Identifying functions and prognostic biomarkers of network motifs marked by diverse chromatin states in human cell lines.

Epigenetic modifications play critical roles in modulating gene expression, yet their roles in regulatory networks in human cell lines remain poorly characterized. We integrated multiomics data to construct directed regulatory networks with nodes and edges labeled with chromatin states in human cell lines. We observed extensive association of diverse chromatin states and network motifs. The gene expression analysis showed that diverse chromatin states of coherent type-1 feedforward loop (C1-FFL) and incoherent type-1 feedforward loops (I1-FFL) contributed to the dynamic expression patterns of targets. Notably, diverse chromatin state compositions could help C1- or I1-FFL to control a large number of distinct biological functions in human cell lines, such as four different types of chromatin state compositions cooperating with K562-associated C1-FFLs controlling "regulation of cytokinesis," "G1/S transition of mitotic cell cycle," "DNA recombination," and "telomere maintenance," respectively. Remarkably, we identified six chromatin state-marked C1-FFL instances (HCFC1-NFYA-ABL1, THAP1-USF1-BRCA2, ZNF263-USF1-UBA52, MYC-ATF1-UBA52, ELK1-EGR1-CCT4, and YY1-EGR1-INO80C) could act as prognostic biomarkers of acute myelogenous leukemia though influencing cancer-related biological functions, such as cell proliferation, telomere maintenance, and DNA recombination. Our results will provide novel insight for better understanding of chromatin state-mediated gene regulation and facilitate the identification of novel diagnostic and therapeutic biomarkers of human cancers.

Systemically identifying and prioritizing risk lncRNAs through integration of pan-cancer phenotype associations.

LncRNAs have emerged as a major class of regulatory molecules involved in normal cellular physiology and disease, our knowledge of lncRNAs is very limited and it has become a major research challenge in discovering novel disease-related lncRNAs in cancers. Based on the assumption that diverse diseases with similar phenotype associations show similar molecular mechanisms, we presented a pan-cancer network-based prioritization approach to systematically identify disease-specific risk lncRNAs by integrating disease phenotype associations. We applied this strategy to approximately 2800 tumor samples from 14 cancer types for prioritizing disease risk lncRNAs. Our approach yielded an average area under the ROC curve (AUC) of 80.66%, with the highest AUC (98.14%) for medulloblastoma. When evaluated using leave-one-out cross-validation (LOOCV) for prioritization of disease candidate genes, the average AUC score of 97.16% was achieved. Moreover, we demonstrated the robustness as well as the integrative importance of this approach, including disease phenotype associations, known disease genes and the numbers of cancer types. Taking glioblastoma multiforme as a case study, we identified a candidate lncRNA gene SNHG1 as a novel disease risk factor for disease diagnosis and prognosis. In summary, we provided a novel lncRNA prioritization approach by integrating pan-cancer phenotype associations that could help researchers better understand the important roles of lncRNAs in human cancers.