AF2BIND: Predicting ligand-binding sites using the pair representation of AlphaFold2
Artem Gazizov, Anna Lian, Casper Goverde, Sergey Ovchinnikov, Nicholas Polizzi
Github | Colab | Preprint
Predicting ligand-binding sites, particularly in the absence of previously resolved homologous structures, presents a significant challenge in structural biology. Here, we leverage the internal pairwise representation of AlphaFold2 (AF2) to train a model, AF2BIND, to accurately predict small-molecule-binding residues given only a target protein. AF2BIND uses 20 "bait" amino acids to optimally extract the binding signal in the absence of a small-molecule ligand. We find that the AF2 pair representation outperforms other neural-network representations for bindingsite prediction. Moreover, unique combinations of the 20 bait amino acids are correlated with chemical properties of the ligand.
Old version (from Protein Society meeting)
Github | Poster
The accurate prediction of ligand-binding sites in proteins remains an outstanding challenge, despite its potential to accelerate drug discovery and inform on natural protein function. Here, we train a neural network, AF2BIND, using embedding features from a protein structure prediction model, AlphaFold2, to accurately predict the binding sites of proteins.