BPS2026 – Maximum entropy polymer model predicts the architectural effects of genetic structural variations

Noncoding structural variations (SVs) exert deleterious phenotypic effects by remodeling chromatin architecture. The nonlinear nature of the remodeling process makes case-by-case prediction of SVs’ effects arduous. This study employs MiChroM, a maximum-entropy polymer model, to systematize the prediction of the architectural effects of SVs. MiChroM trains a pairwise potential matrix on a Hi-C contact map to build an in-silico model of a genomic locus. By applying SVs to our simulated polymer, we investigate the architectural mechanisms involved in SV-associated phenotypic alterations. As our procedure predicts the effects of SVs by training on a wild-type contact map, it does not require Hi-C data from SV-bearing tissue. We benchmark our model on six limb-development-associated structural variations at the EPHA4 locus. Our model correctly predicts changes in key ectopic, disease-associated enhancer-promoter interactions. Analysis of structural ensembles reveals architectural reorganization consistent with prior hypotheses for this set of mutations. To enhance predictive capacity, we develop matrix adjustment techniques grounded in polymer physics and chromatin folding theory. We also employ a simultaneous kinetic/thermodynamic model of chromatin folding by simulating chromatin organizing motors on the polymer. Comparisons of the adjusted-thermodynamic and kinetic/thermodynamic approaches highlight fundamental principles of genome architecture organization. With the EPHA4 mutation as a successful benchmark, we turn our model’s predictive capacities to previously unsimulated structural variations. We predict the effects of architectural SVs to the SOX9 locus and of complex contact domain shuffling SVs. Our predicted SV contact maps are of higher resolution than their experimental counterparts while reproducing the expected ectopic interactions. This procedure enables predictions of the architectural effects of SV at high resolution based solely on already-published WT contact map data. It thus opens up new avenues for laboratory and clinical investigation of genetic disease.