Deep Learning-Based Identification of Pathogenicity Genes in Phytophthora infestans Using Time-Series Transcriptomics

Area of Science:

• Plant Pathology
• Computational Biology
• Genomics

Background:

• Potato (Solanum tuberosum L.) is a vital global food crop, with China being a major producer.
• Late blight significantly limits potato production, necessitating strategies for resistance.
• Current methods for identifying pathogenicity genes lack temporal dynamic insights.

Purpose of the Study:

• To introduce an LSTM-Transformer hybrid model for discovering pathogenicity-related genes from gene expression time-series data.
• To identify novel candidate genes and pathways involved in potato late blight.
• To provide a data-driven framework for enhancing potato resistance breeding.

Main Methods:

• Utilized a time-series gene expression dataset of 32,917 genes across 18 samples (3 infection time points x 6 replicates).
• Developed and applied a hybrid LSTM-Transformer deep learning model with a biologically informed temporal-attention architecture.
• Implemented a gene time-series-specific data partitioning strategy and an interpretable deep analysis module.

Main Results:

• Identified 200 high-confidence pathogenicity-related genes from potato late blight infection data.
• Discovered enrichment of these genes in 15 key biological pathways, including plant immunity and stress responses.
• Uncovered potential new roles for candidate genes in defense hormone pathways and cell wall modification.

Conclusions:

• The study provides a powerful AI-driven framework for prioritizing genes from time-series expression data.
• Identified molecular targets and functional groups crucial for potato late blight resistance.
• Offers valuable insights for developing improved, blight-resistant potato cultivars through advanced breeding programs.