Topology-driven completion of chemical data

A system that generates new molecules, said system comprising:

• a memory;
• and
• a processor in communication with said memory, said processor being configured to perform operations comprising:
• receiving molecular data for a plurality of molecules, wherein said molecular data includes molecular features;
• determining a molecular fingerprint for each of said plurality of molecules with an encoder based on said molecular data;
• performing a topological data analysis on said molecular data with said encoder to generate a molecular topological graph, wherein said molecular data includes said molecular fingerprint for each of said plurality of molecules;
• calculating a generative potential score for said plurality of molecules;
• training a scaffold-conditioned variational autoencoder with said generative potential scores, wherein said scaffold-conditioned variational autoencoder is conditioned on an input scaffold;
• identifying one or more lacunae in said molecular topological graph;
• and
• generating, using said scaffold-conditioned variational autoencoder, one or more additional molecules to fill at least one of said one or more lacunae, wherein said one or more additional molecules each contain said input scaffold.

A computer program product for generating new molecules, said computer program product comprising a computer readable storage medium having program instructions embodied therewith, said program instructions executable by a processor to cause said processor perform a function, said function comprising:

• receiving molecular data for a plurality of molecules, wherein said molecular data includes molecular features;
• determining a molecular fingerprint for each of said plurality of molecules with an encoder based on said molecular data;
• performing a topological data analysis on said molecular data with said encoder to generate a molecular topological graph, wherein said molecular data includes said molecular fingerprint for each of said plurality of molecules;
• calculating a generative potential score for said plurality of molecules;
• training a scaffold-conditioned variational autoencoder with said generative potential scores, wherein said scaffold-conditioned variational autoencoder is conditioned on an input scaffold;
• identifying one or more lacunae in said molecular topological graph;
• and
• generating, using said scaffold-conditioned variational autoencoder, one or more additional molecules to fill at least one of said one or more lacunae, wherein said one or more additional molecules each contain said input scaffold.