Toxic Substructure Extraction Using Clustering and Scaffold Extraction

A system for molecular substructure extraction, comprising:

• a computing device comprising a memory and a processor;
• a substructure extraction module comprising a first plurality of programming instructions stored in the memory and operating on the processor, wherein the first plurality of programming instructions causes the computing device to:
• retrieve a plurality of data, wherein the plurality of data comprises molecular structure data and data regarding a property of interest;
• classify each molecule as either having or not having the property of interest in the plurality of data;
• compute the distributed stochastic neighbor embedding representation of the molecular structure data;
• construct a graph representing unclustered sets of molecules;
• optimize the modularity of the graph to produce clustered sets of molecules;
• compute the mean activity of each of the clusters, wherein clusters that do not meet a mean activity threshold are skipped for extraction;
• identify, for each of the active clusters, each molecule in the active cluster as either active or inactive;
• rank the bits of the Morgan Fingerprints respective of each active molecule by how often the bits appear in active molecules compared to inactive molecules;
• and
• output the substructures related to the top ranked Morgan Fingerprint bits.