Efficient Auto-Interpretability of AI Models in Biology
Abstract
Sparse autoencoders (SAEs), and other interpretability methods could turn AI models in Biology and other fields into engines of scientific discovery by explaining the superhuman capabilities of those models. However, a latent is only useful if we know three things: whether it is coherent, whether it can be described, and whether that description has predictive power. These questions are routinely conflated. We assemble them into a single pipeline and report the practical innovations each stage required. First, cross-seed dictionary stability prioritises which latents are worth spending resources to investigate. Second, an intruder-detection task asks whether a latents activating examples share a recognizable pattern. Third, a separate pass proposes a candidate biological description which we convert into falsifiable predictions which can be tested in silico. Deployed on the Boltz-1 Pairformer trunk, stability prioritisation finds interpretable latents using about 4.4 times fewer latent evaluations each, and at 5.2 times lower measured cost, while recovering over half of them, and the external check shows the surfaced motifs are significantly enriched for their claimed annotations. The results also suggest a possible tension: the cross- seed stability might be selecting for some types of features, like structure-related ones, much more than others, such as function-related features.
Related
- Effects of within-season temporal aggregation of weather variables on performance and interpretability in rice yield machine learning models
- LytM family endopeptidase LmdC is controlled by regulated auto processing and proteolytic destruction in Pseudomonas aeruginosa
- Force Induces a Collective Inactivity in Dyneins to Auto-Regulate Bidirectional Transport of Vesicles
- Mechanistic Interpretability Needs a Logic of Inference
- Evo 2 as a classification machine: evidence from in-context learning and mechanistic interpretability
BibTeX
@misc{jedryszek2026efficient,
title = {{Efficient Auto-Interpretability of AI Models in Biology}},
author = {Piotr Jedryszek and Oliver M. Crook},
howpublished = {arXiv},
year = {2026},
month = aug,
doi = {10.48550/arxiv.2608.27754},
eprint = {2608.27754},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2608.27754},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Efficient Auto-Interpretability of AI Models in Biology
AU - Piotr Jedryszek
AU - Oliver M. Crook
PY - 2026
DA - 2026/08/27
PB - arXiv
DO - 10.48550/arxiv.2608.27754
UR - https://arxiv.org/abs/2608.27754
AB - Sparse autoencoders (SAEs), and other interpretability methods could turn AI models in Biology and other fields into engines of scientific discovery by explaining the superhuman capabilities of those models. However, a latent is only useful if we know three things: whether it is coherent, whether it can be described, and whether that description has predictive power. These questions are routinely conflated. We assemble them into a single pipeline and report the practical innovations each stage required. First, cross-seed dictionary stability prioritises which latents are worth spending resources to investigate. Second, an intruder-detection task asks whether a latents activating examples share a recognizable pattern. Third, a separate pass proposes a candidate biological description which we convert into falsifiable predictions which can be tested in silico. Deployed on the Boltz-1 Pairformer trunk, stability prioritisation finds interpretable latents using about 4.4 times fewer latent evaluations each, and at 5.2 times lower measured cost, while recovering over half of them, and the external check shows the surfaced motifs are significantly enriched for their claimed annotations. The results also suggest a possible tension: the cross- seed stability might be selecting for some types of features, like structure-related ones, much more than others, such as function-related features.
ER -