Extended Data Fig. 1: Evaluation of the CNV segmentation and CNV normalization module in NeoLoopFinder.
a–g, We implemented the same generalized additive model (GAM) used by HiNT-CNV to estimate the copy number profile directly from Hi-C. For CNV segmentation, we applied a different algorithm based on Hidden Markov Model (HMM). a, We compared the copy number profiles estimated by NeoLoopFinder with the CNV profiles computed by Control-FREEC with whole genome sequencing (WGS) data. Each dot represents a 25kb bin. Bins with zero reads were excluded from the calculation. b, Similar to a, but for the comparison between NeoLoopFinder and HiNT-CNV. c, The number of CNV segments identified by NeoLoopFinder and HiNT-CNV, with or without WGS support. Only segments with a copy number ratio larger than 1.5 or smaller than 0.3 were considered in the calculation. d, The fraction of WGS-detected CNV segments that are recalled by NeoLoopFinder or HiNT-CNV. e, Comparison of F1 scores between NeoLoopFinder and HiNT-CNV in eight cancer cell lines. f, Comparisons of CNV segments inferred from HiNT-CNV and NeoLoopFinder based on Hi-C data in K562 cells. We compared their results with the CNV segments computed by Control-FREEC with whole genome sequencing (WGS) data. Results from both HiNT-CNV and NeoLoopFinder are similar to Control-FREEC. However, for more fragmented regions (green circles), NeoLoopFinder’s performance is better. g, A similar example in SK-N-MC cells. h–i, Comparison of different Hi-C normalization methods in K562 (Resolution: 10kb). Hi-C contact heatmaps and copy number variation profiles are shown for two example regions: ‘chr22: 22,340,000 – 24,200,000’ (h) and ‘chr9: 130,000,000 – 131,280,000’ (i). The CAIC method is excluded from the analysis due to memory error (Supplementary Table 2).
