Inherent limitations of zero-shot mutation prediction in protein engineering
- 1The University of Texas at Austin
Abstract
Predicting mutation effects is critical for protein engineering. A commonly pursued strategy is zero-shot prediction with deep-learning models, whose performance is typically evaluated using aggregate benchmark scores that obscure multidimensional biological tradeoffs. Here, we evaluate sequence, structure, and hybrid foundation models across curated datasets to expose the practical limitations of zero-shot predictions. We demonstrate that model performance is highly variable and heavily influenced by experimental noise and competing phenotypic goals. Notably, predictions from competing models often correlate more strongly with one another than with experimental outcomes. While current models effectively filter deleterious mutations and enrich high-scoring variants for natural functions, they struggle to identify precise optimal mutations and consistently fail to predict mutations that drive new-to-nature functions. These models learn from evolutionary constraints that tend to oppose the extreme biophysical optimizations required in targeted engineering. Consequently, zero-shot scoring serves as a valuable initial filter but cannot substitute for task-specific prediction.
Related
- Limitations of general protein language models and public TCRpMHC data in training specificity prediction models
- Towards Sparse Causal Features for Zero-shot Mutation Effect Prediction in a Protein Language Model
- From single-sequence structure prediction to protein fitness landscape through a composable, epistasis-aware mutation atlas
- Joint protein-DNA language modeling enables prediction and engineering of DNA recognition specificity
- MPFIL: Multi-Level Protein Feature Integration for Predicting Mutation-Induced Protein-Protein Binding Affinity Changes
BibTeX
@misc{woolley2026inherent,
title = {{Inherent limitations of zero-shot mutation prediction in protein engineering}},
author = {P. R. Woolley and A. Feller and A. O. Ellington and C. O. Wilke},
howpublished = {bioRxiv},
year = {2026},
month = jun,
doi = {10.64898/2026.06.04.730121},
url = {https://www.biorxiv.org/content/10.64898/2026.06.04.730121v3},
note = {CC BY-NC},
}
RIS
TY - UNPB
TI - Inherent limitations of zero-shot mutation prediction in protein engineering
AU - P. R. Woolley
AU - A. Feller
AU - A. O. Ellington
AU - C. O. Wilke
PY - 2026
DA - 2026/06/04
PB - bioRxiv
DO - 10.64898/2026.06.04.730121
UR - https://www.biorxiv.org/content/10.64898/2026.06.04.730121v3
AB - Predicting mutation effects is critical for protein engineering. A commonly pursued strategy is zero-shot prediction with deep-learning models, whose performance is typically evaluated using aggregate benchmark scores that obscure multidimensional biological tradeoffs. Here, we evaluate sequence, structure, and hybrid foundation models across curated datasets to expose the practical limitations of zero-shot predictions. We demonstrate that model performance is highly variable and heavily influenced by experimental noise and competing phenotypic goals. Notably, predictions from competing models often correlate more strongly with one another than with experimental outcomes. While current models effectively filter deleterious mutations and enrich high-scoring variants for natural functions, they struggle to identify precise optimal mutations and consistently fail to predict mutations that drive new-to-nature functions. These models learn from evolutionary constraints that tend to oppose the extreme biophysical optimizations required in targeted engineering. Consequently, zero-shot scoring serves as a valuable initial filter but cannot substitute for task-specific prediction.
ER -