2016/2017 QMeHSS Workshop (10/21/2016)

The October 21st, 2016 Workshop on Quantitative Methods in Education, Health, and the Social Sciences (QMeHSS) will take place from 10:30am - 12:00pm in NORC Conference room 344. Jiebiao Wang from the University of Chicago will be leading this workshop. We hope to see you there for another session of stimulating discussion. NORC is located at 1155 E. 60th Street.

A High-Dimensional Multivariate Selection Model for Proteomics Data with Batch-Level Missingness

Jiebiao Wang

PhD Student, Public Health Sciences, University of Chicago

 

Abstract:

In quantitative proteomics, mass tag labelling techniques, such as isobaric tags for relative and absolute quantitation (iTRAQ), have been widely adopted in mass spectrometry experiments. These techniques allow peptides/proteins from multiple samples of a batch to be quantified in a single experiment, but they come at a cost of severe batch effects and nonignorable missing data occurring at the batch level. Motivated by the iTRAQ proteomics data, we develop a multivariate selection model to jointly analyze multiple peptides of each protein, considering batch effects and batch-level missingness. To facilitate the computation for high-dimensional outcomes, we employ a log link in the missing-data model and introduce an item response theory (IRT) -type random effect structure that reduces the number of variance components to the dimension of outcomes. We use an expectation-maximization (EM) algorithm for parameter estimation. Simulations demonstrate the advantages of the proposed method in reducing estimation bias, controlling type I error rate and improving power in testing, compared to conventional methods. We apply the proposed method to the iTRAQ-based proteomics data from the Clinical Proteomic Tumor Analysis Consortium and identify differentially expressed genes in triple negative breast cancer. The proposed method and algorithm can also be applied to general multivariate analyses based on clustered data with outcome-dependent missingness. This is joint work with Lin Chen, Pei Wang and Donald Hedeker