Bindome releases 306 thousand computationally designed protein binders for 8 296 human proteins
Bindome releases 306 thousand computationally designed protein binders for 8 296 human proteins
The open database provides each candidate’s amino acid sequence, a predicted structure of its interaction with the target, and model metrics. These computational designs are starting points for experiments on specific human proteins.
On August 4, Julius Wenckstern announced the launch of Bindome. On July 30, the EPFL team published a preprint describing the catalog: 306 146 computationally designed candidates for 8 296 proteins, covering 40,9% of the human proteome. The database allows users to find a protein, inspect the predicted complex, and download the data.
A protein functions through its shape and its contacts with other molecules. To test the role of a protein region in a cell, a laboratory selects a reagent that binds to that region and observes what changes. Antibodies often serve this purpose, but each target requires a separate series of experiments to produce and validate them. A protein binder can be smaller than an antibody, and its sequence is known in advance.
The authors started with AlphaFold structural models, which predict a protein’s shape from its sequence. They then identified compact regions within long proteins, selected accessible surfaces, and applied BindCraft, a method that designs a protein sequence to interact with a specified target. They excluded models with poor structural metrics or clashes with the rest of the protein.
Bindome applies computational screening previously tested for BindCraft to thousands of targets. In the BindCraft paper, the authors measured candidate binding to twelve protein targets. For PD-1, 13 of 53 tested designs produced a signal, compared with 7 of 9 for PD-L1. The catalog gives laboratories a set of starting candidates. They must then synthesize them and measure binding, selectivity, and cellular effects.
The catalog changes the first step of an experiment. For each new target, a researcher receives a set of candidates that can be tested in the relevant system. This may be useful in aging research when a genetic screen has already linked a protein to a cellular state: the target’s name can be converted into specific candidates for experimental testing.