Molecular classification reveals the diverse genetic features and prognosis of gastric cancer: a multi-omics consensus ensemble clustering

Background Gastric cancer (GC) is the fifth most common tumor around the world, it is necessary to reveal novel molecular subtypes to guide the selection of patients who may benefit from specific target therapy. Methods Multi-omics data, including RNA-sequence of transcriptomics (mRNA, LncRNA, miRNA), DNA methylation and gene mutation of TCGA-STAD cohort was used for the clustering. Ten classical clustering algorithms were applied to recognize patients with different molecular features via the R package “ MOVICS ”. The activated signaling pathways were evaluated using the single-sample gene set enrichment analysis. The difference distribution of gene mutations, copy number alterations and tumor mutation burden was compared, and potential response to immunotherapy and chemotherapy was assessed as well. Results Two molecular subtypes (CS1 and CS2) were recognized by ten clustering algorithms with further consensus ensembles. Patients in the CS1 group were found to contain a shorter average overall survival time (28.5 vs. 68.9 months, P = 0.016), and progression-free survival (19.0 vs. 63.9 months, P = 0.008) compared to the CS2 group. CS1 group contained more activation of extracellular associated biological process, while CS2 group displayed the activation of cell cycle associated pathways. The significantly higher total mutation numbers and neo antigens were observed in CS2 group, along with the specific mutation of TTN, MUC16 and ARID1A. Higher infiltration of immunocytes were also observed in CS2 group, reflected to the potential benefit from immunotherapy. Moreover, CS2 group also can response to 5-fluorouracil, cisplatin, and paclitaxel. The similar diverse of clinical outcome of CS1 and CS2 groups were successfully validation in external cohorts of GSE62254, GSE26253, GSE15459, and GSE84437. Conclusion Novel insight into the GC subtypes was obtained via integrative analysis of five omics data by ten clustering algorithms, which can provide the idea to the clinical target therapy based on the specific molecular features. ### Competing Interest Statement The authors have declared no competing interest.