Low data drug modeling

The project aims to facilitate the research and development of new drugs by exploring Machine Learning methodology useful for both the generation of new molecules and the prediction of molecule properties. Doing so will involve training deep learning models on a large number of small, heterogeneous datasets, with the objective of transferring learned representations quickly when faced with a new drug-discovery or drug optimization objectives. The trained models will be used for the purposes of predicting molecular properties of new drugs and generating novel molecules with high likelihood of satisfying certain properties. The multi-objective nature of designing new molecules satisfying competing objectives will be approached using techniques from Reinforcement Learning.

Intern: 
Basile Dura
Faculty Supervisor: 
Yoshua Bengio
Project Year: 
2019
Province: 
Quebec
Partner: 
Discipline: 
Program: