3.1 Construction of a risk prediction model for PRGs in NB patients based on the training set in GES49711

In this study, the included GSE 49711 cohort was randomized in a 1:1 ratio into training and testing groups, with 250 cases in the training group and 248 cases in the testing group. Univariate and multivariate COX regression analyses were performed on the 20 included PRGs to screen for independent predictors associated with prognosis. Univariate COX regression analysis revealed 11 genes of prognostic significance: AIM2, CASP8, DFNA5, GSDMA, GSDMB, GSDMD, GZMA, IL-18, IL-1B, NLRC4 and NLRP3 (Fig. 2A). Multivariate COX regression analysis revealed four genes that could be used as independent predictors: GSDMB, IL-18, NLRP3 and AIM2 (Fig. 2B). A 4-gene risk profile was constructed using independent regression coefficients for each gene with the risk score formula: (− 0.30 × GSDMB) + (− 0.46 × IL 18) + (− 0.21 × NLRP3) + (0.56 × AIM2). Based on the risk score formula, the risk score of each patient was calculated, and the patients were divided into high-risk and low-risk groups based on the risk score, and the distribution of the risk score, associated survival data, and heat map of risk gene distribution for each patient were plotted (Fig. 3A). The area under the ROC curve is shown as 0.816 (Fig. 3B). Scatter plots of risk score distributions and survival times were plotted (Fig. 3C). Survival analysis showed that NB patients in the high-risk subgroup had significantly lower survival rates than those in the low-risk subgroup (P < 0.001, Fig. 3D).

Fig. 2

Prognostic model of pyroptosis correlation based on training cohort construction. A Univariate COX regression analysis revealed 11 PRGs associated with prognosis in patients with NB; B Multivariate COX regression analysis revealed four PRGs as independent influencers of prognosis in patients with NB

Fig. 3

Clinical predictive efficacy of risk scoring models. A Risk score distributions, associated survival data, and gene expression heat maps; B Predictive performance of prognostic models; C Scatter plots of risk score distributions and survival times; D Survival analysis of patients in the high-risk group and patients in the low-risk group

3.2 Validating the reliability of risk models using the GSE49711 cohort and test set

To further validate the applicability of the risk prediction model in NB patients, this study was validated using the GSE49711 overall cohort and test cohort. Risk scores were calculated for each sample based on the risk prediction model, and risk score distributions, survival data, and risk gene expression heat maps were plotted for the overall cohort and the test cohort (Fig. 4A, B). The areas under the ROC curves for the overall cohort and the test cohort are 0.807 and 0.805, respectively, proving that the model has good predictive efficacy (Fig. 4C). Kaplan–Meier survival analyses showed that the high-risk subgroup had a significantly lower survival rate than the low-risk subgroup in both the overall cohort and the test cohort (P < 0.001, Fig. 4D).

Fig. 4

Validating the accuracy of risk scoring models. A Overall Cohort-Based Heatmap of Risk Scores, Survival Data, and Risk Gene Expression; B Test Cohort-Based Heatmap of Risk Scores, Survival Data, and Risk Gene Expression; C ROC curves for predicting prognostic efficacy of risk score models. D Survival analysis of patients in the high-risk group and patients in the low-risk group

3.3 Correlation analysis between PRGs and tumor immune microenvironment

In order to investigate the relationship between cellular PRGs and the tumor immune microenvironment, We assessed immune microenvironmental differences between different sub-risk subgroups of NB patients using the ssGSEA and ESTIMATE algorithms. The composite heat map showing the distribution of 28 immune cells in the GSE49711 cohort showed that Estimate score, Stormal score, Immune score, and tumor purity were all significantly different between the high and low risk groups (Fig. 5A). ESTIMATE analysis showed that the low-risk subgroup had significantly higher Estimate score, Stormal score and Immune score (P < 0.001), the high-risk subgroup had higher tumor purity (P < 0.01) (Fig. 5B). Patients with higher Estimate score, Stormal score and Immune score had better survival, while patients with higher tumor purity had poorer survival (Fig. 5C). We also analyzed the correlation between the two subgroups and 28 immune infiltrating cell types, which showed that 24 immune infiltrating cell types were significantly different between the two subgroups (P < 0.05, Fig. 6A). Comparison of correlations between high- and low-risk subgroups and “RNA writer” showed significant correlations with four “RNA writer” (CD274, CD8A, TBX2 and TNF, P < 0.001). (Fig. 6B).

Fig. 5

Correlation analysis between PRGs and immune infiltration scores. A Heat map of differences in immune infiltration in the tumor immune microenvironment between different risk groups; B The analysis of differences between groups of estimate socre, stromal score, Immune score and tumour purity; C The Survival analysis of estimate socre, stromal score, Immune score and tumour purity. (**P < 0.01; ***P < 0.001; ****P < 0.0001)

Fig. 6

Correlation between PRGs and immune infiltrating cells. A Correlation analysis between different risk groups and 28 immune cells; B Correlation analysis between different risk groups and “RNA writer”. (*P < 0.05; **P < 0.01; ***P < 0.001; ns, not statistically significant)

3.4 Correlation of risk stratification with clinical indicators

We collapsed risk scores and clinicopathological characteristics for screening independent influences on prognostic relevance (Table 1). The correlation between different risk groups and clinicopathologic characteristics was assessed, and the composite heat map showed statistically significant differences in fustat, age, MYCN status, INSS, cli-risk and progression for patients in the GSE49711 cohort between the high- and low-risk groups (P < 0.001, Fig. 7A). The independent predictive efficacy of the risk score model and clinicopathologic factors was validated using ROC curves, which showed that the risk score model and the rest of the clinicopathologic factors had good predictive efficacy except for gender (AUC = 0.469) (Fig. 7B). In a univariate COX regression analysis, the results showed that risk scores, age, mycn status, cli-risk, and inss staging of this prognostic model were significantly associated with disease-free survival of the patients (Fig. 7C). In multivariate COX regression analysis, after adjusting for traditional clinical prognostic variables (age, sex, MYCN status, cli-risk, and INSS stage), the model remained as an independent predictor of NB patients, indicating good model stability (P = 0.002, Fig. 7D).

Table 1 Clinicopathological factorsFig. 7

Prognostic significance of differences in clinicopathologic features and risk scores between risk groups. A Differences in clinicopathologic characteristics among risk subgroups in the GSE49711 cohort; B ROC curves for prognostic predictive efficacy between risk score models and different clinicopathologic features; C NB in the GSE49711 cohort univariate Cox regression analysis of risk factors; D Multivariate Cox regression analysis of NB risk factors in the GSE49711 cohort. (*P < 0.05; **P < 0.01; ***P < 0.001)

3.5 Construction and validation of alignment diagram based on risk scores and clinical prognostic indicators

The results of the prognostic analysis showed that age, mycn status, cli-risk, inss staging, and prognostic model risk score could be used as independent prognostic indicators for patients with NB (P < 0.05). Combining the above risk model scores and clinical metrics, this study constructed alignment diagram to expand clinical usability (Fig. 8A). A risk score was assigned to each patient by summing the weighted scores for each risk factor, with higher total scores corresponding to worse survival outcomes. The AUC values for predicting 5-, 7.5-, and 10-year survival were 0.843, 0.802, and 0.797, respectively (Fig. 8B). The calibration curves show good performance of the predictive model (Fig. 8C).

Fig. 8

Combined risk score and clinicopathologic indicators to construct a nomogram diagram and validate its accuracy. A nomogram diagram constructed by combining risk scores and clinicopathologic indicators; B Time-dependent ROC curves were used to assess the predictive performance of nomogram diagram in 5-, 7.5- and 10-year survival; C Calibration curves for assessing the predictive accuracy of nomogram diagram

3.6 Protein interaction network of PRGs in NB and expression of risk genes in NB cell lines

In this study, we used the String online database and Cytoscape software to analyse the initial inclusion of 20 significantly differentiated protein–protein interaction networks of PRGs to facilitate a better understanding of the biological processes in which they are involved and their impact on disease (Fig. 9). The results show the interaction relationship between the 20 PRGs proteins and the level of expression. By analysing the protein–protein interaction network, new protein interactions can be discovered, new biological functions of proteins can be revealed, and potential drug targets can be searched. To ensure that subsequent in vitro cellular experiments could be performed in the laboratory, we obtained gene expression matrices of the 4 risk genes in different NB cell lines from the CCLE dataset. The results showed that the 4 risk genes were expressed to different degrees in different NB cell lines (Fig. 10).

Fig. 9

Protein–protein interaction networks of PRGs. The darkness of the colour at the node and the size of the node represent the role of the protein in the interaction network, with darker colours and larger nodes indicating that it is most closely associated with related proteins in the interaction network

Fig. 10

The expression distribution of mRNA in different cell lines. The abscissa represents the expression distribution of mRNA and the ordinate represents different cell lines, different colors and the size of dots represent expression. A AIM2 expression; B GSDMB expression; C IL-18 expression; D NLRP3 expression