Quantifying epistasis between two sets of signaling pathway genes by canonical correlation analysis

Abstract

This paper focuses on three aspects related to the conceptualization and application of canonical correlation analysis (CCA) as a statistical model in genetic study: 1) partial canonical correlation analysis to control the effect of population stratification, 2) epistasis analysis for case control study and 3) backward or stepwise canonical analysis to provide more insight into the dynamics of the complex genetic network. We applied these methods to simulated data and to a genetic pathway study of Hirschsprung disease. The results suggest the DLL3 gene from the NOTCH pathways interacted with the PTCH1 gene from the SHH pathway. The first canonical correlation is 0.444 (p-value from the Wilks Lamda test is 8.40E-06) among the cases and 0.157 (p-value 0.236) among the control. The difference between the correlations is significant (p-value 0.00011), after Fisher z transformation.