Detecting patient subgroups using reduced set of disease-related markers with iterative pruning Principal Component Analysis (ipPCA)

Genetic markers such as Single Nucleotide Polymorphisms (SNPs) can be used to find subgroups of populations or patients with carefully selected clustering algorithms. The iterative pruning principal component analysis (ipPCA) has been shown to be a powerful tool to identify fine substructures within general populations based on SNP profiles. Usually, SNPs contributing to such profiles have passed rigorous quality control procedures, similar to the ones used for GWAs. Alternatively, attention is restricted to a smaller subset such as PCA-correlated SNPs.

Here, we applied ipPCA on real-life data consisting of the 163 known inflammatory-bowel disease (IBD) associated loci in 13,400 healthy individuals and 29,500 IBD (16,902 Crohn’s disease (CD), and 12,598 ulcerative colitis (UC)) patients from the IIBDGC. Prior to clustering by ipPCA, in each group separately, we regressed out the first five Principal Components (PCs) that were computed from a filtered panel of genome-wide SNPs, to account for general population strata. Next, we applied ipPCA on the healthy group, to learn about the presence of a population-specific partitioning in controls. Then we performed three subphenotype analyses: CD only, UC only and the combined group of CD and UC patients (IBD). For each patient subgroup analysis and for the ipPCA analysis on controls, we highlighted and compared the key SNP drivers.

CD patients could be molecularly reclassified in two groups, and similar for UC patients. The combined patient group could be subdivided in four groups. Finally, we compared demographic and clinical features among the different groups and looked for meaningful characterizations of adjusted patient clusters by performing pathway analysis on driver genes.