Exploring Gene Signatures in Different Molecular Subtypes of Gastric Cancer (MSS/ TP53+, MSS/TP53-): A Network-based and Machine Learning Approach

Abstract

Gastric cancer (GC) is one of the leading causes of cancer mortality, worldwide. Molecular understanding of GC’s different subtypes is still dismal and it is necessary to develop new subtype-specific diagnostic and therapeutic approaches. Therefore developing comprehensive research in this area is demanding to have a deeper insight into molecular processes, underlying these subtypes. In this study, a three-step methodology was developed to identify important genes and subnetworks in two subtypes of GC (TP53+ and TP53-). First, weighted gene co-expression network analysis was performed to explore co-expressed gene modules in both subtypes. Afterward, the relationship of each module with the tumor pathological stage (as a clinical trait indicating tumor progression) was studied by decision tree machine learning algorithm and the best predicting module was selected for further analysis (modules with 241 genes for TP53+ and 1441 genes for TP53- were identified). Subsequently, a motif exploring and motif ranking analysis was implemented to explore three-member signature gene motifs in the selected modules' biological network. These motifs may have key regulatory roles in the studied GC subtypes. Motif members of TP53- mostly contain MAPK signaling pathway genes which show their key role in this subtype of GC. In the case of the TP53+ subtype, our findings demonstrated that alternative splicing and SNARE proteins could prompt the initiation and advancement of the disease. These findings can be used to develop new diagnostic and therapeutic approaches based on the personalized medicine concept. This methodology could be implemented to unravel underlying mechanisms and pathways in other complex phenotypes and diseases.

Keywords

• gastric cancer
• molecular subtypes
• weighted gene co-expression network analysis
• decision tree
• network analysis