Novel Deep Learning Strategy to Better Predict Pharmacological Properties of Candidate Drugs and Focus Discovery Efforts

Collaborative Drug Discovery, Inc Mathematics, 2021–22

Liaison(s): Philip Cheung ’96, Mike Bowles, Rachel Schibler ’20, Alex Clark, Peter Gedeck
Advisor(s): Lisette de Pillis
Students(s): Mia Taylor, Alex Bishka, Jorge Canedo, Bryan Uribe, Jack Weiler, Daniel Yang

Collaborative Drug Discovery, Inc. (CDD) is a cheminformatics company that has been developing an automated drug discovery application to identify molecules predicted to be safe and effective drugs. CDD’s existing approach represents molecules as a sequence of characters. This character string representation contains artifacts that do not reflect chemical properties of the molecule, thereby obfuscating the real chemical properties that the model must learn. The team has investigated, prototyped, and tested two molecule generation models that represent the molecule as a graph rather than as a character string. With these models, the team has been able to generate chemically valid novel molecules.