ESTIMATION AND INFERENCE IN METABOLOMICS WITH NON-RANDOM MISSING DATA AND LATENT FACTORS

Area of Science:

• Biostatistics
• Bioinformatics
• Metabolomics

Background:

• High-throughput metabolomics data frequently contain missing observations and are influenced by unobserved factors.
• Existing analytical methods often fail to address both missing data and latent factors concurrently, potentially compromising results.

Purpose of the Study:

• To introduce MetabMiss, a novel statistical method designed to handle both non-random missing data and unobserved latent factors in high-throughput metabolomics datasets.
• To provide a computationally tractable approach for investigating the complex relationships between the metabolome and multiple phenotypes.

Main Methods:

• Developed MetabMiss, a statistically rigorous methodology.
• The method does not necessitate the specification of a likelihood function for missing data.
• Designed for computational efficiency when analyzing large-scale metabolomics data.

Main Results:

• Demonstrated the accuracy of MetabMiss using both simulated and real-world metabolomics data.
• Provided theoretical proof of the method's asymptotic correctness with increasing sample size and number of metabolites.

Conclusions:

• MetabMiss offers a robust solution for analyzing metabolomics data with missing observations and latent factors.
• The method enhances the reliability of estimators and facilitates the exploration of metabolome-phenotype associations.