Protein-protein interactions are fundamental to the regulation of cellular life, governing essential pathways in signal transduction and immune responses, and control most biological functions, including pathological processes like cancer, Alzheimer or COVID. Despite their paramount biomedical importance, systematically predicting whether two proteins will associate, calculating their binding free energy, and accurately estimating their association and dissociation rates remains an outstanding challenge in computational biophysics. The scientific community nowadays studies PPI by means of all-atom Molecular Dynamics simulations, producing a huge amount of data in the process. To extract information, dimensionality reduction algorithms are employed. This project aims at transcending standard dimensionality reduction by shifting the optimization paradigm from spatial compression to dynamical minimization.