BackgroundRNA binding protein (RBP) plays a crucial role in tumorigenesis at post-transcriptional level in various cancer types. Nevertheless, the role of RBPs in liver hepatocellular carcinoma (LIHC) remains obscure. We attempted to uncover the association between RBPs and the prognosis of LIHC patients. MethodsWe analyzed the transcriptome and corresponding clinical data of LIHC patients from the cancer genome atlas (TCGA) (training cohort) and international cancer genome consortium (ICGC) (validating cohort) database with a series of bioinformatics methods. Differently expressed RNA-binding proteins (DERBPs) were screened and subjected to functional enrichment analysis and co-expression network establishment. Overall survival (OS) related DERBPs and our prognosis risk model were confirmed by univariate, LASSO and multivariate regression analysis in training cohort. Survival analysis, Receiver operating characteristic curve (ROC) and nomogram were conducted in both training and validating groups to confirm the performance of our model. Human protein atlas (HPA) database and Kaplan-Meier plotter were used to verify the expression and prognostic significance of the hub RBPs respectively.Results There were 330 RBPs were found significantly different in TCGA. Functional analysis indicated most of the DERBPs were majored in RNA processing, alternative splicing and metabolism, etc. 6 RBPs (UPF3B, MRPL54, ZC3H13, DHX58, PPARGC1A, EIF2AK4) were recognized as OS related and enrolled into our prognostic model. Survival analysis showed the risk signature was negatively correlated with the OS of LIHC patients in both training (p = 5.808e-06) and validating (p = 3.38e-03) groups. The area under curves (AUC) of the receiver operating characteristics (ROC) curve in training and validating cohorts was 0.756, 0.781 respectively which indicating the good performance of our model. The risk signature was an independent hazardous factor in multivariate COX regression analysis either in TCGA (HR = 1.626;95% CI 1.394 -1.897, p < 0.001) or ICGC (HR = 1.939;95% CI 1.324 -2.838, p < 0.001). Nomogram and calibration curve indicated our model had best performance in predicting 3-year survival rate.ConclusionsWe constructed a six-RBPs based risk signature model which had moderate efficiency in LIHC patients’ prognosis forecasting which may assist practitioners to make better decision in the management of LIHC.