Protein models find simpler search directions to improve evolution predictions
Linear Fitness Subspace in Protein Language Models Enables Sample-Efficient Directed Evolution
Artificial IntelligenceMachine Learning
Summary
Finding better proteins for medicine or industry can be very slow and expensive because testing many variants takes a lot of effort. The authors found that changes in protein language model representations follow a simpler pattern when linked to how well proteins perform a task, making it easier to predict good variants from few examples. They created a method that focuses on this simple pattern, allowing faster and more accurate searches for high-performing proteins using fewer tests. This approach worked better than existing methods on many protein fitness tests.
What this means in practice
- •For biotech engineers: Optimize protein variants efficiently with fewer lab tests by using learned compact directions from protein language model changes.
- •For pharmaceutical developers: Improve screening of therapeutic protein candidates by guiding experimental searches using assay-specific low-dimensional representations.
Authors
SiYuan Ma, Canran Xiao, Zikai Xiao, Albert Gao, Liang He, Xuan-Yu Wang, Shuying Cao, Xiaojun Jia
Abstract
Model-guided directed evolution seeks to identify high-fitness protein variants under limited oracle budgets. Protein language models (PLMs) provide rich representations for this task, but task-agnostic zero-shot scores can be misaligned with a target assay, while supervised search in high-dimensional embedding spaces can make surrogate modeling and uncertainty estimation sample-inefficient. We propose the Linear Fitness Subspace (LFS) hypothesis: within mutation-induced residue-level representation changes, a compact, assay-specific set of directions makes fitness variation linearly accessible from few labeled variants. This is a local, supervision-recoverable statement rather than a claim that protein fitness landscapes or global PLM geometry are universally linear. Building on this observation, we introduce Subspace-Guided Evolutionary Search (SGES), which estimates an LFS from a small initial sample and performs surrogate modeling, uncertainty estimation, and acquisition in the learned subspace. Across 10 core ProteinGym assays, 87 extended static-validation assays, and an 18-assay budgeted-search evaluation, SGES improves fitness prediction and search efficiency over zero-shot PLMs and recent ML-guided protein optimization baselines. Controlled comparisons with PCA, random projections, label-shuffled PLS, classical mutation features, and acquisition ablations further isolate the benefit of a fitness-aligned site-delta coordinate.