Imputation for transcription factor binding predictions based on deep learning.

Area of Science:

• Genomics
• Molecular Biology
• Bioinformatics

Background:

• Transcription factor (TF) binding patterns are crucial for understanding gene regulatory networks.
• ChIP-seq is the gold standard for TF binding data but is resource-intensive, limiting coverage across TF-cell line combinations.
• Existing data represent a small fraction of potential TF-cell line interactions.

Purpose of the Study:

• To develop a computational method for accurately predicting cell-specific TF binding.
• To leverage existing ChIP-seq data to infer binding patterns for uncharacterized TF-cell line combinations.
• To enable prediction of TF binding and its functional consequences, such as enhancer activity and SNP impact.

Main Methods:

• Utilized a deep neural network architecture with a multi-task learning framework.
• Trained the model on a small fraction (4%) of available TF-cell line ChIP-seq data.
• Employed TFImpute to predict TF binding across various TF-cell line combinations.

Main Results:

• TFImpute achieved comparable accuracy to existing methods for TF-cell line combinations with available ChIP-seq data.
• TFImpute demonstrated superior accuracy for TF-cell line combinations lacking ChIP-seq data.
• The model successfully predicted cell line-specific enhancer activities (K562, HepG2) and the impact of SNPs on TF binding.

Conclusions:

• TFImpute offers an effective computational approach to predict cell-specific transcription factor binding.
• The method significantly expands the utility of existing ChIP-seq data for studying gene regulation.
• TFImpute has implications for predicting functional genomic elements and the effects of genetic variations.