Functional Categorization of Disease Genes Based on Spectral Graph Theory and Integrated Biological Knowledge

Interaction of multiple genetic variants is a major challenge in the development of effective treatment strategies for complex disorders. Identifying the most promising genes enhances the understanding of the underlying mechanisms of the disease, which, in turn leads to better diagnostic and therapeutic predictions. Categorizing the disease genes into meaningful groups even helps in analyzing the correlated phenotypes which will further improve the power of detecting disease-associated variants. Since experimental approaches are time consuming and expensive, computational methods offer an accurate and efficient alternative for analyzing gene-disease associations from vast amount of publicly available genomic information. Integration of biological knowledge encoded in genes are necessary for identifying significant groups of functionally similar genes and for the sufficient biological elucidation of patterns classified by these clusters. The aim of the work is to identify gene clusters by utilizing diverse genomic information instead of using a single class of biological data in isolation and using efficient feature selection methods and edge pruning techniques for performance improvement. An optimized and streamlined procedure is proposed based on spectral clustering for automatic detection of gene communities through a combination of weighted knowledge fusion, threshold-based edge detection and entropy-based eigenvector subset selection. The proposed approach is applied to produce communities of genes related to Autism Spectrum Disorder and is compared with standard clustering solutions.