BPS2026 – Efficient comparison of protein conformational ensembles

Comparing protein conformational ensembles is of central importance in structural biology, yet available methods can be computationally prohibitive for large ensembles. We present PROTHON, which represents an ensemble as a vector of per-residue probability distribution functions of a local structural quantity such as the C-beta contact number. Dissimilarity between two ensembles is computed as the mean of Jensen-Shannon distances between corresponding per-residue distributions, yielding both a global score and a residue-resolved dissimilarity vector; local differences are assessed by resampling and a Mann-Whitney U test. This representation is alignment-free and scales efficiently. We validate PROTHON on molecular-dynamics-generated ubiquitin ensembles spanning folded and partially folded macrostates, where it orders the ensembles sensibly and avoids non-monotonic artifacts seen with RMSD-based comparisons. We further compare experimentally derived ensembles of the tau K18 domain (an intrinsically disordered protein), where PROTHON confirms previous studies that the ensembles are similar. PROTHON is efficient, delivering substantially shorter wall-clock times and reduced compute requirements in our benchmarks. It is available as an open-source Python package with code and example data. The framework generalizes to other local order parameters (e.g., per-residue SASA, virtual C-alpha bond, and torsion angles) and supports motif-level as well as global comparisons.