{"paper":{"title":"Robust Inference-Time Steering of Protein Diffusion Models via Embedding Optimization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Optimizing the conditional embedding steers protein diffusion models to fit experimental constraints more robustly than coordinate perturbation.","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Jiequn Han, Luhuan Wu, Minhuan Li, Pilar Cossio","submitted_at":"2026-02-05T04:13:33Z","abstract_excerpt":"A core challenge in structural biophysics is generating biomolecular conformations that are both physically plausible and consistent with experimental measurements. While sequence-to-structure diffusion models provide powerful priors, posterior sampling methods steer generation by perturbing atomic coordinates with gradients from experimental likelihoods. However, when the target lies in a low-density region of the prior, these methods require aggressive upweighting of the likelihood that can destabilize sampling and be sensitive to hyperparameters. We propose EmbedOpt, an inference-time steer"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"EmbedOpt matches coordinate-based posterior sampling baselines on sparse distance constraints and outperforms them on cryo-electron microscopy map fitting, including real, noisy experimental ones. Furthermore, EmbedOpt's smooth optimization behavior yields robustness to hyperparameters spanning two orders of magnitude and enables comparable performance with fewer diffusion steps.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That updating the conditional embedding reliably shifts the structural prior to satisfy experimental constraints without introducing non-physical artifacts or losing the model's learned coevolutionary knowledge.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"EmbedOpt optimizes the conditional embedding of protein diffusion models at inference time to shift the structural prior toward experimental constraints, outperforming coordinate-based posterior sampling on cryo-EM fitting while remaining robust across hyperparameter ranges.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Optimizing the conditional embedding steers protein diffusion models to fit experimental constraints more robustly than coordinate perturbation.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"5266571e7d75618f6c113ef363a33bb10406e04d4a9f1aec03452c5e7d6b4834"},"source":{"id":"2602.05285","kind":"arxiv","version":2},"verdict":{"id":"6f1849d1-b03c-4762-bcf1-fdd3874b7a4a","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T07:32:26.540811Z","strongest_claim":"EmbedOpt matches coordinate-based posterior sampling baselines on sparse distance constraints and outperforms them on cryo-electron microscopy map fitting, including real, noisy experimental ones. Furthermore, EmbedOpt's smooth optimization behavior yields robustness to hyperparameters spanning two orders of magnitude and enables comparable performance with fewer diffusion steps.","one_line_summary":"EmbedOpt optimizes the conditional embedding of protein diffusion models at inference time to shift the structural prior toward experimental constraints, outperforming coordinate-based posterior sampling on cryo-EM fitting while remaining robust across hyperparameter ranges.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That updating the conditional embedding reliably shifts the structural prior to satisfy experimental constraints without introducing non-physical artifacts or losing the model's learned coevolutionary knowledge.","pith_extraction_headline":"Optimizing the conditional embedding steers protein diffusion models to fit experimental constraints more robustly than coordinate perturbation."},"references":{"count":15,"sample":[{"doi":"10.1101/2025.01.08","year":2025,"title":"Diffusion Posterior Sampling for General Noisy Inverse Problems","work_id":"083ab9fb-05f0-41e1-9628-982017eac344","ref_index":1,"cited_arxiv_id":"2209.14687","is_internal_anchor":true},{"doi":"","year":null,"title":"arXiv preprint arXiv:2310.01110 (2023)","work_id":"0b0f610f-c2cd-42e6-8880-800c6336ef32","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Fadini, A., Li, M., McCoy, A. J., Terwilliger, T. C., Read, R. J., Hekstra, D., and AlQuraishi, M. Alphafold as a prior: Experimental structure determination conditioned on a pretrained neural network","work_id":"be3dc1c0-bb32-458e-af4d-84438b740fe5","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Manifold preserv- ing guided diffusion","work_id":"05205a7e-1f41-41c9-a877-8549235593dc","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Classifier-Free Diffusion Guidance","work_id":"acf2c588-c088-4a6c-938e-150ad7c666d7","ref_index":5,"cited_arxiv_id":"2207.12598","is_internal_anchor":true}],"resolved_work":15,"snapshot_sha256":"3e45ae5b131d088a05e7ff743e74b98317b7f76ae95e870f8ca5b54deffb1371","internal_anchors":3},"formal_canon":{"evidence_count":2,"snapshot_sha256":"2287ceadcbedacc2abf4f864dede13ec3764bac9fb38c1389433cd6fb3f32f38"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}