Analysis of Incrementally Generated Clusters in Biological Networks Using Graph-Theoretic Filters and Ontology Enrichment

Since the explosive influx of biological data obtained from high-throughput medical instruments, the ability to leverage the currently available data to extract useful knowledge has become one of the most challenging problems in biomedical research. The analysis of such data is particularly complex not only due to its massive size but also due to its heterogeneity and inherent noise associated with several data gathering steps. The utilization of biological networks to model and integrate large-scale heterogeneous biomedical data continues to grow, especially with the systems biology approach taking center stage in many bioinformatics applications. Although loaded with biologically relevant signals, correlation networks do contain noise and are too large for simple data mining tools. In this project, we implement different types of filters to reduce the network size and sort out signals from noise. We propose a new approach for generating various filters that iterate on sub graphs along a spectrum between spanning tree and chordal filters. We show how different network filters incrementally obtain various clusters along this spectrum to maintain structural and domain-relevant components of the original network, while reducing noise. We test the proposed approach using gene expression levels obtained from diabetes and yeast datasets and compare the filtered networks with original networks using ontology enrichment. The obtained results support our main hypothesis that the filters conserve important elements from the original networks while uncovering new biologically significant clusters. However, results analyzing maintained and uncovered biologically significant hubs were inconclusive.