A Web-based Software System Utilizing Consensus Networks to Infer Gene Interactions

摘要

In this dissertation, various methods from the biology and computer science disciplines were integrated to create a web-based software system with which gene interactions can be inferred, hypothesized, and visualized. Characterizing how genes interact with one another in a biological system is a difficult and complex process. Choosing which data to gather, gathering data, correctly interpreting data, and representing findings in a readable and unbiased manner can take an excessive amount of time and resources. In addition, with the increased use of biotechnologies such as microarray and next generation sequencing, as well as the growing wealth of knowledge within publications available through PubMed, the amount of data available to infer gene interactions grows rapidly. A software system is needed that utilizes prior published knowledge and experimental data to infer and hypothesize the relationships among genes. In this research, I developed a software system that utilizes data sources in the public domain, the KEGG pathway database, PubMed, and the Genomic Data Commons as prior knowledge for Bayesian inference learning of gene interactions. Integrating these data sources into a single local data hub for use in biological studies, this system creates integrated networks used by biologists to validate their suspected gene interactions and generated novel hypotheses.;This research first focused on developing a software platform through which the data from PubMed and KEGG can be collected, cleaned, formatted, and stored. I developed a parallelized software for Bayesian inference learning in order to create consensus networks from multiple Bayesian networks using the collected data. Through a sequence of computational experiments, I confirmed that with the use of a small set of randomized topologies for the K2 algorithm, comparable consensus Bayesian networks can be created. As a result, large networks can be learned with limited computational power. I showed that consensus networks created using this software with PubMed data can be used to infer known interactions among genes within a biological pathway without relying on the direct input of expert biologists. Additionally, I show that consensus networks exhibit the ability to hypothesize novel interactions among genes. To incorporate the vast amount of next generation sequencing data, I developed a module that collects, processes, and discretizes RNA-Seq gene expression data. This module was used and tested to process lung cancer gene expression data from the Genomic Data Commons in order to create and analyze consensus networks. To facilitate the understanding and interpretation of the constructed consensus networks, I introduced the concept of edge resolution and implemented it in the visualization module. Finally, a software system was developed, combining modules for data acquisition, consensus network creation, and network visualization for multiple studies simultaneously.