BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Hierarchical Protein Structure Representation Learning via Topolog
 ical Deep Learning - Zhiyu Wang
DTSTART:20250729T161500Z
DTEND:20250729T173000Z
UID:TALK234460@talks.cam.ac.uk
CONTACT:Pietro Lio
DESCRIPTION:Protein representation learning (PRL) is crucial for understan
 ding structure-function relationships\, yet current sequence- and graph-ba
 sed methods fail to capture the hierarchical organization inherent in prot
 ein structures. We introduce Topotein\, a comprehensive framework that app
 lies topological deep learning to PRL through the novel Protein Combinator
 ial Complex (PCC) and Topology-Complete Perceptron Network (TCPNet). Our P
 CC represents proteins at multiple hierarchical levels---from residues to 
 secondary structures to complete proteins---while preserving geometric inf
 ormation at each level. TCPNet employs SE(3)-equivariant message passing a
 cross these hierarchical structures\, enabling more effective capture of m
 ulti-scale structural patterns. Through extensive experiments on four PRL 
 tasks\, TCPNet consistently outperforms state-of-the-art geometric graph n
 eural networks. Our approach demonstrates particular strength in tasks suc
 h as fold classification which require understanding of secondary structur
 e arrangements\, validating the importance of hierarchical topological fea
 tures for protein analysis.
LOCATION:Computer Laboratory\, William Gates Building\, Lecture Theatre 1
END:VEVENT
END:VCALENDAR