rMisbeta: A robust missing value imputation approach in transcriptomics and metabolomics data.

Area of Science:

• Bioinformatics
• Computational Biology
• Data Science

Background:

• Transcriptomics and metabolomics data frequently exhibit missing values and outliers, hindering downstream statistical analysis.
• Existing imputation methods often lack robustness against outliers, leading to performance degradation.

Purpose of the Study:

• To develop a robust iterative approach for simultaneous imputation of missing values and handling of outliers.
• To evaluate the proposed method's performance against established imputation techniques.

Main Methods:

• A novel robust iterative approach utilizing minimum beta divergence estimators.
• Comparison with six imputation methods (Zero, KNN, robust SVD, EM, RF, WLSA) using simulated and real datasets.
• Performance evaluation using ten indices including accuracy, sensitivity, specificity, and computational runtime.

Main Results:

• The proposed method demonstrates superior performance in handling various rates of outliers and missing values compared to traditional methods.
• The approach maintains comparable performance to other methods when outliers are absent.
• The method is accurate, simple, and computationally efficient.

Conclusions:

• The developed robust iterative method is highly recommended for large-scale transcriptomics and metabolomics data analysis.
• An R package, rMisbeta, is available for public use, facilitating the application of this technique.