Pyroptosis-Derived Long Noncoding RNA Profiles Reveal a Novel Signature for Evaluating the Prognosis of Patients With Lung Adenocarcinoma

Abstract

PURPOSE

Long noncoding RNAs (lncRNAs) were recently implicated in modifying pyroptosis. Nonetheless, pyroptosis-related lncRNAs and their possible clinical relevance persist largely uninvestigated in lung adenocarcinoma (LUAD).

MATERIALS AND METHODS

A sum of 921 samples were collected from three independent data sets. We obtained pyroptosis-related genes from both the Molecular Signatures Database and relevant literature sources and used four machine learning techniques, comprising stepwise Cox, ridge regression, least absolute shrinkage and selection operator, and random forest. Multiple bioinformatics approaches were used to further investigate the underlying mechanisms.

RESULTS

In total, 39 differentially expressed pyroptosis genes were identified by comparing normal and tumor samples. Correlation analysis revealed 933 pyroptosis-related lncRNAs. Furthermore, univariate Cox regression determined 11 lncRNAs that exhibited stable associations with prognosis in the three cohorts, which were used to construct the pyroptosis-derived lncRNA signature. After analyzing the optimal results from four machine learning algorithms, we ultimately selected random forest to develop the pyroptosis-derived lncRNA signature. This signature was proven to be an independent prognostic factor and exhibited robust performance in three cohorts.

CONCLUSION

We provided novel insight and established a pyroptosis-derived lncRNA signature for patients with LUAD, exhibiting strong predictive capabilities in both the training and validation sets.

INTRODUCTION

Lung cancer, the leading cause of cancer-related fatalities, ranks second in the incidence of tumors according to Global Cancer Observatory.¹ Notably, lung adenocarcinoma (LUAD) exhibits unique pathological characteristics because of its high metastasis and invasion tendencies.^2,3 Although the use of novel therapies such as targeted therapy and immunotherapy continues to grow, treatment outcomes remain considerably poor because of tumor heterogeneity and individual differences.^4,5 Undoubtedly, molecular differences contribute significantly to heterogeneity.⁶ As a result, there exists a pressing demand for a new, sensitive molecular signature capable of predicting the prognosis of patients with LUAD, which would enable the implementation of more precise treatment strategies.

CONTEXT

• Key Objective

• In the context of lung adenocarcinoma (LUAD), where prognosis remains a challenge because of tumor heterogeneity, this study aimed to identify and develop a pyroptosis-derived long noncoding RNA (lncRNA) signature for predicting patient outcomes.

• Knowledge Generated

• We analyzed 921 samples from three independent data sets, identifying 39 differentially expressed pyroptosis-related genes and 933 pyroptosis-related lncRNAs in LUAD. Through rigorous analysis and machine learning, we established an 11-lncRNA signature significantly associated with prognosis, exhibiting robust predictive capabilities in three cohorts.

• Relevance

• This pyroptosis-derived lncRNA signature offers potential clinical application by enabling the stratification of LUAD patients into responders and nonresponders to treatment. Precision in patient management could spare nonresponders from unnecessary toxicity and treatment delays, ultimately improving LUAD patient care.

Pyroptosis is a mode of programmed cell death that occurs alongside an associated inflammatory reaction.⁷ An increasing number of investigations have reported that cell pyroptosis bears a strong connection to the onset and advancement of tumors in melanoma,⁸ breast cancer,⁹ colorectal cancer,¹⁰ gastric cancer,¹¹ hepatocellular carcinoma,¹² and lung cancer.¹³ Specifically, cell pyroptosis was associated with prognosis and sensitivity to chemotherapy in lung cancer. For instance, the expression of Gasdermin D (GSDMD) was upregulated in non–small cell lung cancer (NSCLC), and high expression of GSDMD indicated a poor prognosis of LUAD.¹⁴ Xi et al¹⁵ revealed that GSDMD colocalized with granzyme B near immune synapses, and the deficiency in GSDMD attenuated the cytolytic power of CD8⁺ T cells, indicating that GSDMD was pivotal in the immune response to lung tumor cells. Despite emerging research in lung cancer, the precise functions of pyroptosis in LUAD remain largely unknown.

Accumulating evidence has shown the pivotal functions of long noncoding RNAs (lncRNAs) in the biological processes of cancer.¹⁶ Furthermore, a multitude of lncRNAs are involved in the development and advancement of lung cancer.^17,18 For example, overexpression of the lncRNA DGCR5 facilitates LUAD progression by targeting miR-22-3p.¹⁹ Interestingly, lncRNAs can indirectly regulate cell pyroptosis. LncRNA-XIST, an oncogene, is upregulated in NSCLC. Knockdown of lncRNA-XIST impeded the progression of NSCLC by activating miR-335/SOD2/ROS axis-mediated pyroptosis.²⁰ However, no studies have reported on pyrosis-related lncRNAs predicting the prognosis of LUAD.

This study aimed to develop a pyroptosis-derived lncRNA signature using various machine learning algorithms to forecast the prognosis of patients with LUAD. The multiomics characteristics and mutational landscape of this signature were further delineated. Furthermore, the latent mechanisms of this signature were decoded. Overall, this lncRNA signature derived from pyroptosis holds potential as a valuable prognostic biomarker for LUAD.

MATERIALS AND METHODS

Sources of Data Sets and Preprocessing

A sum of 921 LUAD samples were collected, including The Cancer Genome Atlas (TCGA; n = 568), GSE31210 (n = 226),^21,22 and GSE50081 (n = 127).²³ The TCGA-LUAD cohort served as the training set, whereas GSE50081 and GSE31210 data sets were used as validation sets. The TCGA data included 510 primary LUAD samples and 58 normal lung tissue samples, with 497 patients with LUAD having clinical and survival information. RNA sequencing data for TCGA-LUAD and the associated clinical information were acquired from the University of California Santa Cruz Xena database.²⁴ Survival data were acquired from the cBioPortal website.²⁵ The TCGA-LUAD count data were subjected to variance stabilizing transformation by the DESeq2 package.²⁶ The microarray data of the GSE31210 and GSE50081 data sets were downloaded from the Gene Expression Omnibus²⁷ and normalized by quantile normalization. Both the GSE31210 and GSE50081 data sets were based on the GPL570 platform.

Pyroptosis-related genes were retrieved from the Molecular Signatures Database of gene set enrichment analysis (GSEA²⁸). In addition, we extracted other pyroptosis-related genes from the literature.^29-32 A sum of 49 pyroptosis genes expressed in LUAD were identified and listed in the Data Supplement (Table S1). According to gene annotations in GENCODE,³³ lncRNAs from three data sets were extracted and intersected, and 3,394 common lncRNAs were obtained.

Pyroptosis-Related LncRNAs

The differential expression of pyroptosis genes (DEPGs) was ascertained via T-test (P < .05) on processed TCGA count data for normal and tumor samples using the ggpubr package. The identification of pyroptosis-related lncRNAs was based on correlation analysis between DEPGs and lncRNA expression profiles in TCGA-LUAD samples via the limma package.³⁴ LncRNAs with an absolute value of correlation >0.3 and P < .001 were defined as pyroptosis-related lncRNAs.³⁵

Differentially Expressed Pyroptosis-Related LncRNAs

Differential expression analysis was performed using the TCGA count data. Genes with count values of at least one in at least 50% of samples were retained. The differentially expressed pyroptosis-related lncRNAs (DEPlncRNAs) between normal and tumor samples were identified using the DESeq2 package.²⁶ Fold change >1.5 or <2/3 and adjusted P < .05 were set as the cutoff criteria.³⁶

Univariate Cox Regression Analysis

We conducted univariate Cox regression analysis on TCGA-LUAD, GSE50081, and GSE31210 data sets to identify DEPlncRNAs with stable prognostic significance. The survival package was used for univariate Cox regression analysis. DEPlncRNAs with both P < .05 and those with a consistent hazard ratio direction were chosen.

Construction and Validation of a Pyroptosis-Derived LncRNA Signature

The processed TCGA-LUAD data were used as the training set to construct the prognostic signature. Stepwise Cox regression,³⁷ ridge regression,³⁸ least absolute shrinkage and selection operator (LASSO),³⁹ and random forest⁴⁰ were used to develop the prognostic risk signature via the survival, glmnet, and randomForestSRC packages, respectively. The optimal method was selected through receiver operating characteristic (ROC) and decision curve analysis (DCA).⁴¹ The prognostic risk signature was used to calculate the risk score for each patient in both the training and validation sets. All samples were arbitrarily split into high- and low-risk categories on the basis of the median score. Subsequently, we validated the obtained results using GSE50081 and GSE31210, and the specific methods are detailed in the Data Supplement.

Evaluation of Immune Infiltration and Immunotherapy Prediction

The ESTIMATE algorithm in the estimate package of the R software was used. Four scoring forms, namely, the immune score, stromal score, ESTIMATE score, and tumor purity, were calculated and compared between the high- and low-risk groups by using the T test method. The landscape of immune infiltration was examined using the TIMER, MCPcounter, and ssGSEA algorithms. In addition, the CIBERSORT tool was used for the assessment of the content of 22 tumor-infiltrating immune cells. A sum of 47 immune checkpoint genes were gathered from literature sources.⁴² P < .05 was deemed statistically significant.

Biological Function Assay for Representative LncRNA in LUAD Cell Line

Our study selected a representative lncRNA and further investigated the biological function of lncRNA in human LUAD cell line A549, including cell proliferation, wound healing, and trans-well assays. Detailed descriptions of materials, methods and steps were added in the Data Supplement (Methods).

Statistical Analysis and Visualization

Statistical analysis and visualization graphics were performed using R software version 4.3.0. Coexpression correlation analysis between lncRNAs and pyroptosis gene expression profiles was performed using Pearson's correlation. The Wilcoxon test was used for comparing the two groups. All the results with P < .05 were deemed statistically significant.

RESULTS

Identification and Univariate Cox Regression Analysis of the DEPlncRNAs

The study flowchart is presented in Figure 1. The boxplot was used to depict the expression landscape of 49 pyroptosis related genes in the TCGA-LUAD cohort, among which 39 pyroptosis genes were significantly differentially expressed (Fig 2A). A sum of 3,394 overlapping lncRNAs were extracted from the intersection of the TCGA-LUAD, GSE50081, and GSE31210 data sets. Subsequently, 933 pyroptosis-related lncRNAs were identified by correlation analysis (Data Supplement, Table S2). After satisfying the screening criteria, a sum of 530 DEPlncRNAs were identified from the TCGA-LUAD cohort, with 365 showing upregulation and 165 showing downregulation (Fig 2B; Data Supplement, Table S3). Finally, univariate Cox regression analysis indicated that 11 lncRNAs were significantly correlated with overall survival (OS) in both training and validation sets, suggesting that these 11 lncRNAs had stable prognostic significance (Figs 2C-2E). These lncRNAs served as candidate genes for constructing the risk signature.

FIG 1.

The flowchart of this study. CNV, copy-number variation; DCA, decision curve analysis; DEPlncRNA, differentially expressed pyroptosis-related lncRNA; GSEA, gene set enrichment analysis; GSVA, gene set variation analysis; K-M, Kaplan-Meier; LASSO, least absolute shrinkage and selection operator; lncRNA, long noncoding RNA; LUAD, lung adenocarcinoma; ROC, receiver operating characteristics; TCGA, The Cancer Genome Atlas; TMB, tumor mutation burden.

FIG 2.

Differential expression of pyroptosis genes and lncRNAs and univariate Cox regression analysis of lncRNAs (A) Boxplot of pyroptosis genes between normal and tumor tissues. P values are shown as *P < .05; **P < .01; ***P < .001. (B) Volcano plot of pyroptosis-related lncRNAs. (C-E) Forest plot of prognosis-related DEPlncRNAs common to the three data sets (TCGA-LUAD, GSE50081, GSE31210) via univariate Cox regression. DEPlncRNA, differentially expressed pyroptosis-related lncRNA; FDR, false discovery rate; HR, hazard ratio; lncRNA, long noncoding RNA; LUAD, lung adenocarcinoma; TCGA, The Cancer Genome Atlas.

Construction and Validation of the Pyroptosis-Derived LncRNA Signature

We assigned the TCGA-LUAD group as the training data set, using GSE50081 and GSE31210 as validation data sets to refine the selection of prognosis-related pyroptosis lncRNAs and establish a risk signature. We adopted four machine learning approaches to screen out an optimal risk signature. First, we adopted random forest to establish the risk signature. Figure 3A illustrates the connection between the error rate and the quantity of taxonomic trees. The relative importance order of the 11 lncRNA out-of-bag scores is displayed in Figure 3B. LncRNAs with relative importance >0.2 were identified to fit the final signature. Second, LASSO was used. LASSO identified seven pyroptosis-related lncRNAs (RP5-907D15.4, LINC00857, TMPO-AS1, RP11-890B15.2, RP5-1092A3.4, CTC-429P9.3, and LINC00968) that were included in the classifier (Figs 3C and 3D). Third, we applied ridge regression to establish the risk signature. Ridge regression analysis was applied to all 11 pyroptosis-associated lncRNAs when constructing the risk model (Figs 3E and 3F). Ultimately, we used stepwise multivariate Cox regression for the creation of a risk signature. And in step seven, we obtained the minimum akaike information criterion value with a five-variable model (Figs 3G and 3H).

FIG 3.

Construction of the pyroptosis-related lncRNA signature (A) Error rate for the data on the basis of the number of trees. (B) Out-of-bag relative importance values for predictors. (C and D) Determination of the number of factors by LASSO analysis. (E and F) Determination of the number of factors by ridge regression analysis. (G) AIC value of stepwise multivariate Cox regression analysis. (H) Coefficients of the model genes obtained by stepwise multivariate Cox regression analysis. AIC, akaike information criterion; LASSO, least absolute shrinkage and selection operator; lncRNA, long noncoding RNA.

Subsequently, we used DCA to evaluate the clinical efficacy of the four risk signatures. In the 1-, 3-, and 5-year DCA, the net benefit of the random forest–based risk signature was notably superior to that of the other machine learning-based risk signatures (Figs 4A, 4B, and 4D). In addition, the AUCs of the random forest–based risk signature were 0.995, 0.994, and 0.973 for the 1-, 3-, and 5-year ROC curves, respectively (Fig 4G). However, the AUCs of the other risk signatures were all <0.7 (Figs 4C, 4E, and 4F). Consequently, we adopted random forest to develop a seven pyroptosis-derived lncRNA signature (LINC00857, RP5-907D15.4, RP11-303E16.2, TMPO-AS1, LINC00968, RP11-890B15.2, and RP11-932O9.10). The results of the validation set (Data Supplement, Figs S1 and S2) demonstrate that the pyroptosis-based lncRNA signature serves as a distinct prognostic element for patients with LUAD.

FIG 4.

DCA and time-dependent ROC analysis in LUAD. (A, B, and D) DCA curves of the four models for 1-, 3-, and 5-year overall survival in LUAD. (C) Time-dependent ROC analysis of the stepwise multivariate Cox regression model in the TCGA-LUAD cohort. (E) Time-dependent ROC analysis of the ridge regression model in the TCGA-LUAD cohort. (F) Time-dependent ROC analysis of the LASSO model in the TCGA-LUAD cohort. (G) Time-dependent ROC analysis of the random forest model in the TCGA-LUAD cohort. (H) Time-dependent ROC analysis of the random forest model in the GSE50081 cohort. (I) Time-dependent ROC analysis of the random forest model in the GSE31210 cohort. AUC, area under the curve; DCA, decision curve analysis; LASSO, least absolute shrinkage and selection operator; LUAD, lung adenocarcinoma; ROC, receiver operating characteristics; TCGA, The Cancer Genome Atlas.

Biological Function Assays for Representative LncRNA in LUAD Cell Line

Our study selected a representative lncRNA TMPO-AS1 to test the biological function in LUAD cell line. Two specific siRNAs (Si1 and Si2) were designed to target and downregulate the expression of TMPO-AS1 in the A549 cell line. The wound healing assay showed that suppressing TMPO-AS1 hindered the migration of LUAD cells (Fig 5A). Besides, trans-well assays demonstrated that the migratory and invasive performances were decreased with the reduction of TMPO-AS1 (Fig 5B). Moreover, the growth curves of CCK8 indicated that downregulated TMPO-AS1 inhibited the proliferation power of LUAD cells (Fig 5C). Taken together, TMPO-AS1 facilitated the proliferation, invasion, and metastasis of LUAD.

FIG 5.

The result of biological function assays for representative lncRNA in LUAD cell line. (A) Wound healing, (B) trans-well, (C) CCK8 assays. LncRNA, long noncoding RNA; LUAD, lung adenocarcinoma; NC, negative control; OD, optical density.

Evaluation of Immune Infiltration and Immunotherapy Prediction

First, we used the ESTIMATE algorithm to compare the extent of immune cell infiltration between the high- and low-risk groups. The tumor purity was significantly higher in the group with a higher risk (Fig 6A; P < .05). Conversely, both the ESTIMATE score and immune score were significantly lower in the group with a higher risk (Fig 6B; P < .05; Fig 6C; P < .01). Although the stromal score was also lower in the high-risk group, the difference was not statistically significant (Fig 6D).

FIG 6.

Analysis of immune infiltration and immune checkpoints in LUAD. (A) Violin plot of tumor purity in the high- and low-risk groups. (B) Violin plot of the ESTIMATE score in the high- and low-risk groups. (C) Violin plot of the immune score in the high- and low-risk groups. (D) Violin plot of stromal scores in the high- and low-risk groups. (E) Heatmaps of immune cell components in the high- and low-risk groups. (F) Boxplot of immune cell fractions in the high- and low-risk groups. (G) The proportions plot of 22 immune cell subtypes in the high- and low-risk groups. (H) Boxplot of immune checkpoint gene expression profiles in the high- and low-risk groups. P values are shown as *P < .05; **P < .01; ***P < .001. LUAD, lung adenocarcinoma.

Furthermore, the heatmap exhibited the landscape of immune infiltration of the high- and low-risk groups, suggesting that the low-risk group had a greater immune cell infiltration (Fig 6E). The boxplot depicted the fraction of 22 TIIC subtypes in the groups with high- or low-risk scores (Fig 6F). The percentages of M0 macrophages (P < .001), activated mast cells (P < .05), resting NK cells (P < .01), and activated memory CD4⁺ T cells (P < .01) were significantly elevated in patients with a higher risk. However, the proportions of resting dendritic cells (P < .01), resting mast cells (P < .001), and resting memory CD4⁺ T cells (P < .001) were significantly higher in the group with a lower risk. Figure 6G shows the fraction of 22 TIIC subtypes in the groups with different risk scores.

Finally, we presented the expression profiles of 45 immune checkpoint genes in the groups with high- or low-risk scores (Fig 6H). The expression levels of the BTLA, CD200R1, CD27, CD28, CD40LG, CD48, IDO2, KIR3DL1, LAIR1, NRP1, TNFRSF14, TNFRSF25, and TNFSF15 genes were notably elevated in the group with low risk. However, only the expression of the CD274, CD276, CD70, TNFRSF9, TNFSF4, and TNFSF9 genes were considerably elevated in the group with high risk.

DISCUSSION

The prognostic signature of pyroptosis genes has recently been reported in cancers such as gastric cancer,¹¹ esophageal adenocarcinoma,⁴³ and ovarian cancer.⁴⁴ In particular, Lin et al⁴⁵ established a pyroptosis gene-related prognostic signature for LUAD and identified that the lncRNA KCNQ1OT1/Mir-335-5p/NLRP1/NLRP7 axis as regulators of LUAD progression. In addition, the signatures of lncRNAs related to immunity, iron metabolism, and ferroptosis have also been reported.^46-48 However, it is important to highlight that despite the extensive research on pyroptosis-related lncRNAs, including studies within the context of LUAD, our research stands out because of its unique significance. What sets our study apart is the integration of multiple algorithms, providing a more reliable and accurate prognostic prediction model. Additionally, we explore the biological functions of one lncRNA (TMPO-AS1), providing direction for further research. This research serves as a foundation for further investigations into the intricate interplay between pyroptosis-related lncRNAs and LUAD progression, with potential implications for the development of targeted therapies.

In this study, lncRNAs expressed in all three independent cohorts were selected. An analysis of correlations was carried out to discover lncRNAs associated with pyroptosis. To improve the sensitivity of gene signature detection, we selected lncRNAs that exhibited significant differential expression within the TCGA-LUAD group for further analysis. To improve the stability of the prognostic signature, we included lncRNAs with prognostic significance in three independent cohorts for model construction. To obtain the best prognostic model, we adopted four methods to construct the prognostic model. As the model constructed by random forest had the best discrimination and net benefit rate, we finally adopted random forest to construct the prognostic risk signature. LncRNAs with relative importance >0.2 were selected for the prognostic signature. Moreover, this pyroptosis-derived lncRNA signature also showed strong predictive power in two independent validation sets. The Kaplan-Meier analysis of OS and relapse-free survival (RFS) suggested that the high-risk group exhibited a notably poor prognosis and was more prone to relapse in all three cohorts. In addition, multivariate prognostic analysis indicated that the pyroptosis-derived lncRNA signature was an independent prognostic factor in OS and RFS of LUAD in three independent cohorts.

TMPO-AS1 is one of the seven pyroptosis-related lncRNAs. Our study findings demonstrate that TMPO-AS1 boosts the proliferation, invasion, and metastasis of LUAD cells, in keeping with prior investigations.⁴⁹ Study shows that TMPO-AS1 promoted NSCLC proliferation and metastasis via the miR-204-3p/ERBB2 axis.⁴⁹ Further investigation into the regulatory effects of these seven lncRNAs on pyroptosis in LUAD is warranted.

A growing number of studies have confirmed that immune infiltrates influence the cancer progression.⁵⁰ Liu et al⁵¹ reported that increased infiltration of M0 macrophages in early LUAD was associated with poor prognosis. Activation of lung mast cells promoted angiogenesis and inflammation.⁵² In addition, activated mast cells were positively associated with cancer-associated cachexia.⁵³ Li et al⁵⁴ reported that the activation or rest of mast cells and CD4⁺ memory T cells in LUAD was significantly different between ever- and never-smokers, and the activation of these two immune cell subsets was correlated with a poor survival outcome. By contrast, Zhang et al⁵⁵ reported that resting dendritic cells, resting mast cells, and resting CD4⁺ memory T cells were negatively associated with risk scores and prognosis in a LUAD ferroptosis-related gene signature. Our results suggest significantly higher M0 macrophage infiltrates in patients with higher risk and significantly higher resting dendritic cell infiltrates in the other group. Interestingly, the activation and rest of CD4⁺ memory T cells and mast cells exhibited significant differences between the patients with different level of risk score.

Our findings accounted for the poorer prognosis in the high-risk group. This outcome can be attributed to several significant factors (Data Supplement, Fig S4). First, the high-risk group displays notable enrichment in pathways associated with cell proliferation, DNA replication, and DNA repair. This heightened cellular activity may lead to accelerated tumor growth and expansion, whereas the enhanced DNA repair mechanisms could contribute to resistance against treatment-induced damage. Second, our study reveals an enrichment of pathways linked to immune suppression, including the p53 signaling pathway, within the high-risk group. This suggests that the immune system may be less effective at recognizing and eliminating tumor cells in this subgroup, thereby hastening disease progression. Finally, our biological process analysis indicates increased metabolic activity in the high-risk group, particularly in vitamin metabolism. This may imply a greater reliance on obtaining essential nutrients and growth factors by tumor cells, promoting rapid tumor growth. Furthermore, immunotherapy, especially immune checkpoint inhibitors, has been approved for patients with lung cancer, but the efficacy of immunotherapy is still not satisfactory.⁵⁶ Therefore, we screened immune checkpoint genes that were significantly differentially expressed in patients with different level of risk score, providing potential targets for immunotherapy and requiring further study.

The construction of a prognostic risk signature aims to guide rational therapeutic strategies. Therefore, we used the pRRophetic approach to predict the drug treatment sensitivity of various drugs in the groups with high- or low-risk scores (Data Supplement, Fig S3). These drugs included targeted and chemotherapy drugs, both those in clinical use and those in scientific research, as well as both classic and rare drugs. Our aim was to provide a variety of potential treatment options for clinicians and researchers in clinical and future research.

The study of pyroptosis-derived lncRNA signatures is a new approach, which may open a novel direction for future research. However, a few limitations should be acknowledged. The training set and validation set data in our study were all retrieved from publicly accessible databases, and further in-house validation should be collected as test sets. The precise mechanism through which pyroptosis-associated lncRNAs modulate pyroptosis in LUAD warrants further investigation using in vivo and in vitro studies.

In conclusion, we established a pyroptosis-derived lncRNA signature for patients with LUAD by performing extensive and systematic bioinformatics analysis. The pyroptosis-associated lncRNA signature demonstrated robust predictive capabilities in both the training and validation groups and served as an independent prognostic factor. Functional analysis revealed the underlying mechanism of the signature. Drug sensitivity prediction offered potentially effective drugs but needs to be further validated.

DATA SHARING STATEMENT

The original data presented in the study are available for free on the following websites: GEO, https://www.ncbi.nlm.nih.gov/geo/; UCSC Xena database, https://xenabrowser.net/datapages/; cBioPortal, http://www.cbioportal.org/; GSEA, http://www.gsea-msigdb.org/gsea/msigdb/search.jsp; MSigDB, http://software.broadinstitute.org/gsea/index.jsp.

AUTHOR CONTRIBUTIONS

Conception and design: Xinwei Han

Provision of study materials or patients: Shutong Liu

Collection and assembly of data: Yuqing Ren, Lifeng Li, Chunwei Li, Xiaoyong Ge, Xinwei Han

Data analysis and interpretation: Yuhao Ba, Shutong Liu, Zhengpan Wei, Nannan Zhao, Tong Qiao, Yuyuan Zhang, Siyuan Weng, Hui Xu, Xiaoyong Ge, Xinwei Han

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/po/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).

No potential conflicts of interest were reported.