Protein optimization improves multiple traits with modular diffusion steering
Multi-Objective Protein Optimization via Modular Diffusion Steering
Machine Learning
Summary
Designing proteins that work well in several ways at once is hard because there are so many possible protein sequences. The authors created a method called MODS that uses a smart sampling approach to guide protein design toward different tradeoffs between traits. MODS works by combining predictions from multiple models and adjusting goals to explore a variety of protein possibilities. It works with different underlying models and quickly updates its strategies without needing to relearn everything. When tested on several protein design problems, MODS matched or beat strong alternatives.
What this means in practice
- •For biopharmaceutical developers: Design protein candidates that balance multiple desired properties efficiently and iteratively during drug discovery pipelines.$Commercial implications: Enables creation of multi-trait-optimized proteins for therapeutic use, accelerating commercial drug candidate optimization with less experimental testing.
- •For biotech protein engineers: Optimize enzymes or proteins for various functional properties in fewer experimental rounds by steering generative models toward different trait tradeoffs.
Authors
Meir Roketlishvili, Felix Huber, Wenda Chu, Yisong Yue, Raul Astudillo
Abstract
Protein engineering often requires improving several properties simultaneously, motivating the search for high-performing sequences that represent useful fitness tradeoffs. Discovering these sequences is challenging because the design space is combinatorially large and experimental campaigns often permit only a few rounds of batched measurements. We introduce \textit{MODS}, a modular framework for multi-objective protein optimization via inference-time diffusion steering. For each proposal, MODS samples augmented Chebyshev scalarization weights and a fitness-surrogate ensemble member to define a scalar steering reward. Varying the weights directs generation toward different objective tradeoffs, while varying ensemble members diversifies the predictions used for guidance. Because MODS expresses multi-objective acquisition through scalar rewards, it can be paired with different diffusion models and compatible steering procedures. After each evaluation batch, MODS updates the surrogates without retraining the generative model. We evaluate MODS on six protein optimization tasks using predictive oracles trained on experimental assay data, with two diffusion models and two steering procedures. MODS achieves competitive final hypervolume across all six tasks with both diffusion models, and substantially outperforms the best evaluated baseline on several tasks. Our results demonstrate a flexible and effective approach to multi-objective protein optimization with limited feedback.