System and Method for Molecular Reconstruction from Molecular Probability Distributions

A computer system comprising:

• a hardware memory, wherein the computer system is configured to execute software instructions stored on nontransitory machine-readable storage media that:
• train a plurality of molecule models, one for each molecule in a dataset of molecules, wherein each molecule model comprises every substructure in the respective molecule, and wherein each substructure is represented as an embedding;
• use contrastive optimization across all the plurality of molecule models to form clusters of similar substructures;
• assign signatures to each embedding, wherein similar substructure embeddings have similar signatures and less-similar substructure embeddings have less-similar signatures;
• receive a probability distribution of a molecule;
• predict a set of molecular descriptors from the probability distribution using a trained substructure processing model, wherein the set of molecular descriptors comprises at least one of the following:
• substructure centroids, substructure dimensions, substructure directions, substructure embeddings, or some combination thereof;
• wherein the substructure embeddings are determined by comparative signature analysis with the substructure embeddings generated by the embeddings module;
• encode the set of molecular descriptors into a tensor, wherein the tensor fully describes the molecule;
• use the tensor as input to a junction tree connector model to predict a junction tree node structure for the molecule;
• use the tensor to predict the atomic assignment of each atom in each substructure of the molecule;
• connect the substructures together to form a valid molecule using the predicted junction tree node structure and atomic assignments;
• and
• produce a molecular string which is fully representative of the molecule from the molecular probability distribution.