Gaussian Mixture Modeling Extensions for Improved False Discovery Rate Estimation in GC-MS Metabolomics

Area of Science:

• Metabolomics
• Analytical Chemistry
• Bioinformatics

Background:

• Reliable small molecule identification is essential for metabolomics research.
• Gas chromatography-mass spectrometry (GC-MS) is a common technique for metabolite identification.
• Current GC-MS workflows lack methods to quantify identification error rates, leading to unknown risks of false discoveries.

Purpose of the Study:

• To propose and evaluate a model-based framework for estimating the false discovery rate (FDR) in GC-MS based metabolite identification.
• To incorporate similarity scores and experimental data for more accurate FDR estimation.
• To assess the impact of reference library size on FDR estimation accuracy.

Main Methods:

• Developed a model-based framework extending traditional mixture modeling to estimate FDR.
• Incorporated both spectral similarity scores and experimental metadata into the FDR estimation.
• Applied the framework to identification lists from 548 diverse samples and compared performance against Gaussian Mixture Models (GMM).
• Conducted simulations to evaluate the effect of reference library size on FDR estimation accuracy.

Main Results:

• The proposed framework demonstrated significant improvements in FDR estimation compared to GMM.
• Median absolute estimation error (MAE) decreased by 12% to 70% across various sample types and complexities.
• Performance improvements were generally consistent across different library sizes.
• FDR estimation accuracy decreased with smaller reference compound libraries.

Conclusions:

• The developed model-based framework provides a robust method for quantifying identification error rates in GC-MS metabolomics.
• This approach reduces the risk of false metabolite identifications, enhancing the reliability of metabolomics studies.
• Accurate FDR estimation is crucial, and performance is influenced by the comprehensiveness of reference libraries.