Integrated machine learning survival framework for consensus modeling in a large multicenter cohort of NSCLC resistant to aumolertinib

Integrated machine learning survival framework for consensus modeling in a large multicenter cohort of NSCLC resistant to aumolertinib | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Integrated machine learning survival framework for consensus modeling in a large multicenter cohort of NSCLC resistant to aumolertinib Xiao Wu, Yang Lu, Yongping Li, Annan Zhu, Tingting Wang, Fang Su, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6311145/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 28 Oct, 2025 Read the published version in Scientific Reports → Version 1 posted 10 You are reading this latest preprint version Abstract Patients with advanced non-small cell lung cancer (NSCLC) harboring epidermal growth factor receptor (EGFR) mutations often benefit from third-generation tyrosine kinase inhibitors (TKIs), such as aumolertinib (AUM). However, the development of drug resistance significantly limits the clinical efficacy of AUM. To address this, we established an in vitro model of AUM-resistant cell lines and performed RNA sequencing to identify resistance-associated differentially expressed genes. Using machine learning, we constructed an AUM resistance-related prognostic signature (ARRPS). Our results demonstrated that ARRPS effectively predicts the prognostic risk of patients. Notably, for patients with high ARRPS scores, the addition of CD-437 or TPCA-1 to conventional AUM treatment may help overcome drug resistance. These findings suggest that ARRPS serves as both a prognostic tool and a guide for personalized treatment strategies, potentially optimizing the clinical management of NSCLC patients. Biological sciences/Computational biology and bioinformatics Biological sciences/Computational biology and bioinformatics/Computational models NSCLC Resistance prognostic signature aumolertinib Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Introduction The problem of drug resistance to cancer treatment has long been recognized as a key factor in cancer treatment failure[ 1 ]. For example, resistance to epidermal growth factor receptor (EGFR) tyrosine kinase inhibitors (TKIs) greatly contributes to treatment failure and cancer recurrence in patients with EGFR-mutant non-small cell lung cancer (NSCLC)[ 2 ]. Among the multiple mechanisms of resistance to EGFR-TKIs, T790M, a secondary mutation in EGFR exon 20, is thought to be the most prevalent cause of acquired resistance to first- and second-generation EGFR-TKIs, and a number of third-generation EGFR TKIs have been designed to irreversibly target EGFRs with both T790M-resistant mutations and EGFR-activating mutations. Among them, aumolertinib (AUM) is a representative compound widely approved for standard second-line treatment of EGFR T790M-mutant NSCLC and first-line treatment of EGFR-activating mutant NSCLC, with significant preclinical and clinical efficacy but inevitably acquired resistance. Multiple mechanisms have been reported, including EGFR mutations[ 2 ], activation of alternative pathways[ 3 ], and aberrant downstream signaling[ 4 ]. However, the investigation of AUM resistance mechanisms remains an urgent problem, which simultaneously attracts people to explore new resistance mechanisms and find potential therapeutic strategies for AUM-resistant NSCLC. In order to be able to better find the mechanisms of resistance to AUM as well as to have an accurate prediction of the prognosis of NSCLC patients, we developed and validated a robust signature consisting of a small set of genes using large-scale data. Using 10 machine learning algorithms, they were transformed into 100 combinations that were further executed in 5 independent cohorts. The best-performing ARRPS among the 100 model types was finally identified. In this model, patients with a high ARRPS showed a higher risk of death, relapse, and disease progression[ 5 ]. The ARRPS was independent of traditional clinical signatures (e.g., American Joint Committee on Cancer (AJCC) staging) and molecular signatures (e.g., TP53 mutations) and additionally demonstrated superior predictive performance in these variables. Comparisons with published signatures show favorable performance of ARRPS. Additionally, patients with high ARRPS scores may benefit from 2 potential therapeutic agents: CD-437 and TPCA-1. Methods 1.1. Induction of AUM-resistant NSCLC cell lines At the beginning, we modeled acquired resistance to AUM by deriving polyclonal acquired resistant cell lines on the basis of stepwise dose escalation over a period of six months followed by maintenance in 10µM of drug over 6 weeks[ 6 ], that is well representative of the clinic, and performed RNA whole transcriptome sequencing of this model using parental cells as a control to screen for AUM-resistant genes. 1.2. Generation of ARRPS We extracted data from the Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/ ) (GSE50081, GSE30219, GSE72094 and GSE132313) and The Cancer Genome Atlas (TCGA, http://portal.gdc.cancer.gov/ ) (TCGA-LUAD) in which five independent lung adenocarcinoma datasets were retrospectively extracted. Sample selection criteria were as follows (1) primary cancer tissues that had not received any adjuvant therapy and (2) survival information and RNA expression data were available. In the TCGA-LUAD project, we obtained single nucleotide variants (SNV) and copy number variants (CNV) from the TCGA portal. Prognostic genes were screened using Cox regression analysis, and genes with an adjusted P-value < 0.05 and a consistent hazard ratio (HR) direction in the four cohorts were considered to be robust genes associated with overall survival (OS). The intersection of differentially expressed genes in AUM-resistant cell lines with the above obtained genes was used to identify robust genes associated with AUM resistance and OS. To develop a consensus ARRPS with high accuracy and stable performance, we integrated 10 machine learning algorithms including Random Survival Forest (RSF), Elastic Network (Enet), Lasso regression, Ridge regression, Stepwise Cox regression, Cox boosting, Cox partial least squares regression (plsRcox), Supervised Principal Component Analysis (SuperPC), Generalized Augmented Regression Modeling (GBM) and Survival Support Vector Machine (survival-svm). In this study, the TCGA-LUAD cohort was used as the training set, while the other four cohorts were used as the test set. We calculated the c-index of each feature in all cohorts, where the feature with the highest average c-index was considered as the best feature[ 7 ]. 1.3. Validating the prognostic value of ARRPS Patients in the five cohorts were categorized into high and low ARRPS subgroups based on the median ARRPS score. Kaplan-Meier curves and Cox regression analysis were used to assess the prognostic value of ARRPS. Subjects' work characteristics (ROC) curves were plotted to assess the predictive accuracy of ARRPS. Subsequently, we calculated the C-index to compare the superiority of ARRPS with a variety of common clinical characteristics (including gender, age, TNM stage and grading) in predicting overall survival (OS) in patients with LUAD. 1.4. Collection and calculation of LUAD published signatures To compare the predictive performance of ARRPS with these published features, we searched PubMed on October 1, 2024, for published articles on prognostic models[ 8 – 11 ]. Subsequently, we calculated risk scores for the 5 cohorts based on the genes and coefficients provided in the articles and assessed their prognostic performance in LUAD by one-way Cox analysis and C-index. 1.5. Collection and calculation of LUAD published signatures To compare the predictive performance of ARRPS with these published signatures, we searched PubMed for published articles on prognostic models as of October 1, 2024 Subsequently, we calculated risk scores for the 5 cohorts based on the genes and coefficients provided in the articles and evaluated their prognostic performance in LUAD by a one-way Cox analysis and C-index. 1.6. Genomic alteration landscape To compare the predictive performance of ARRPS with these published features, we searched PubMed on October 1, 2024, for published articles on prognostic models. Subsequently, we calculated risk scores for the 5 cohorts based on the genes and coefficients provided in the articles and assessed their prognostic performance in LUAD by one-way Cox analysis and C-index. ( https://cloud.genepattern.org/gp/pages/index.jsf ). 1.7. Gene set variation analysis (GSVA) Gene sets were obtained from the Molecular Signatures Database (MSigDB, https://www.gsea-msigdb.org/gsea/msigdb/index.jsp ). Gene set variance analysis (GSVA) methods were used to quantify the values of each metric in each sample. The limma package was used to identify pathways that were significantly altered between the high and low ARRPS groups[ 12 ]. 1.8. Comprehensive analyses based on immune cell infiltration We explored the correlation between ARRPS scores and immune infiltration. We obtained immune infiltration data for TCGA-LUAD from the TIMER 2.0 web server ( http://timer.cistrome.org/ )[ 13 ], which covers immune infiltration data obtained by seven methods, namely, CIBERSORT, CIBERSORT-ABS, EPIC, MCPCOUNTER, QUANTISEQ, TIMER and XCELL[ 14 – 17 ]. 1.9. Development and validation of potential therapeutic agents We obtained drug sensitivity data for ARRPS from the CTRP and PRISM recombinant datasets. Based on the Wilcoxon rank-sum test, we analyzed the difference in drug response between the high ARRPS group (top 10%) and the low ARRPS group (bottom 10%) and set a threshold of log2FC < 0.05 to identify compounds with lower AUC values in the high ARRPS group. Next, we applied Spearman's correlation analysis to further screen compounds whose AUC values were negatively correlated with ARRPS (setting a threshold R < − 0.3)[ 18 – 21 ]. 1.10. Tumor xenograft models All animals were housed according to the guidelines of the relevant institutions and approved by the Animal Care Committee of Bengbu Medical College. The BALB/c nude mice (male, 6 weeks old) were purchased from Unilever (Beijing, China). The BALB/c nude mice were kept in a specific pathogen-free (SPF)-grade animal laboratory. Nude mice were injected subcutaneously with 5×10 6 cells on the right side and randomly grouped after the tumors grew to 100–150 cubic millimeters. Tumor volume was measured every two days. Tumor volume was calculated as length × width 2 × 0.5. After three weeks, all mice were executed, and tumors were harvested and weighed. 1.11 Cell lines and culture conditions The HCC827 and H1975 cell line was sourced from American Type Culture Collection (Manassas, VA, USA). All media were supplemented with 10% fetal bovine serum (FBS; Gibco), 100 U/ml penicillin, and 100 µg/ml streptomycin. The cells were incubated at 37°C in a humidified 5% CO2 atmosphere. The authenticity of the human cell lines used in this study was ascertained through short tandem repeat (STR) profiling. 1.12 RNA sequencing RNA Sequencing and data collection were performed by Wei Huan Biotechnology (China) Co. Total RNA was extracted using TRIzol® reagent (Invitrogen, #15596018) according to the manufacturer's instructions, and the samples were sent to Shanghai Wei Huan Biotechnology (Shanghai, China) for RNA purification, reverse transcription, library construction, and sequencing. After quality control of raw reads, clean reads were mapped using HISAT2 software and further assembled using StringTiein. Cellular pathway analysis was performed using GSEA. 1.13 Statistical analysis Statistical analyses were performed using GraphPad Prism 8, and data are expressed as mean ± SEM. Two-tailed unpaired Student's t-test and one- or two-way ANOVA were used to analyze differences between groups. Results 2.1. Construction of the ARRPS The ARRPS construction process is shown in Fig. 1 . HCC827 cells resistant to AUM were obtained through six months of induction with a resistance index of 3.35, which is a good representation of the clinical situation (Fig. 2 A). We performed RNA whole transcriptome sequencing of this model using parental cells as a control to screen for AUM resistance genes. We found 2987 up-regulated genes and 3410 down-regulated genes in resistant HCC827 cells (Fig. 2 B-C). Through analysis, we obtained 20 genes significantly associated with overall survival and alectinib resistance, including ANLN, ASPM, BUB1B, CDC20, CDC25C, CDC6, CDCA2, CDCA3, CDKN3, CENPE, CENPF, FOXM1, GINS4, KIF2C, MELK, MYBL2, NEK2, PRC1, SHCBP1, and TOP2A.Based on the 10-fold cross-validation framework, we used 10 machine learning algorithms and their constituent combinations for a total of 100 methods for modeling. The results show that the combined algorithm of lasso + RSF (including 12 genes) has the highest average C-index and is considered the optimal modeling approach (Fig. 2 D). We found that these 12 genes were highly expressed in cancer tissues (Fig. 2 E), and that these 12 genes had a high ability to discriminate between diagnosing LUAD and non-tumors (Fig. 2 F). To evaluate the prognostic performance of ARRPS, we categorized LUAD patients into high and low ARRPS groups according to the median value. KM curves showed that in the training cohort, mortality was significantly higher in the high-risk group than in the low-risk group (TCGA-LUAD, Fig. 2 G), and the other four validation cohorts showed the same trend of patients in the high-ARRPS group having worse OS (Fig. 2 H-K). For the five cohorts, the ROC curves demonstrated the high ability of ARRPS to predict patient OS (Fig. 3 A-E). Subsequently, comparing the ability of ARRPS with multiple clinical features in predicting OS in patients with LUAD showed that ARRPS was significantly more accurate than these features, including age, gender, smoking BI, TNM staging, KRAS, EGFR, STK11, and TP53 mutations or new events (Fig. 3 F-J). Meanwhile, we also investigated other clinical outcomes of TCGA-LUAD, including DSS, PFI, and DFI, and the results showed that patients with high ARRPS had worse DSS, PFI, and DFI, and ARRPS was an independent risk factor for DSS, PFI, and DFI in TCGA-LUAD. ARRPS also had high accuracy in predicting TCGA-LUAD DSS, PFI, and DFI (Fig. 4 A-J). In multiple cohorts, as described above, our ARRPS model shows robust pre-prediction performance. To further verify whether ARRPS has good predictive performance compared to published signatures, we collected 40 published signatures. We then calculated the c-index for each of the 40 signatures in each of the five cohorts and compared them to ARRPS. The results show that the c-index of ARRPS is significantly higher than that of other signatures in the TCGA-LUAD, GSE30219, GSE50081, and GSE72094 cohorts, except for the GSE13213 cohort (Fig. S1 A). This suggests that ARRPS maintains a relatively strong predictive power even when compared to other signatures. Taken together, the results of Kaplan-Meier survival analyses, Cox regression analyses, ROC curves, and C-indexes of the five cohorts consistently showed that the ARRPS accurately and robustly predicted the prognosis of patients with LUAD, suggesting that the ARRPS may become an attractive tool to serve clinical practice. 2.2. ARRPS shows substantial correlation with immune-relate To explore the immune status reflected by the ARRPS profile score, we analyzed the differences between high and low ARRPS profile scores in immune-infiltrating cells, which were concentrated in T cells, B cells, macrophages, and NK cells (Fig. 5 A). 2.3. Differences in pathway enrichment between high and low ARRPS Signaling pathway enrichment shows the differences involved in the groups with high and low ARRPS feature scores, and the results indicate that the group with high ARRPS feature scores mainly focuses on many signaling pathways that promote tumor growth, such as cell cycle and DNA damage repair and metabolism-related pathways, such as KEGG_DRUG_METABOLISM_CYTOCHROME_P450, KEGG_ARACHIDONIC_ACID_METABOLISM, KEGG_LINOLEIC_ACID_METABOLISM, KEGG_ETHER_LIPID_METABOLISM, KEGG_ALPHA_LINOLENIC_ACID_METABOLISM, and KEGG_TAURINE_AND_HYPOTAURINE_METABOLISM (Fig. 5 B-E). 2.4. Analysis of genomic variation between patients in the high and low ARRPS groups Study of the TCGA-LUAD SNV data found significant differences in SNV between patients in the high and low ARRPS groups. Patients in the high ARRPS group had higher mutation frequencies in NPAP1, TSHZ3, PCDHB8, PTPRT, GRM8, ANK1, KLHL4, TRPC4, COL4A2, LAMA4, and BCHE, whereas patients in the low ARRPS group had higher mutation frequencies in CPS1, COL19A1, PLCB4, etc. In terms of CNV, several loci had high CNV frequencies in patients in the high ARRPS group, including 3q26.2(Amp), 12q15(Amp), 17p13.3(Del), 17p11.2(Del), 6q26(Del), 18q22.2(Del), 18q21.1(Del), 22q13.32(Del), 21q21.1(Del), 1p13.1(Del), and 1p36.11(Del) (Fig. S5A). ZBTB4, XAF1, ZMYM2, FOXO1, TNFSF11, ZNF236, BCL2, MAP2K3RBFA, PIK3C3, RBX1, ZNF24, NCF4, SLC5A1 CNV events with more frequencies in patients in the high ARRPS group (Fig. S2A). In addition, we identified multiple gene pairs with SNV co-occurence, including CDH10-MYLK, MRC2-NPAP1, and KIF1A-TSHZ3 (Fig. S3A). 2.5. CD-437 and TPCA-1 are drug candidates for patients with high ARRPS Patients with a high ARRPS have a poorer prognosis and a higher risk of developing resistance to AUM. In order to be able to use the ARRPS as a reliable prognostic risk assessment toolkit, we performed a screening of effective therapeutic agents. The drug screening flowchart is shown in (Fig. 6 A). First, differential drug response analysis was performed on the two groups of patients with high and low ARRPS scores to screen compounds of high ARRPS scores patients with low AUC values, and then Spearman correlation analysis was performed to screen compounds with significant negative correlation between AUC values and ARRPS (r < -0.3, Log 2 FC < -0.01), and finally 15 compounds were screened, including 6 CTRP derived compounds (3-Cl-AHPC, cabozantinib, CD-437, mitomycin, nakiterpiosinand StemRegenin 1) and 9 PRISM-derived compounds (broxyquinoline, combretastatin-A-4, deferasirox, frentizole, imiquimod, napabucasin, norepinephrine, tanespimycin, and TPCA-1)( Fig. 6 B-E). Finally, by CMap database and searching the previous publications, we found that CD-437 and TPCA-1, with high CMap scores and few studies in LUAD, are drugs with high potential for the treatment of LUAD (Fig. 6 F-G). Both CD-437 and TPCA-1 alone significantly inhibited the viability of drug-resistant cell lines. We then sought to combine both with AUM to explore the sensitivity of drug-resistant cell lines to AUM. We used the CCK8 assay to detect the antiproliferative effects of CD-437 and ASK120067 in combination with AUM on drug-resistant cell lines and expressed them as a dose-response matrix and a synergy scoring matrix (Fig. 7 A-C). Synergy scores were calculated by SynergyFinder using the HSA model [ 22 , 23 ]. It was found that this combination exerted a stronger antiproliferative effect. The same therapeutic effect was subsequently detected using clone formation (Fig. 7 D), not also. In addition to that, this combination effect significantly inhibited the growth of xenograft tumors in nude mice (Fig. 7 E-J, Fig. S6A-B). In conclusion, these results of ours surface that the two compounds screened by high and low ARRPS can effectively inhibit the growth of NSCLC. Discussion Resistance to EGFR-TKI has long been a major impediment to targeted therapy for EGFR-mutant NSCLC[ 24 – 26 ]. The conventional solution is to increase the dose or add additional other chemotherapeutic agents. However, this is not a good solution. Immunotherapy has a more expensive cost and serious adverse effects. Therefore, exploring a new biomarker to effectively treat EGFR mutant NSCLC is also necessary. To achieve this goal, we built multiple prognostic models based on AUM resistance-associated differential genes using a combination of 10 machine learning algorithms. Subsequently, the 12-gene ARRPS, derived from the combination of lasso + GBM, emerged as superior. As a linear regression model, Lasso regression selects features during model training and can be used for feature selection and dimensionality reduction. It also improves the generalization ability of the model and avoids overfitting. However, in this study, its direct application of this model produces not much accuracy, achieving an average c-index of 0.655. GBM has the advantages of good interpretability and wide applicability, but it lacks the inherent ability to downscale and carries the risk of overfitting. To address these challenges, we combine the advantages of CoxBoost and GBM. It turns out that this combined approach proves to be the most effective, outperforming other models in our study. In our study, ARRPS was found to be a reliable prognostic risk assessment tool. This also gives us a greater interest in the screening of therapeutic agents for patients with high ARRPS. It turned out that CD-437 and TPCA-1 had good ability to inhibit drug-resistant cell lines, and we combined them with AUM, which also showed good synergistic anti-tumor ability, both in vivo and in vitro. In clinical practice, we can obtain biopsy samples from diagnosed NSCLC patients and perform RNA-seq analysis to elucidate their transcriptome profiles. Subsequently, we can reconcile these profiles with the meta-LUAD cohort and calculate the ARRPS. Based on the ARRPS threshold, NSCLC patients can be categorized into a low ARRPS group and a high ARRPS group. Conventional therapy is sufficient for patients in the low ARRPS group, while those in the high ARRPS group may benefit from the combination of AUM with CD-437 and TPCA-1. However, the limitations of our study are worth mentioning. Our retrospective approach has its drawbacks, emphasizing the need for subsequent prospective trials to validate the predictive ability of ARRPS. OTS167 inhibits MELK, but targeting MELK deserves to be explored[ 27 – 30 ]. Although our findings, combined with animal studies, suggest the efficacy and safety of CD-437 and TPCA-1, rigorous evaluation by additional preclinical models, such as OSA patient-derived xenografts and transgenic mouse models, is necessary. In summary, this study introduced ARRPS as a potential tool for identifying high-risk subgroups of patients with LUAD and identified therapeutic candidates CD-437 and TPCA-1 targeting high ARRPS. These results provide a promising therapeutic avenue and represent a potential complementary treatment for high-risk LUAD patients. Abbreviations NSCLC Non-small cell lung cancer EGFR-TKI epidermal growth factor receptor-tyrosine kinase inhibitor AUM Aumolertinib ARRPS Aumolertinib resistance-related prognostic signature LUAD lung adenocarcinoma. Declarations Acknowledgements Not applicable. A uthor contributions Xiao Wu performed most of the experiments, Yongxia Chen and Guolei Song analyzed data, Yang Lu, Yongping Li and Annan Zhu contributed to experiments. Tingting Wang, Zishu Wang and Fang Su designed experiments, supervised the study, and wrote the manuscript. Funding This study was supported by Natural Science key project of Bengbu Medical College (No. 2021byzd064). Data availability The datasets generated and/or analyzed during the current study are available in the NCBI BioProject repository under accession number PRJNA1248195 (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1248195). The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request. Ethics approval and consent to participate The animal experiments were conducted in accordance with ARRIVE 2.0 guidelines (Animal Research: Reporting of In Vivo Experiments, https://arriveguidelines.org). All animal experimental procedures were approved by the Experimental Animal Teaching and Research Committee of Bengbu Medical University (2023-522). All methods were performed in accordance with the relevant guidelines and regulations Consent for publication All authors consent to publication of this article. Declaration of competing interest The authors declare no conflict of interest. References Passaro A, Jänne PA, Mok T, Peters S: Overcoming therapy resistance in EGFR-mutant lung cancer. Nat Cancer 2021, 2: 377-391. Zhang T, Qu R, Chan S, Lai M, Tong L, Feng F, Chen H, Song T, Song P, Bai G, et al: Correction: Discovery of a novel third-generation EGFR inhibitor and identification of a potential combination strategy to overcome resistance. Mol Cancer 2023, 22: 139. Pan Z, Wang K, Wang X, Jia Z, Yang Y, Duan Y, Huang L, Wu ZX, Zhang JY, Ding X: Cholesterol promotes EGFR-TKIs resistance in NSCLC by inducing EGFR/Src/Erk/SP1 signaling-mediated ERRα re-expression. Mol Cancer 2022, 21: 77. Shih PC: The role of the STAT3 signaling transduction pathways in radioresistance. Pharmacol Ther 2022, 234: 108118. Liu Z, Liu L, Weng S, Guo C, Dang Q, Xu H, Wang L, Lu T, Zhang Y, Sun Z, Han X: Machine learning-based integration develops an immune-derived lncRNA signature for improving outcomes in colorectal cancer. Nat Commun 2022, 13: 816. Shah KN, Bhatt R, Rotow J, Rohrberg J, Olivas V, Wang VE, Hemmati G, Martins MM, Maynard A, Kuhn J, et al: Aurora kinase A drives the evolution of resistance to third-generation EGFR inhibitors in lung cancer. Nat Med 2019, 25: 111-118. Liu Z, Guo C, Dang Q, Wang L, Liu L, Weng S, Xu H, Lu T, Sun Z, Han X: Integrative analysis from multi-center studies identities a consensus machine learning-derived lncRNA signature for stage II/III colorectal cancer. EBioMedicine 2022, 75: 103750. Yang Z, Yan G, Zheng L, Gu W, Liu F, Chen W, Cui X, Wang Y, Yang Y, Chen X, et al: YKT6, as a potential predictor of prognosis and immunotherapy response for oral squamous cell carcinoma, is related to cell invasion, metastasis, and CD8+ T cell infiltration. Oncoimmunology 2021, 10: 1938890. Wang L, Liu Z, Zhu R, Liang R, Wang W, Li J, Zhang Y, Guo C, Han X, Sun Y: Multi-omics landscape and clinical significance of a SMAD4-driven immune signature: Implications for risk stratification and frontline therapies in pancreatic cancer. Comput Struct Biotechnol J 2022, 20: 1154-1167. Tan Z, Lei Y, Xu J, Shi S, Hua J, Zhang B, Meng Q, Liu J, Zhang Y, Wei M, et al: The value of a metabolic reprogramming-related gene signature for pancreatic adenocarcinoma prognosis prediction. Aging (Albany NY) 2020, 12: 24228-24241. Yuan Q, Ren J, Li L, Li S, Xiang K, Shang D: Development and validation of a novel N6-methyladenosine (m6A)-related multi- long non-coding RNA (lncRNA) prognostic signature in pancreatic adenocarcinoma. Bioengineered 2021, 12: 2432-2448. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES, Mesirov JP: Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A 2005, 102: 15545-15550. Li T, Fu J, Zeng Z, Cohen D, Li J, Chen Q, Li B, Liu XS: TIMER2.0 for analysis of tumor-infiltrating immune cells. Nucleic Acids Res 2020, 48: W509-w514. Chrétien S, Zerdes I, Bergh J, Matikas A, Foukakis T: Beyond PD-1/PD-L1 Inhibition: What the Future Holds for Breast Cancer Immunotherapy. Cancers (Basel) 2019, 11 . Wang J, Sanmamed MF, Datar I, Su TT, Ji L, Sun J, Chen L, Chen Y, Zhu G, Yin W, et al: Fibrinogen-like Protein 1 Is a Major Immune Inhibitory Ligand of LAG-3. Cell 2019, 176: 334-347.e312. Janakiram M, Chinai JM, Zhao A, Sparano JA, Zang X: HHLA2 and TMIGD2: new immunotherapeutic targets of the B7 and CD28 families. Oncoimmunology 2015, 4: e1026534. Bindea G, Mlecnik B, Tosolini M, Kirilovsky A, Waldner M, Obenauf AC, Angell H, Fredriksen T, Lafontaine L, Berger A, et al: Spatiotemporal dynamics of intratumoral immune cells reveal the immune landscape in human cancer. Immunity 2013, 39: 782-795. Yang C, Zhang H, Chen M, Wang S, Qian R, Zhang L, Huang X, Wang J, Liu Z, Qin W, et al: A survey of optimal strategy for signature-based drug repositioning and an application to liver cancer. Elife 2022, 11 . Yang K, Dinasarapu AR, Reis ES, Deangelis RA, Ricklin D, Subramaniam S, Lambris JD: CMAP: Complement Map Database. Bioinformatics 2013, 29: 1832-1833. Maeser D, Gruener RF, Huang RS: oncoPredict: an R package for predicting in vivo or cancer patient drug response and biomarkers from cell line screening data. Brief Bioinform 2021, 22 . Geeleher P, Cox N, Huang RS: pRRophetic: an R package for prediction of clinical chemotherapeutic response from tumor gene expression levels. PLoS One 2014, 9: e107468. Ianevski A, Giri AK, Aittokallio T: SynergyFinder 3.0: an interactive analysis and consensus interpretation of multi-drug synergies across multiple samples. Nucleic Acids Res 2022, 50: W739-w743. Zheng S, Wang W, Aldahdooh J, Malyutina A, Shadbahr T, Tanoli Z, Pessia A, Tang J: SynergyFinder Plus: Toward Better Interpretation and Annotation of Drug Combination Screening Datasets. Genomics Proteomics Bioinformatics 2022, 20: 587-596. Fu Y, Zhang Y, Lei Z, Liu T, Cai T, Wang A, Du W, Zeng Y, Zhu J, Liu Z, Huang JA: Abnormally activated OPN/integrin αVβ3/FAK signalling is responsible for EGFR-TKI resistance in EGFR mutant non-small-cell lung cancer. J Hematol Oncol 2020, 13: 169. Nilsson MB, Yang Y, Heeke S, Patel SA, Poteete A, Udagawa H, Elamin YY, Moran CA, Kashima Y, Arumugam T, et al: CD70 is a therapeutic target upregulated in EMT-associated EGFR tyrosine kinase inhibitor resistance. Cancer Cell 2023, 41: 340-355.e346. Peng S, Wang R, Zhang X, Ma Y, Zhong L, Li K, Nishiyama A, Arai S, Yano S, Wang W: EGFR-TKI resistance promotes immune escape in lung cancer via increased PD-L1 expression. Mol Cancer 2019, 18: 165. Tang Q, Li W, Zheng X, Ren L, Liu J, Li S, Wang J, Du G: Correction: MELK is an oncogenic kinase essential for metastasis, mitotic progression, and programmed death in lung carcinoma. Signal Transduct Target Ther 2024, 9: 186. Li B, Yan J, Phyu T, Fan S, Chung TH, Mustafa N, Lin B, Wang L, Eichhorn PJA, Goh BC, et al: MELK mediates the stability of EZH2 through site-specific phosphorylation in extranodal natural killer/T-cell lymphoma. Blood 2019, 134: 2046-2058. Li Z, Zhou H, Zhai X, Gao L, Yang M, An B, Xia T, Du G, Li X, Wang W, Jin B: MELK promotes HCC carcinogenesis through modulating cuproptosis-related gene DLAT-mediated mitochondrial function. Cell Death Dis 2023, 14: 733. Tang B, Zhu J, Shi Y, Wang Y, Zhang X, Chen B, Fang S, Yang Y, Zheng L, Qiu R, et al: Tumor cell-intrinsic MELK enhanced CCL2-dependent immunosuppression to exacerbate hepatocarcinogenesis and confer resistance of HCC to radiotherapy. Mol Cancer 2024, 23: 137. Additional Declarations No competing interests reported. Supplementary Files 2.supplementary.docx Cite Share Download PDF Status: Published Journal Publication published 28 Oct, 2025 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 03 Jun, 2025 Reviews received at journal 02 Jun, 2025 Reviews received at journal 16 May, 2025 Reviewers agreed at journal 12 May, 2025 Reviewers agreed at journal 09 May, 2025 Reviewers invited by journal 24 Apr, 2025 Editor assigned by journal 24 Apr, 2025 Editor invited by journal 11 Apr, 2025 Submission checks completed at journal 11 Apr, 2025 First submitted to journal 26 Mar, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6311145","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":448017722,"identity":"60632e63-373e-4c00-b74f-7e323f60f95c","order_by":0,"name":"Xiao Wu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA7UlEQVRIiWNgGAWjYBACgwMMDEAkIcfGcLD9xwcDGztitdgY8zEebpCcUZCWTJQWIEhLnMd8vEGa58MhxgaCWm7kGB74UXE4sY3tYIOxjcEBZgb2w0c34NNidiMt4WDPmcPGbTwHG5JzDO7wMfCkpd3AryX5wAHetsOybRIHGw7nGDxjZpDgMSOgJbHh4N+2w4xt8g8bmy0MDjM2ENJiD7TlMG9bmmIbw8FmZgZitFieeZZwWOaMjTEwXtoYewzSktkI+cXgeI7xxzcVEnLyDcefMfz4Y2PHz374GF4tmICNNOWjYBSMglEwCrABAOlKVr0enDj4AAAAAElFTkSuQmCC","orcid":"","institution":"The First Affiliated Hospital of Bengbu Medical University","correspondingAuthor":true,"prefix":"","firstName":"Xiao","middleName":"","lastName":"Wu","suffix":""},{"id":448017723,"identity":"65e0a362-accf-41fc-9a8a-2e610040985f","order_by":1,"name":"Yang Lu","email":"","orcid":"","institution":"Shandong University of Traditional Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Yang","middleName":"","lastName":"Lu","suffix":""},{"id":448017724,"identity":"735f85fe-f60f-4f56-a1b4-e79c78e4b969","order_by":2,"name":"Yongping Li","email":"","orcid":"","institution":"Bengbu Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yongping","middleName":"","lastName":"Li","suffix":""},{"id":448017725,"identity":"10b27032-1560-40f7-92e2-1ce82cec4008","order_by":3,"name":"Annan Zhu","email":"","orcid":"","institution":"Northeastern University","correspondingAuthor":false,"prefix":"","firstName":"Annan","middleName":"","lastName":"Zhu","suffix":""},{"id":448017726,"identity":"9ece7e99-8fed-46a8-80fc-72cc0f98514f","order_by":4,"name":"Tingting Wang","email":"","orcid":"","institution":"The First Affiliated Hospital of Bengbu Medical University","correspondingAuthor":false,"prefix":"","firstName":"Tingting","middleName":"","lastName":"Wang","suffix":""},{"id":448017727,"identity":"757a403b-b7b9-4943-883e-eac87c903ce0","order_by":5,"name":"Fang Su","email":"","orcid":"","institution":"The First Affiliated Hospital of Bengbu Medical University","correspondingAuthor":false,"prefix":"","firstName":"Fang","middleName":"","lastName":"Su","suffix":""},{"id":448017728,"identity":"bd848af1-7120-4cc4-aeef-4f52231a7b44","order_by":6,"name":"Zishu Wang","email":"","orcid":"","institution":"The First Affiliated Hospital of Bengbu Medical University","correspondingAuthor":false,"prefix":"","firstName":"Zishu","middleName":"","lastName":"Wang","suffix":""},{"id":448017729,"identity":"f87aea52-9e37-4c20-b945-006c1171b006","order_by":7,"name":"Yongxia Chen","email":"","orcid":"","institution":"The First Affiliated Hospital of Bengbu Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yongxia","middleName":"","lastName":"Chen","suffix":""},{"id":448017730,"identity":"c831f0bd-1074-4f47-b745-6300494e95e8","order_by":8,"name":"Guolei Song","email":"","orcid":"","institution":"The First Affiliated Hospital of Bengbu Medical University","correspondingAuthor":false,"prefix":"","firstName":"Guolei","middleName":"","lastName":"Song","suffix":""}],"badges":[],"createdAt":"2025-03-26 10:08:06","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6311145/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6311145/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-025-20159-7","type":"published","date":"2025-10-28T15:58:10+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":82122279,"identity":"49cb910d-77b3-4f99-8077-cfbf67678ecb","added_by":"auto","created_at":"2025-05-07 03:27:58","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":439955,"visible":true,"origin":"","legend":"\u003cp\u003eThe flowchart for building ARRPS.\u003c/p\u003e","description":"","filename":"image1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6311145/v1/1ff25b0fbd1b0f67c03d46b9.jpeg"},{"id":82122283,"identity":"a6d7c38e-61d5-474c-b409-8930f8174cdb","added_by":"auto","created_at":"2025-05-07 03:27:58","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":928431,"visible":true,"origin":"","legend":"\u003cp\u003eA. IC50 of HCC827 parental and resistant. B-C. RNA sequencing of HCC827 parental and resistant, differential genes were represented using volcano (B) and heat maps (C). D. The c-index of 100 combinations of machine learning algorithms in 5 test cohorts. E. Expression of 12 genes in LUAD. F. Diagnostic ROC curves for 12 genes. G-K. Kaplan-Meier analysis of TCGA-LUAD, GSE72094, GSE50081, GSE30219 and GSE13213 datasets.\u003c/p\u003e","description":"","filename":"image2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6311145/v1/80dc8d83e3da3447238658fe.jpeg"},{"id":82122280,"identity":"498a7b43-2b52-4401-9f4b-e593054ff267","added_by":"auto","created_at":"2025-05-07 03:27:58","extension":"jpeg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":379052,"visible":true,"origin":"","legend":"\u003cp\u003eA-E. ROC curve analysis of ARRPS over time in the TCGA-LUAD, GSE72094, GSE50081, GSE30219, and GSE13213 cohorts. F-J. C-index when ARRPS is compared with the ability of multiple clinical characteristics to predict OS in LUAD patients\u003c/p\u003e","description":"","filename":"image3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6311145/v1/f71039e68930dc5546a53764.jpeg"},{"id":82122284,"identity":"a55e01c9-dfeb-4d65-81a2-35ca01402cc6","added_by":"auto","created_at":"2025-05-07 03:27:58","extension":"jpeg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":332543,"visible":true,"origin":"","legend":"\u003cp\u003eA-L. The relationship between ARRPS and DSS was analyzed through Roc curves, Kaplan-Meier, univariate COX regression, and multivariate COX regression. (A-D). The relationship between ARRPS and DFI was analyzed through Roc curves, Kaplan-Meier, univariate COX regression, and multivariate COX regression (E-H). The relationship between ARRPS and PFI was analyzed through Roc curves, Kaplan-Meier, univariate COX regression, and multivariate COX regression (I-L).\u003c/p\u003e","description":"","filename":"image4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6311145/v1/37c087e397812c04a2945023.jpeg"},{"id":82123817,"identity":"0574905c-804a-47c9-b768-04a4e8beb662","added_by":"auto","created_at":"2025-05-07 03:35:58","extension":"jpeg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":741680,"visible":true,"origin":"","legend":"\u003cp\u003eA. Differences in immune background between high and low ARRPS scores. B-E. Differences in pathway enrichment between high and low ARRPS scores\u003c/p\u003e","description":"","filename":"image5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6311145/v1/7b3829df52ee236b369c4e14.jpeg"},{"id":82123814,"identity":"250eeb5b-5fbd-465a-919e-b841a98ed65d","added_by":"auto","created_at":"2025-05-07 03:35:58","extension":"jpeg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":337463,"visible":true,"origin":"","legend":"\u003cp\u003eA. Schematic representation of the strategy to marshal potential therapeutic agents with higher drug sensitivity in the high ARRPS group. B-C. Differential drug response analysis of Prism-derived compounds (B) and CTRP-derived compounds（C）showed that. D-E. Results of Spearman correlation analysis of the relative inhibitory effects of CTRP (D) and PRISM-derived compounds (E). F-G. Screening of thera-peutic agents by CMap.\u003c/p\u003e","description":"","filename":"image6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6311145/v1/c42b345de739b4a1c465a896.jpeg"},{"id":82125079,"identity":"250b3284-b1b6-4a69-a2fe-fe271f4107b9","added_by":"auto","created_at":"2025-05-07 03:43:58","extension":"jpeg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":645986,"visible":true,"origin":"","legend":"\u003cp\u003eA. Schematic chemical structures of CD-437 and TPCA-1. B-C The effects of CD-437 and TPCA-1 in combination with AUM on cell viability were examined using CCK8 and are shown as a dose-response matrix (B) and synergism score matrix (C). D. Effects on colony formation under combination therapy. E-F. Antitumor effects of CD-437 (E) and TPCA-1 (F) in combination with AUM including. G-H. Tumor volume changes. I-J. Tumor tissue weight.\u003c/p\u003e","description":"","filename":"image7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6311145/v1/37b5e0357174f0b374e0f569.jpeg"},{"id":95040014,"identity":"9ce4ef27-ccea-49f3-9998-164257e2e9b3","added_by":"auto","created_at":"2025-11-03 16:07:25","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":6035473,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6311145/v1/1aed3563-7bbe-4ce5-b5fb-d5b0e44bdd6b.pdf"},{"id":82122294,"identity":"ae4ef69d-a242-4636-a98e-ab52d7c95fc0","added_by":"auto","created_at":"2025-05-07 03:27:58","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":2239683,"visible":true,"origin":"","legend":"","description":"","filename":"2.supplementary.docx","url":"https://assets-eu.researchsquare.com/files/rs-6311145/v1/8d53619cae87f7df89289c84.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Integrated machine learning survival framework for consensus modeling in a large multicenter cohort of NSCLC resistant to aumolertinib","fulltext":[{"header":"Introduction","content":"\u003cp\u003eThe problem of drug resistance to cancer treatment has long been recognized as a key factor in cancer treatment failure[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. For example, resistance to epidermal growth factor receptor (EGFR) tyrosine kinase inhibitors (TKIs) greatly contributes to treatment failure and cancer recurrence in patients with EGFR-mutant non-small cell lung cancer (NSCLC)[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Among the multiple mechanisms of resistance to EGFR-TKIs, T790M, a secondary mutation in EGFR exon 20, is thought to be the most prevalent cause of acquired resistance to first- and second-generation EGFR-TKIs, and a number of third-generation EGFR TKIs have been designed to irreversibly target EGFRs with both T790M-resistant mutations and EGFR-activating mutations. Among them, aumolertinib (AUM) is a representative compound widely approved for standard second-line treatment of EGFR T790M-mutant NSCLC and first-line treatment of EGFR-activating mutant NSCLC, with significant preclinical and clinical efficacy but inevitably acquired resistance. Multiple mechanisms have been reported, including EGFR mutations[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e], activation of alternative pathways[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e], and aberrant downstream signaling[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. However, the investigation of AUM resistance mechanisms remains an urgent problem, which simultaneously attracts people to explore new resistance mechanisms and find potential therapeutic strategies for AUM-resistant NSCLC.\u003c/p\u003e \u003cp\u003eIn order to be able to better find the mechanisms of resistance to AUM as well as to have an accurate prediction of the prognosis of NSCLC patients, we developed and validated a robust signature consisting of a small set of genes using large-scale data. Using 10 machine learning algorithms, they were transformed into 100 combinations that were further executed in 5 independent cohorts. The best-performing ARRPS among the 100 model types was finally identified. In this model, patients with a high ARRPS showed a higher risk of death, relapse, and disease progression[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. The ARRPS was independent of traditional clinical signatures (e.g., American Joint Committee on Cancer (AJCC) staging) and molecular signatures (e.g., TP53 mutations) and additionally demonstrated superior predictive performance in these variables. Comparisons with published signatures show favorable performance of ARRPS. Additionally, patients with high ARRPS scores may benefit from 2 potential therapeutic agents: CD-437 and TPCA-1.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003e1.1. Induction of AUM-resistant NSCLC cell lines\u003c/h2\u003e \u003cp\u003eAt the beginning, we modeled acquired resistance to AUM by deriving polyclonal acquired resistant cell lines on the basis of stepwise dose escalation over a period of six months followed by maintenance in 10\u0026micro;M of drug over 6 weeks[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e], that is well representative of the clinic, and performed RNA whole transcriptome sequencing of this model using parental cells as a control to screen for AUM-resistant genes.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003e1.2. Generation of ARRPS\u003c/h3\u003e\n\u003cp\u003eWe extracted data from the Gene Expression Omnibus (GEO, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.ncbi.nlm.nih.gov/geo/\u003c/span\u003e\u003cspan address=\"https://www.ncbi.nlm.nih.gov/geo/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) (GSE50081, GSE30219, GSE72094 and GSE132313) and The Cancer Genome Atlas (TCGA, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://portal.gdc.cancer.gov/\u003c/span\u003e\u003cspan address=\"http://portal.gdc.cancer.gov/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) (TCGA-LUAD) in which five independent lung adenocarcinoma datasets were retrospectively extracted. Sample selection criteria were as follows (1) primary cancer tissues that had not received any adjuvant therapy and (2) survival information and RNA expression data were available. In the TCGA-LUAD project, we obtained single nucleotide variants (SNV) and copy number variants (CNV) from the TCGA portal. Prognostic genes were screened using Cox regression analysis, and genes with an adjusted P-value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 and a consistent hazard ratio (HR) direction in the four cohorts were considered to be robust genes associated with overall survival (OS). The intersection of differentially expressed genes in AUM-resistant cell lines with the above obtained genes was used to identify robust genes associated with AUM resistance and OS. To develop a consensus ARRPS with high accuracy and stable performance, we integrated 10 machine learning algorithms including Random Survival Forest (RSF), Elastic Network (Enet), Lasso regression, Ridge regression, Stepwise Cox regression, Cox boosting, Cox partial least squares regression (plsRcox), Supervised Principal Component Analysis (SuperPC), Generalized Augmented Regression Modeling (GBM) and Survival Support Vector Machine (survival-svm). In this study, the TCGA-LUAD cohort was used as the training set, while the other four cohorts were used as the test set. We calculated the c-index of each feature in all cohorts, where the feature with the highest average c-index was considered as the best feature[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e].\u003c/p\u003e\n\u003ch3\u003e1.3. Validating the prognostic value of ARRPS\u003c/h3\u003e\n\u003cp\u003ePatients in the five cohorts were categorized into high and low ARRPS subgroups based on the median ARRPS score. Kaplan-Meier curves and Cox regression analysis were used to assess the prognostic value of ARRPS. Subjects' work characteristics (ROC) curves were plotted to assess the predictive accuracy of ARRPS. Subsequently, we calculated the C-index to compare the superiority of ARRPS with a variety of common clinical characteristics (including gender, age, TNM stage and grading) in predicting overall survival (OS) in patients with LUAD.\u003c/p\u003e\n\u003ch3\u003e1.4. Collection and calculation of LUAD published signatures\u003c/h3\u003e\n\u003cp\u003eTo compare the predictive performance of ARRPS with these published features, we searched PubMed on October 1, 2024, for published articles on prognostic models[\u003cspan additionalcitationids=\"CR9 CR10\" citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. Subsequently, we calculated risk scores for the 5 cohorts based on the genes and coefficients provided in the articles and assessed their prognostic performance in LUAD by one-way Cox analysis and C-index.\u003c/p\u003e\n\u003ch3\u003e1.5. Collection and calculation of LUAD published signatures\u003c/h3\u003e\n\u003cp\u003eTo compare the predictive performance of ARRPS with these published signatures, we searched PubMed for published articles on prognostic models as of October 1, 2024 Subsequently, we calculated risk scores for the 5 cohorts based on the genes and coefficients provided in the articles and evaluated their prognostic performance in LUAD by a one-way Cox analysis and C-index.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e1.6. Genomic alteration landscape\u003c/h2\u003e \u003cp\u003eTo compare the predictive performance of ARRPS with these published features, we searched PubMed on October 1, 2024, for published articles on prognostic models. Subsequently, we calculated risk scores for the 5 cohorts based on the genes and coefficients provided in the articles and assessed their prognostic performance in LUAD by one-way Cox analysis and C-index. (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://cloud.genepattern.org/gp/pages/index.jsf\u003c/span\u003e\u003cspan address=\"https://cloud.genepattern.org/gp/pages/index.jsf\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003e1.7. Gene set variation analysis (GSVA)\u003c/h3\u003e\n\u003cp\u003eGene sets were obtained from the Molecular Signatures Database (MSigDB, \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.gsea-msigdb.org/gsea/msigdb/index.jsp\u003c/span\u003e\u003cspan address=\"https://www.gsea-msigdb.org/gsea/msigdb/index.jsp\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). Gene set variance analysis (GSVA) methods were used to quantify the values of each metric in each sample. The limma package was used to identify pathways that were significantly altered between the high and low ARRPS groups[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e].\u003c/p\u003e\n\u003ch3\u003e1.8. Comprehensive analyses based on immune cell infiltration\u003c/h3\u003e\n\u003cp\u003eWe explored the correlation between ARRPS scores and immune infiltration. We obtained immune infiltration data for TCGA-LUAD from the TIMER 2.0 web server (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://timer.cistrome.org/\u003c/span\u003e\u003cspan address=\"http://timer.cistrome.org/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e)[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e], which covers immune infiltration data obtained by seven methods, namely, CIBERSORT, CIBERSORT-ABS, EPIC, MCPCOUNTER, QUANTISEQ, TIMER and XCELL[\u003cspan additionalcitationids=\"CR15 CR16\" citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e].\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e1.9. Development and validation of potential therapeutic agents\u003c/h2\u003e \u003cp\u003eWe obtained drug sensitivity data for ARRPS from the CTRP and PRISM recombinant datasets. Based on the Wilcoxon rank-sum test, we analyzed the difference in drug response between the high ARRPS group (top 10%) and the low ARRPS group (bottom 10%) and set a threshold of log2FC\u0026thinsp;\u0026lt;\u0026thinsp;0.05 to identify compounds with lower AUC values in the high ARRPS group. Next, we applied Spearman's correlation analysis to further screen compounds whose AUC values were negatively correlated with ARRPS (setting a threshold R\u0026thinsp;\u0026lt;\u0026thinsp;\u0026minus;\u0026thinsp;0.3)[\u003cspan additionalcitationids=\"CR19 CR20\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e1.10. Tumor xenograft models\u003c/h2\u003e \u003cp\u003eAll animals were housed according to the guidelines of the relevant institutions and approved by the Animal Care Committee of Bengbu Medical College. The BALB/c nude mice (male, 6 weeks old) were purchased from Unilever (Beijing, China). The BALB/c nude mice were kept in a specific pathogen-free (SPF)-grade animal laboratory. Nude mice were injected subcutaneously with 5\u0026times;10\u003csup\u003e6\u003c/sup\u003e cells on the right side and randomly grouped after the tumors grew to 100\u0026ndash;150 cubic millimeters. Tumor volume was measured every two days. Tumor volume was calculated as length \u0026times; width\u003csup\u003e2\u003c/sup\u003e \u0026times; 0.5. After three weeks, all mice were executed, and tumors were harvested and weighed.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e1.11 Cell lines and culture conditions\u003c/h2\u003e \u003cp\u003eThe HCC827 and H1975 cell line was sourced from American Type Culture Collection (Manassas, VA, USA). All media were supplemented with 10% fetal bovine serum (FBS; Gibco), 100 U/ml penicillin, and 100 \u0026micro;g/ml streptomycin. The cells were incubated at 37\u0026deg;C in a humidified 5% CO2 atmosphere. The authenticity of the human cell lines used in this study was ascertained through short tandem repeat (STR) profiling.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e1.12 RNA sequencing\u003c/h2\u003e \u003cp\u003eRNA Sequencing and data collection were performed by Wei Huan Biotechnology (China) Co. Total RNA was extracted using TRIzol\u0026reg; reagent (Invitrogen, #15596018) according to the manufacturer's instructions, and the samples were sent to Shanghai Wei Huan Biotechnology (Shanghai, China) for RNA purification, reverse transcription, library construction, and sequencing. After quality control of raw reads, clean reads were mapped using HISAT2 software and further assembled using StringTiein. Cellular pathway analysis was performed using GSEA.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003e1.13 Statistical analysis\u003c/h2\u003e \u003cp\u003eStatistical analyses were performed using GraphPad Prism 8, and data are expressed as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SEM. Two-tailed unpaired Student's t-test and one- or two-way ANOVA were used to analyze differences between groups.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003e2.1. Construction of the ARRPS\u003c/h2\u003e \u003cp\u003eThe ARRPS construction process is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. HCC827 cells resistant to AUM were obtained through six months of induction with a resistance index of 3.35, which is a good representation of the clinical situation (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA). We performed RNA whole transcriptome sequencing of this model using parental cells as a control to screen for AUM resistance genes. We found 2987 up-regulated genes and 3410 down-regulated genes in resistant HCC827 cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB-C). Through analysis, we obtained 20 genes significantly associated with overall survival and alectinib resistance, including ANLN, ASPM, BUB1B, CDC20, CDC25C, CDC6, CDCA2, CDCA3, CDKN3, CENPE, CENPF, FOXM1, GINS4, KIF2C, MELK, MYBL2, NEK2, PRC1, SHCBP1, and TOP2A.Based on the 10-fold cross-validation framework, we used 10 machine learning algorithms and their constituent combinations for a total of 100 methods for modeling. The results show that the combined algorithm of lasso\u0026thinsp;+\u0026thinsp;RSF (including 12 genes) has the highest average C-index and is considered the optimal modeling approach (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eD). We found that these 12 genes were highly expressed in cancer tissues (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eE), and that these 12 genes had a high ability to discriminate between diagnosing LUAD and non-tumors (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eF). To evaluate the prognostic performance of ARRPS, we categorized LUAD patients into high and low ARRPS groups according to the median value. KM curves showed that in the training cohort, mortality was significantly higher in the high-risk group than in the low-risk group (TCGA-LUAD, Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eG), and the other four validation cohorts showed the same trend of patients in the high-ARRPS group having worse OS (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eH-K). For the five cohorts, the ROC curves demonstrated the high ability of ARRPS to predict patient OS (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA-E). Subsequently, comparing the ability of ARRPS with multiple clinical features in predicting OS in patients with LUAD showed that ARRPS was significantly more accurate than these features, including age, gender, smoking BI, TNM staging, KRAS, EGFR, STK11, and TP53 mutations or new events (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eF-J). Meanwhile, we also investigated other clinical outcomes of TCGA-LUAD, including DSS, PFI, and DFI, and the results showed that patients with high ARRPS had worse DSS, PFI, and DFI, and ARRPS was an independent risk factor for DSS, PFI, and DFI in TCGA-LUAD. ARRPS also had high accuracy in predicting TCGA-LUAD DSS, PFI, and DFI (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003eA-J). In multiple cohorts, as described above, our ARRPS model shows robust pre-prediction performance. To further verify whether ARRPS has good predictive performance compared to published signatures, we collected 40 published signatures. We then calculated the c-index for each of the 40 signatures in each of the five cohorts and compared them to ARRPS. The results show that the c-index of ARRPS is significantly higher than that of other signatures in the TCGA-LUAD, GSE30219, GSE50081, and GSE72094 cohorts, except for the GSE13213 cohort (Fig. \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003eA). This suggests that ARRPS maintains a relatively strong predictive power even when compared to other signatures. Taken together, the results of Kaplan-Meier survival analyses, Cox regression analyses, ROC curves, and C-indexes of the five cohorts consistently showed that the ARRPS accurately and robustly predicted the prognosis of patients with LUAD, suggesting that the ARRPS may become an attractive tool to serve clinical practice.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003e2.2. ARRPS shows substantial correlation with immune-relate\u003c/h2\u003e \u003cp\u003eTo explore the immune status reflected by the ARRPS profile score, we analyzed the differences between high and low ARRPS profile scores in immune-infiltrating cells, which were concentrated in T cells, B cells, macrophages, and NK cells (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eA).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003e2.3. Differences in pathway enrichment between high and low ARRPS\u003c/h2\u003e \u003cp\u003eSignaling pathway enrichment shows the differences involved in the groups with high and low ARRPS feature scores, and the results indicate that the group with high ARRPS feature scores mainly focuses on many signaling pathways that promote tumor growth, such as cell cycle and DNA damage repair and metabolism-related pathways, such as KEGG_DRUG_METABOLISM_CYTOCHROME_P450, KEGG_ARACHIDONIC_ACID_METABOLISM, KEGG_LINOLEIC_ACID_METABOLISM, KEGG_ETHER_LIPID_METABOLISM, KEGG_ALPHA_LINOLENIC_ACID_METABOLISM, and KEGG_TAURINE_AND_HYPOTAURINE_METABOLISM (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003eB-E).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003e2.4. Analysis of genomic variation between patients in the high and low ARRPS groups\u003c/h2\u003e \u003cp\u003eStudy of the TCGA-LUAD SNV data found significant differences in SNV between patients in the high and low ARRPS groups. Patients in the high ARRPS group had higher mutation frequencies in NPAP1, TSHZ3, PCDHB8, PTPRT, GRM8, ANK1, KLHL4, TRPC4, COL4A2, LAMA4, and BCHE, whereas patients in the low ARRPS group had higher mutation frequencies in CPS1, COL19A1, PLCB4, etc. In terms of CNV, several loci had high CNV frequencies in patients in the high ARRPS group, including 3q26.2(Amp), 12q15(Amp), 17p13.3(Del), 17p11.2(Del), 6q26(Del), 18q22.2(Del), 18q21.1(Del), 22q13.32(Del), 21q21.1(Del), 1p13.1(Del), and 1p36.11(Del) (Fig. S5A). ZBTB4, XAF1, ZMYM2, FOXO1, TNFSF11, ZNF236, BCL2, MAP2K3RBFA, PIK3C3, RBX1, ZNF24, NCF4, SLC5A1 CNV events with more frequencies in patients in the high ARRPS group (Fig. S2A). In addition, we identified multiple gene pairs with SNV co-occurence, including CDH10-MYLK, MRC2-NPAP1, and KIF1A-TSHZ3 (Fig. S3A).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec21\" class=\"Section2\"\u003e \u003ch2\u003e2.5. CD-437 and TPCA-1 are drug candidates for patients with high ARRPS\u003c/h2\u003e \u003cp\u003ePatients with a high ARRPS have a poorer prognosis and a higher risk of developing resistance to AUM. In order to be able to use the ARRPS as a reliable prognostic risk assessment toolkit, we performed a screening of effective therapeutic agents. The drug screening flowchart is shown in (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eA). First, differential drug response analysis was performed on the two groups of patients with high and low ARRPS scores to screen compounds of high ARRPS scores patients with low AUC values, and then Spearman correlation analysis was performed to screen compounds with significant negative correlation between AUC values and ARRPS (r \u0026lt; -0.3, Log\u003csub\u003e2\u003c/sub\u003eFC \u0026lt; -0.01), and finally 15 compounds were screened, including 6 CTRP derived compounds (3-Cl-AHPC, cabozantinib, CD-437, mitomycin, nakiterpiosinand StemRegenin 1) and 9 PRISM-derived compounds (broxyquinoline, combretastatin-A-4, deferasirox, frentizole, imiquimod, napabucasin, norepinephrine, tanespimycin, and TPCA-1)( Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eB-E). Finally, by CMap database and searching the previous publications, we found that CD-437 and TPCA-1, with high CMap scores and few studies in LUAD, are drugs with high potential for the treatment of LUAD (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eF-G). Both CD-437 and TPCA-1 alone significantly inhibited the viability of drug-resistant cell lines. We then sought to combine both with AUM to explore the sensitivity of drug-resistant cell lines to AUM. We used the CCK8 assay to detect the antiproliferative effects of CD-437 and ASK120067 in combination with AUM on drug-resistant cell lines and expressed them as a dose-response matrix and a synergy scoring matrix (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eA-C). Synergy scores were calculated by SynergyFinder using the HSA model [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. It was found that this combination exerted a stronger antiproliferative effect. The same therapeutic effect was subsequently detected using clone formation (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eD), not also. In addition to that, this combination effect significantly inhibited the growth of xenograft tumors in nude mice (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003eE-J, Fig. S6A-B). In conclusion, these results of ours surface that the two compounds screened by high and low ARRPS can effectively inhibit the growth of NSCLC.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eResistance to EGFR-TKI has long been a major impediment to targeted therapy for EGFR-mutant NSCLC[\u003cspan additionalcitationids=\"CR25\" citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. The conventional solution is to increase the dose or add additional other chemotherapeutic agents. However, this is not a good solution. Immunotherapy has a more expensive cost and serious adverse effects. Therefore, exploring a new biomarker to effectively treat EGFR mutant NSCLC is also necessary. To achieve this goal, we built multiple prognostic models based on AUM resistance-associated differential genes using a combination of 10 machine learning algorithms. Subsequently, the 12-gene ARRPS, derived from the combination of lasso\u0026thinsp;+\u0026thinsp;GBM, emerged as superior.\u003c/p\u003e \u003cp\u003eAs a linear regression model, Lasso regression selects features during model training and can be used for feature selection and dimensionality reduction. It also improves the generalization ability of the model and avoids overfitting. However, in this study, its direct application of this model produces not much accuracy, achieving an average c-index of 0.655. GBM has the advantages of good interpretability and wide applicability, but it lacks the inherent ability to downscale and carries the risk of overfitting. To address these challenges, we combine the advantages of CoxBoost and GBM. It turns out that this combined approach proves to be the most effective, outperforming other models in our study.\u003c/p\u003e \u003cp\u003eIn our study, ARRPS was found to be a reliable prognostic risk assessment tool. This also gives us a greater interest in the screening of therapeutic agents for patients with high ARRPS. It turned out that CD-437 and TPCA-1 had good ability to inhibit drug-resistant cell lines, and we combined them with AUM, which also showed good synergistic anti-tumor ability, both in vivo and in vitro. In clinical practice, we can obtain biopsy samples from diagnosed NSCLC patients and perform RNA-seq analysis to elucidate their transcriptome profiles. Subsequently, we can reconcile these profiles with the meta-LUAD cohort and calculate the ARRPS. Based on the ARRPS threshold, NSCLC patients can be categorized into a low ARRPS group and a high ARRPS group. Conventional therapy is sufficient for patients in the low ARRPS group, while those in the high ARRPS group may benefit from the combination of AUM with CD-437 and TPCA-1. However, the limitations of our study are worth mentioning. Our retrospective approach has its drawbacks, emphasizing the need for subsequent prospective trials to validate the predictive ability of ARRPS. OTS167 inhibits MELK, but targeting MELK deserves to be explored[\u003cspan additionalcitationids=\"CR28 CR29\" citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. Although our findings, combined with animal studies, suggest the efficacy and safety of CD-437 and TPCA-1, rigorous evaluation by additional preclinical models, such as OSA patient-derived xenografts and transgenic mouse models, is necessary.\u003c/p\u003e \u003cp\u003eIn summary, this study introduced ARRPS as a potential tool for identifying high-risk subgroups of patients with LUAD and identified therapeutic candidates CD-437 and TPCA-1 targeting high ARRPS. These results provide a promising therapeutic avenue and represent a potential complementary treatment for high-risk LUAD patients.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cdiv class=\"DefinitionList\"\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eNSCLC\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eNon-small cell lung cancer\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eEGFR-TKI\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eepidermal growth factor receptor-tyrosine kinase inhibitor\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eAUM\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eAumolertinib\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eARRPS\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003eAumolertinib resistance-related prognostic signature\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv class=\"DefinitionListEntry\"\u003e \u003cdiv class=\"Term\"\u003eLUAD\u003c/div\u003e \u003cdiv class=\"Description\"\u003e \u003cp\u003elung adenocarcinoma.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eA\u003c/strong\u003e\u003cstrong\u003euthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eXiao Wu performed most of the experiments, Yongxia Chen and Guolei Song analyzed data, Yang Lu, Yongping Li and Annan Zhu contributed to experiments. Tingting Wang, Zishu Wang and Fang Su designed experiments, supervised the study, and wrote the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was supported by Natural Science key project of Bengbu Medical College (No. 2021byzd064).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets generated and/or analyzed during the current study are available in the NCBI BioProject repository under accession number PRJNA1248195 (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1248195). The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe animal experiments were conducted in accordance with ARRIVE 2.0 guidelines (Animal Research: Reporting of In Vivo Experiments, https://arriveguidelines.org). All animal experimental procedures were approved by the Experimental Animal Teaching and Research Committee of Bengbu Medical University (2023-522). All methods were performed in accordance with the relevant guidelines and regulations\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll authors consent to publication of this article.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDeclaration of competing interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no conflict of interest.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003ePassaro A, J\u0026auml;nne PA, Mok T, Peters S: \u003cstrong\u003eOvercoming therapy resistance in EGFR-mutant lung cancer.\u003c/strong\u003e \u003cem\u003eNat Cancer \u003c/em\u003e2021, \u003cstrong\u003e2:\u003c/strong\u003e377-391.\u003c/li\u003e\n\u003cli\u003eZhang T, Qu R, Chan S, Lai M, Tong L, Feng F, Chen H, Song T, Song P, Bai G, et al: \u003cstrong\u003eCorrection: Discovery of a novel third-generation EGFR inhibitor and identification of a potential combination strategy to overcome resistance.\u003c/strong\u003e \u003cem\u003eMol Cancer \u003c/em\u003e2023, \u003cstrong\u003e22:\u003c/strong\u003e139.\u003c/li\u003e\n\u003cli\u003ePan Z, Wang K, Wang X, Jia Z, Yang Y, Duan Y, Huang L, Wu ZX, Zhang JY, Ding X: \u003cstrong\u003eCholesterol promotes EGFR-TKIs resistance in NSCLC by inducing EGFR/Src/Erk/SP1 signaling-mediated ERR\u0026alpha; re-expression.\u003c/strong\u003e \u003cem\u003eMol Cancer \u003c/em\u003e2022, \u003cstrong\u003e21:\u003c/strong\u003e77.\u003c/li\u003e\n\u003cli\u003eShih PC: \u003cstrong\u003eThe role of the STAT3 signaling transduction pathways in radioresistance.\u003c/strong\u003e \u003cem\u003ePharmacol Ther \u003c/em\u003e2022, \u003cstrong\u003e234:\u003c/strong\u003e108118.\u003c/li\u003e\n\u003cli\u003eLiu Z, Liu L, Weng S, Guo C, Dang Q, Xu H, Wang L, Lu T, Zhang Y, Sun Z, Han X: \u003cstrong\u003eMachine learning-based integration develops an immune-derived lncRNA signature for improving outcomes in colorectal cancer.\u003c/strong\u003e \u003cem\u003eNat Commun \u003c/em\u003e2022, \u003cstrong\u003e13:\u003c/strong\u003e816.\u003c/li\u003e\n\u003cli\u003eShah KN, Bhatt R, Rotow J, Rohrberg J, Olivas V, Wang VE, Hemmati G, Martins MM, Maynard A, Kuhn J, et al: \u003cstrong\u003eAurora kinase A drives the evolution of resistance to third-generation EGFR inhibitors in lung cancer.\u003c/strong\u003e \u003cem\u003eNat Med \u003c/em\u003e2019, \u003cstrong\u003e25:\u003c/strong\u003e111-118.\u003c/li\u003e\n\u003cli\u003eLiu Z, Guo C, Dang Q, Wang L, Liu L, Weng S, Xu H, Lu T, Sun Z, Han X: \u003cstrong\u003eIntegrative analysis from multi-center studies identities a consensus machine learning-derived lncRNA signature for stage II/III colorectal cancer.\u003c/strong\u003e \u003cem\u003eEBioMedicine \u003c/em\u003e2022, \u003cstrong\u003e75:\u003c/strong\u003e103750.\u003c/li\u003e\n\u003cli\u003eYang Z, Yan G, Zheng L, Gu W, Liu F, Chen W, Cui X, Wang Y, Yang Y, Chen X, et al: \u003cstrong\u003eYKT6, as a potential predictor of prognosis and immunotherapy response for oral squamous cell carcinoma, is related to cell invasion, metastasis, and CD8+ T cell infiltration.\u003c/strong\u003e \u003cem\u003eOncoimmunology \u003c/em\u003e2021, \u003cstrong\u003e10:\u003c/strong\u003e1938890.\u003c/li\u003e\n\u003cli\u003eWang L, Liu Z, Zhu R, Liang R, Wang W, Li J, Zhang Y, Guo C, Han X, Sun Y: \u003cstrong\u003eMulti-omics landscape and clinical significance of a SMAD4-driven immune signature: Implications for risk stratification and frontline therapies in pancreatic cancer.\u003c/strong\u003e \u003cem\u003eComput Struct Biotechnol J \u003c/em\u003e2022, \u003cstrong\u003e20:\u003c/strong\u003e1154-1167.\u003c/li\u003e\n\u003cli\u003eTan Z, Lei Y, Xu J, Shi S, Hua J, Zhang B, Meng Q, Liu J, Zhang Y, Wei M, et al: \u003cstrong\u003eThe value of a metabolic reprogramming-related gene signature for pancreatic adenocarcinoma prognosis prediction.\u003c/strong\u003e \u003cem\u003eAging (Albany NY) \u003c/em\u003e2020, \u003cstrong\u003e12:\u003c/strong\u003e24228-24241.\u003c/li\u003e\n\u003cli\u003eYuan Q, Ren J, Li L, Li S, Xiang K, Shang D: \u003cstrong\u003eDevelopment and validation of a novel N6-methyladenosine (m6A)-related multi- long non-coding RNA (lncRNA) prognostic signature in pancreatic adenocarcinoma.\u003c/strong\u003e \u003cem\u003eBioengineered \u003c/em\u003e2021, \u003cstrong\u003e12:\u003c/strong\u003e2432-2448.\u003c/li\u003e\n\u003cli\u003eSubramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES, Mesirov JP: \u003cstrong\u003eGene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles.\u003c/strong\u003e \u003cem\u003eProc Natl Acad Sci U S A \u003c/em\u003e2005, \u003cstrong\u003e102:\u003c/strong\u003e15545-15550.\u003c/li\u003e\n\u003cli\u003eLi T, Fu J, Zeng Z, Cohen D, Li J, Chen Q, Li B, Liu XS: \u003cstrong\u003eTIMER2.0 for analysis of tumor-infiltrating immune cells.\u003c/strong\u003e \u003cem\u003eNucleic Acids Res \u003c/em\u003e2020, \u003cstrong\u003e48:\u003c/strong\u003eW509-w514.\u003c/li\u003e\n\u003cli\u003eChr\u0026eacute;tien S, Zerdes I, Bergh J, Matikas A, Foukakis T: \u003cstrong\u003eBeyond PD-1/PD-L1 Inhibition: What the Future Holds for Breast Cancer Immunotherapy.\u003c/strong\u003e \u003cem\u003eCancers (Basel) \u003c/em\u003e2019, \u003cstrong\u003e11\u003c/strong\u003e.\u003c/li\u003e\n\u003cli\u003eWang J, Sanmamed MF, Datar I, Su TT, Ji L, Sun J, Chen L, Chen Y, Zhu G, Yin W, et al: \u003cstrong\u003eFibrinogen-like Protein 1 Is a Major Immune Inhibitory Ligand of LAG-3.\u003c/strong\u003e \u003cem\u003eCell \u003c/em\u003e2019, \u003cstrong\u003e176:\u003c/strong\u003e334-347.e312.\u003c/li\u003e\n\u003cli\u003eJanakiram M, Chinai JM, Zhao A, Sparano JA, Zang X: \u003cstrong\u003eHHLA2 and TMIGD2: new immunotherapeutic targets of the B7 and CD28 families.\u003c/strong\u003e \u003cem\u003eOncoimmunology \u003c/em\u003e2015, \u003cstrong\u003e4:\u003c/strong\u003ee1026534.\u003c/li\u003e\n\u003cli\u003eBindea G, Mlecnik B, Tosolini M, Kirilovsky A, Waldner M, Obenauf AC, Angell H, Fredriksen T, Lafontaine L, Berger A, et al: \u003cstrong\u003eSpatiotemporal dynamics of intratumoral immune cells reveal the immune landscape in human cancer.\u003c/strong\u003e \u003cem\u003eImmunity \u003c/em\u003e2013, \u003cstrong\u003e39:\u003c/strong\u003e782-795.\u003c/li\u003e\n\u003cli\u003eYang C, Zhang H, Chen M, Wang S, Qian R, Zhang L, Huang X, Wang J, Liu Z, Qin W, et al: \u003cstrong\u003eA survey of optimal strategy for signature-based drug repositioning and an application to liver cancer.\u003c/strong\u003e \u003cem\u003eElife \u003c/em\u003e2022, \u003cstrong\u003e11\u003c/strong\u003e.\u003c/li\u003e\n\u003cli\u003eYang K, Dinasarapu AR, Reis ES, Deangelis RA, Ricklin D, Subramaniam S, Lambris JD: \u003cstrong\u003eCMAP: Complement Map Database.\u003c/strong\u003e \u003cem\u003eBioinformatics \u003c/em\u003e2013, \u003cstrong\u003e29:\u003c/strong\u003e1832-1833.\u003c/li\u003e\n\u003cli\u003eMaeser D, Gruener RF, Huang RS: \u003cstrong\u003eoncoPredict: an R package for predicting in vivo or cancer patient drug response and biomarkers from cell line screening data.\u003c/strong\u003e \u003cem\u003eBrief Bioinform \u003c/em\u003e2021, \u003cstrong\u003e22\u003c/strong\u003e.\u003c/li\u003e\n\u003cli\u003eGeeleher P, Cox N, Huang RS: \u003cstrong\u003epRRophetic: an R package for prediction of clinical chemotherapeutic response from tumor gene expression levels.\u003c/strong\u003e \u003cem\u003ePLoS One \u003c/em\u003e2014, \u003cstrong\u003e9:\u003c/strong\u003ee107468.\u003c/li\u003e\n\u003cli\u003eIanevski A, Giri AK, Aittokallio T: \u003cstrong\u003eSynergyFinder 3.0: an interactive analysis and consensus interpretation of multi-drug synergies across multiple samples.\u003c/strong\u003e \u003cem\u003eNucleic Acids Res \u003c/em\u003e2022, \u003cstrong\u003e50:\u003c/strong\u003eW739-w743.\u003c/li\u003e\n\u003cli\u003eZheng S, Wang W, Aldahdooh J, Malyutina A, Shadbahr T, Tanoli Z, Pessia A, Tang J: \u003cstrong\u003eSynergyFinder Plus: Toward Better Interpretation and Annotation of Drug Combination Screening Datasets.\u003c/strong\u003e \u003cem\u003eGenomics Proteomics Bioinformatics \u003c/em\u003e2022, \u003cstrong\u003e20:\u003c/strong\u003e587-596.\u003c/li\u003e\n\u003cli\u003eFu Y, Zhang Y, Lei Z, Liu T, Cai T, Wang A, Du W, Zeng Y, Zhu J, Liu Z, Huang JA: \u003cstrong\u003eAbnormally activated OPN/integrin \u0026alpha;V\u0026beta;3/FAK signalling is responsible for EGFR-TKI resistance in EGFR mutant non-small-cell lung cancer.\u003c/strong\u003e \u003cem\u003eJ Hematol Oncol \u003c/em\u003e2020, \u003cstrong\u003e13:\u003c/strong\u003e169.\u003c/li\u003e\n\u003cli\u003eNilsson MB, Yang Y, Heeke S, Patel SA, Poteete A, Udagawa H, Elamin YY, Moran CA, Kashima Y, Arumugam T, et al: \u003cstrong\u003eCD70 is a therapeutic target upregulated in EMT-associated EGFR tyrosine kinase inhibitor resistance.\u003c/strong\u003e \u003cem\u003eCancer Cell \u003c/em\u003e2023, \u003cstrong\u003e41:\u003c/strong\u003e340-355.e346.\u003c/li\u003e\n\u003cli\u003ePeng S, Wang R, Zhang X, Ma Y, Zhong L, Li K, Nishiyama A, Arai S, Yano S, Wang W: \u003cstrong\u003eEGFR-TKI resistance promotes immune escape in lung cancer via increased PD-L1 expression.\u003c/strong\u003e \u003cem\u003eMol Cancer \u003c/em\u003e2019, \u003cstrong\u003e18:\u003c/strong\u003e165.\u003c/li\u003e\n\u003cli\u003eTang Q, Li W, Zheng X, Ren L, Liu J, Li S, Wang J, Du G: \u003cstrong\u003eCorrection: MELK is an oncogenic kinase essential for metastasis, mitotic progression, and programmed death in lung carcinoma.\u003c/strong\u003e \u003cem\u003eSignal Transduct Target Ther \u003c/em\u003e2024, \u003cstrong\u003e9:\u003c/strong\u003e186.\u003c/li\u003e\n\u003cli\u003eLi B, Yan J, Phyu T, Fan S, Chung TH, Mustafa N, Lin B, Wang L, Eichhorn PJA, Goh BC, et al: \u003cstrong\u003eMELK mediates the stability of EZH2 through site-specific phosphorylation in extranodal natural killer/T-cell lymphoma.\u003c/strong\u003e \u003cem\u003eBlood \u003c/em\u003e2019, \u003cstrong\u003e134:\u003c/strong\u003e2046-2058.\u003c/li\u003e\n\u003cli\u003eLi Z, Zhou H, Zhai X, Gao L, Yang M, An B, Xia T, Du G, Li X, Wang W, Jin B: \u003cstrong\u003eMELK promotes HCC carcinogenesis through modulating cuproptosis-related gene DLAT-mediated mitochondrial function.\u003c/strong\u003e \u003cem\u003eCell Death Dis \u003c/em\u003e2023, \u003cstrong\u003e14:\u003c/strong\u003e733.\u003c/li\u003e\n\u003cli\u003eTang B, Zhu J, Shi Y, Wang Y, Zhang X, Chen B, Fang S, Yang Y, Zheng L, Qiu R, et al: \u003cstrong\u003eTumor cell-intrinsic MELK enhanced CCL2-dependent immunosuppression to exacerbate hepatocarcinogenesis and confer resistance of HCC to radiotherapy.\u003c/strong\u003e \u003cem\u003eMol Cancer \u003c/em\u003e2024, \u003cstrong\u003e23:\u003c/strong\u003e137.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"NSCLC, Resistance, prognostic signature, aumolertinib","lastPublishedDoi":"10.21203/rs.3.rs-6311145/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6311145/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003ePatients with advanced non-small cell lung cancer (NSCLC) harboring epidermal growth factor receptor (EGFR) mutations often benefit from third-generation tyrosine kinase inhibitors (TKIs), such as aumolertinib (AUM). However, the development of drug resistance significantly limits the clinical efficacy of AUM. To address this, we established an in vitro model of AUM-resistant cell lines and performed RNA sequencing to identify resistance-associated differentially expressed genes. Using machine learning, we constructed an AUM resistance-related prognostic signature (ARRPS). Our results demonstrated that ARRPS effectively predicts the prognostic risk of patients. Notably, for patients with high ARRPS scores, the addition of CD-437 or TPCA-1 to conventional AUM treatment may help overcome drug resistance. These findings suggest that ARRPS serves as both a prognostic tool and a guide for personalized treatment strategies, potentially optimizing the clinical management of NSCLC patients.\u003c/p\u003e","manuscriptTitle":"Integrated machine learning survival framework for consensus modeling in a large multicenter cohort of NSCLC resistant to aumolertinib","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-05-07 03:27:53","doi":"10.21203/rs.3.rs-6311145/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-06-03T08:27:33+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-06-02T12:33:49+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-05-16T19:31:59+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"207966000612447387448576361331321934895","date":"2025-05-13T00:49:51+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"250489450872538687649386127654116402184","date":"2025-05-09T14:45:31+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-04-24T12:54:30+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-04-24T12:40:51+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2025-04-11T07:55:56+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-04-11T04:25:00+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2025-03-26T09:52:25+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"f66ebbeb-ab83-4345-8ff8-273341f6dc83","owner":[],"postedDate":"May 7th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":47686899,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":47686900,"name":"Biological sciences/Computational biology and bioinformatics/Computational models"}],"tags":[],"updatedAt":"2025-11-03T16:02:28+00:00","versionOfRecord":{"articleIdentity":"rs-6311145","link":"https://doi.org/10.1038/s41598-025-20159-7","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2025-10-28 15:58:10","publishedOnDateReadable":"October 28th, 2025"},"versionCreatedAt":"2025-05-07 03:27:53","video":"","vorDoi":"10.1038/s41598-025-20159-7","vorDoiUrl":"https://doi.org/10.1038/s41598-025-20159-7","workflowStages":[]},"version":"v1","identity":"rs-6311145","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6311145","identity":"rs-6311145","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}