Frontiers in Genetics (Feb 2022)
Exploring the Common Mechanism of Fungal sRNA Transboundary Regulation of Plants Based on Ensemble Learning Methods
Abstract
Studies have found that pathogenic fungi and plants have sRNA transboundary regulation mechanisms. However, no researchers have used computer methods to carry out comprehensive studies on whether there is a more remarkable similarity in the transboundary regulation of plants by pathogenic fungi. In this direction, high-throughput non-coding sRNA data of three types of fungi and fungi-infected plants for 72 h were obtained. These include the Magnaporthe, Magnaporthe oryzae infecting Oryza sativa, Botrytis cinerea, Botrytis cinerea infecting Solanum lycopersicum, Phytophthora infestans and Phytophthora infestans infecting Solanum tuberosum. Research on these data to explore the commonness of fungal sRNA transboundary regulation of plants. First, using the big data statistical analysis method, the sRNA whose expression level increased significantly after infection was found as the key sRNA for pathogenicity, including 355 species of Magnaporthe oryzae, 399 species of Botrytis cinerea, and 426 species of Phytophthora infestans. Secondly, the target prediction was performed on the key sRNAs of the above three fungi, and 96, 197, and 112 core nodes were screened out, respectively. After functional enrichment analysis, multiple GO and KEGG_Pathway were obtained. It is found that there are multiple identical GO and KEGG_Pathway that can participate in plant gene expression regulation, metabolism, and other life processes, thereby affecting plant growth, development, reproduction, and response to the external environment. Finally, the characteristics of key pathogenic sRNAs and some non-pathogenic sRNAs are mined and extracted. Five Ensemble learning algorithms of Gradient Boosting Decision Tree, Random Forest, Adaboost, XGBoost, and Light Gradient Boosting Machine are used to construct a binary classification prediction model on the data set. The five indicators of accuracy, recall, precision, F1 score, and AUC were used to compare and analyze the models with the best parameters obtained by training, and it was found that each model performed well. Among them, XGBoost performed very well in the five models, and the AUC of the validation set was 0.86, 0.93, and 0.90. Therefore, this model has a reference value for predicting other fungi’s key sRNAs that transboundary regulation of plants.
Keywords
