Predicting transcription factor binding in single cells through deep learning.

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Understanding genome-wide transcription factor (TF) binding is crucial for deciphering biological processes.
• Current methods struggle to profile TF binding at the single-cell level, limiting insights into cellular heterogeneity.

Purpose of the Study:

• To introduce scFAN (single-cell factor analysis network), a deep learning model for predicting genome-wide TF binding profiles in individual cells.
• To enable the study of cellular identity and heterogeneity through chromatin accessibility.

Main Methods:

• scFAN utilizes deep learning, pretrained on bulk ATAC-seq, DNA sequence, and ChIP-seq data.
• It predicts TF binding in individual cells using single-cell ATAC-seq data.
• A novel "TF activity score" metric was developed to characterize cells.

Main Results:

• scFAN accurately predicts TF binding profiles at the single-cell level.
• Analysis of sequence motifs within predicted binding peaks validates the model's predictions.
• Predicted TF activity scores reliably capture distinct cell identities.
• The model successfully identifies cell types based on chromatin accessibility.

Conclusions:

• scFAN provides a powerful new approach for dissecting TF binding dynamics at single-cell resolution.
• The TF activity score offers a robust method for cell type identification and characterization.
• This work advances the study of cellular heterogeneity and gene regulation in individual cells.