TSPTFBS

TSPTFBS predicts transcription factor binding sites (TFBS) across plant species using deep convolutional neural networks trained on DAP-seq data and transfer learning to homologous transcription factors.

Key Features:

• Deep Convolutional Neural Network (DeepCNN): Uses a DeepCNN architecture to train 265 prediction models for Arabidopsis thaliana TFBSs on DAP-seq datasets.
• Transfer Learning: Adapts Arabidopsis-trained models to homologous transcription factors in other plant species, validated on ten TFs across Oryza sativa, Zea mays, and Glycine max.
• Performance Superiority: Achieves higher accuracy than gkm-SVM and MEME for Arabidopsis TFBS prediction and learns known motifs and cooperative TF motifs supported by protein-protein interaction evidence.
• Data Source: Leverages DAP-seq datasets as the high-resolution source of DNA–protein interaction data for model training.

Scientific Applications:

• Plant gene regulation: Enables prediction of TFBSs across species to study transcriptional regulatory mechanisms in plants.
• Comparative genomics: Supports identification of conserved and species-specific TFBSs for cross-species regulatory comparisons.
• Evolutionary biology: Facilitates analysis of transcription factor binding evolution and regulatory divergence among plant lineages.
• Functional genomics: Assists identification of regulatory elements for genes of interest in species lacking extensive experimental TFBS data.

Methodology:

Train DeepCNN models on DAP-seq datasets to generate 265 Arabidopsis TFBS prediction models and adapt these models via transfer learning to homologous TFs in other species, with validation on ten TFs across Oryza sativa, Zea mays, and Glycine max.