Abstract: Systems and methods are described for forming a decision tree by considering multiple possible alternative tests or thresholds as well as the possible alternative branches from each test or threshold. This method of making decision trees may be referred to as “look-ahead” or “chart-ahead” recursive partitioning because tests or thresholds can be selected not only for their own ability to divide items into different classes, but also for the ability of future tests or thresholds, enabled by the present tests or thresholds, to further divide items into classes. In preferred embodiments, the method of making the decision tree includes generating clusters of items having similar descriptors and properties, and then evaluating various tests or thresholds for their ability to isolate clusters. Also disclosed are methods of using such decision trees, particularly for the purpose of predicting the unknown properties of molecules.

Abstract: Molecules are represented as one dimensional strings of atoms having a linear position and a width. One dimensional representations are compared to assess the likelihood of similarity in chemical behavior.

Abstract: A method predicts at least one property of a candidate molecule. A set of reference molecules is classified as either possessing or not possessing at least one property. A subset of the molecules is selected from which a plurality of marker molecules is chosen. The marker molecules are used to predict whether the candidate molecule possesses at least one property.

Abstract: The disclosure relates to a method of estimating the polar component of the solvation energy for a molecule embedded in different media. In one embodiment, the molecule is partially embedded in a membrane. For an atom of the molecule, the polar component of the atom's solvation energy is represented as a combination of at least a self-energy term and a screening-effect term. The self-energy term represents the contribution to the atom's polar component made by the membrane and the molecule's other atoms located inside the membrane. The screening-effect term represents the typically negative contribution to the atom's polar component made by the molecule's other atoms located outside the membrane. An analytical function is used to calculate the self-energy term.

Abstract: A high-throughput molecular docking facility is presented for screening combinatorial libraries to identify binding ligands and ultimately pharmaceutical compounds. The facility employs a pre-docking conformational search to generate multiple solution conformations of a ligand. The molecular docking facility includes: generating a binding site image of the protein, the binding site image having multiple hot spots; matching hot spots of the binding site image to atoms in at least one solution conformation of the multiple solution conformations of the ligand to obtain at least one ligand position relative to the protein in a ligand-protein complex formation; and optimizing the at least one ligand position while allowing translation, orientation and rotatable bonds of the ligand to vary, and while holding the protein fixed.

Abstract: This application describes systems and methods of forming a decision tree that simultaneously classifies multiple properties of items, and systems and methods of using the decision tree to simultaneously predicting multiple unknown properties of an item. In one embodiment, the decision tree has a top generic layer of nodes that are generic to all properties, a middle K layer of nodes that split on a property type descriptor K, and a bottom specific layer of nodes that are specific to some properties. In addition to simultaneously predicting multiple properties of new items, the decision tree may also aid users in uncovering relationships between properties and in determining what descriptors may represent a generic quality, and what descriptors may represent property-specific activity.

Abstract: The present teachings describe a solubility prediction approach that is based, in part, upon quantitative-structure property relationships (QSPR). In various embodiments, the solubility prediction methodology comprises identifying a reduced descriptor set from which a solubility equation is derived. The descriptors used in the solubility equation are based upon simplified structural and molecular features which facilitate researcher understanding and solubility model refinement. Moreover, the present teachings provide solubility prediction relationships that may be used to rapidly screen large libraries or collections of compounds.

Abstract: A method and system for indexing powder diffraction data are disclosed comprising choosing a maximum impurity peak tolerance level for a crystallography data search, choosing a range of number of calculated peaks for possible indexing solutions having a minimum number of peaks and a maximum number of peaks, selecting a crystal system to search, selecting powder extinction classes to search for indexing solutions, performing an exhaustive unit cell search of each of the selected powder extinction classes using a successive dichotomy approach to determine a set of indexing results, and ranking the obtained solutions according to likelihood.