Omics data discovery agents: Agent-supported retrieval, reanalysis, and synthesis of published omics data
4 affiliations
- 1Department of Computational Biomedicine, Cedars-Sinai Medical Center, Los Angeles, California, United States of America.
- 2Advanced Clinical Biosystems Research Institute, Cedars-Sinai Medical Center, Los Angeles, California, United States of America.
- 3Board of Governors Innovation Center, Cedars-Sinai Medical Center, Los Angeles, California, United States of America.
- 4Smidt Heart Institute, Cedars-Sinai Medical Center, Los Angeles, California, United States of America.
Abstract
The biomedical literature contains a vast collection of omics studies, yet most published data remain functionally inaccessible for computational reuse. When raw data are deposited in public repositories, essential information for reproducing reported results is dispersed across main text, supplementary files, and code repositories, and in the rarer cases where intermediate data (e.g., protein abundance files) are shared, their location is irregular. Here we present an agentic framework for the agent-supported retrieval, reanalysis, and synthesis of published omics data. The system employs large language model (LLM) agents with access to tools for fetching omics studies, extracting article metadata, identifying and downloading published data, executing containerized quantification pipelines, and synthesizing results across studies. Applied at corpus scale, the pipeline catalogued dataset references across thousands of PubMed Central articles; we report these as descriptive system outputs rather than as a validated measure of extraction accuracy. Using model context protocol (MCP) servers to expose containerized analysis tools, the agents retrieved and re-quantified data in five end-to-end reanalyses spanning data-dependent and data-independent proteomics and bulk RNA-seq. All five reanalyses completed, each with documented human guidance and workflow accommodations, and reproduced the authors’ deposited abundances with high per-sample correlation (0.85–0.997) and strongly concordant differentially expressed features (fold-change Spearman 0.88–0.91), with no direction reversals among features called differentially expressed in both analyses; residual differences in significant-feature lists were attributable to threshold placement, tool-version, and preprocessing differences rather than to the underlying quantities. We further demonstrate that agents can identify semantically similar studies, judge data compatibility, and synthesize findings across studies, including a random-effects meta-analysis that recovered consistent protein regulation in liver fibrosis. Rather than a validated benchmark of literature-wide performance, this work is a feasibility demonstration together with an auditable, reusable toolset, establishing a foundation for prospective evaluation of automated omics-data reuse. Author summary: Every year, tens of thousands of biological studies measure the abundance of thousands of proteins or genes at once, so-called omics experiments, and deposit the resulting data in public archives. In principle these data can be reused to answer new questions or to check whether published findings hold up. In practice, reuse is rare: the information needed to make sense of a dataset, such as which samples are which, how the data were processed, and with which software, is scattered across the article, its supplementary files, and code repositories, and the raw data often must be reprocessed with specialized tools before it can be compared. We built ODDA, a system in which artificial-intelligence language-model agents read the scientific literature, locate the associated data, and reprocess it using the same kinds of containerized tools the original authors used, all through a controlled, auditable interface. Across five studies spanning proteomics and RNA sequencing, ODDA reproduced the authors’ published measurements closely and combined results across studies. We share the software so that others can build on, and rigorously test, automated data reuse.
Related
- Bridging MS2 Spectra and Chemical Space: Advances in Spectral Similarity, Molecular Retrieval, and De Novo Structure Discovery
- Information-Dense Synthesis for Molecular Discovery
- Quantitative reanalysis of classic pigeon homing experiments rejects a liver-based magnetic compass based on superparamagnetism
- A Shared Epithelial Barrier Program Across Mucocutaneous Lichen Planus: A Compartment-Aware Multi-Cohort Transcriptomic Reanalysis
- Gastric involvement in multiple myeloma-associated AL amyloidosis: a case with serial endoscopy and a pooled analysis of 30 published cases.
BibTeX
@article{hutton2026omics,
title = {{Omics data discovery agents: Agent-supported retrieval, reanalysis, and synthesis of published omics data}},
author = {Alexandre Hutton and Jesse G Meyer},
journal = {PLOS Computational Biology},
year = {2026},
month = oct,
doi = {10.1371/journal.pcbi.1014822},
url = {https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1014822},
note = {CC BY},
}
RIS
TY - JOUR
TI - Omics data discovery agents: Agent-supported retrieval, reanalysis, and synthesis of published omics data
AU - Alexandre Hutton
AU - Jesse G Meyer
PY - 2026
DA - 2026/10/06
JO - PLOS Computational Biology
DO - 10.1371/journal.pcbi.1014822
UR - https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1014822
AB - The biomedical literature contains a vast collection of omics studies, yet most published data remain functionally inaccessible for computational reuse. When raw data are deposited in public repositories, essential information for reproducing reported results is dispersed across main text, supplementary files, and code repositories, and in the rarer cases where intermediate data (e.g., protein abundance files) are shared, their location is irregular. Here we present an agentic framework for the agent-supported retrieval, reanalysis, and synthesis of published omics data. The system employs large language model (LLM) agents with access to tools for fetching omics studies, extracting article metadata, identifying and downloading published data, executing containerized quantification pipelines, and synthesizing results across studies. Applied at corpus scale, the pipeline catalogued dataset references across thousands of PubMed Central articles; we report these as descriptive system outputs rather than as a validated measure of extraction accuracy. Using model context protocol (MCP) servers to expose containerized analysis tools, the agents retrieved and re-quantified data in five end-to-end reanalyses spanning data-dependent and data-independent proteomics and bulk RNA-seq. All five reanalyses completed, each with documented human guidance and workflow accommodations, and reproduced the authors’ deposited abundances with high per-sample correlation (0.85–0.997) and strongly concordant differentially expressed features (fold-change Spearman 0.88–0.91), with no direction reversals among features called differentially expressed in both analyses; residual differences in significant-feature lists were attributable to threshold placement, tool-version, and preprocessing differences rather than to the underlying quantities. We further demonstrate that agents can identify semantically similar studies, judge data compatibility, and synthesize findings across studies, including a random-effects meta-analysis that recovered consistent protein regulation in liver fibrosis. Rather than a validated benchmark of literature-wide performance, this work is a feasibility demonstration together with an auditable, reusable toolset, establishing a foundation for prospective evaluation of automated omics-data reuse. Author summary: Every year, tens of thousands of biological studies measure the abundance of thousands of proteins or genes at once, so-called omics experiments, and deposit the resulting data in public archives. In principle these data can be reused to answer new questions or to check whether published findings hold up. In practice, reuse is rare: the information needed to make sense of a dataset, such as which samples are which, how the data were processed, and with which software, is scattered across the article, its supplementary files, and code repositories, and the raw data often must be reprocessed with specialized tools before it can be compared. We built ODDA, a system in which artificial-intelligence language-model agents read the scientific literature, locate the associated data, and reprocess it using the same kinds of containerized tools the original authors used, all through a controlled, auditable interface. Across five studies spanning proteomics and RNA sequencing, ODDA reproduced the authors’ published measurements closely and combined results across studies. We share the software so that others can build on, and rigorously test, automated data reuse.
SN - 1553-7358
SN - 1553-734X
ER -