DNA methylation analysis has been applied to determine the primary site of cancer; however, robust and accurate prediction of cancer types with a minimum number of sites is still a significant scientific challenge. To build an accurate and robust cancer type prediction tool with a minimum number of DNA methylation sites, we internally benchmarked different DNA methylation site selection and ranking procedures, as well as different classification models. We used The Cancer Genome Atlas dataset (26 cancer types with 8296 samples) to train and test models and used an independent dataset (17 cancer types with 2738 samples) for model validation. A deep neural network model using a combined feature selection procedure (named MethyDeep) can predict 26 cancer types using 30 methylation sites with superior performance compared with the known methods for both primary and metastatic cancers in independent validation datasets. In conclusion, MethyDeep is an accurate and robust cancer type predictor with the minimum number of DNA methylation sites; it could help the cost-effective clarification of cancer of unknown primary patients and the liquid biopsy-based early screening of cancers.

The human genome is widely transcribed, with part of these transcribed regions producing stably expressed protein-coding or non-coding RNAs. Long intergenic non-coding RNAs (lincRNAs) are significantly differentially expressed in various cell lines and tissues. However, the influence of their transcription events remains unclear. In this study, we constructed a human genomic interaction network and found frequent interactions between lincRNA genes and protein-coding genes that are highly related to the occupancy of RNA polymerase II on the lincRNA gene. Interestingly, in the human genome interaction networks, the degree of lincRNA genes was significantly higher than that of protein-coding genes. The promoter regions of the protein-coding genes interacting with the lincRNA genes are enriched with R-loop structures, indicating that lincRNA may influence the target genes through R-loop structures. These promoters were enriched in more transcription factor binding sites. Furthermore, the whole network and sub-network could be utilized to explore potential biomarkers of leukemia. We found that zinc finger protein 668 (ZNF668), eosinophil granule ontogeny transcript (EGOT), and glutamate metabotropic receptor 7 (GRM7) could serve as novel biomarkers for acute myeloid leukemia (LMAL). Pasireotide acetate (CAS No. 396091-76-2) represents a potential drug for LMAL patients. These results suggested that potential biomarkers and corresponding drugs for cancer could be identified based on lincRNA–promoter network/sub-network topological parameters.