Improved Metabolite Prediction Using Microbiome Data-Based Elastic Net Models.

Area of Science:

• Microbiome research
• Metabolomics
• Bioinformatics

Background:

• Microbiome data is abundant, but matched metabolomics data is scarce, hindering biological activity insights.
• Existing metabolomics profiling methods like HUMAnN have limitations in accuracy and predicting individual metabolites.

Purpose of the Study:

• To develop and evaluate a novel method (ENVIM) for predicting metabolites from microbiome data.
• To address the gap in accurately inferring metabolomics from microbial taxonomy or metagenomics.

Main Methods:

• Developed ENVIM, a novel metabolite prediction method based on the elastic net model (ENM) with added variable importance (VI) scores.
• Applied and evaluated ENVIM using metagenomic and metatranscriptomic data from oral and gut microbiome studies.
• Compared ENVIM's performance against the standard ENM strategy used in MelonnPan.

Main Results:

• ENVIM demonstrated superior metabolite predictive accuracy compared to MelonnPan across oral and gut microbiome datasets.
• Metabolite prediction was more accurate in gut microbiome settings than in oral microbiome settings.
• Identified well-predicted metabolites, including trehalose, maltose, stachyose, and ribose, in the supragingival microbiome.

Conclusions:

• The ENVIM method offers improved metabolite prediction accuracy from microbiome data, particularly when using metatranscriptomics.
• ENVIM enhances the understanding of microbial and host biological activities by enabling metabolite inference.
• This tool is valuable for multi-omics studies with limited or missing metabolomics data.