Uncovering hidden gene-trait patterns
through biclustering analysis of the UK Biobank

Milton Pividori, Suraju Sadeeq, Arjun Krishnan, Barbara E. Stranger, Christopher R. Gignoux

Department of Biomedical Informatics, University of Colorado Anschutz Medical Campus

Visualization of biclusters relationships

For this example, we binarized the z-scores matrix using a threshold of 10 (corresponding to a p-value of 1.5E-23). From all bicluster pairs having in common more than 20% of the cells, we only kept the larger one. In this case we searched for the trait "Medication: insulin", where five biclusters were identified and plotted below. For each bicluster, genes are shown with slightly larger and blue circles, and traits with smaller circles of different colors according to their type: orange for diseases, green for medications, red for haematological assays and purple for other phenotypes. According to our definition, all traits and genes belonging to the same bicluster are associated with z-score >10, and here we used lines to connect the same traits/genes that are also present in other biclusters, and hovering the mouse over a gene/trait draws a red line to easily identify in which other bicluster that trait/gene is also present.