Free-energy calculations and machine learning uncover diverse p38α–MK2 molecular glues
- Glue-type inhibitors that bridge p38α and MK2 act longer than classic active-site blockers, but the ternary-complex mechanism makes rational design hard.
- The team combined free energy perturbation with machine learning to rank compounds drawn from varied chemical sources in silico.
- This yielded several structurally distinct, potent, selective glues; two leads showed good exposure and cut TNFα in an LPS mouse model after oral dosing.
💡 Points Based on the abstract
A team would next point this FEP-plus-ML triage at its own induced-proximity target, but first ask for the missing control: how did the ranking beat plain docking or a similarity screen on hit rate? For glues I'd also want cooperativity (ternary versus binary affinity) and residence time reported, since those, not potency alone, explain the durable effect.
Diverse Computational Strategies Enable the Discovery of p38α−MK2 Molecular Glues ↗