AlloDyn: Integrating Molecular Dynamics and Generative Conformational Ensembles for Allosteric Site Prediction

AlloDyn is a machine learning pipeline for predicting allosteric binding sites in proteins. It combines structural pocket features (fpocket) with dynamic features derived from conformational ensembles generated by AlphaFlow or molecular dynamics simulations. The XGBoost classifier is trained on a curated dataset of proteins with known allosteric modulators and evaluated on the D24 benchmark (24 entries). This deposit contains: - AlloDyn source code (GitLab archive, tag v1.0.0): inference pipeline, feature extraction scripts, training scripts, benchmark evaluation tools, and documentation. The code is also available at: https://gitlab.inria.fr/vpryakhi/AlloDyn - Supplementary material (PDF): supplementary figures and tables for the accompanying publication