Development of a clinical decision support system for classification of genomic data with TREE-based machine learning methods

Aim: The aim of this study is to develop a software for bioinformatic analysis of large data obtained as a result of genomic analysis of kidney tissue samples taken from rats with nephrotoxicity and without pathology with MTX and to visualize the results. In addition, it was aimed to model the data with tree-based learning methods, one of the machine learning methods, and to determine possible biomarkers for the diagnosis of nephrotoxicity by providing the interpretability of the model with explainable artificial intelligence methods. Material and Method: In this study, genomic data obtained from an experimental setup created by taking 20 female Wistar Albino rats were used. R programming language was used to perform bioinformatic analysis. Decision trees, Random Forest, AdaBoost, Bagged CART and XGBoost models from tree-based machine learning methods were used in the models. Local Interpretable Model-Agnostic Explanations and SHapley Additive Explanations methods were used to improve the interpretability of the XGBoost model. Python programming language was used in the analysis of models and explainable artificial intelligence methods. Results: The genomic dataset used in the study includes 16,386 lncRNA expressions. According to the results of bioinformatics analysis, 35 of the lncRNAs for nephrotoxicity and control groups showed up-expression, while 17 of them showed down-expression. As a result of the models made with lncRNAs selected by Boruta variable selection, the XGBoost method has been the most successful machine learning method according to performance criteria. As a result of SHAP, the top three most important candidates for predictive biomarkers for Nephrotoxicity were RNA-XR_591534.3 (LOC103691816), RNA-XR_351582.4 (LOC102555118), RNA-XR_005499541.1 (LOC120099962). Conclusion: As a result of the bioinformatic analysis, models and modeling interpretability performed in this study, possible genomic biomarkers for nephrotoxicity were determined by using lncRNA expression data of rats with nephrotoxicity and rats in the control group. Keywords: Explainable Artificial Intelligence, Tree-Based Learning, Decision Support System, Genomics, Machine Learning, Classification, Artificial Intelligence