PlantCTCIP: Chromatin Interaction Prediction Using Convolutional Neural Network and Transformer in Plants.

Area of Science:

• Genomics
• Computational Biology
• Plant Science

Background:

• Chromatin interactions are crucial for gene expression and phenotypic traits by linking regulatory elements to target genes.
• Predicting these interactions is key to understanding gene regulation in plants.

Purpose of the Study:

• To develop and evaluate PlantCTCIP, a deep learning model for predicting plant chromatin interactions.
• To construct genome-wide chromatin interaction maps for maize, rice, cotton, and wheat.
• To identify regulatory motifs, transcription factors, and their networks influencing chromatin interactions.

Main Methods:

• Utilized Convolutional Neural Networks and Transformer architectures for the PlantCTCIP model.
• Performed genome-wide chromatin interaction mapping for four plant species.
• Validated predictions using Hi-C experiments and analyzed motif and transcription factor enrichment.

Main Results:

• PlantCTCIP significantly improved chromatin interaction prediction accuracy (14.56% AUC for proximal promoter interaction, 9.6% for distal promoter interaction).
• Identified key motifs enriched in eQTLs and open chromatin regions.
• Analyzed TF enrichment and synergistic networks affecting promoter interactions across species.

Conclusions:

• PlantCTCIP offers a superior method for predicting plant chromatin interactions, aiding in gene regulatory mechanism analysis.
• The model assists in identifying distal regulatory elements and functional sites, supporting crop breeding and genetic improvement.
• Provides novel insights into gene regulation and facilitates intelligent breeding strategies for diverse crops.