Sparse Structure Equation Models and Integer Programming for Joint Imaging and Genomic Data Analysis and Its Application to Kidney Renal Clear Cell Carcinoma

• Activity Number: 359
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Nan Lin* and Panpan Wang and Yun Zhu and Momiao Xiong
• Companies: and The University of Texas Health Science Center and Tulane University and The University of Texas Health Science Center
• Keywords: structure equation ; integer programming ; causal inference ; imaging-genomic analysis ; RNA-seq data ; Computed Tomography

Abstract

The structure of biomedical images of the tissue of human body is strongly shaped by the genetic and expression variation. Imaging genomics uses measure of imaging as endophenotype and attempts to find the associated genomic variants. However, it is a lack of methods for imaging-genomic analysis to test the association of the genomic variants with the image region of interest. To address this limitation, we develop sparse structure equation models (SEMs) for the integrated analysis of the imaging and the genomic data and integer programming (IP) for exact learning of expression causal network. Specifically, we first use super-voxel and cluster analysis techniques to segment the original image volume into a number of regions based on the texture similarity. Then sparse SEMs are used to encode structure of imaging and genomic data and IP is used to infer expression causal network. By applying the models to 38 patients with both the kidney cancer computerized tomography (CT) and RNA-seq data. After dividing the CT images into 34 regions, we found that 11 genes from the Hedgehod signaling pathway relate to the CT images and 6 of the 11 genes point to two tumor sites.