The open-source tool Lacuna searches proteins for hidden pockets where small molecules can bind
The open-source tool Lacuna searches proteins for hidden pockets where small molecules can bind
In a preprint published on August 18, Clayton Moore described Lacuna, an open-source tool for finding these pockets. In its standard mode, Lacuna generates 20 closely related protein conformations. In the tests, the median processing time on a single CPU core was 2.6 seconds per chain.
Proteins continually undergo small changes in shape, whereas a structural model captures only one conformation. Some pockets open only during these movements. A search based on a single conformation can therefore miss a site that could accommodate a small molecule.
In an evaluation using 925 binding maps, GNINA, a program that predicts how a molecule fits into a protein, reproduced the molecule’s position in fewer than 5% of cases when given a protein conformation in which a loop covered the pocket.
Starting from a single structure, Lacuna generates 20 closely related conformations using normal modes, a simplified model of small, coordinated movements among parts of a protein. In each conformation, the program identifies cavities, groups matching cavities into a single site, and ranks the sites by how often they opened, the volume they reached, and how far they moved. This process turns brief pocket openings into a short list of sites for further calculations.
The ranking of this list affects the result. CryptoBench is a dataset containing pairs of structures of the same protein, one without a bound molecule and one with a bound molecule. The author kept the detected candidates unchanged and varied only their ranking. The previous analytical rule placed the correct site among the top five candidates in 17.8% of cases. A linear model based on 23 properties of each cavity and its behavior across conformations increased this proportion to 55.6%.
Clayton Moore describes the result as follows:
“The detector was already identifying the correct pocket for most targets missed by the analytical rule, but it ranked that pocket lower in the results.”
Lacuna records the selected sites in a report. On request, it also prepares search regions for AutoDock Vina, which predicts how a molecule fits into a protein, and constraints for Boltz, a model that predicts the joint structure of a protein and a molecule.
For large movements between protein domains and loop rearrangements, Lacuna can use molecular dynamics, Boltz-2, or an existing set of structures.