Proper imputation of missing values in proteomics datasets for differential expression analysis.

Area of Science:

• Biomedical research
• Proteomics
• Mass spectrometry

Background:

• Label-free shotgun proteomics is crucial for protein identification and quantification in biomedical research.
• Data-dependent acquisition (DDA) mass spectrometry datasets often contain significant missing values (MVs), complicating downstream analysis.
• Existing imputation methods lack validation on real-world proteomic data, and their impact on differential expression analysis is unclear.

Purpose of the Study:

• To characterize the composition of missing values (MVs) in public label-free shotgun proteomics datasets.
• To develop realistic simulated datasets that mimic the MV profiles of real-world data.
• To evaluate the performance of various imputation methods on differential expression analysis using these simulated datasets.

Main Methods:

• Analysis of public DDA proteomics datasets to identify MV patterns.
• Creation of simulated datasets reflecting real-world MV characteristics.
• Comparative assessment of popular imputation techniques on simulated data.
• Evaluation of imputation impact on differential expression analysis.

Main Results:

• Characterization of MV composition in diverse real-life proteomic datasets.
• Development of a novel simulation approach for MVs.
• Demonstration of varying impacts of imputation methods on differential expression results.
• Identification of imputation strategies suitable for different MV profiles.

Conclusions:

• Understanding MV composition in real proteomic data is essential for effective imputation.
• The choice of imputation method significantly influences differential expression analysis outcomes.
• Recommendations are provided for selecting appropriate imputation methods for label-free shotgun proteomics data, extending beyond DDA datasets.