Title:	Modelling of transient protein-protein interactions relevant for human health
DNr:	LiU-compute-2024-43
Project Type:	LiU Compute
Principal Investigator:	Björn Wallner <bjorn.wallner@liu.se>
Affiliation:	Linköpings universitet
Duration:	2024-11-18 – 2025-12-01
Classification:	10203
Keywords:

Abstract

Building from the transformative change AlphaFold has had on structural biology by enabling prediction on par with experimental data. Despite its success in predicting protein structures with unprecedented accuracy, AlphaFold is constrained by the static and single-conformation nature of its training data. We propose a more advanced AI method that can predict dynamic ensembles of structures, thereby providing a more realistic representation of macromolecules. The project will start from our recent improvements to the AlphaFold model, AFsample, which was ranked at the very top in the latest CASP15 competition, and involve the following aims: 1. Ensemble: Investigate how to utilize sampling in AlphaFold to optimally predict local structural ensembles of disordered loops and other heterogeneous regions in macromolecular experimental data sets from crystallography and cryo-EM. 2. Dynamics: Explore structural dynamics by retraining a network with a similar architecture as the AlphaFold network but with the added ability to reason over dynamic 3D structures. 3. Density: Develop a system to produce probability distributions directly from raw experimental data and a sampling method to generate populations of sample structures from these probability distributions. These advancements could unlock new applications and bring predicted structures (ensembles) closer to reality.