The association estimators' effect on the integration of proteomic and gene expression data for gene network inference

In this thesis, the effects of association estimators, which have a significant influence on gene network inference methods and used to determine molecular interactions, on the integration of different biological data types were examined. Gene expression and proteomic data for all cancer types used in this thesis were provided from The Cancer Proteome Atlas (TCPA). Firstly, the effect of the correlation-based association estimators on the analysis of proteomic data from sixteen different cancer types was examined by using Gene Network Inference (GAI) methods that are frequently used in the literature. Furthermore, attempts were made to detect the hub genes/proteins in the gene-gene/protein-protein interaction subnetworks associated with the disease by using proteomic data from five different cancer types, which are commonly seen according to American Cancer Society data. During this process, the mutual information (MI) and correlation based nine association estimators, which are commonly used in the literature, were compared. The disease-gene association integration platform (DisGeNET) and the Molecular Signature Database (MSigDB) were used as the gold standard for measuring the performance of the association estimators. The disease-associated pathways were compared with the as-generated co-expression networks and the Fisher's exact test was used to assess the association estimators' performance. Based on the Spearman and Pearson correlation approaches used for the estimation of regulatory networks in the weighted correlation network analysis (WGCNA), the MI-based association estimators' performance was observed to be higher. The best average success rate for five cancer types is 60% for the correlation-based methods, while for the MI-based methods it is 71% for James-Stein Shrinkage (Shrink), and 64% for Schurmann-Grassberger (SG). Integration of the inferred networks was then conducted by using the gene expression and proteomic data. Finally, for each cancer type, hub genes and inferred subnets are presented for the investigations of researchers and biologists.