Do protein representations encode variant function? Diagnosing residue-level protein attributions via embedding-space gradients
- 1University of Toronto
Abstract
A common workflow in protein engineering involves building fitness predictors on protein language models (pLMs) to estimate a variant's fitness and interpret which residues drive it. Attribution methods such as integrated gradients address this second goal by assigning residues contributions to the variant's predicted fitness. However, without ground truth for residue importance, it is unclear whether these attributions reflect what the predictor learned about function. Here, we introduce IGLOO, a framework that scores residue-level attributions against synthetic fitness landscapes with known per-residue importance. Across experimental axes, high predictive accuracy did not guarantee faithful attributions. We identify model interventions that more than double attribution skill scores and raise TOPSIS by 29%, with similar patterns holding in real -amylase and {beta}-lactamase datasets. IGLOO provides a protocol for probing what pLMs learn and a guide for choosing defaults when using residue attributions to pick sequence edits and build more trustworthy protein models.
Related
- Embedding response surfaces in kinetic models predicts microalgae growth across light and nutrient gradients
- Distinct gradients of cortical architecture capture visual representations and behavior across the lifespan
- Gene Embedding Hub: a standardized resource for human gene representations
- Diagnosing Neurodegenerative Diseases by Label-Free 3-D Imaging of Intestinal Samples
- Do Perturbation Predictors Learn Gene Interactions? Diagnosing Extrapolation in Agentic Screening
BibTeX
@misc{barghout2026protein,
title = {{Do protein representations encode variant function? Diagnosing residue-level protein attributions via embedding-space gradients}},
author = {R. A. Barghout and S. Wu and W. Chau and J. M. Barraza-Chavez and P. Driza and R. Mahadevan and B. Sanchez-Lengeling},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.21.760763},
url = {https://www.biorxiv.org/content/10.64898/2026.10.21.760763v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Do protein representations encode variant function? Diagnosing residue-level protein attributions via embedding-space gradients
AU - R. A. Barghout
AU - S. Wu
AU - W. Chau
AU - J. M. Barraza-Chavez
AU - P. Driza
AU - R. Mahadevan
AU - B. Sanchez-Lengeling
PY - 2026
DA - 2026/10/21
PB - bioRxiv
DO - 10.64898/2026.10.21.760763
UR - https://www.biorxiv.org/content/10.64898/2026.10.21.760763v1
AB - A common workflow in protein engineering involves building fitness predictors on protein language models (pLMs) to estimate a variant's fitness and interpret which residues drive it. Attribution methods such as integrated gradients address this second goal by assigning residues contributions to the variant's predicted fitness. However, without ground truth for residue importance, it is unclear whether these attributions reflect what the predictor learned about function. Here, we introduce IGLOO, a framework that scores residue-level attributions against synthetic fitness landscapes with known per-residue importance. Across experimental axes, high predictive accuracy did not guarantee faithful attributions. We identify model interventions that more than double attribution skill scores and raise TOPSIS by 29%, with similar patterns holding in real -amylase and {beta}-lactamase datasets. IGLOO provides a protocol for probing what pLMs learn and a guide for choosing defaults when using residue attributions to pick sequence edits and build more trustworthy protein models.
ER -