Prediction of regulatory motifs from human Chip-sequencing data using a deep learning framework

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Identifying transcription factor binding sites (TFBS) and cis-regulatory motifs is crucial for understanding gene regulation.
• Existing methods face challenges in accurately predicting these sites and uncovering complex protein-DNA interactions.

Purpose of the Study:

• To develop and validate a novel deep learning method, DEep Sequence and Shape mOtif (DESSO), for enhanced cis-regulatory motif prediction.
• To explore the utility of integrating DNA shape features into motif prediction models.
• To identify novel protein-protein-DNA tethering interactions.

Main Methods:

• Developed DESSO, a deep neural network and binomial distribution model for motif prediction.
• Applied DESSO to 690 human ENCODE ChIP-sequencing datasets.
• Integrated DNA shape features into the DESSO framework.
• Analyzed transcription factor (TF) binding and identified tethering interactions in K562 cells.

Main Results:

• DESSO demonstrated superior performance compared to existing tools like DeepBind in motif prediction.
• Identified 61 putative protein-protein-DNA tethering interactions among 100 TFs in K562 cells.
• DNA shape information significantly improved predictive power and revealed new shape motif information for human TFs.

Conclusions:

• DESSO offers an advanced deep learning framework for accurate TFBS and motif identification.
• Integration of DNA shape features enhances the structural analysis of TF binding sites.
• DESSO facilitates the discovery of novel TF binding mechanisms and interactions.