Benchmarking AlphaFold2 in the context of protein dynamics and transient protein interactions
Title: Benchmarking AlphaFold2 in the context of protein dynamics and transient protein interactions
SNIC Project: SNIC 2021/5-373
Project Type: SNIC Medium Compute
Principal Investigator: Björn Wallner <bjorn.wallner@liu.se>
Affiliation: Linköpings universitet
Duration: 2021-08-30 – 2022-05-01
Classification: 10203 10610 10601
Keywords:

Abstract

The release of AlphaFold2 (AF2) code for structure prediction mid July as completely changed the playing field for structural biology. We can now predict structure with an accuracy on par with experimental structures. This is great news for us since we need good starting structures to predict interactions (see below). However, to use AlphaFold2 we need to benchmark and evaluate its performance for our specific problems. In the current project we will investigate how AF2 can be utilized to predict dynamics and interactions. AF2 was developed for single protein chain prediction, but preliminary studies seem to indicate that it can used to predict interaction between two protein chains by simply connecting them with a flexible linker (something which is also commonly done experimentally). However, it still unclear exactly how to do this optimally. The current proposal have the following aims: 1. Investigate how AF2 perform on a data set of protein-peptide interactions that we have used previously and compare it to state-of-the-art. 2. Investigate how AF2 can be used to predict alternative conformation. Initial studies seem to suggest that most often AF2 converge to one single structure. We want to see if we can enable more variation in the prediction to capture alternate conformations. We will use data from CoDNaS (http://ufq.unq.edu.ar/codnas/), a database of Conformational Diversity of Native State in proteins and see if AF2 can be used to generate structural ensembles for protein sequences that are known to have multiple energy minimas. 3. Integrate AF2 structure prediction in our current prediction protocol that previous has relied on using known protein structures. Fulfilling these aims is absolutely crucial to our other projects which aims at understanding biology and human health at the molecular level.