Fig. 5: In silico perturbation-based Hi-C experiment on pancreatic cancer metastasis.

a EpiVerse pipeline for PANC experiment reproduction: Schematic diagram illustrating the EpiVerse process from constructing virtual epigenetic signals to chromatin 3D structure modeling, using various histone modifications and CTCF as inputs to accurately replicate chromatin structures. b Comparison between Imputation Hi-C and experimental Hi-C: Heatmaps indicating the Hi-C interaction matrices with substantial variations across chromosome 15. The top row presents the ground truth Hi-C matrices for Capan-1 and PANC, and their differential heatmap, while the bottom row displays imputed data. c Enhanced predictive performance with additional tracks: A bar graph comparing the Spearman correlation coefficient of EpiVerse’s predictive accuracy for Capan-1 and PANC-1 datasets. The graph highlights improvements as additional epigenetic tracks are integrated: 1 track (CTCF), 2 tracks (CTCF + H3K4me3), 4 tracks (CTCF + H3K4me3 + H3K9me3 + H3K27ac), and 6 tracks (CTCF + H3K4me3 + H3K9me3 + H3K27ac + H3K27me3 + H3K36me3). d Proportion of differentially expressed genes in regions enriched with chromatin Interaction and enhancer activity  changes: A gene identification curve showing the proportion of differentially expressed genes located in regions enriched for both top enhancer and Hi-C signal differences in Capan-1, compared against random regions. EpiVerse consistently identifies a higher proportion of relevant genes, outperforming the random baseline. e Enriched GO terms of differentially expressed genes ranked by chromatin Interaction changes: GSEA of the top-ranked differentially expressed genes between Capan-1 and PANC-1 cells, revealing enrichment in GO terms associated with cell motility, metastasis, and apoptotic processes.