Neither random nor censored: estimating intensity-dependent probabilities for missing values in label-free proteomics

Area of Science:

• Proteomics
• Biomedical Research
• Statistical Modeling

Background:

• Mass spectrometry proteomics is vital for biomedical research.
• Missing values in peptide quantification limit its utility.
• Distinguishing missing data types (MCAR, MAR, MNAR) is crucial.

Purpose of the Study:

• To develop statistical models for estimating peptide detection probabilities.
• To assess the impact of missing values on statistical information recovery.
• To improve differential expression analysis in proteomics.

Main Methods:

• Proposed statistical models and algorithms to estimate detection probabilities.
• Modeled detection probability as a logit-linear function of intensity.
• Developed a probability model to infer unobserved intensities.
• Incorporated the detection probability model into a likelihood-based differential expression approach.

Main Results:

• Missing value process is intermediate between MAR and censoring.
• Detection probability approaches 100% for high intensities, indicating few intensity-unrelated missing values.
• The proposed model recovers statistical power for differential expression analysis.
• Imputation methods showed poor performance, reducing power or increasing false discovery rates.

Conclusions:

• The developed probability model accurately estimates peptide detection probabilities in mass spectrometry proteomics.
• This approach enhances statistical power in differential expression analysis compared to traditional methods and imputation.
• The findings provide a robust framework for handling missing data in proteomics.