GSimp: A Gibbs sampler based left-censored missing value imputation approach for metabolomics studies

Area of Science:

• Metabolomics
• Bioinformatics
• Statistical Analysis

Background:

• Left-censored missing values are prevalent in targeted metabolomics.
• These values are often missing not at random (MNAR), complicating analysis.
• Existing imputation methods are insufficient for MNAR data in metabolomics.

Purpose of the Study:

• To develop a practical imputation method for left-censored MNAR data in metabolomics.
• To evaluate the performance of the developed method against existing approaches.

Main Methods:

• Developed an iterative Gibbs sampler based imputation approach (GSimp).
• Compared GSimp with three other methods using two real-world and one simulation metabolomics dataset.
• Utilized a dedicated imputation evaluation pipeline for performance assessment.

Main Results:

• GSimp demonstrated superior imputation accuracy compared to other methods.
• The method improved observation distribution and univariate/multivariate analyses.
• GSimp enhanced statistical sensitivity in metabolomics data analysis.
• A parallel version of GSimp was created for large-scale datasets.

Conclusions:

• GSimp is a robust and accurate method for imputing left-censored MNAR values in targeted metabolomics.
• The developed method offers significant advantages over existing techniques for metabolomics data processing.
• GSimp provides a valuable tool for improving the reliability of metabolomics statistical analyses.