Common genes regulating multiple abiotic stress response in rice: An analysis using supervised and unsupervised machine learning models

Rice response to abiotic stresses is generally understood through comparison of gene expression and gene mapping between contrasting tolerant and susceptible genotypes. Machine learning uses large datasets and trains the data through building efficient models for pattern identification and feature importance (gene) prediction. This work explores the potential of using machine learning tools in identifying gene expression models common to different abiotic stress response in rice. For this, 146 rice microarray gene expression samples comprised of drought (57), salinity (35), cold (33) and heat (20) were categorized into two classes namely control, stress and analyzed using supervised (Random forest) and unsupervised (Cluster) machine learning algorithms. The best random forest trained model showed an accuracy of ~ 79% in the classification analysis with an ROC curve area of 0.7963. Besides, genes involved in ABA pathway, flowering, and secondary metabolites (Two PP2Cs, three expressed proteins, Flowering Promoter Factor-like, Terpene Synthase, Gamete EXpressed) were identified as common set of genes for multiple abiotic stress response in rice. Moreover, validation of these eight genes through q-PCR under vegetative stage drought stress identified intrinsic drought tolerance mechanism and drought responsiveness in drought tolerant and rainfed upland, aerobic, and irrigated rice varieties.