Polymer physics and AI-based methods reveal the mechanisms of chromatin folding in health and disease

Olimpo, Matteo, Pierno, Florinda, Carluccio, Ciro, Marotta, Gabriele, Nicodemi, Mario, Conte, Mattia
Biophysical Reviews · Springer Nature · 2026-11-20 · DOI: 10.1007/s12551-026-01480-5

Abstract

The eukaryotic genome is folded into a functionally consequential three-dimensional architecture whose organizing principles have recently attracted substantial interest. Two physical mechanisms dominate current thinking: active loop extrusion by motor complexes such as SMC (Structural Maintenance of Chromosomes), and thermodynamic phase separation driven by homotypic affinities between chromatin binding sites and their associated binding factors. Here we review how polymer-physics models of chromatin, combined with machine-learning inference, provide a quantitative and predictive description of chromosome folding. We focus on the Strings and Binders (SBS) model and on PRISMR, a physics-informed inference method that learns the arrangement of binding sites along a chromatin filament from contact data alone, and on the Loop-Extrusion model. We discuss how these approaches are validated against bulk contact maps, single-cell imaging and multi-way contact data, how they compare with purely sequence-based deep-learning predictors, and how they reconcile loop extrusion with phase separation at the single-molecule level. Finally, we show how the same type of chromatin models explain the formation of extrachromosomal DNA (ecDNA) condensates in cancer cells, where it accounts quantitatively for the selective amplification of oncogene regulatory contacts beyond copy-number effects.

Read at publisher →

Similar articles