Identifying metastatic biomarkers of colorectal cancer: Machine learning modeling and experimental validation

Abstract

Background Colorectal cancer (CRC) liver metastasis is responsible for the majority of CRC-related deaths. Early detection of metastasis is crucial for improving patient outcomes but can be delayed due to a lack of symptoms. In this research, we aimed to investigate for CRC metastasis related biomarkers by employing machine learning (ML) approach and experimental validation. Methods Gene expression profile of CRC patients with liver metastasis was obtained using GSE41568 dataset and the differentially expressed genes between primary and metastatic samples were screened. Subsequently, we carried out feature selection to identify most relevant DEGs using LASSO and Penalized-SVM methods. DEGs commonly selected by these methods were selected for further analysis. Finally, the experimental validation was done through qRT-PCR. Results 11 genes were commonly selected by SCAD and P-SVM algorithms among which seven had prognostic value in colorectal cancer. It was found that the expression of MMP3 gene decreases in stage IV of colorectal cancer compared to other stages (p-value < 0.01). Also, the expression of WNT11gene increases significantly in this stage(p-value < 0.001). It was also found that expression of WNT5a, TNFSF11 and MMP3 is significantly lower, and the expression level of WNT11 is significantly higher in liver metastasis samples compared to primary tumors. Conclusion In summary, this study has identified a set of potential biomarkers for CRC metastasis using ML algorithms. The findings of this research may provide new insights into the identification of biomarkers for CRC metastasis and may lead to new therapeutic strategies for the treatment of this disease.