Passer au contenu

/ Department of Computer Science and Operations Research

Je donne

Rechercher

Huiyu Cai's Predoc III Presentation

Dear all / Bonjour à tous,


We are happy to invite you to  Huiyu Cai's Predoc III defense on Friday, August, 25nd, at 10:00 am (hybrid).


Vous êtes cordialement invité.e.s à la présentation du sujet de recherche de Huiyu Cai, le vendredi 25 août, à 10h00 (hybride)


Title: Modeling interaction in complex biomolecular systems with geometric deep learning

Date: August 25nd, 2023 - 10:00am-12:00pm (Montreal time)

Location: Auditorium 1 - 6650 rue Saint Urbain

 

Jury

PrésidentWolf, Guy/ Nie, Jian-Yun
Directeur de rechercheTang, Jian
Membre
Wolf, Guy/ Nie, Jian-Yun

 

Abstract

Modeling interactions in complex biomolecular systems is fundamental to biological sciences and therapeutics. From energy transfer to cell signaling, from DNA replication to immune responses, most biological functions are facilitated by a series of complex interactions among various ions, small molecules and macromolecules. In the realm of therapeutics, effectively obstructing disease-causing biological pathways necessitates precise prediction of drug-target interactions.

In this proposal, we comprehensively review existing literature in both geometric deep learning and biomolecular interaction modeling, particularly in applications such as protein-ligand docking and antibody affinity maturation. We highlight the superior potential of geometric deep learning for comprehending the intricate interplay in biomolecular systems due to its symmetry-awareness, accuracy, and efficiency. Furthermore, we explore diverse considerations critical for the design of geometric deep learning models, identify limitations to current approaches, and underscore promising research directions that might shape the future of this field.

We demonstrate our capability across these perspectives with two recent projects. (1) E3Bind, an end-to-end equivariant network designed for protein ligand docking. E3Bind iteratively updates the ligand pose through careful consideration of the geometric constraints in docking and the local context of the binding site. (2) GearBind, a pretrainable geometric graph neural network engineered for antibody affinity maturation. Notably, our GearBind-based pipeline successfully enhanced the affinity of antibody CR3022 to the spike protein of the SARS-CoV-2 Omicron strain by a factor of 17. Based on these projects, we outline our future work on biomolecular interaction learning. We anticipate that this work will not only advance the computational modeling of complex biomolecular systems but also contribute to high-impact applications in biological discovery and drug design.