Gene Panel Characteristics Discovery and Analysis by Biomedical Literature Mining

• Main Authors: Chen-RueiLiu, 劉宸睿
• Other Authors: Jung-Hsien Chiang
• Format: Others
• Language: en_US
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/xq3ew6

碩士 === 國立成功大學 === 資訊工程學系 === 106 === A gene panel test is a targeted tumor sequencing test in using of detecting gene mutations in both rare and common cancers. The testing result allows doctors to quickly find out whether a patient’s tumor carries clinically useful mutations and to match patients with available therapies or clinical trials that will most benefit them. For a certain gene panel, there are about hundreds of genes that have been selected into gene panel, but usually we have no idea why and how those genes are selected. In order to have a better understanding of the gene panel, we developed a biomedical literature mining pipeline which can analyze the function of gene panels and the genes in it. Our study used the gene panel test developed by Memorial Sloan Kettering Cancer Center, MSK-IMACTTM, as example. For the biomedical explainability, our study utilized biomedical literature mining method to perform characteristic analysis on the gene panel. We want to dig out the useful information in the large-scale corpus by machine learning algorithms. However, most of machine learning algorithms can provide good precision and recall, but the results are hard to interpret. Therefore, we chose decision tree and topic modeling to analyze the literatures related to human genes since decision trees can provide clear decision-making process and the result of topic modeling is great to be interpreted in biomedical concepts. The experiment result shows that our study can not only represent the certain genes in a rational manner and is able to find different characteristics of the gene panel. Besides, we can make an appropriate biomedical explanation on both the results of decision tree and topic modeling and verify them by a manual curated pathway database. In case studies, we also find that decision tree and topic modeling have similar results. We hope that our study can help doctors making decisions and help bioinformatics researchers understanding more details about gene panels.