RiceSNP-BST: a deep learning framework for predicting biotic stress-associated SNPs in rice.

Area of Science:

• Genomics
• Plant Science
• Bioinformatics

Background:

• Rice faces significant biotic stresses, necessitating the identification of genetic resistance.
• Accurate and rapid identification of single-nucleotide polymorphisms (SNPs) is vital for developing disease-resistant rice varieties.
• Limited high-quality rice genotype data has historically hindered SNP research.

Purpose of the Study:

• To develop an advanced computational framework for predicting SNPs associated with rice biotic stress traits (BST-associated SNPs).
• To improve the accuracy and efficiency of identifying genetic variations linked to biotic stress resistance in rice.
• To enhance the biological interpretability of SNP prediction models in rice.

Main Methods:

• Developed RiceSNP-BST, an automatic architecture search framework integrating multidimensional features for SNP prediction.
• Utilized convolutional neural networks for extracting structural and local features from DNA sequences.
• Employed causal inference and feature extraction from DNA sequences to improve model interpretability.

Main Results:

• RiceSNP-BST demonstrated higher precision compared to state-of-the-art methods in predicting BST-associated SNPs.
• The model showed robust performance on independent test sets and cross-species datasets.
• Extracted features and causal inference enhanced the biological interpretability of the SNP prediction.

Conclusions:

• RiceSNP-BST offers a powerful and precise approach for predicting rice SNPs related to biotic stress resistance.
• The framework advances genome prediction in rice and supports the development of improved crop varieties.
• A user-friendly web server for RiceSNP-BST is available to facilitate broader genomic research.