Revealing Hi-C subcompartments by imputing inter-chromosomal chromatin interactions

Abstract

Higher-order genome organization and its variation in different cellular conditions remain poorly understood. Recent high-coverage genome-wide chromatin interaction mapping using Hi-C has revealed spatial segregation of chromosomes in the human genome into distinct subcompartments. However, subcompartment annotation, which requires Hi-C data with high sequencing coverage, is currently only available in the GM12878 cell line, making it impractical to compare subcompartment patterns across cell types. Here we develop a computational approach, SNIPER (Subcompartment iNference using Imputed Probabilistic ExpRessions), based on denoising autoencoder and multilayer perceptron classifier to infer subcompartments using typical Hi-C datasets with moderate coverage. SNIPER accurately reveals subcompartments using moderate coverage Hi-C datasets and outperforms an existing method that uses epigenomic features in GM12878. We apply SNIPER to eight additional cell lines and find that chromosomal regions with conserved and cell-type specific subcompartment annotations have different patterns of functional genomic features. SNIPER enables the identification of subcompartments without high-coverage Hi-C data and provides insights into the function and mechanisms of spatial genome organization variation across cell types.

Fig. 1

Overview of SNIPER. a Flowchart of SNIPER’s training procedure. b SNIPER denoising autoencoder. Rows of the low coverage Hi-C probability map are used in the input layer. Weights are optimized using binary cross-entropy (BCE) loss between the reconstructed and ground truth contact probabilities. c SNIPER neural network classifier is trained using latent variables from b as input and optimized using cross-entropy between predictions and the original annotations based on the high-coverage Hi-C data in Rao et al.

Fig. 2

SNIPER performance in GM12878. a Confusion matrix between SNIPER predictions and the original subcompartment annotation based on the high-coverage Hi-C in Rao et al.. b Precision-recall curve and AUPR values for the prediction of each subcompartment. c Histone mark and replication timing fold change profiles constructed for SNIPER results (top) and the Gaussian HMM subcompartment calls based on the full dataset (bottom). Fold change of an epigenetic mark in each subcompartment is defined as the median signal of the mark divided by the median signal across all subcompartments. Source data are available in the Source Data file

Fig. 3

SNIPER predictions correlate with various functional genomic data. a Reconstruction of the inter-chromosomal Hi-C contact matrix in IMR90. This example between chromosomes 2 and 3 shows that SNIPER imputes missing contacts in the sparse matrix, recovers subcompartment-specific contact patterns, and predicts annotations that correlate with DNA replication timing Repli-seq, H3K27ac ChIP-seq, and RNA-seq (FPKM). b Normalized histone mark signal changes at the boundary between A2 (left) and B1 (right) in GM12878, IMR90, and K562. c Subcompartment distribution in K562 SON TSA-seq deciles for the SNIPER K562 subcompartments (left) and the Rao et al. GM12878 subcompartments (right). Source data are available in the Source Data file

Fig. 4

SNIPER allows comparisons of subcompartments across different cell types. a (Top) Distribution of thirteen conservation states in the genome. (Bottom) Distribution of subcompartment regions in each conservation state. b UCSC Genome Browser shot displaying the information content (IC), cross cell type predictions, H3K27ac fold change, and smoothed Repli-seq wavelets of each 100 kb region. Subcompartments offering the most information, associated with taller bars in the IC track, are more conserved across nine cell types. Highlighted color boxes show three regions with distinct patterns across cell types. c For each conservation state, we show Repli-seq signal distribution in the most frequent subcompartment annotations among nine cell types at each 100 kb region. Regions in conservation state 1 are the most conserved, reflected by low variance of Repli-seq signal in each subcompartment. Regions in the NC state are the most dynamic, suggesting multiple annotations for a single chromatin region and high variance in Repli-seq signal. d Hi-C reconstructions across cell types in chr18 (47.1–47.9 Mb) where IMR90 is A2-specific and other cell lines are predicted as B3. Source data are available in the Source Data file