General Multimodal Protein Design Enables DNA-Encoding of Chemistry

DISCO is a multimodal deep generative model that co-designs protein sequences and 3D structures.

It co-designs sequence and structure around arbitrary biomolecules, including reactive intermediates.

DISCO designed diverse heme enzymes with novel active-site geometries.

Those enzymes catalyze carbene-transfer reactions: alkene and spirocyclopropanation, B–H and C(sp3)–H insertions.

Designed enzymes show activities exceeding those of engineered enzymes, and random mutagenesis confirmed improvements via directed evolution; code: https://github.com/DISCO-design/DISCO.