iDHSs-HO: a high-order correlation method for identifying DNase I hypersensitive sites

DNase I hypersensitivity sites (DHSs) are critical elements in DNA regulatory mechanisms, and their accurate identification holds substantial significance for biomedical research. While traditional experimental methods can identify DHSs, they are often hindered by high costs and time demands. With the progress of machine learning, computational approaches have drawn growing interest. In this work, we introduce a machine learning framework that leverages dinucleotide physicochemical properties to recognize DHSs. To capture comprehensive correlation information, a sliding window strategy was implemented. Moreover, two distinct techniques-clustering and principal component analysis-were utilized to derive high-order correlation characteristics. To minimize the impact of redundant and noisy information, the F-score method was applied for feature selections. Subsequently, the selected features were fed into a support vector machine to distinguish DHSs from non-DHSs. Experimental outcomes demonstrated that the clustering-based approach obtained superior performance, with classification accuracies 0.9902 and 0.9896 on two independent datasets. Compared with earlier studies using the same datasets, our proposed method also yielded improved results. This suggests that our model is effective for DHSs identification. For academic purposes, we provide the data and codes used in current work at https://figshare.com/articles/online_resource/iDHSs-HO/31860025.