Joint sequence-structure diffusion or flow matching models show superior performance when designing structure at the beginning and sequence near the end

Summary

Joint sequence/structure diffusion models or flow matching models show best performance when structure design is done at the beginning and sequence design is left to the end (1–2). DISCO specifically found that including entropy-adaptive temperature annealing throughout the diffusion process improved both the number of viable clusters and the designability fraction, particularly when the sequence sampling temperature was fixed at a specific value (3).

Figures

Ref (3)

See also

• Coupling sidechain and main chain prediction or design does not always lead to improvements

1.

Didi K, Zhang Z, Zhou G, Reidenbach D, Cao Z, Cha S, et al. Scaling Atomistic Protein Binder Design with Generative Pretraining and Test-Time Compute. 2026; Available from: https://arxiv.org/abs/2603.27950

2.

Stark H, Faltings F, Choi M, Xie Y, Hur E, O’Donnell T, et al. BoltzGen: Toward Universal Binder Design. openRxiv; 2025. Available from: https://doi.org/10.1101/2025.11.20.689494

3.

Rector-Brooks J, Lambert T, Skreta M, Roth D, Long Y, Li Z-Q, et al. General Multimodal Protein Design Enables DNA-Encoding of Chemistry. 2026; Available from: https://arxiv.org/abs/2604.05181

4.

Qiu K, Wang X, Zhang Z, Zhou H, Ma W-Y. Demystifying Multimodal Biomolecular Co-design With Intrinsic Geodesic Coupling. 2026; Available from: https://arxiv.org/abs/2606.01628