Joint co-clustering: Co-clustering of genomic and clinical bioimaging data

For better understanding the genetic mechanisms underlying clinical observations, and better defining a group of potential candidates for protein-family-inhibiting therapy, it is interesting to determine the correlations between genomic, clinical data and data coming from high resolution and fluorescent microscopy. We introduce a computational method, called joint co-clustering, that can find co-clusters or groups of genes, bioimaging parameters and clinical traits that are believed to be closely related to each other based on the given empirical information. As bioimaging parameters, we quantify the expression of growth factor receptor EGFR/erb-B family in non-small cell lung carcinoma (NSCLC) through a fully-automated computer-aided analysis approach. This immunohistochemical analysis is usually performed by pathologists via visual inspection of tissue samples images. Our fully-automated techniques streamlines this error-prone and time-consuming process, thereby facilitating analysis and diagnosis. Experimental results for several real-life datasets demonstrate the high quantitative precision of our approach. The joint co-clustering method was tested with the receptor EGFR/erb-B family data on non-small cell lung carcinoma (NSCLC) tissue and identified statistically significant co-clusters of genes, receptor protein expression and clinical traits. The validation of our results with the literature suggest that the proposed method can provide biologically meaningful co-clusters of genes and traits and that it is a very promising approach to analyse large-scale biological data and to study multi-factorial genetic pathologies through their genetic alterations.