#3 · Structural bio & protein design

MD and cofolding models probe conformational states in kinases, Aβ, and HIV-1 Env

The common thread is the use of explicit conformational ensembles, generated by MD or enhanced sampling, to define functional states and to evaluate them against ligands, mutations, or membrane context. KinConfBench tests whether cofolding models (Boltz-2, Chai-1, Protenix) recover ligand-induced kinase conformational states across 2,225 human kinase chains, a benchmark that depends on distinguishing conformations rather than single structures. The KRAS study converts MD-derived features into logistic regression, random forest, SVM, and Bayesian network classifiers of resistance linked to secondary mutations. DyME scales MD-based high-throughput mutagenesis with solvated simulations and comparative analysis for protein engineering and recognition mimicry. The Aβ work uses T-REMD ensembles and self-docking to derive an amyloidogenic-propensity descriptor for transient aggregation-prone states, and the HIV-1 Env study shows glycans modulating membrane tilting in microsecond all-atom simulations corroborated by cryo-ET. The new element across these items is the direct use of conformational ensembles as predictive or design-relevant descriptors, though scope and scale vary considerably.