Abstract: Objective · To explore the molecular classification and prognostic prediction of gastric cancer based on nonnegative matrix factorization (NMF).  Methods · Cases of gastric cancer were acquired from Gene Expression Omnibus (GEO). Expression profiling of lncRNA was performed by using a lncRNA-mining approach. NMF model was built with Consensus Cluster Plus package. The relationship among NMF subgroups and clinical relevance was assessed.  Results · According to the molecular classification based on NMF, samples were divided into three subgroups. Significant difference was observed in relapse state, lymph node ratio, Lauren classification, TNM stage and age of onset among three subgroups. High-risk group was identified with shortest relapse time by survival analysis both in GSE62254 and GSE15459. Multivariate Cox proportional-hazards regression showed that NMF model based molecular classfication could be regarded as an independent risk factor for gastric cancer. Gene set variance analysis (GSVA) and gene set enrichment analysis (GSEA) showed that the high-risk subgroup was enriched in several tumor development pathways.  Conclusion · Based on NMF model, the molecular classification of gastric cancer can be used for treatment decision and prognostic prediction.

Key words: gastric cancer, molecular classification, prognosis, bioinformatics