Land, TX), Charles Anthony Eldering (Furlong, PA)
Application Number: 11/747,912
International Classification: G06F 17/30 (20060101); G06F 7/00 (20060101);
