Yilin Chen

This PhD project aims to develop Al-driven methods for modeling flexible protein-peptide interactions and designing functional peptides. Protein-peptide binding often involves dynamic interfaces, peptide flexibility, and disorder-to-order transitions, which are not fully captured by static docking approaches. The project will focus on generative modeling of protein-peptide binding ensembles, where target conformational ensembles are used to identify dynamic interaction hotspots and guide peptide design. In the first stage, machine learning models will be developed to predict protein-peptide binding structures, interaction patterns, and binding interfaces, with particular attention to flexible and intrinsically disordered regions. In the second stage, target-aware generative models will be developed for de novo peptide generation and lead optimization, aiming to produce peptides with desirable binding, specificity, and biophysical properties. The generated candidates will be evaluated through structural validation, affinity prediction, hotspot consistency, and multi-stage filtering. The long-term goal is to build an end- to-end computational pipeline integrating interaction modeling, peptide generation, structural validation, and candidate ranking. This pipeline will support both de novo peptide design and optimization of existing peptide leads. Potential application scenarios include neurodegenerative disease targets involving intrinsically disordered proteins or regions, such as Aẞ, Tau, and a-synuclein, as well as cancer-related dynamic extracellular receptors such as c-MET. The project is expected to contribute new computational tools for peptide therapeutic discovery and provide mechanistic insights into dynamic protein-peptide recognition.