Structure prediction

Diffusion-based structure prediction can be guided by backpropagating to the conditioning embeddings rather than the atomic coordinates directly, and such embeddings can be re-refined in subsequent iterations

Diego del Alamo

— 2 min read

Diffusion-based biomolecular structure prediction, which is used in latest-generation methods like AlphaFold3[1] and BioEmu[2], can be guided or steered into specific conformations by backpropagating to the conditioning representations rather than the atomic coordinates being diffused [3][4]. This was recently shown by two methods, EmbedOpt and IT-Optimization.

Pasted image 20260306093401.png

There are two arguments for doing this over standard Diffusion Posterior Sampling[5] (DPS), which operates directly on the elements being subject to diffusion (here, atomic coordinates). First, DPS tends to distort the outputs and send them off the manifold of realistic conformations. This is consistent with my own experience with cryoBoltz[6], which uses DPS; often, for example, chainbreaks are introduced to occupy spurious density clouds.

Pasted image 20260220170148.png

Second, the conditioning embeddings can be re-used and re-refined in subsequent runs, leading to even greater improvements. The figure below shows how a single run of Protenix (labeled AF3) can match or even exceed guidance of conditioning representations, but is fundamentally limited to a single diffusion roll-out; subsequent inference must start from scratch.

Pasted image 20260306093452.png

As the recent report by Nvidia discusses[7], such guidance approaches begin to blend together elements of hallucination-based and diffusion-based protein modeling.

References:

  1. Abramson, J., Adler, J., Dunger, J., Evans, R., Green, T., Pritzel, A., Ronneberger, O., Willmore, L., Ballard, A. J., Bambrick, J., Bodenstein, S. W., Evans, D. A., Hung, C.-C., O’Neill, M., Reiman, D., Tunyasuvunakool, K., Wu, Z., Žemgulytė, A., Arvaniti, E., … Jumper, J. M. (2024). Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature, 630(8016), 493–500. https://doi.org/10.1038/s41586-024-07487-w ↩︎

  2. Lewis, S., Hempel, T., Jiménez-Luna, J., Gastegger, M., Xie, Y., Foong, A. Y. K., Satorras, V. G., Abdin, O., Veeling, B. S., Zaporozhets, I., Chen, Y., Yang, S., Foster, A. E., Schneuing, A., Nigam, J., Barbero, F., Stimper, V., Campbell, A., Yim, J., … Noé, F. (2025). Scalable emulation of protein equilibrium ensembles with generative deep learning. Science, 389(6761). https://doi.org/10.1126/science.adv9817 ↩︎

  3. Li, M., Han, J., Cossio, P., & Wu, L. (2026). Robust Inference-Time Steering of Protein Diffusion Models via Embedding Optimization (Version 1). arXiv. https://doi.org/10.48550/ARXIV.2602.05285 ↩︎

  4. Maddipatla, A., Rzayev, A., Pegoraro, M., Pacesa, M., Schanda, P., Marx, A., Vedula, S., & Bronstein, A. M. (2026). Inference-time optimization for experiment-grounded protein ensemble generation (Version 2). arXiv. https://doi.org/10.48550/ARXIV.2602.24007 ↩︎

  5. Chung, H., Kim, J., Mccann, M. T., Klasky, M. L., & Ye, J. C. (2022). Diffusion Posterior Sampling for General Noisy Inverse Problems (Version 4). arXiv. https://doi.org/10.48550/ARXIV.2209.14687 ↩︎

  6. Raghu, R., Levy, A., Wetzstein, G., & Zhong, E. D. (2025). Multiscale guidance of protein structure prediction with heterogeneous cryo-EM data (Version 2). arXiv. https://doi.org/10.48550/ARXIV.2506.04490 ↩︎

  7. Didi, K., Zhang, Z., Zhou, G., Reidenbach, D., Cao, Z., Cha, S., Geffner, T., Dallago, C., Tang, J., Bronstein, M. M., Steinegger, M., Kucukbenli, E., Vahdat, A., & Kreis, K. (2026). Scaling atomistic protein binder design with generative pretraining and test-time compute. The Fourteenth International Conference on Learning Representations. https://openreview.net/forum?id=qmCpJtFZra ↩︎

Read more