High-Dimensional Factor Analysis with Sparse Procrustean Rotation for Gene Discovery and Genetic Risk Assessment

• Activity Number: 45
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Randy Carter*+ and Netsanet Imam
• Companies: University At Buffalo and Virginia Bioinformatics Institute
• Keywords: high dimensional data ; partial Procrustean target ; sparse Procrustean objective function ; Gauss-Newton optimization ; cilia ; ciliopathic genes

Abstract

Gene discovery studies often produce high dimensional data (p>n). For example, we are interested in the expression of 1781 genes localized in the primary cilia of retinal cells from 120 rats. 52 of these genes are known to be associated with cilia dysfunction. Our goal is to discover candidate ciliopathic genes for further study. Assuming the existence of a latent construct that is associated with ciliopathy, a high dimensional factor analysis (HDFA) was developed and was applied for initial factor extraction. Varimax rotation failed to produce a rotated factor with the desired interpretation. Sparse Procrustean rotation (SPR) was developed to find a rotated structure with a factor on which many of the 52 characteristic genes load most highly. 35 of the 52 (67%) loaded most highly on a factor (f1) identified by SPR. Furthermore, 13/14 IFT genes, which are thought to interact with several of the characteristic genes to perform critical cilia function, also loaded most highly on f1. 608 other gene expression variables loaded most highly on f1 and are candidates for further investigation. HDFA and SPR, together, are promising methods for gene discovery and genetic risk assessment.