Predicting Relative Populations of Protein Conformations without a Physics Engine Using AlphaFold2

TL;DR

This paper introduces a method leveraging AlphaFold 2 to predict the relative populations of protein conformations and mutation effects without using a physics engine, achieving over 80% accuracy in tests.

Contribution

It demonstrates how AlphaFold 2 can be adapted to predict conformational populations and mutation-induced changes, extending its capabilities beyond single structure prediction.

Findings

Achieved over 80% accuracy in predicting population changes.

Successfully applied to proteins with varying sequence data availability.

Provided a fast, cost-effective alternative to traditional methods.

Abstract

This paper presents a novel approach for predicting the relative populations of protein conformations using AlphaFold 2, an AI-powered method that has revolutionized biology by enabling the accurate prediction of protein structures. While AlphaFold 2 has shown exceptional accuracy and speed, it is designed to predict proteins' single ground state conformations and is limited in its ability to predict fold switching and the effects of mutations on conformational landscapes. Here, we demonstrate how AlphaFold 2 can directly predict the relative populations of different conformations of proteins and even accurately predict changes in those populations induced by mutations by subsampling multiple sequence alignments. We tested our method against NMR experiments on two proteins with drastically different amounts of available sequence data, Abl1 kinase and the granulocyte-macrophage colony-stimulating factor, and predicted changes in their relative state populations with accuracies in excess of 80%. Our method offers a fast and cost-effective way to predict protein conformations and their relative populations at even single point mutation resolution, making it a useful tool for pharmacology, analyzing NMR data, and studying the effects of evolution.