Evaluating Proteomics Imputation Methods with Improved Criteria.

Area of Science:

• Proteomics
• Quantitative Biology
• Biotechnology

Background:

• Quantitative proteomics experiments using tandem mass spectrometry often yield numerous missing values.
• These missing values impede reproducibility, diminish statistical power, and complicate cross-sample/experiment comparisons.
• Existing imputation methods are frequently used but often perform poorly, with prior evaluations focusing on simplistic error metrics.

Purpose of the Study:

• To evaluate common missing value imputation methods in quantitative proteomics.
• To assess performance using practical, downstream-centric criteria: identifying differentially expressed peptides, generating new quantitative peptides, and improving the peptide lower limit of quantification.
• To compare methods across diverse experimental types and acquisition strategies (data-dependent and data-independent acquisition).

Main Methods:

• Evaluation of imputation methods using downstream-centric criteria.
• Benchmarking against data-dependent and data-independent acquisition data.
• Comparison of performance based on peptide quantification and differential expression analysis.

Main Results:

• Imputation did not consistently improve the identification of differentially expressed peptides.
• Imputation successfully identified new quantitative peptides and enhanced the peptide lower limit of quantification.
• The MissForest method demonstrated superior performance across the evaluated downstream-centric criteria.

Conclusions:

• While imputation can enhance peptide quantification and lower detection limits, its benefit for differential expression analysis is not guaranteed.
• MissForest emerges as a top-performing imputation method for proteomics data based on practical utility.
• There is a critical need for imputation methods that adequately incorporate peptide quantification variance.