BIPSPI+: Enhancing partner-specific binding site prediction with better data

BIPSPI (xgBoost Interface Prediction of Specific Partner Interactions) is a partner specific
binding site predictor that employs the XGBoost algorithm. This new version, BIPSPI+, has been trained on
newer and larger datasets compiled from the PDB, severely improving its performance.

BIPSPI+ can be employed to predict partner-specific binding sites given two atomic models,
two sequences, or one sequence and one structure.

BIPSPI+ is distributed as a Docker image and as a GitHub repository. Installation is only required in
the latter case. Complete guide can be found in:

• DOCKER image
• GitHub repository

New features

• Homo-complexes and hetero-complex specific models, boosting performance especially when predicting homo-complexes.
• Sequence vs structure mode. Original version could only be executed with two sequences or two structures. Now one sequence and one structure could be used as input as well.
• Atomatic correlated mutations (optional)