(1100-29) Plant germplasm refinement and identifying representative traits for comprehensive analysis using machine learning tools

Body of Abstract: Plant germplasm of the National Agrobiodiversity Center of RDA, Korea, has about 30,000 collections of pepper in the database of the seed bank. We carried out the refinement of the plant germplasm DB to bring better usage. We used the pepper germplasm DB as a model, which is one of the largest collections of RDA and is valuable economy-wise. It has 16 traits as primary characteristics, such as flowering period, plant height, and fruit shape, and 68 traits as in-depth characteristics, such as disease resistance, detailed morphology of the plant, and spiciness. These traits are composed of measurable and unstructured data, and some genetic resources are not genetically fixed and show segregation in some characteristics; therefore, the decision of representative data of a particular cultivar is critical to analyze compared with other cultivars. We decide on representative data based on the distribution pattern of the traits. Then, we applied machine learning tools to find an optimum model to understand pepper germplasm comprehensively. To extract innate characteristics of the cultivar through the machine learning algorithms, we first converted each feature’s values of the pepper data into proper numerical data based on their characteristics. For missing values, either performing median imputation or filtering was applied depending on the mission portion. After the data preprocessing phase, classification, clustering, important feature extraction, and correlation analyses were performed. In classification and clustering analysis, traditional SVM (Support Vector Machine), RF (Random Forest), K-means Clustering, and DBSCAN methods were used. For the important feature extraction phase, we utilized logistic regression-based feature selection and SVM-RFE. Each of these analyzed results was then visualized and delivered for further analysis.

This work was supported by the Agenda project (PJ014307012023) of the Rural Development Administration, Republic of Korea.