SCW: building the whole-genome 3D structures based on extremely sparse single-cell Hi-C data

The study of three-dimensional (3D) genome structures at the single-cell level is crucial for understanding cell-to-cell variability. However, it is challenging to reconstruct the 3D structures of the whole genome based on single-cell Hi-C data because of the sparseness of the single-cell Hi-C data and the complexity of the problem.

To address this, we developed a new computational method, named SCW (single-cell whole-genome), to build the high-resolution 3D genome structures of the whole genome based on extremely sparse single-cell Hi-C data (either zeros or ones in the Hi-C matrix). We evaluated our reconstructed 3D genome structures on various types of cells, checked the fitness of the reconstructed 3D structures to the single-cell Hi-C data, cross-validated the reconstructed structures with FISH data, bulk Hi-C data, and gene expression data, and then compared SCW with the state-of-the-art tools Nuc_dynamics, Hickit, and Tensor-FLAMINGO. SCW achieved better robustness as Nuc_dynamics failed on extremely sparse Hi-C matrices that contained only zeros or ones. The Pearson Correlation between our reconstructed 3D structure and the FISH data can reach 0.63, which is higher than the structures built by Hickit. SCW can also build better structures compared to Tensor-FLAMINGO based on multiple evaluations, particularly with the 20 Kbp resolution. Both the intra- and inter-chromosomal contact patterns are maintained in our reconstructed 3D structures, which also match the findings from single-cell gene expression data.

SCW enables high-precision whole-genome 3D reconstruction from extremely sparse single-cell Hi-C data. SCW outperforms existing tools in structural accuracy and robustly maintains intra/inter-chromosomal contacts. Its versatility is validated across diverse cell types.