Scalable Incremental Clustering for Tandem Mass Spectra in Untargeted Metabolomics

Area of Science:

• Analytical Chemistry
• Computational Biology
• Metabolomics

Background:

• Tandem mass spectrometry (MS/MS) is crucial for untargeted metabolomics, generating vast amounts of spectral data.
• Current clustering methods face computational and memory limitations, hindering analysis of large-scale and continuously growing MS/MS datasets.
• These limitations impact long-term studies and public data repositories, necessitating scalable solutions.

Purpose of the Study:

• To present a novel, scalable clustering framework for processing large-scale MS/MS metabolomics data.
• To address the computational and memory bottlenecks associated with existing clustering algorithms.
• To enable efficient analysis of continuously growing MS/MS datasets in long-term studies and public repositories.

Main Methods:

• Developed an incremental clustering framework designed for MS/MS metabolomics data.
• Implemented a novel spectrum pooling strategy to maintain clustering performance by propagating local density structure across data batches.
• Evaluated performance using database-search-based metrics on proteomics datasets and the MS1-retention time (MS-RT) method on metabolomics datasets.

Main Results:

• The incremental clustering approach achieved performance comparable to state-of-the-art methods in terms of cluster purity and completeness.
• Demonstrated scalability by successfully clustering 368 million spectra into millions of clusters within 10,000 CPU hours.
• Traditional methods failed to complete this task due to excessive memory or time requirements, highlighting the scalability advantage.

Conclusions:

• The proposed scalable clustering framework offers a practical solution for analyzing large-scale, continuously growing MS/MS metabolomics datasets.
• The method is suitable for integration into public metabolomics platforms like GNPS2, enhancing data accessibility and analysis capabilities.
• This advancement overcomes critical limitations in current metabolomics data processing, enabling more comprehensive downstream analyses.