Integrating differential pathway analysis with Monte Carlo cross-validation reveals seed cross-talks in non-small...

The objective of this study was to identify seed pathway cross-talks in non-small cell lung carcinoma (NSCLC), and to reveal potential pathological mechanism at molecular level systematically.

Differentially expressed genes (DEGs) between NSCLC and normal controls were identified using quantile- adjusted conditional maximum likelihood (QCML) method. Subsequently, differential pathways (DPs) enriched by DEGs were determined according to the Ingenuity Pathways Analysis )IPA) pathways and Fisher's exact test. A discriminating score )DS) was computed for each pair of DPs also called as cross-talk, and random forest )RF) algorithm was implemented to investigated hub cross-talks. Finally, global cross-talks with repeated times > 5 were calculated by Monte Carlo Cross-Validation )MCCV). By taking intersections between hub cross-talks and global crosstalks, we obtained seed cross-talks.

We obtained 122 DEGs and 5 DPs between NSCLC samples and normal controls. Based on DS and RF algorithm, 5 hub cross-talks with best area under the curve )AUC) were identified, of which Agranulocyte Adhesion and Diapedesis, and IL-17A Signaling in Fibroblasts were the best with AUC=0.996. After intersected with global cross-talks, we gained 2 seed cross-talks (Agranulocyte Adhesion and Diapedesis, Granulocyte Adhesion and Diapedesis and Agranulocyte Adhesion and Diapedesis, Glutathione Redox Reactions I).

Two seed cross-talks were identified and validated by MCCV, which may give insights for revealing pathological mechanism and potential biomarkers for target therapy in NSCLC.