A six-gene model using an artificial neural network to predict regional lymph node metastasis after neo-adjuvant chemoradiotherapy for locally advanced rectal cancer

Abstract

Objective To screen out the potential gene to predict regional lymph node metastasis after neoadjuvant chemoradiotherapy (CRT) for locally advanced rectal cancer (LARC) and develop a 6-gene model using an artificial neural network (ANN). Methods The gene expression profiles (GSE46862) of locally advanced rectal cancer undergoing preoperative chemoradiotherapy from 64 specimens (21 with ypN- and 43 with ypN+ ) were downloaded from the gene expression omnibus (GEO) database. The differentially expressed genes were identified to screen out the potential biomarkers through the Gene-Cloud of Biotechnology Information (GCBI) platform. The top 6 genes were screened out for building model. An ANN model was trained and validated using the SPSS Modeler software. The study samples were allocated randomly into the training sample group and testing sample group with a 7∶3 ratio. The training samples and testing samples were respectively used for building an ANN model and independent back-substitution test. Observation indicators: (1) screening results of differentially expressed genes; (2) analysis results of ANN model. The receiver operating characteristic (ROC) curve was drawn and the area under the curve (AUC) was calculated to evaluate the predictive abilities of ANN and each biomarker. Results (1) Screening results of differentially expressed genes: A total of 50 genes were screened. Six top genes included IL6, AKR1B1, AREG, SELE, ROBO1 and CD274. (2) Analysis results of ANN model: Six top genes were selected to construct a three-layer ANN model with a 7-5-2 structure. The IL6 made the greatest effect on the ANN model, followed by ROBO1, AKR1B1, AREG, CD274 and SELE. The AUC was 0.929. The sensitivity and specificity of ANN model were 96.7% and 85.7%, and accuracy of training samples was 93.2%. In the independent back-substitution test, sensitivity and specificity were 92.3% and 85.7%, and accuracy of testing samples was 90.0%. Conclusion The prediction ANN model based on multiple molecular markers (IL6, ROBO1, AKR1B1, AREG, CD274 and SELE) for regional lymph node metastases in LARC patients after CRT would be beneficial in selecting potential candidates for rectum-preserving surgery following CRT for LARC. Key words: Rectal neoplasms; Neoadjuvant radiotherapy; Neoadjuvant chemotherapy; Regional lymph node; Artificial neural network