Work

Asking the Data: Conversational Interrogation of LC–MS/MS Proteomics Data Sets with an LLM Agent

Year: 2026

Type: preprint

Abstract: ABSTRACT Proteomics workflows are designed to answer predefined questions, leaving analytical problems that emerge after first results difficult to investigate without additional bioinformatics work. We introduce data set interrogation, a conversational framework in which a researcher uses a large language model (LLM) agent to translate emerging questions into executable tests, check the answers against controls, and keep the code, inputs, and decisions in a dated record. Applied retrospectively to four LC–MS/MS data sets, the framework exposed problems that changed the interpretation of three in-house studies: a search database lacking one of two symbionts, a swapped pair of tandem mass tag (TMT) channels, statistics inflated by non-host proteins, and false deamidation assignments caused by selection of the 13 C isotope peak as monoisotopic; the same question asked of every data set revealed failed or partial cysteine alkylation in all three. A public TMT data set was reanalyzed from raw files in 24 min, a feasibility measure without a human comparison; among spectra identified by both workflows, 99.2% received the same peptide assignment as in Proteome Discoverer, and protein fold changes correlated (r = 0.93). In seeded tests, a purpose-built channel-assignment check detected all 216 between-strain channel swaps and none of 90 within-strain swaps. The framework couples natural-language interaction with executable analyses, explicit controls, and traceable decisions, enabling researchers to investigate questions not anticipated by the original pipeline. These results establish feasibility in a limited retrospective setting with one researcher; broader evaluation across users, models, and data sets is required to determine general performance.

Source: bioRxiv (Cold Spring Harbor Laboratory)

DOI: 10.64898/2026.10.22.760949