A combined test for feature selection on sparse metaproteomics data-an alternative to missing value imputation.

Area of Science:

• Proteomics
• Bioinformatics
• Statistical Analysis

Background:

• Metaproteomics data analysis faces challenges with high proportions of missing values.
• Imputation methods for missing values rely on assumptions about missingness mechanisms (at random or not at random).

Purpose of the Study:

• To develop and evaluate a novel univariate feature selection method for multi-class metaproteomics comparisons.
• To overcome limitations of imputation-based methods by implicitly handling various missingness mechanisms.

Main Methods:

• A univariate selection method combining a test for association between missingness and classes with a test for differences in observed intensities.
• Quantitative and qualitative comparisons against imputation-based feature selection methods using experimental and simulated data.

Main Results:

• The proposed method demonstrated robust performance across various simulated missingness scenarios, unlike imputation-based methods.
• Feature ranking and selection showed significant divergence among different imputation-based methods on experimental data.
• The combined test correlated reasonably with other methods and maintained efficiency across all tested scenarios.

Conclusions:

• The proposed univariate method offers a more reliable approach to feature selection in metaproteomics, particularly when dealing with complex missing data patterns.
• This method provides a valuable alternative to imputation-based strategies, enhancing the interpretability and reproducibility of metaproteomics studies.