일시: 2025. 11. 26. (수요일), 오후 5시

장소: 제4공학관 D504호

Speaker: Minkyung Baek, Ph.D. (Department of Biological Sciences, Seoul National University, Republic of Korea)

Title: AI-based protein structure prediction and design

Abstract:

With the increase in bio big data, data-driven approaches using AI techniques have made considerable progress in protein structure prediction and protein design. In this talk, I’ll present an AI-based protein structure prediction method, RoseTTAFold, and its utilization to design de novo proteins having desired functions. RoseTTAFold is a three-track attention-based neural network which transforms and integrates information at the sequence level, the distance map level, and the 3D coordinate level to generate accurate protein structures. It enabled high accuracy protein modeling and protein-protein interaction prediction. We further extended our approach to develop a general AI framework for protein design including de novo protein binder design. We transformed RoseTTAFold to a diffusion model that have had considerable success in generative modeling of images and languages. The resulting RFdiffusion model achieves outstanding performance on protein design.

Lab: https://biosci.snu.ac.kr/cssb