Diffusion-based structure prediction can be steered by modifying the conditioning embeddings rather than the latent space, and such embeddings can be used for subsequent iterations

Summary

Diffusion-based structure prediction can be steered into specific conformations by modifying conditioning embeddings rather than the latent-space embeddings used for diffusion (1,2). This has the added advantage of being reused, and therefore facilitating improvements, in sequential diffusion runs. This was done using cryo-EM and NMR data and was shown to slightly outperform standard Diffusion Posterior Sampling.

Figures

Figures from (2)

Ref (1)

1.

Li M, Han J, Cossio P, Wu L. Robust Inference-Time Steering of Protein Diffusion Models via Embedding Optimization. 2026; Available from: https://arxiv.org/abs/2602.05285

2.

Maddipatla A, Rzayev A, Pegoraro M, Pacesa M, Schanda P, Marx A, et al. Inference-time optimization for experiment-grounded protein ensemble generation. 2026; Available from: https://arxiv.org/abs/2602.24007