Deep Neural Network-Mining of Rice Drought-Responsive TF-TAG Modules by a Combinatorial Analysis of ATAC-Seq and...

Area of Science:

• Plant Molecular Biology
• Computational Biology
• Genomics

Background:

• Drought stress critically impacts rice growth and yield.
• Previous research focused on individual transcription factors, with limited understanding of gene regulatory networks under multi-factor stress.

Purpose of the Study:

• To develop a machine learning model for predicting transcription factor binding sites.
• To identify drought-responsive transcription factors and their target genes in rice using an integrated approach.

Main Methods:

• Compiled a comprehensive dataset of transcription factors and binding sites from plant genomes (rice, Arabidopsis, barley) using the JASPAR database.
• Developed TFBind, a nine-layer convolutional deep neural network using PyTorch for transcription factor binding prediction.
• Integrated Weighted Gene Co-expression Network Analysis (WGCNA) with ATAC-seq data to identify transcription factors associated with open chromatin regions under drought stress.

Main Results:

• Identified 15 drought-responsive transcription factors linked to open chromatin regions.
• TFBind model predicted direct binding for 81% of transcription factors to opened genes.
• Enrichment analysis revealed target genes involved in protein transport, protein allocation, and nitrogen compound transport.

Conclusions:

• The developed TFBind model is a valuable tool for predicting transcription factor-target gene interactions in biological processes.
• This study provides insights into the regulatory mechanisms of drought response in rice, highlighting key transcription factors and their roles.