Abstract: Having known the 3D structures of chromosomos is of great importance to the understanding of gene expression and regulation in nuclei. Hi-C technology has been developed to capture genome-wide interactions and generate contact frequency data. Based on the characteristic of the sparse and noisy interaction sampling in Hi-C data, a MBO-based method with parameter optimization, named VMBO, was proposed to predict a 3D chromatin structure. First, for converting the interaction frequency to spatial distance between two chromosome fragments the conversion factor was optimized by golden section search. Second, manifold based optimization (MBO) was applied to reconstruct a consensus 3D structure. The VMBO accuracy and robustness were validated on both simulation data and real Hi-C data. The results of structure similarity measures, root mean squared deviation and distance Spearman correlation coefficient, indicate that the proposed method can well reconstruct 3D chromatin structures.