Vilya-2: Unlocking structure prediction across a broad chemical space

Sub-angstrom accuracy for non-canonical peptides that existing computational tools do not model well.

Read the technical report
arrow-right

Co-folding models break down at the point where peptides acquire the non-canonical chemistry they need to reach the clinic. Vilya-2 is built for that regime.

Modeling peptide–target complexes

Vilya-2 resolves a peptide–target complex through diffusion, without templates or co-evolutionary input utilizing a single unified atomic representation. Vilya-2 accurately predicts the interactions of Zolucatetide, Daraxonrasib, Icotrokinra, and Enlicitide with their respective therapeutic protein targets.

Generalization from small molecules to miniproteins

Because Vilya-2 represents every molecule as atoms and bonds, its performance extends to small molecules and miniproteins of 2–3 kDa — several-fold larger than any molecule in its training set — across both canonical and non-canonical chemistries.

Interested?

We are actively using Vilya-2 in multiple internal programs. Want access? Contact us!

Email
arrow-right
Abstract, blurred view of a laboratory with a scientist in the background and lab equipment in the foreground.

Open positions

Interested in joining us to push the boundary of AI-based design of macrocyclic peptides? Explore the open positions!

Work with us
arrow-right
Scientist in protective gear writing notes while working in a laboratory.

Next