Interpretable single-cell transcription factor prediction based on deep learning with attention mechanism.

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Predicting transcription factor binding sites (TFBS) is crucial for understanding gene transcription control.
• Existing deep learning methods for TFBS prediction require improvement, especially when using single-cell ATAC-seq data and attention mechanisms.

Purpose of the Study:

• To develop an interpretable deep learning method, IscPAM, for predicting transcription factor binding in single cells.
• To improve the efficiency and accuracy of TFBS prediction using single-cell ATAC-seq data.

Main Methods:

• IscPAM utilizes a deep learning model with an attention mechanism and a convolution neural network for feature extraction.
• The model is pre-trained on ATAC-seq, ChIP-seq, and DNA sequence data, and then applied to single-cell ATAC-seq data for TF binding graph prediction.
• Interpretability is validated through ablation experiments and sensitivity analysis.

Main Results:

• IscPAM achieves faster training and prediction times due to its embedded attention mechanism.
• The model efficiently predicts whole-genome transcription factor combinations in single cells.
• The method demonstrates interpretability in predicting TF binding.

Conclusions:

• IscPAM offers an efficient and interpretable approach for predicting single-cell transcription factor binding.
• This tool can advance the study of cellular heterogeneity through chromatin accessibility analysis.
• IscPAM has potential applications in understanding gene regulation in the context of related diseases.