A deep learning method for replicate-based analysis of chromosome conformation contacts using Siamese neural networks

Area of Science:

• Genomics
• Molecular Biology
• Computational Biology

Background:

• Genome organization in nuclear space is crucial for biological processes.
• Chromosome conformation capture (3C) techniques like Hi-C and Micro-C generate genome-wide contact maps.
• Conventional analysis methods may miss biological information due to technical noise.

Purpose of the Study:

• To develop a novel deep learning method for analyzing chromatin conformation capture data.
• To effectively distinguish biological variations from technical noise in contact maps.
• To provide a more comprehensive understanding of genome architecture.

Main Methods:

• A replicate-based deep learning approach using a Siamese network configuration.
• Training the network to differentiate technical noise from biological signals in Hi-C and Micro-C data.
• Evaluating the method's performance against image similarity metrics.

Main Results:

• The deep learning method successfully distinguishes technical noise from biological variation.
• Extracted features from perturbed Hi-C maps reflect the roles of cohesin and CTCF in genome organization.
• Learned distance metrics correlate with cohesin and CTCF binding densities, indicating biological relevance.

Conclusions:

• The developed deep learning method is a powerful tool for exploring chromosome conformation capture data.
• This approach offers improved sensitivity and accuracy in identifying biologically significant features in genome architecture.
• The method facilitates deeper insights into the functional roles of proteins like cohesin and CTCF in genome organization.