Information-Content-Informed Kendall-Tau Correlation Methodology: Interpreting Missing Values in Metabolomics as...

Area of Science:

• Metabolomics
• Bioinformatics
• Statistical Analysis

Background:

• Current correlation measures struggle with missing data, often discarding or imputing it.
• Missing values in metabolomics frequently result from left-censorship (below detection limits), containing valuable information.
• This missingness is not random and represents data at the lower end of the distribution.

Purpose of the Study:

• To develop a novel methodology that incorporates left-censored missing data into correlation analysis.
• To leverage the information contained within missing values in metabolomics datasets.
• To improve the accuracy and interpretability of correlation measures in the presence of missing data.

Main Methods:

• Proposed the information-content-informed Kendall-tau (ICI-Kt) methodology.
• Developed a statistical test to identify left-censored missing values.
• Integrated left-censored values into the Kendall-tau coefficient calculation, adding interpretable information.
• Implemented calculations for theoretical maxima and pairwise completeness.

Main Results:

• Demonstrated that ICI-Kt effectively includes left-censored missing data as interpretable information.
• Showcased improved determination of outlier samples using the ICI-Kt methodology.
• Achieved enhanced feature-feature network construction in metabolomics datasets.
• Validated findings using simulated data and over 700 experimental datasets from Metabolomics Workbench.

Conclusions:

• The ICI-Kt methodology offers a robust approach to handling missing data in metabolomics.
• Provides parallel R and Python implementations for fast, large-scale calculations.
• ICI-Kt methods are available as open-source R package and Python module on GitHub.