A structure-derived contact-network responsiveness atlas of human proteins

Abstract

Protein structures encode non-local contact organization, but static coordinates do not directly quantify how a contact network responds when effective stabilizing interactions are strengthened or weakened. Here we introduce Contact-Network Responsiveness (CNR), a structure-derived framework that converts residue-level protein coordinates into density-controlled and topology-corrected response descriptors. The method is structure-source agnostic and can be applied to experimentally determined PDB structures, AlphaFold models, or other predicted structures; here, human AlphaFold models serve as the high-coverage structural substrate. Across 22,167 valid human protein structures, hydrophobic non-local contact density defined a nearly exact Bethe mean-field baseline for the conformational susceptibility threshold. A graph-aware residue-level extension then revealed systematic topology-dependent deviations from this density-only prediction. We define a topology correction ratio, [Formula], which separates topology-facilitated, density-dominated and topology-suppressed contact-network response regimes. CNR descriptors were associated with curated DisProt disorder annotations and broad-coverage UniProt/MobiDB-lite disorder fractions, supporting the interpretation that CNR captures a structural organization axis related to non-local contact availability and responsiveness.