Repository scale classification and decomposition of tandem mass spectral data

Area of Science:

• Metabolomics
• Bioinformatics
• Computational Biology

Background:

• Existing studies link molecular features to biological phenotypes but are limited to single phenotypes.
• Repository-scale metabolomics data presents challenges due to diverse sample types and data collection standards.
• Current methods are insufficient for analyzing large, heterogeneous metabolomics datasets.

Purpose of the Study:

• To develop MetSummarizer, a novel method for predicting biological phenotypes and raw ingredient composition.
• To demonstrate the utility of aggregating diverse metabolomic datasets for enhanced prediction accuracy.
• To address the challenge of data heterogeneity and standard-specific biases in multi-laboratory metabolomics data.

Main Methods:

• Developed MetSummarizer, a computational method for phenotype and composition prediction.
• Aggregated diverse metabolomic datasets from various sources.
• Implemented a classification step to detect and discard standard-specific features for unbiased results.

Main Results:

• MetSummarizer accurately predicts biological phenotypes of environmental and host-oriented samples.
• High accuracy was achieved in predicting the raw ingredient composition of complex foods.
• Data aggregation significantly improved prediction accuracy compared to single-dataset analyses.

Conclusions:

• MetSummarizer offers a robust solution for analyzing repository-scale metabolomics data.
• The method effectively handles data heterogeneity and standard-specific biases.
• This approach advances the analysis of complex biological mixtures and foodomics data.