Hi-C data

We obtained the Hi-C data of K562 cells from [12]. Both intra-chromosomal and inter-chromosomal interactions were processed at 25kb resolution and VC_SQRT normalization was applied to the Hi-C contact matrices. Hi-C data extraction and normalization were performed using Juicer [46]. For intra-chromosomal contact maps, we calculated the log2 ratio between observed (O) over expected (E) interactions (i.e., O/E) for each pair of interactions. The rationale is to consider genomic loci (not necessarily close on 1D distance) that share spatial localization with higher than expected genome-wide Hi-C interaction patterns to facilitate the identification of compartmentalization. Our simulation evaluation also supports this rationale (Additional file 1: Supplementary Results). For inter-chromosomal interactions, the expected number of interactions was set to be uniformly distributed between genomic loci on different chromosomes. For each chromosome, we fitted a Weibull distribution for Hi-C contacts and kept those interactions with p value < 1E −5 as significant interactions for subsequent steps as input to the SPIN method. For each inter-chromosomal interaction, we also used p value < 1E −5 as the cutoff for significant interactions, but for each pair (i,j), we required that all neighboring pairs between i±1 and j±1 should be also significant to increase the reliability of added edge.