High throughput chromosome conformation capture technologies such as Hi-C,have made it possible to survey the 3D genome structure.However,due to inherent limitations of the technologies,it remains a major challenge to obtain 3D profiles at kilobase resolution with low cost.Here,we present CISD,a computational method to precisely identify chromatin interaction sites at kilobase resolution from MNase-seq data,and CISD_loop,a CISD based chromatin-chromatin interaction (CCI) prediction method from low resolution Hi-C data.The methods are built on a hypothesis that the CCIs may result in a characteristic nucleosome arrangement pattern flanking the interaction sites.We show that the predictions of CISD and CISD_loop overlap closely with ChIA-PET anchors and loops,respectively.Moreover,the methods trained in one cell type can be applied to other cell types with high accuracy.The validity of the methods were further supported by 3C-experiments at 5kb resolution.Finally,we demonstrate that only modest amounts of MNase-seq and Hi-C data are sufficient to achieve ultrahigh resolution map of CCIs.The predictive power of CISD/CISD_loop supports the hypothesis that CCIs induce local nucleosome patterns that may serve as probes for the 3D dynamics of the genome.Thus,our method will facilitate precise and systematic investigations of the interactions between distal regulatory elements on a larger scale than hitherto have been possible.