Analysis of copy number variations with machine learning technics

Copy number variations, that are losses or gains of particular DNA regions, are found about two percent in the human genome. This group which belongs to structural variations can be found in the healthy population and sometimes losses or gains specific regions can lead to clinical manifestations. Interpretation of clinical and genetic data, which also requires family work to interpret clinically the detected copy number variation, does not always provide accurate information. Increasing application of machine learning algorithms is getting widely used in the medical field and becoming increasingly important, especially in areas with large data such as genetics. The aim of this study is to use machine learning algorithms in clinical interpretation of copy number variations. Machine learning models were created using multiple multi-class machine learning algorithms with the ClinVar data set. In the model made with a smaller data set, the average accuracy of the multi-class decision forest is 0.96, while in the main data set, the accuracy of the multi-class decision forest is 0.86. In this study, it could be shown that the most common identified variants were interpreted with higher accuracy and the model was more successful when the two classes benign and pathogenic were separated. This preliminary study demonstrated that a machine learning model could be used to automate diagnosis and in the clinical interpretation of copy number variants.