iCancer-Pred: A tool for identifying cancer and its type using DNA methylation

• Published in: Genomics (San Diego, Calif.) Calif.), 2022-11, Vol.114 (6), p.110486-110486, Article 110486
• Main Authors: Lin, Weizhong, Hu, Siqin, Wu, Zhicheng, Xu, Zhaochun, Zhong, Yu, Lv, Zhe, Qiu, Wangren, Xiao, Xuan
• Format: Article
• Language: English
• Subjects: DNA Methylation; elastic net; Epigenomics; feature selection; Humans; Neoplasms - genetics; Predicting cancer types
• ISSN: 0888-7543; 1089-8646
• DOI: 10.1016/j.ygeno.2022.110486

DNA methylation is an important epigenetics, which occurs in the early stages of tumor formation. And it also is of great significance to find the relationship between DNA methylation and cancer. This paper proposes a novel model, iCancer-Pred, to identify cancer and classify its types further. The datasets of DNA methylation information of 7 cancer types have been collected from The Cancer Genome Atlas (TCGA). The coefficient of variation firstly is used to reduce the number of features, and then the elastic network is applied to select important features. Finally, a fully connected neural network is constructed with these selected features. In predicting seven types of cancers, iCancer-Pred has achieved an overall accuracy of over 97% accuracy with 5-fold cross-validation. For the convenience of the application, a user-friendly web server: http://bioinfo.jcu.edu.cn/cancer or http://121.36.221.79/cancer/ is available. And the source codes are freely available for download at https://github.com/Huerhu/iCancer-Pred. •A high accurate predictor is proposed to classify cancer and its type.•A feature selection mold is designed to choose the significate DNA methylation sites related cancer.•Discovering genes related cancer.