Accounting for multiple imputation-induced variability for differential analysis in mass spectrometry-based...

Area of Science:

• Proteomics
• Bioinformatics
• Statistical analysis

Background:

• Label-free quantitative proteomics often involves imputing missing values.
• Current imputation methods may not adequately account for the uncertainty introduced by imputation.
• This can lead to biased downstream analyses and inaccurate variability estimation.

Purpose of the Study:

• To develop a rigorous multiple imputation strategy for label-free quantitative proteomics.
• To improve the estimation of parameter variability by accounting for imputation uncertainty.
• To enhance the accuracy of differential analysis in proteomics datasets.

Main Methods:

• A multiple imputation strategy using Rubin's rules for unbiased variability estimation.
• Bayesian hierarchical models to moderate the imputation-based variance estimator.
• Integration of the moderated estimator into moderated t-test statistics for differential analysis.
• A workflow (mi4p) applicable at both peptide and protein levels, including a protein-level aggregation step.

Main Results:

• The mi4p methodology was compared against the limma workflow in the DAPAR R package.
• Evaluations were performed on both simulated and real-world proteomics datasets.
• mi4p demonstrated superior overall performance, particularly in terms of F-Score, compared to DAPAR.
• A trade-off between sensitivity and specificity was observed, characteristic of differential analysis.

Conclusions:

• The mi4p workflow provides a more accurate and less biased approach to differential analysis in label-free quantitative proteomics.
• Accounting for imputation uncertainty is crucial for reliable downstream statistical inference.
• mi4p offers improved performance over existing state-of-the-art methods for proteomics data analysis.