Posterior inference of Hi-C contact frequency through sampling

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Hi-C is a key technique for studying 3D genome structure, generating contact frequency matrices.
• Current Hi-C data analysis provides point estimations without uncertainty quantification, limiting downstream analysis accuracy.
• Existing methods for analyzing Hi-C contact maps do not account for inherent data uncertainties.

Purpose of the Study:

• To develop a computational tool, HiCSampler, for inferring the uncertainty of Hi-C interaction frequencies.
• To enable downstream analyses of Hi-C data with robust uncertainty measurements.
• To improve the reliability of analyses such as TAD and loop annotation.

Main Methods:

• Developed HiCSampler, a novel algorithm that infers the posterior distribution of interaction frequencies.
• Exploited dependencies between neighboring loci in Hi-C contact maps to improve inference.
• Utilized posterior predictive checks to validate the accuracy of inferred interaction frequencies.

Main Results:

• HiCSampler reliably infers the posterior distribution of interaction frequencies from Hi-C data.
• Posterior predictive checks confirmed HiCSampler's ability to generate highly predictive chromosomal interaction frequencies.
• Summary statistics from HiCSampler provide quantifiable uncertainty measures for Hi-C experiments.

Conclusions:

• HiCSampler addresses the critical need for uncertainty quantification in Hi-C data analysis.
• The inferred samples from HiCSampler are compatible with existing downstream analysis tools.
• This method allows for uncertainty-aware analyses of 3D genome organization, enhancing biological insights.