ProTEUS is a mutation-resolved world model that combines sequence, evolutionary, and structural representations with fast all-atom prediction and closed-loop feedback to guide protein evolution and improve mutant activity.
Protein engineering requires navigating rugged fitness landscapes, where small sequence changes can produce large functional differences, but experimental screening can only sample a small fraction of sequence space. Thus, a central challenge is to efficiently traverse the large combinatorial space of mutations in silico. We introduce ProTEUS, a mutation-resolved world model for closed-loop protein evolution. Pretrained on deep mutational scanning assays, ProTEUS learns a joint representation of sequence, evolutionary, and structural information using MutationFormer, a mutation-biased attention architecture. Variants are represented as states in this joint latent space, and mutations…
world modelmutation-resolved world modelpretrainedjoint representationMutationFormermutation-biased attention architecturejoint latent spacephysics-constrained all-atom predictorexplicit geometric constraint enforcementAlphaFold3
- Upvotes
- 4
- Authors
- 8
- Published
- Nov 9, 2026
Summary and abstract excerpt via Hugging Face Papers ↗ · arXiv.