Home Department: Computer Science
Mentor: Grant Rotskoff (Chemistry)

“Evaluating Protein Structure Prediction Models’ Conformational Ensemble Prediction Ability"

Protein structure prediction models like AlphaFold offer useful but incomplete insights: in reality, proteins rapidly interchange between many conformations, of which structure prediction models only capture a static representation. The Rotskoff lab’s project probes whether these models, beyond their static predictions, ‘know’ about conformational ensembles. Ethan will specifically examine diffusion-based models, for which numerical methods can be used to calculate the probabilities that models assign alternative conformations. Ethan will compare his results to ground-truth probabilities determined through molecular dynamics simulations. In a time where AI methodologies are increasingly applied to physical sciences, it’s ever-important to rigorously investigate when and where they’re truly reliable.