AUGMENTED DOUBLY ROBUST POST-IMPUTATION INFERENCE FOR PROTEOMIC DATA.

Area of Science:

• Proteomics
• Bioinformatics
• Statistical Biology

Background:

• Mass spectrometry proteomics generates quantitative data crucial for understanding molecular mechanisms.
• High proportions of missing values in proteomics datasets pose significant analytical challenges.
• Ignoring imputation errors can introduce systematic bias into downstream analyses.

Purpose of the Study:

• To develop a robust statistical framework for valid and efficient inference in proteomics data analysis.
• To address the challenge of missing values in quantitative proteomics.
• To improve the accuracy of downstream analyses by mitigating imputation-induced bias.

Main Methods:

• Proposed a statistical framework inspired by doubly robust estimators.
• Integrated machine learning tools, specifically variational autoencoders, for enhanced imputation quality.
• Employed a parametric model for propensity score estimation to debias imputed outcomes.
• Ensured compatibility with the double machine learning framework, providing provable properties.

Main Results:

• Simulation studies demonstrated empirical superiority over existing procedures.
• The method successfully utilized imputed data for additional discoveries in single-cell and Alzheimer's Disease proteomics.
• Maintained good control of false positives despite leveraging imputed data.

Conclusions:

• The proposed doubly robust framework offers a statistically sound approach for proteomics data analysis.
• This method effectively handles missing data, improving the quality of insights derived from proteomics experiments.
• The framework enables meaningful discoveries while ensuring analytical rigor and reliability.