Fold-Conditioned De Novo Binder Design via AlphaFold2-Multimer Hallucination.
Rustamov, K. R.; Baev, A. Y.
Show abstract
De novo protein binder design has been revolutionized by deep learning methods, yet controlling binder topology remains a challenge. We introduce a fold-conditioned AlphaFold2-Multimer hallucination framework - FoldCraft - guided by a contact map similarity loss, enabling precise generation of binders with user-defined structural folds. This single loss function enforces fold-specific geometry while implicitly optimizing AlphaFold confidence metrics. We demonstrate the methods versatility by designing binders with six distinct topologies. Compared to RFdiffusion, FoldCraft yields higher structural and binding confidence. Applied to VHH nanobody design against four therapeutically relevant targets, our method outperforms RFAntibody in AlphaFold3-based evaluations. FoldCraft offers a general, efficient strategy for structure-guided binder design, expanding the accessible fold space for protein engineering and enabling robust nanobody generation with improved success rates.
Matching journals
The top 5 journals account for 50% of the predicted probability mass.
Similar papers in this journal
Similar papers in this journal
- Deep learning assessment of nativeness and pairing likelihood for antibody and nanobody design with AbNatiV2 94%
- BioPhi: A platform for antibody design, humanization and humanness evaluation based on natural antibody repertoires and deep learning 94%
- Towards generalizable prediction of antibody thermostability using machine learning on sequence and structure features 94%
Similar papers in this journal
- To pack or not to pack: revisiting protein side-chain packing in the post-AlphaFold era 94%
- SPRI: Structure-Based Pathogenicity Relationship Identifier for Predicting Effects of Single Missense Variants and Discovery of Higher-Order Cancer Susceptibility Clusters of Mutations 94%
- InversePep: Diffusion-Driven Structure-Based Inverse Folding for Functional Peptides 94%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.