Kernel weighted least square approach for imputing missing values of metabolomics data.

Area of Science:

• Analytical Chemistry
• Bioinformatics
• Systems Biology

Background:

• Mass spectrometry is a high-throughput technique for large-scale metabolomic analyses.
• Metabolomic data matrices often contain missing values and outliers from technical or biological sources.
• Existing imputation methods fail to address outliers, compromising data accuracy.

Purpose of the Study:

• To develop a novel missing data imputation technique that simultaneously handles missing values and outliers.
• To evaluate the performance of the proposed method against conventional and recent imputation techniques.

Main Methods:

• Development of a kernel weight function-based imputation method.
• Evaluation using artificially generated and experimentally measured metabolomics data.
• Comparison in the presence and absence of varying rates of outliers.

Main Results:

• The proposed kernel weight-based method effectively resolves both missing values and outliers.
• Performance evaluations demonstrate the superiority of the new technique over existing alternatives.
• Validation on both artificial and real-world metabolomics datasets confirms its effectiveness.

Conclusions:

• The kernel weight-based imputation technique offers a superior solution for metabolomics data with missing values and outliers.
• This method enhances the accuracy and reliability of metabolomic data analysis.
• An R package is available for user convenience, facilitating the adoption of this advanced imputation technique.