Inference-time optimization for experiment-grounded protein ensemble generation

Sai Advaith Maddipatla, Anar Rzayev, Marco Pegoraro, Ailie Marx, Martin Pacesa, Paul Schanda, Sanketh Vedula, Alexander Bronstein

Keywords: protein generative models, alphafold3, inference-time optimization, experiment-grounded ensemble generation, design confidence metrics

TL;DR: We introduce an inference-time optimization method that improves protein ensemble accuracy by optimizing latent variables and combining AlphaFold3 and force-field priors. It improves data fit, and reveals flaws in iPTM-based confidence metrics.

Abstract: Protein function relies on dynamic conformational ensembles, yet models like AlphaFold3 (AF3) often fail to produce ensembles that match experimental data. Recent experiment-guided approaches address this by steering the reverse diffusion process, but remain limited by fixed sampling horizons and sensitivity to initialization, often yielding thermodynamically implausible results. We introduce a general inference-time optimization framework that instead optimizes latent representations to maximize ensemble likelihood, removing dependence on diffusion length, reducing initialization bias, and enabling flexible incorporation of external constraints. Further, we present sampling schemes for generating Boltzmann-weighted ensembles by combining AF3 structural priors with force-field energies. Across X-ray and NMR benchmarks, our method improves diversity, physical energy, and agreement with experimental data, often surpassing deposited PDB structures. Finally, ipTM maximization shows that perturbing AF3 embeddings can artificially inflate model confidence, exposing a vulnerability in current design metrics and suggesting ways to reduce false discoveries in binder engineering.

Submission Number: 98