The Midpoint Mixed Model with a Missingness Mechanism (M5): A Likelihood-Based Framework for Quantification of Mass Spectrometry Proteomics Data (Preprint)

TL;DR

This paper introduces a novel likelihood-based statistical model for proteomics data that effectively accounts for nonignorable missingness and peptide-level matched pairs, significantly improving protein quantification accuracy.

Contribution

It presents the first model to incorporate nonignorable missing data mechanisms while utilizing peptide-level matched pairs in proteomics analysis.

Findings

Model reduces mean squared error by 35% compared to median ratio estimates.

Simulation shows basic ANOVA estimates have 371% higher MSE than median ratio estimates.

Application to breast cancer data increases estimated proteins by 22%.

Abstract

Statistical models for proteomics data often estimate protein fold changes between two samples, A and B, as the average peptide intensity from sample A divided by the average peptide intensity from sample B. Such average intensity ratios fail to take full advantage of the experimental design which eliminates unseen confounding variables by processing peptides from both samples under identical conditions. Typically this structure is exploited through the estimation of a protein ratio as the median ratio of matched peptide intensities. This simple solution fails to account for a substantial missing data bias which has led to the development of more sophisticated average intensity models. Here we develop the first statistical model that accounts for nonignorable missingness while utilizing peptide level matched pairs across samples. Our simulation analysis shows that models which fail to utilize peptide level ratios, su.er astonishing losses to accuracy with basic ANOVA estimates having an average MSE 371% higher than median ratio estimates. In turn, median ratio estimates have an average MSE 35% higher than our model estimates. An analysis of breast cancer data reinforces these relationships and shows that our model is capable of increasing the number of proteins estimated by 22%.