Biological reasoning models often ignore their biological inputs in tasks
When Do Biological Reasoning Models Use Their Biological Inputs?
Machine Learning
Summary
Some computer models try to solve biology problems by combining information from large language models and biology-specific data. This study tested six such models on tasks about DNA, proteins, and cells to see if they really use their biological information. The researchers found that some models mostly rely on language text rather than the biology data, even when that biology data contains useful info. Other models do use biological inputs more to perform better. Overall, just adding biological data to these models doesn’t guarantee they use it to improve their answers.
What this means in practice
- •For biotech software developers: Avoid overestimating biological input impact when building reasoning tools that combine language and biological models to improve reliability.
- •For bioinformatics platform teams: Design diagnostic tools by prioritizing text inputs or foundation models selectively, as biological data may not always enhance model predictions.
Authors
Ada Fang, Nikitha Thoduguli, Lukas Fesser, Hanlin Zhang, Sham M. Kakade, Marinka Zitnik
Abstract
Biological reasoning models use post-training to connect LLMs to biological foundation model representations and biological text. Their benchmark accuracy is taken as evidence that LLMs reason over these inputs. We test this assumption in six biological reasoning models across DNA, protein, and single-cell tasks. We perturb one biological input while holding the query and other inputs fixed, construct evidence conflicts that pair the foundation model representation of one genome, protein, or cell with the text of another, fit linear probes to the representations the language model receives, and analyze reasoning traces against the biological inputs. Evo2 and ESM3 contribute little to BioReason and BioReason-Pro performance on the evaluated tasks. Shuffling the DNA sequence barely changes BioReason disease prediction accuracy, and in evidence conflicts the two models follow the text in 97.9% and 99.7% of cases. Linear probes trained on the Evo2 and ESM3 representations predict the task targets, so these foundation models encode information relevant to the task, but provide limited overall performance improvement to BioReason and BioReason-Pro. In contrast, foundation model inputs contribute to ChatNT, Prot2Text-V2, and CellWhisperer performance, and differentially expressed genes in the gene sentence contribute to Cell2Sentence-Scale performance. Across SFT and RL checkpoints of BioReason-Pro and 42 BioReason checkpoints, increases in accuracy do not imply greater performance contributions from biological inputs. BioReason traces misstate nucleotide changes, while BioReason-Pro traces describe functions omitted from final predictions under evidence conflicts. We find that current post-training strategies do not ensure that foundation model representations contribute to task performance.