DEL-Flow: DEL selection data steer a flow-matching generator to MDM2–p53 macrocyclic peptides with confirmed activity

  • Two rounds of DNA-encoded library selection on symmetric then asymmetric macrocyclic peptides pointed to an enriched half-scaffold, which was kept fixed.
  • The selection signals, generated conformations and MDM2 pocket poses trained a pocket-conditioned SE(3)-equivariant flow-matching model that varies side chains around the macrocyclic core.
  • From two templates, 40 candidates were generated; two were prioritized and resynthesized off DNA, and both inhibited MDM2–p53 in fluorescence polarization, in the micromolar IC50 range.
  • The authors conclude that DEL signals plus pocket geometry can guide generation of macrocycles not present in the library.
💡 Points Based on the abstract

Using DEL enrichment as the training prior rather than scarce IC50 data is the clever part, and it answers the usual data-scarcity problem for macrocycles. But micromolar hits from 2 of 40 candidates is a modest return when the DEL itself already produced enriched binders; the question is how it compares with simply picking more library members. Check selectivity against MDM4 and the cell activity next.

DEL-Flow: AI-Driven Structural Generation and Activity Validation of MDM2-Targeting Macrocyclic Peptides ↗