Probabilistic Modeling of Antibody Structural Dynamics

NeurIPS 2025 Workshop FPI Submission32 Authors

Track: Open Problems

Keywords: Antibody dynamics, Probabilistic inference, Posterior sampling, Molecular simulations, Conformational ensembles, Uncertainty quantification, Generative models, Bayesian modeling, Rare-event sampling, Data integration

TL;DR: Efficiently sampling and inferring antibody conformational ensembles with quantified uncertainty can transform therapeutic design and advance probabilistic modeling in high-dimensional, multimodal systems.

Abstract: Antibody function depends on a distribution of structural conformations, with different states exhibiting distinct binding properties. Accurately characterizing these ensembles is crucial for therapeutic design; however, current experimental data are sparse and noisy, and molecular simulations are computationally expensive. This frames antibody dynamics as a posterior inference and sampling problem: efficiently exploring high-dimensional conformational space and estimating probable structures with quantified uncertainty. Solving this problem would transform antibody design, advance machine learning for high-dimensional, multimodal inference, and bridge theory and experiment with rigorous structural confidence estimates.

Submission Number: 32