Machine learning and BP neural network revealed abnormal B cell infiltration predicts the survival of lung cancer patients

Abstract

FAM83A gene is related to the invasion and metastasis of various tumors. However, the abnormal immune cell infiltration associated with the gene is poorly understood in the pathogenesis and prognosis of NSCLC. Based on the TCGA and GEO databases, we used COX regression and machine learning algorithms (CIBERSORT, random forest, and back propagation neural network) to study the prognostic value of FAM83A and immune infiltration characteristics in NSCLC. High FAM83A expression was significantly associated with poor prognosis of NSCLC patients (p = 0.00016), and had excellent prognostic independence. At the same time, the expression level of FAM83A is significantly related to the T, N, and Stage. Subsequently, based on machine learing strategies, we found that the infiltration level of naive B cells was negatively correlated with the expression of FAM83A. The low infiltration of naive B cells was significantly related to the poor overall survival rate of NSCLC (p = 0.0072). In addition, Cox regression confirmed that FAM83A and naive B cells are risk factors for the prognosis of NSCLC patients. The nomogram combining FAM83A and naive B cells (C-index = 0.748) has a more accurate prognostic ability than the Stage (C-index = 0.651) system. Our analysis shows that abnormal infiltration of naive B cells associated with FAM83A is a key factor in the prognostic prediction of NSCLC patients.

Keywords: COX regression; back propagation (BP) neural network; immune infiltration; prognostic prediction; random forest.