Classification of gene expression data using PCA-based fault detection and identification

This paper introduces a simple and robust method for the classification of significantly expressed genes in high-throughput microarray measurements of a cell's transcriptome. The technique has its origins in PCA-based fault detection and isolation (FDI) systems engineering. PCA-FDI is a data-driven procedure that can be used to isolate gene expression profiles associated with anomalous cell function by projecting target assays onto a 'residual' subspace orthogonal to a set of PCA coordinates extracted from microarray data collected under normative cell conditions. The method is robust to noise and disturbances, and is insensitive to natural variation due to nominal cell functioning. The approach is demonstrated on a sequence of simulated gene regulatory network (GRN) time-series expression profiles.