Fig. 1: Spectral decomposition of trans Hi-C data identifies distinct interaction profiles.

a, Left, map of trans Hi-C in HCT116 and heatmaps of leading trans eigenvectors; right, magnification of a 35-Mb intrachromosomal region of chr2. b, Scatter plots of all 50-kb genomic bins projected onto (left) a UMAP embedding of E1–E9 and (right) the E1–E2 subspace, colored by cluster identity (k-means, k = 8). c, E1 versus E2 scatter plots colored, from left to right, by point density, GC content and distance from the centromere. The fourth plot traces the trajectory of a single chromosome arm (chr1p) from centromere to telomere. d, Heatmaps of mean signal intensity of functional genomics features (rows) for each 50-kb genomic bin (column), grouped into Hi-C-derived clusters as in b. Top to bottom: GC content, distance from centromere, TSA-seq for SON, two-stage Repli-seq (Early/Late), fraction of methylated CpGs derived from WGBS, LaminB1 DamID-seq and Protect-seq. Clusters (I–VIII) are ordered by ascending Protect-seq signal and within each cluster bins are sorted by distance from the centromere. Three pairs of clusters are combined for a total of five IPGs indicated in the lower row of colored bars. E1–E9 are displayed in between. e, Left, violin density plots of total RNA expression per IPG represented as log₁₀ transcripts per million (TPM) with internal miniature box plots. White circles mark the median, box bounds represent the interquartile range and whiskers extend by a factor of 1.5 for TPMs from IPGs A1 (n = 8,233 genes), A2 (n = 18,390), B0 (n = 5,369), B1 (n = 8,862), B4 (n = 3,200). Right, gene density in genes per megabase in each IPG. Same sample sizes as on the left. f, Pairwise mean observed/expected contact frequency between IPGs at 50 kb in cis (left) and trans (right). g, Distribution of SNIPER subcompartment label assignments³⁶ of HCT116 B₄ loci across various cell types. FC, fold change; O/E, observed/expected.