Schroedinger Bridges and Generative Models

Conventional molecular discovery relies on high-throughput screening strategies (HTS) to identify molecules that match a specific set of user-defined properties. Unfortunately, due to the enormous size of chemical spaces (10^60 molecules in the case of the drug-like chemical space), HTS is usually performed over pre-built chemical libraries of moderate size, often by employing computational/AI methods to increase the screening speed and the size of investigated libraries. Generative models (GM) are neural networks that can be used to generate novel molecules with desired profiles, and thus represent a powerful alternative to property-based HTS for exploring the chemical space. However, current GMs are challenging to design and train with high-dimension molecular representations, large molecules and large databases. We propose to develop next-generation GMs for molecular discovery that will alleviate existing challenges and unlock their potential for applications in drug and material discovery.
This project has an important learning component for the students visiting uOttawa. It combines state-of-art techniques in computational chemistry, optimal transport theory and deep generative models to tackle fundamental problems in computer-aided molecular discovery.

Faculty Supervisor:

Gentile Francesco

Student:

Partner:

Kharkiv National University of Radio Electronics

Discipline:

Computer science

Sector:

Artificial Intelligence; Biotechnology; Pharmaceuticals

University:

University of Ottawa

Program: