Abstract: This disclosure provides a machine learning technique to predict a protein characteristic. A first training set is created that includes, for multiple proteins, a target feature, protein sequences, and other information about the proteins. A first machine learning model is trained and then used to identify which of the features are relevant as determined by feature importance or causal relationships to the target feature. A second training set is created with only the relevant features. Embeddings generated from the protein sequences are also added to the second training set. The second training set is used to train a second machine learning model. The first and second machine learning models may be any type of regressors. Once trained, the second machine learning model is used to predict a value for the target feature for an uncharacterized protein. The model of this disclosure provides 91% accuracy in predicting an ideal digestibility score.

Abstract: This disclosure provides a data-driven and scalable method to discover cause-and-effect relationships in data from natural systems that include sparse data sets. This technique can learn a causal graph from heterogenous data sources by combining embeddings from real data and embeddings from simulated data generated by process-based models. The causal graph is used for what-if analysis in out-of-distribution settings. One application is understanding the factors that affect soil carbon. A causal model created by these techniques can be used to discover cause-and-effect relationships that affect soil carbon. This model has applications such as forecasting soil carbon for a future time point to help inform farm practices. Farm practices, like tilling, may be modified in response to predictions provided by the model.

Abstract: A deep learning system is used to predict crop characteristics from inputs that include crop variety features, environmental features, and field management features. The deep learning system includes domain-specific modules for each category of features. Some of the domain-specific modules are implemented as convolutional neural networks (CNN) while others are implemented as fully-connected neural networks. Interactions between different domains are captured with cross attention between respective embeddings. Embeddings from the multiple domain-specific modules are concatenated to create a deep neural network (DNN). The prediction generated by the DNN is a characteristic of the crop such as yield, height, or disease resistance. The DNN can be used to select a crop variety for planting in a field. For a crop that is planted, the DNN may be used to select a field management technique.

Abstract: Techniques for optically detecting a subject chemical species within an atmospheric environment are disclosed. Image data is obtained representing multispectral imagery of a geographic region captured through the atmospheric environment. The image data includes an array of band-specific intensity values for each of a plurality of spectral bands, including a sample spectral band having increased sensitivity to the subject chemical species as compared to a plurality of reference spectral bands. A background reflectance map is generated that includes an array of inter-band intensity values in which each inter-band intensity value represents a filtered combination of band-specific intensity values of albedo-normalized arrays for a grouped subset of the plurality of reference spectral bands. The albedo-normalized array of band-specific intensity values for the sample spectral band is compared to the background reflectance map to obtain an index array of intensity variance values for the subject chemical species.

Abstract: A method for source attribution comprises receiving measurements of a chemical species at a spatially distributed sensor array for a given set of spatially positioned emission sources in a physical environment using a dispersion model. Based on the received measurements, a concentration field is mapped from the emission sources to the sensor array using a forward operator. For each emission source, a likelihood data set is evaluated at least by fitting an emission rate of the chemical species using a regression model based on the mapped concentration field and real-world, runtime measurements from the sensor array. A posterior data set is evaluated based at least on the evaluated likelihood data set and historical data for the physical environment. For each sensor of the sensor array, estimated emission rates and contribution rankings for emission sources are determined and output based on the evaluation of the posterior data set.

Abstract: A method for pollutant sensor placement for pollutants from point sources is described. Data about environmental characteristics for a geographic region are received from a plurality of environmental sensors. The geographic region includes pollutant sources that emit a pollutant. The received data from one or more of the plurality of environmental sensors are transformed into common data having a common spatial and temporal discretization across the geographic region. Predicted emission plumes are generated for the pollutant sources within the geographic region that identify pollutant detection regions for the pollutant when the pollutant is emitted by the pollutant sources using the common data. Sensor locations for a plurality of pollutant sensors are greedily selected across the common spatial and temporal discretization according to a number of predicted emission plumes that are detectable by the plurality of pollutant sensors.