DeepChIA-PET: Accurately predicting ChIA-PET from Hi-C and ChIP-seq with deep dilated networks.

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Chromatin interaction analysis by paired-end tag sequencing (ChIA-PET) reveals genome-wide interactions mediated by specific DNA-binding proteins.
• ChIA-PET experiments are crucial for understanding chromatin folding and transcription regulation.
• Limited availability of ChIA-PET datasets necessitates computational methods to predict interactions using more common data types like Hi-C and ChIP-seq.

Purpose of the Study:

• To develop a computational method, DeepChIA-PET, for accurately predicting ChIA-PET interactions.
• To leverage Hi-C and ChIP-seq data to infer ChIA-PET interactions, reducing the need for wet-lab experiments.
• To validate the model's performance across different cell types and protein factors.

Main Methods:

• A supervised deep learning approach, DeepChIA-PET, was developed.
• The model utilizes Hi-C and ChIP-seq data as input to predict ChIA-PET interactions.
• Deep models were trained using CTCF-mediated ChIA-PET data from GM12878 cells, featuring 40 dilated residual convolutional blocks.

Main Results:

• DeepChIA-PET significantly outperformed Peakachu, a random forest-based method, when using only Hi-C data.
• Incorporating ChIP-seq data improved classification performance, though Hi-C data played a more dominant role.
• The model accurately predicted CTCF-mediated ChIA-PET in GM12878 and HeLa cells, as well as non-CTCF interactions (RNAPII, RAD21) in various cell types.

Conclusions:

• DeepChIA-PET is an accurate computational tool for predicting ChIA-PET interactions.
• The method effectively predicts interactions mediated by various chromatin-associated proteins across different cell types.
• DeepChIA-PET facilitates the study of chromatin interactions by leveraging widely available Hi-C and ChIP-seq data.