EpiXFormer: a cross-attention neural network for predicting cell type-specific transcription factor binding sites

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Transcription factors (TFs) regulate gene expression and cell identity by binding to specific DNA sequences.
• TF binding sites (TFBSs) are cell type-specific, influenced by epigenomic context.
• Experimental TFBS profiling is costly and impractical for broad cell type coverage.

Purpose of the Study:

• To develop a computational method for accurate cell type-specific TFBS prediction.
• To leverage epigenomic data and deep learning for TFBS identification.
• To provide a scalable framework for understanding TF binding dynamics.

Main Methods:

• Developed EpiXFormer, a transformer-based neural network.
• Modeled proximal and distal epigenomic features influencing DNA-binding protein (DBP) binding.
• Incorporated TF motifs and potential co-occurring protein information.

Main Results:

• EpiXFormer demonstrated exceptional performance in predicting TFBSs across diverse cell types.
• The model effectively integrated epigenomic information to predict DBP binding.
• EpiXFormer successfully inferred pioneer factors and delineated cell type-specific TF functions.

Conclusions:

• EpiXFormer provides a robust and scalable computational framework for cell type-specific TFBS prediction.
• The tool facilitates the characterization and interpretation of multimodal genomic data.
• EpiXFormer can be readily applied to new cell types for enhanced genomic analysis.