Machine learning models for predicting lymph node and distant metastases in colorectal cancer

Background: Colorectal cancer (CRC) is the third most common malignancy in the world and metastasis is responsible for a major proportion of the cancer-related deaths in CRC patients.Aims: To construct machine learning models for predicting lymph node and distant metastases in colorectal cancer and analyze biological functions features of metastasis-related genes.Methods: RNA-seq and miRNA-seq data as well as corresponding clinical data from colon adenocarcinoma (COAD) and rectum adenocarcinoma (READ) were obtained from The Cancer Genome Atlas (TCGA) database. The differentially expressed RNAs (DE-RNAs) in non-LNM (N0) and LNM (N1/N2) as well as non-distant metastases (M0) and distant metastases (M1) were analyzed. Six machine learning models including logistic regression (LR), random forest (RF), support vector machine (SVM), Catboost, gradient boosting decision tree (GBDT), and artificial neural network (NN) were constructed to predict cancer metastasis and the feature genes of the optimal model were further analyzed by functional enrichment, protein-protein interaction (PPI) network, and drug-target analyses.Results: Differential RNA expression profiles of LNM and non-LNM as well as M0 vs. M1 were observed in both COAD and READ samples. NN model was determined to be the optimal model for predicting distant metastases, while Catboost and LR models were the optimal models for predicting LNM in COAD and READ samples, respectively. PPI analysis indicated that KIR2DL4, chemokine-related genes CXCL9/10/11/13 and CCL25, and gamma-aminobutyric acid (GABA) receptor genes (GABRR1, GABRB2 and GABRA3) were key genes in metastasis. In addition, atorvastatin and eszopiclone were identified as potential therapeutic agents as they target these genes.Conclusions: We constructed six machine learning models for predicting colorectal cancer metastases and identify the optimal model. We analyzed biological functions features of metastasis-related RNAs in colorectal cancer.