Selecting and combining biomarkers by gene expression profiles for colon cancer classification

In order to improve colorectal cancer (CRC) stratification, researches made before were using biomarkers, biomarker combinations or gene expression profile (GEP) clustering individually for patient classification. This study was trying to use both biomarker and GEP for the colon cancer (CC) classification. GEPs were adopted as an approach in selecting and combining biomarkers for patient classification. A practical research was made based on public dataset GSE40967, which contained GEP data of 566 CC patients, messages of tumor-node-metastasis (TNM) staging, biomarker (DNA mismatch repair [MMR], CpG island methylator phenotype [CIMP], chromosomal instability [CIN], KRAS, BRAF, TP53) statuses, patient sex, tumor location, chemotherapy status, and relapse-free survival (RFS). For each of the biomarkers, differentially expressed genes (DEGs) between two biomarker-statuses of patients were identified. The biomarker with the most DEGs was selected first and was used to classify patients. In each obtained class (or subclass), the remained biomarkers were used again to identify DEGs, to select next biomarker, to classify patients, and so on. These selected biomarkers were combined successively for the further classification. For stage II patients, MMR, CIN and KRAS were selected successively; for stage III patients, MMR and KRAS were selected successively. As a result, seven leaf-classes, II/dMMR, II/pMMR/CIN-, II/pMMR/CIN+/KRAS-wild-type, II/pMMR/CIN+/KRAS-mutant, III/dMMR, III/pMMR/KRAS-wild-type, and III/pMMR/KRAS-mutant were obtained. Significant RFS rate differences were observed between the leaf classes, especially in the case of stage III and non-chemotherapy, which validates the effectiveness of the classification. Large inner-stage differences were observed between the leaf classes of stage III in the case of non-chemotherapy, which might be helpful for future staging.