Controllable protein design through Feynman-Kac steering

2025-11-12

AI Summary

Existing diffusion models struggle to incorporate non-differentiable functional constraints—such as binding affinity or sequence composition—into protein generation. Method: We propose the Feynman–Kac guidance framework, the first approach to embed non-differentiable, multi-objective reward signals directly into diffusion inference, enabling model-agnostic controllable protein design. Our method jointly optimizes structural energy and functional metrics by integrating diffusion sampling, ProteinMPNN-based sequence redesign, and all-atom refinement. Contribution/Results: On multiple binder design tasks targeting diverse proteins, our method significantly improves predicted interface binding energy accuracy (mean gain of 1.8 kcal/mol), generates high-diversity, high-fidelity structures, and achieves low computational overhead with strong generalization across targets. Crucially, it overcomes the differentiability barrier inherent in conventional gradient-based control, establishing a general, function-driven paradigm for de novo protein design.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Sampling/Simulation-based Search
• Humans and AI: Game Design — Procedural Content Generation & Storytelling

Application Category

• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Diffusion-based models have recently enabled the generation of realistic and diverse protein structures, yet they remain limited in their ability to steer outcomes toward specific functional or biochemical objectives, such as binding affinity or sequence composition. Here we extend the Feynman-Kac (FK) steering framework, an inference-time control approach, to diffusion-based protein design. By coupling FK steering with structure generation, the method guides sampling toward desirable structural or energetic features while maintaining the diversity of the underlying diffusion process. To enable simultaneous generation of both sequence and structure properties, rewards are computed on models refined through ProteinMPNN and all-atom relaxation. Applied to binder design, FK steering consistently improves predicted interface energetics across diverse targets with minimal computational overhead. More broadly, this work demonstrates that inference-time FK control generalizes diffusion-based protein design to arbitrary, non-differentiable, and reward-agnostic objectives, providing a unified and model-independent framework for guided molecular generation.

Problem

• Steering protein generation toward specific functional objectives like binding affinity
• Guiding sampling toward desirable structural features while maintaining diversity
• Enabling simultaneous generation of both sequence and structure properties

Innovation

• Feynman-Kac steering guides diffusion-based protein design
• Rewards computed using ProteinMPNN and all-atom relaxation
• Framework generalizes to non-differentiable objectives