Conformational sampling and docking with AlphaFold
Title: Conformational sampling and docking with AlphaFold
DNr: Berzelius-2024-152
Project Type: LiU Berzelius
Principal Investigator: Björn Wallner <>
Affiliation: Linköpings universitet
Duration: 2024-05-01 – 2024-11-01
Classification: 10601


Proteins are key players in virtually all biological events and accomplish their function as part of larger protein complexes. Unfortunately, compared to structure determination of individual proteins, structural characterization protein-protein interactions is much more complex and is a major challenge in structural biology. For transient interactions it is even more difficult since the interactions only last for a short period of time they are hard to detect, and thus, even more challenging to study in molecular detail. Yet these transient interactions are key for regulating complex signaling networks that determine normal cell fate or lead to diseases such as cancer, autoimmunity, cardiovascular and neurological diseases. In this proposal, we want to utilize the AlphaFold AI software for protein structure prediction developed by DeepMind to study protein-protein interactions involving multiple dynamic and disordered partners. Over the last year, we have adapted AlphaFold to better sample the conformational space. It is now possible to sample more conformational states to capture the dynamics and functional nature of proteins. We will continue developing this protocol. In the coming month, we will also explore the possibility of retraining AlphaFold to tailor it to our specific application areas. A complete retrain of AlphaFold is beyond the scope of this small allocation (it would require the complete Berzelius for a couple of weeks at least). Thus, what we are aiming for here is to see if we can picky back on the learned representations in AlphaFold and retrain, for instance, the structural module to improve the selection of different conformational states.