Seek and You Shall Fold

2025-11-17

AI Summary

This work addresses the fundamental challenge of incorporating non-differentiable experimental data—such as NMR chemical shifts, NOE restraints, and pairwise distances—into protein generative modeling. We propose the first gradient-free, general-purpose framework that couples continuous diffusion models with a customized genetic algorithm to enable conditional sampling under arbitrary non-differentiable experimental objective functions. Our method achieves, for the first time, direct chemical shift–guided protein structure generation, thereby overcoming the strict differentiability assumption inherent in existing generative models. Validated on multimodal NMR constraints, our generated structures exhibit high fidelity to experimental observations and uncover systematic physical inconsistencies in mainstream structure prediction tools. This work establishes a new paradigm for automated, experiment-driven protein modeling using non-differentiable biophysical data.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Diffusion Models for Vision
• Natural Language Processing: Generation

Application Category

• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Accurate protein structures are essential for understanding biological function, yet incorporating experimental data into protein generative models remains a major challenge. Most predictors of experimental observables are non-differentiable, making them incompatible with gradient-based conditional sampling. This is especially limiting in nuclear magnetic resonance, where rich data such as chemical shifts are hard to directly integrate into generative modeling. We introduce a framework for non-differentiable guidance of protein generative models, coupling a continuous diffusion-based generator with any black-box objective via a tailored genetic algorithm. We demonstrate its effectiveness across three modalities: pairwise distance constraints, nuclear Overhauser effect restraints, and for the first time chemical shifts. These results establish chemical shift guided structure generation as feasible, expose key weaknesses in current predictors, and showcase a general strategy for incorporating diverse experimental signals. Our work points toward automated, data-conditioned protein modeling beyond the limits of differentiability.

Problem

• Incorporating non-differentiable experimental data into protein modeling
• Enabling chemical shift integration in generative structure prediction
• Developing gradient-free guidance for experimental data conditioning

Innovation

• Genetic algorithm couples diffusion generator with black-box objectives
• Enables chemical shift guided protein structure generation
• Overcomes non-differentiability limitation in experimental data integration