Modeling and visualization for characteristics extraction of mutations by radiation using machine learning

Kanzaki, Norie   ; Shimada, Mikio*; Yanagihara, Hiromi*

After ionizing radiation exposure, several types of DNA damage occur in mammalian cells. Several studies have been reported that the mutation type depends on the radiation type and the cell type. Thus, it is important to characterize the type of mutation caused by irradiation for understanding of the biological effects of radiation. This study is a foundational investigation of construction of data analysis method dedicated to characteristics extraction of mutations. Two datasets were hypothetically made by reference to previous researches. The data was composed of eight items, namely two types of transition, four types of transversion, insertion, and deletion. The datasets were first analyzed by Self-Organizing Maps. We categorized and predicted the characteristics of mutations. This result was compared with those of other multivariate analysis methods. The data with similar mutation patterns was closely located on the output map, indicating the variation of mutations given by different radiation types.