In-silico Tandem Mass Spectral Libraries for the Blood Exposome and Metabolomics Workbench Databases: Application of Gemini Large Language Model

Abstract

Human body contains endogenous metabolites, nutrients and xenobiotics chemicals. These chemicals can be detected in a biospecimen using high-resolution mass spectrometry instruments, yielding metabolomic and exposomic datasets. Mass spectral libraries are critical components of these workflows to annotate experimental data with chemical information. In this data descriptor, we report mass spectral libraries for human-relevant small molecules, that are generated using Google Gemini large-language model. Using an optimized chain-of-thought prompt and instructions, we have generated mass spectra for each compound at high/medium/low energy fragmentation level. Each spectrum was reported in the NIST MSP text format for easy to use by metabolomics software. These libraries cover 91,231 unique 2D structures from the Blood Exposome Database and the Metabolomics Workbench data repository. Technical validation was shown by comparing the Gemini-generated spectra against NIST 23 mass spectral database, which demonstrated 55% accuracy. LLMs represent a useful and accessible tool for expanding in-silico spectral libraries to annotate compounds in the HRMS data for human biospecimens and may accelerate the compound annotation in metabolomics and exposomics.

Lifecycle

• biorxiv v1 2026-10-25 source ↗

Discussion

No qualifying discussion yet.