Development of computer-aided diagnosis system based on ensemble learning methods: Application on genomic technologies

Aim: The aim of this study is to develop web-based software for bioinformatic analysis of large data obtained by performing genomic analysis of liver tissue samples from rats with cisplatin hepatotoxicity and rats without pathology. In addition, as a result of modeling the data with community learning methods, it was aimed to determine possible biomarkers for diagnosis/early diagnosis related to hepatotoxicity. Material and Method: Genomic data obtained from an experimental setup created by taking 20 female Sprague-Dawley rats were used in the study. The web-based software developed for bioinformatics analysis is designed using the Shiny library, which allows interactive web-based applications to be designed based on the R programming language, and ggplot2, ggrepel, DT, shinyWidgets, shinyLP, shinydashboard, and limma the packages. In the models, bagging, boosting, stacking, and XGBoost models from ensemble learning methods were used. Results: The genomic dataset used in the study contains 16,386 expressions. According to the results of the bioinformatics analysis, 589 lncRNAs for hepatotoxicity and control groups showed different expressions in the groups. Of these, 450 showed up-expression, while 139 showed down-expression. As a result of modeling with lncRNAs selected by the variable selection, the XGBoost model had the highest performance metrics. Conclusion: With this study, software was developed that allows bioinformatic analysis. In addition, as a result of bioinformatic analysis and modeling, potential genomic biomarkers for hepatotoxicity were determined using lncRNA expression data of rats with and without hepatotoxicity. Keywords: Computer Aided Diagnosis System, Genomics, Machine Learning, Classification, Ensemble Learning, Artificial Intelligence