Machine Learning-Based identification of resistance genes associated with sunflower broomrape

Abstract

Background: Sunflowers (Helianthus annuus L.), a vital oil crop, are facing a severe challenge from broomrape (Orobanche cumana), a parasitic plant that seriously jeopardizes the growth and development of sunflowers, limits global production and leads to substantial economic losses, which urges the development of resistant sunflower varieties.

Results: This study aims to identify resistance genes from a comprehensive transcriptomic profile of 103 sunflower varieties based on gene expression data and then constructs predictive models with the key resistant genes. The least absolute shrinkage and selection operator (LASSO) regression and random forest feature importance ranking method were used to identify resistance genes. These genes were considered as biomarkers in constructing machine learning models with Support Vector Machine (SVM), K-Nearest Neighbours (KNN), Logistic Regression (LR), and Gaussian Naive Bayes (GaussianNB). The SVM model constructed with the 24 key genes selected by the LASSO method demonstrated high classification accuracy (0.9514) and a robust AUC value (0.9865), effectively distinguishing between resistant and susceptible varieties based on gene expression data. Furthermore, we discovered a correlation between key genes and differential metabolites, particularly jasmonic acid (JA).

Conclusion: Our study highlights a novel perspective on screening sunflower varieties for broomrape resistance, which is anticipated to guide future biological research and breeding strategies.

Keywords: Feature selection; Machine learning; Resistance genes; Sunflower broomrape.

Fig. 1

Sunflower phenotypes and results of WGCNA analyses. A Schematic diagram of the phenotypes of resistant and susceptible varieties after infestation with broomrape and pictures of roots parasitized by broomrape. B Bar plot showing the number of susceptible and resistant varieties. C Plot of phenotypes of different classifications. D Volcano plot of differential genes of edgeR analysis. Blue represents the down-regulation of gene expression, and red represents the up-regulation of gene expression. E The scale-free fit index for various soft-thresholding powers (β) and the mean connectivity for various soft-thresholding powers. F Dendrogram of genes clustered via the dissimilarity measure. G Heatmap of the correlation between module eigengenes and traits. Grey module indicates genes that do not match any module. H Scatter plot between GS and MM in the brown module

Fig. 2

Performance of models constructed using shared genes. A A total of 32 shared genes were identified between key module genes and DEGs. B The GO enrichment results from shared genes, GO terms with False Discovery Rate (FDR) < 0.05 were considered significant. C Model prediction performance of train dataset. D Model prediction performance of test dataset

Fig. 3

Prediction model was constructed using key genes identified through the LASSO method. A LASSO regression algorithm. B Number of features for different lambda values. C Model prediction performance of training dataset. D Model prediction performance of test dataset. E Expression of 24 key genes in the two groups. *, **, and *** represent statistical significance at p < 0.05, p < 0.01, and p < 0.001, respectively

Fig. 4

Prediction model constructed by Random Forest importance rank. A The top 30 feature importance ranking. B Model prediction performance of training dataset. C Model prediction performance of test dataset. D, E Partial dependency graph of the top 2 features of the selected features. F The contour plot visually demonstrates the influence of the top 2 genes on the predicted values of LR

Fig. 5

The performance of the machine learning model. A The performance of the machine learning model on the training datasets. B The performance of the machine learning model on the test datasets

Fig. 6

The results of metabolic analysis. A PCA plot. B Volcano plot of differential metabolites. C Heatmap showed the results of the clustering analysis of differential metabolites. D Correlation between genes from LASSO and differential metabolites. Diamonds represented key genes from LASSO; red lines represented positive correlation; green lines represented negative correlation