Identification and validation of drought-responsive genes in rice using machine learning and candidate gene association mapping

초록

Rice, a staple crop relied upon by nearly half of world population, is vital for maintaining global food security. Drought stress dramatically decreases rice yields and causes substantial economic losses across Asia. Gene expression plays a crucial role in predicting gene function. Machine learning leverages large datasets to train models that efficiently identify patterns and predict the significance of specific genes as features. In this study, machine learning and candidate gene association mapping were employed to identify and validate genes related to drought stress response in rice. The analysis utilized 146 microarray samples of rice, categorized into salinity (35), drought (57), heat (20), and cold (33) conditions. These samples were divided into two groups: the data from both control and stress conditions were analyzed using machine learning techniques, including supervised algorithms like Random Forest and unsupervised methods such as clustering. The most effective Random Forest achieved an accurateness of approximately 79%. In addition, eight genes such as PP2Cs, gamete expressed, FPFL4, TPS20, and three expressed proteins were identified which is induced in drought stress. All genes, except one, were upregulated in rice cultivars exhibiting varying degrees of stress tolerance. Candidate gene association mapping further identified five SNPs within three genes including TPS20, PP72, and FPFL4 that are linked to root traits under drought stress during the vegetative stage. This study successfully applied machine learning and candidate gene association mapping to enhance understanding of the genetic basis of drought tolerance in rice. Therefore, integrating these insights into rice breeding programs could facilitate the creation of more resilient cultivars, thereby increasing food security and economic stability in drought-prone regions.

키워드