Chromosome3D: reconstructing three-dimensional chromosomal structures from Hi-C interaction frequency data using distance geometry simulated annealing.

Badri Adhikari,Jianlin Cheng,Tuan Trieu

doi:10.1186/s12864-016-3210-4

Abstract

BackgroundReconstructing three-dimensional structures of chromosomes is useful for visualizing their shapes in a cell and interpreting their function. In this work, we reconstruct chromosomal structures from Hi-C data by translating contact counts in Hi-C data into Euclidean distances between chromosomal regions and then satisfying these distances using a structure reconstruction method rigorously tested in the field of protein structure determination.ResultsWe first evaluate the robustness of the overall reconstruction algorithm on noisy simulated data at various levels of noise by comparing with some of the state-of-the-art reconstruction methods. Then, using simulated data, we validate that Spearman’s rank correlation coefficient between pairwise distances in the reconstructed chromosomal structures and the experimental chromosomal contact counts can be used to find optimum conversion rules for transforming interaction frequencies to wish distances. This strategy is then applied to real Hi-C data at chromosome level for optimal transformation of interaction frequencies to wish distances and for ranking and selecting structures. The chromosomal structures reconstructed from a real-world human Hi-C dataset by our method were validated by the known two-compartment feature of the human chromosome organization. We also show that our method is robust with respect to the change of the granularity of Hi-C data, and consistently produces similar structures at different chromosomal resolutions.ConclusionChromosome3D is a robust method of reconstructing chromosome three-dimensional models using distance restraints obtained from Hi-C interaction frequency data. It is available as a web application and as an open source tool at http://sysbio.rnet.missouri.edu/chromosome3d/.Electronic supplementary materialThe online version of this article (doi:10.1186/s12864-016-3210-4) contains supplementary material, which is available to authorized users.

Highlights

Reconstructing three-dimensional structures of chromosomes is useful for visualizing their shapes in a cell and interpreting their function
Inspired by the computational techniques such as Crystallography & NMR System (CNS) suite [14,15,16,17,18] used for reconstructing protein structures from atom-atom distances measured by X-ray crystallography and nuclear magnetic resonance (NMR), in this work, we introduce a distance geometry simulated annealing (DGSA) based method to reconstruct three-dimensional chromosomal structures from the chromosomal wish distances derived from chromosomal interaction frequencies
Reconstruction using simulated datasets In order to evaluate our method’s reconstruction on noisy datasets, we tested our method on the simulated datasets with noise, and compared its performance with five existing distance-based methods implemented in Pastis [9] and ShRec3D [12], including three classic multidimensional scaling methods (metric multidimensional scaling (MDS), non-metric multidimensional scaling (NMDS) and ShRec3D), two statistical methods using a Poisson distribution (PM1 and PM2)

Summary

Introduction

Reconstructing three-dimensional structures of chromosomes is useful for visualizing their shapes in a cell and interpreting their function. Fluorescence in situ hybridization (FISH) is used to study the 3D organization of chromosomes and genomes [1,2,3]. Due to the low throughput and low resolution of FISH data, it cannot be used to study the organization of chromosomes and genomes at a finer and larger scale. The chromosomal contacts generated by Hi-C data can be used to infer 3D structures of chromosomes and genomes. A typical Hi-C experiment produces a matrix of interaction frequencies (IFs) between pairs of loci at a granularity defined in terms of resolution. An interaction frequency matrix is often termed as chromosomal contact matrix

Objectives

Methods

Results

Conclusion