Modelling protein complexes with crosslinking mass spectrometry and deep learning

Protein structure prediction has been revolutionised by AlphaFold2, an algorithm that uses deep learning and evolutionary information to predict accurate models from the primary sequence. However, proteins are dynamic entities that interact with other molecules, undergo conformational changes, and remain difficult to predict in some cases. We have developed AlphaLink, a modified version of the AlphaFold2 algorithm that synergistically incorporates experimental distance restraint information into its network architecture (manuscript submitted, data presented below). By employing sparse experimental contacts as anchor points, AlphaLink improves on the performance of AlphaFold2 in predicting challenging targets. The noise-tolerant framework for integrating data in protein structure prediction presented here opens a path to accurate characterisation of protein structures from in-cell data. Having addressed the challenge of predicting single proteins is encouraging also for predicting protein complexes. However, while AlphaFold2 and AlphaFold-multimer are closely related, they are different algorithms. Accordingly, AlphaLink can model the structure of single proteins but not the structure of protein complexes. The aim of the proposed project is to add experimental distance restraints to the AlphaFold-multimer framework for structure modelling of protein complexes.