中央研究院 資訊科學研究所

In the past several decades, transcriptomics analysis has become prevalent in biology and medical science. Differential expression approach is the most frequently performed and well-known approach to analyze trnascriptomics data and has been successful in identifying disease genes. However, information provided by transcriptomics is far beyond differential expression analysis can do when incorporating with other biological or medical data. Moreover, after investigation of differential expression, scientists realized that disease are more complicated and not only associated with a few differentially expressed genes. Therefore, integrative network analysis becomes another approach to investigate mechanisms of cellular processes or disease development. Through integrating transcriptome data derived from multiple cancer genomes, we recently identified a rewired regulatory feedback loop between STAT1 and miR-155-5p, which is consistently activated in cancers and might promote carcinogenesis via regulating cancer immunoediting mechanism; and discovered two miRNA-regulated onco-modules ssociated with poor survival outcomes of cancer patients and involved in mitosis and DNA replication. Additionally, through investigating the human gene regulatory network, we identified another regulatory feedback loop between HNF4A and NR2F2 highly bridging the network. Further analysis of their transcriptome profiles suggested that the disturbance of this feedback loop might protect tumor cells from apoptotic processes. On the other hand, combining gene expression profiles with survival information of patients, we could simulate the experimental processes used to identified essential genes to discover the influential genes to patient survival. We found that the survival influential gene sets were highly heterogeneous between cancers. More importantly, the pan-cancer harmful genes could play essential-like roles in tumor cells and therefore deleterious to patient survival. We also suggested 10 survival influential genes as promising target of cancer therapy. Furthermore, using gene expression profiles across patients as features, we proposed a ensemble learning model to identify candidate genes associated with TNFA-activated pathways. Our approach performs outstandingly (averaged area under curve > 0.9) to predict the promising genes responded to TNFA in 16 cancer types. Moreover, the predicted genes are significantly enriched in differentially expressed genes, oncogenic signatures, and the biological processes responded to TNFA. These results suggest that the genes predicted by our model might play critical roles in cell survival and carcinogenesis through TNFA pathway. However, even though the transcriptomics data has been widely used and studied nowadays, there are still much more applications, such as precision medicine, being waiting us to discover.

【國立陽明交通大學 生物醫學資訊研究所】