Sequence context decodes multi-state conformational heterogeneity from crystallographic B-factors

Protein function is governed by conformational ensembles, yet the overwhelming majority of experimentally determined structures capture only static snapshots, leaving the multi-state heterogeneity that underlies biological activity largely inaccessible. X-ray crystallography, the dominant source of these structures, does carry one experimental proxy for motion, the atomic B-factor, but crystal packing and refinement artifacts frequently obscure its link to solution-state dynamics. Here we show that sequence context is sufficient to decode a protein's multi-state conformational heterogeneity from this seemingly noisy signal. We introduce the B-Factor Corrector (BFC), a fine-tuned protein language model that reframes B-factor analysis from a purely physical calculation into a sequence-context-aware translation task. By leveraging deep contextual embeddings to decouple intrinsic conformational heterogeneity from lattice constraints, BFC recovers the residue-wise heterogeneity ranking derived from crystallographic structural ensembles, reproducing the normalized profile at a Pearson correlation of 0.83 while correcting the distortions that analytical formulas leave in the raw signal. We further demonstrate the utility of this sequence-inferred signal across structural-biology applications: it provides an integrable prior for molecular docking that is statistically equivalent to experimental B-factors, unmasks cryptic conformational heterogeneity essential for chaperone function, and guides the energetic refinement of antibody interfaces to resolve steric conflicts. These results suggest that the static structures of the Protein Data Bank can be effectively repurposed to decode the conformational heterogeneity latent within them.