With the news saturated by the opportunities presented by AI, it’s easy to assume that AI alone holds all of the answers to our problems in computer-aided drug discovery. Indeed, the release of the ground-breaking protein structure prediction tool AlphaFold2, was an exciting development that increased the accuracy and speed of de-novo structure prediction. Nevertheless, it would be foolish to throw away our work on physics-based methods in favour of AlphaFold2 since significant challenges remain; including the fact that biomolecules exist as structural ensembles rather than static structures and that these structural ensembles change upon binding to other biomolecules or ligands.

In this newsletter we choose to highlight two recent advancements in AlphaFold; AlphaFold-latest,⁽¹⁾ reported by the DeepMind AlphaFold Team, and AlphaFold2-RAVE, published by Vani et al.⁽²⁾  Excitingly, AlphaFold-latest has extended the functionality of AlphaFold2 to include protein-ligand structure prediction that rivals the current best docking tools. AlphaFold2-RAVE, on the other hand, utilises both AlphaFold2 and RAVE (Re-weighted Auto-encoded Variational Bayes for Enhanced Sampling), a molecular dynamics enhanced sampling methodology, to increase the conformational diversity of the structure predictions while enforcing thermodynamic relevance with Boltzmann-weighting. The development of both of these open-source tools is an exciting step forward for the applications of AI in drug discovery, and the integration of Boltzmann-weighting in AlphaFold2-RAVE serves as a reminder that physics-based methods are still relevant and needed.

Figure from ref 2)

• Google DeepMind AlphaFold Team and Isomorphic Labs Team. Performance and structural coverage of the latest, in-development AlphaFold model. 2023

• Vani, B.P et al. AlphaFold2-RAVE: From Sequence to Boltzmann Ranking. .J. Chem. Theory Comput. 2023, 19, 14, 4351–4354