Simulating Tandem Mass Spectra for Small Molecules using a General-Purpose Large-Language Model.

Area of Science:

• Computational chemistry
• Analytical chemistry
• Bioinformatics

Background:

• Mass spectrometry is crucial for identifying compounds in biological samples.
• Existing methods for spectral library generation often require extensive domain-specific training.
• The Blood Exposome Database contains valuable compound information.

Purpose of the Study:

• To demonstrate a practical application of the Google Gemini large language model (LLM) for simulating tandem mass spectra.
• To assess the feasibility of using a general-purpose LLM for spectral simulation without domain-specific training.
• To explore the potential of LLMs in expanding in-silico spectral libraries for metabolomics and exposomics.

Main Methods:

• Utilized the Google Gemini LLM to simulate tandem mass spectra for compounds present in the Blood Exposome Database.
• Evaluated the model's performance in generating chemically relevant fragmentation patterns.
• Did not employ any domain-specific model training for chemical fragmentation.

Main Results:

• The Google Gemini LLM successfully simulated tandem mass spectra for compounds from the Blood Exposome Database.
• The results suggest that chemical fragmentation knowledge is implicitly encoded within the Gemini model.
• The approach demonstrated the capability of a general-purpose LLM for this task.

Conclusions:

• General-purpose LLMs like Google Gemini offer a practical and accessible method for simulating mass spectra.
• This approach can significantly expand in-silico spectral libraries.
• LLMs have the potential to accelerate compound annotation in mass spectrometry-based metabolomics and exposomics research.