The Midpoint Mixed Model with a Missingness Mechanism: A Likelihood-Based Framework for Relative Quantification Mass Spectrometry Experiments

• Activity Number: 139
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Jonathon O'Brien* and Bahjat Qaqish and Harsha Gunawardena and Joseph Ibrahim
• Companies: and The University of North Carolina at Chapel Hill and The University of North Carolina and The University of North Carolina
• Keywords: Proteomics ; Missing Data ; Random Effects ; Gibbs Sampler ; Skew Normal

Abstract

A key component of a proteomics experiment is the estimation of relative protein abundance from peptide level measurements. Two statistical features of this inferential step are matched pairs data and non-ignorable missingness. Software often estimates proteins using a complete case analysis of peptide ratios. While missing data models usually fail to match peptides across samples. Here we develop the first statistical model that accounts for both critical features. Our simulation analysis shows that models based on average intensity suffer large losses to accuracy with basic ANOVA estimates having an average MSE 371% higher than median ratio estimates. In turn the method of medians, which ignores missing data, has an average MSE 35% higher than estimates from our M5 model. Analysis of tumor data reinforces these relationships and shows how our M5 model improves depth of discovery by enabling a 22% increase in the number of proteins estimated. Our assessment of models based on average intensity suggest that they should be categorically avoided. When compared with the method of medians, our model provides an alternative with improved accuracy and enhanced depth of discovery.