Background Since there are inextricably connections among molecules in the biological networks, it would be a more efficient and accurate research strategy to screen microRNA (miRNA) markers combining with miRNA-mRNA regulatory networks. The independent regulation mode is more “fragile” and “influential” than the co-regulation mode. miRNAs can be used as biomarkers if they can independently regulate hub genes with important roles in the PPI network, simultaneously the expression products of the regulated hub genes play important roles in the signaling pathways of related tissue diseases. Methods We collected miRNA expression of non-small cell lung cancer (NSCLC) from The Cancer Genome Atlas (TCGA) database and the Gene Expression Omnibus (GEO) database. Volcano plot and signal-to-noise ratio (SNR) methods were used to obtain significant differentially expressed (SDE) miRNAs from the TCGA database and GEO database, respectively. A human miRNA-mRNA regulatory network was constructed and the number of genes uniquely targeted (NOG) by a certain miRNA was calculated. The area under the curve (AUC) values were used to screen for clinical sensitivity and specificity. The candidate markers were obtained using the criteria of the top five maximum AUC values and NOG ≥ 3. The protein–protein interaction (PPI) network was constructed and independently regulated hub genes were obtained. Gene Ontology (GO) analysis and KEGG pathway analysis were used to identify genes involved in cancer-related pathways. Finally, the miRNA which can independently regulate a hub gene and the hub gene can participate in an important cancer-related pathway was considered as a biomarker. The AUC values and gene expression profile analysis from two external GEO datasets as well as literature validation were used to verify the screening capability and reliability of this marker. Results Fifteen SDE miRNAs in lung cancer were obtained from the intersection of volcano plot and SNR based on the GEO database and the TCGA database. Five miRNAs with the top five maximum AUC values and NOG ≥ 3 were screened out. A total of 61 hub genes were obtained from the PPI network. It was found that the hub gene GTF2F2 was independently regulated by miR-708-5p. Further pathway analysis indicated that GTF2F2 participates in protein expression by binding with polymerase II, and it can regulate transcription and accelerate tumor growth. Hence, miR-708-5p could be used as a biomarker. The good screening capability and reliability of miR-708-5p as a lung cancer marker were confirmed by AUC values and gene expression profiling of external datasets, and experimental literature. The potential mechanism of miR-708-5p was proposed. Conclusions This study proposes a new idea for lung cancer marker screening by integrating microRNA expression, regulation network and signal pathway. miR-708-5p was identified as a biomarker using this novel strategy. This study may provide some help for cancer marker screening.

Finding reliable miRNA markers and revealing their potential mechanisms will play an important role in the diagnosis and treatment of NSCLC. Most existing computational methods for identifying miRNA biomarkers only consider the expression variation of miRNAs or rely heavily on training sets. These deficiencies lead to high false-positive rates. The independent regulatory model is an important complement to traditional models of co-regulation and is more impervious to the dataset. In addition, previous studies of miRNA mechanisms in the development of non-small cell lung cancer (NSCLC) have mostly focused on the post-transcriptional level and did not distinguish between NSCLC subtypes. For the above problems, we improved mainly in two areas: miRNA identification based on both the NOG network and biological functions of miRNA target genes; and the construction of a 4-node directed competitive regulatory network to illustrate the mechanisms. NSCLC was classified as lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUSC) in this work. One miRNA biomarker of LUAD (miR-708-5p) and four of LUSC (miR-183-5p, miR-140-5p, miR-766-5p, and miR-766-3p) were obtained. They were validated using literature and external datasets. The ceRNA-hub-FFL involving transcription factors (TFs), microRNAs (miRNAs), mRNAs, and long non-coding RNAs (lncRNAs) was constructed. There were multiple interactions among these components within the net at the transcriptional, post-transcriptional, and protein levels. New regulations were revealed by the network. Meanwhile, the network revealed the reasons for the previous conflicting conclusions on the roles of CD44, ACTB, and ITGB1 in NSCLC, and demonstrated the necessity of typing studies on NSCLC. The novel miRNA markers screening method and the 4-node directed competitive ceRNA-hub-FFL network constructed in this work can provide new ideas for screening tumor markers and understanding tumor development mechanisms in depth.