Fast decoding cell type-specific transcription factor binding landscape at single-nucleotide resolution

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Understanding cell type-specific transcription factor (TF) binding is vital for biological processes and diseases.
• Experimental TF binding profiling is resource-intensive, limiting comprehensive analysis.
• Existing computational methods lack cell-specific accuracy or high resolution.

Purpose of the Study:

• To develop a novel deep learning approach for predicting TF binding sites at single-nucleotide resolution.
• To improve the accuracy and efficiency of TF binding landscape prediction across diverse cell types.

Main Methods:

• Developed Leopard, a deep learning model utilizing a many-to-many neural network architecture.
• Evaluated Leopard's performance against state-of-the-art methods like Anchor and FactorNet.
• Assessed prediction accuracy using area under the receiver operating characteristic curve (AUROC) and area under the precision recall curve (AUPRC).

Main Results:

• Leopard achieved high predictive performance with an average AUROC of 0.982 and AUPRC of 0.208.
• Significantly outperformed Anchor and FactorNet, improving AUPRC by 19% and 27% at 200-bp resolution.
• Demonstrated a hundredfold to thousandfold speedup compared to existing many-to-one machine learning methods.

Conclusions:

• Leopard provides accurate, high-resolution TF binding site predictions.
• The method offers a significant advancement in computational TF binding analysis.
• Leopard enables faster and more comprehensive exploration of TF binding landscapes.