StressGenePred

StressGenePred predicts stress types and identifies stress-specific biomarker genes from time-series Arabidopsis transcriptome data using a twin neural network architecture with feature embedding and Confident Multiple Choice Learning (CMCL) loss.

Key Features:

• Integrated Analysis of Multiple Stresses: Performs integrated analysis of time-series gene expression data across multiple environmental stresses to capture interactions between stress types.
• Twin Neural Network Model Architecture: Implements a twin neural network with separate biomarker gene discovery and stress type prediction modules that share a common logical layer to reduce training complexity.
• Confident Multiple Choice Learning (CMCL) Loss: Uses CMCL loss to enhance selection of genes that respond specifically to individual stresses.
• High Accuracy in Stress Classification: Achieves higher accuracy in classifying heat, cold, salt, and drought stresses in Arabidopsis compared to limma feature embedding, support vector machines, and random forests.
• Discovery of Known Stress-Related Genes: Identifies known stress-related genes with greater specificity than the Fisher method while also detecting novel biomarkers.

Scientific Applications:

• Plant stress-response research: Enables identification of stress-specific biomarkers and classification of environmental stresses in Arabidopsis time-series transcriptome studies.
• Gene-stress association studies: Supports elucidation of molecular mechanisms underlying responses to heat, cold, salt, and drought stresses.
• Adaptation to phenotype-gene association studies: Provides a framework that can be applied to other phenotype-gene association analyses using time-series expression data.

Methodology:

Integrates time-series transcriptome data using a neural network-based approach combining feature embedding with a twin model architecture (biomarker discovery and stress prediction modules sharing a common logical layer) and CMCL loss.