Authors
Fu, D., Patil, A., Kush, U., Taheriyoun, A. R., Rahnavard, A.
Abstract
Diverse existing approaches and analyses of large-scale biomedical omics datasets along with the complexity of the corresponding statistical analytical workflows pose challenges to the direct application of large language models (LLMs) to data analysis tasks. We present statLens, a model-agnostic AI-driven statistical intelligence framework that integrates LLMs, code and human review. We demonstrate statLens using differential expression analysis across heterogeneous biomedical data as a use case. On 102 benchmark tasks from 74 datasets across six omics modalities, statLens maintained consistently high accuracy (92.5--95.0%) across seven language models and outperformed LLM-only inference (17.5--46.8%), whereas autonomous-agent accuracy varied widely (58.7--99.5%). We also showed that statLens substantially improved the accuracy of the four small language models (92.5--94.4%) over autonomous agents (58.7--92.2%) and LLM-only inference (17.5--43.5%). By decomposing analytical workflows into complementary tasks for code, LLMs and human experts, statLens provides an integrated framework for more accurate, interpretable and reproducible AI-assisted data analysis, particularly in complex omics settings.
Preprint server:
bioRxiv
The authors list and abstract were imported from bioRxiv on 18 Nov 2026.
Advertisement
Stats
- Recommendations n/a n/a positive of 0 vote(s)
- Views 6
- Comments 0