#39 · Epigenomics

Computational models probe methylation memory, enzyme mechanisms and protein states

The common thread across these three preprints is computational inference evaluated against explicit biological constraints rather than new experimental readouts. The methylation study applies Shannon information theory to estimate the capacity of the DNA methylome as an epigenetic memory system and its mutual information with cell identity under biological constraints. EZSolver is a template-free generative framework combining flow matching with bidirectional beam search to predict polar enzymatic mechanisms. The benchmarking study evaluates AlphaFold3, Boltz-2, Chai-1 and BioEmu on four multi-state proteins and reports a persistent bias toward dominant PDB states, with ligands contributing weakly. What is new is quantitative framing of epigenetic memory capacity, template-free mechanism generation, and explicit characterization of conformational bias in structure predictors. Each of these claims currently rests on a single preprint.