Integrating multi-omics analysis and machine learning to identify molecular subtypes and construct prognostic models for lung squamous cell carcinoma

Integrating multi-omics analysis and machine learning to identify molecular subtypes and construct prognostic models for lung squamous cell carcinoma | 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 Integrating multi-omics analysis and machine learning to identify molecular subtypes and construct prognostic models for lung squamous cell carcinoma Ya Dong, Xiang Zhang, Yuhan Wang, Tao Xu This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-4432088/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 9 You are reading this latest preprint version Abstract LUSC had a high morbidity and mortality rate in China, resulting in high social burdens. Most LUSC patients are already in the highly advanced cancer stage at diagnosis, and the clinical treatment is relatively difficult and the prognosis is relatively poor. The progress and application of sequencing technologies and machine learning algorithms offer new therapeutic perspectives and survival opportunities for LUSC patients. First, we gained multi-omics data on LUSC from the TCGA and GEO databases and performed batch effect. A total of ten different clustering methods were adopted to conduct multiomics consensus ensemble analysis. Then, we combined the integration analysis with ten machine learning algorithms to develop a CMLS. Besides, we explored the immune landscape and immunotherapeutic response of LUSC. Lastly, we identified potential therapeutic agents in LUSC. We independently identified two subtypes (CS1 and CS2) from ten multi-omics integrated clustering algorithms and CS2 showed the most favourable survival outcome among all subtypes. Subsequently, we identified 24 PRGs based on markers between subtypes and constructed CMLS using ten machine learning algorithms. In the TCGA-LUSC and META-LUSC sets, patients with high-CMLS group had a poorer clinical prognosis. Moreover, we evaluated the immunological landscape of LUSC using "IOBR" package. Low-CMLS group exhibited significantly higher levels of immune cell infiltration, including NK cells, T cells and B cells, suggesting that they may have better survival outcomes. CMLS might predict the prognosis and immune response of LUSC patients in some degree. In conclusion, our study provided novel ways to optimise the clinical diagnosis and therapeutic approaches of LUSC. Biological sciences/Cancer Health sciences/Oncology Lung squamous cell carcinoma Multiomics analysis Machine learning algorithm Consensus machine learning-driven signature Prognosis Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Introduction LUSC had a high morbidity and mortality rate in China, resulting in high social burdens. Most LUSC patients are already in the highly advanced cancer stage at diagnosis, and the clinical treatment is relatively difficult and the prognosis is relatively poor. The progress and application of sequencing technologies and machine learning algorithms offer new therapeutic perspectives and survival opportunities for LUSC patients. First, we gained multi-omics data on LUSC from the TCGA and GEO databases and performed batch effect. A total of ten different clustering methods were adopted to conduct multiomics consensus ensemble analysis. Then, we combined the integration analysis with ten machine learning algorithms to develop a CMLS. Besides, we explored the immune landscape and immunotherapeutic response of LUSC. Lastly, we identified potential therapeutic agents in LUSC. We independently identified two subtypes (CS1 and CS2) from ten multi-omics integrated clustering algorithms and CS2 showed the most favourable survival outcome among all subtypes. Subsequently, we identified 24 PRGs based on markers between subtypes and constructed CMLS using ten machine learning algorithms. In the TCGA-LUSC and META-LUSC sets, patients with high-CMLS group had a poorer clinical prognosis. Moreover, we evaluated the immunological landscape of LUSC using "IOBR" package. Low-CMLS group exhibited significantly higher levels of immune cell infiltration, including NK cells, T cells and B cells, suggesting that they may have better survival outcomes. CMLS might predict the prognosis and immune response of LUSC patients in some degree. In conclusion, our study provided novel ways to optimise the clinical diagnosis and therapeutic approaches of LUSC. Lung cancer (LC) is the most harmful malignant tumour in the world, and the incidence and mortality rates of LC in China account for about one third of those in the world 1 .As the major histological subtype of NSCLC, LUSC accounts for about 20–35% of the incidence of LC 2 . LUSC is mainly derived from the proximal bronchus and has a greater risk of invading large vessels 3 . The risk of LUSC is much higher in men than in women and the tumour mutation burden is positively correlated with increasing age 4 . Smoking, a recognised risk factor, has a strong association with the incidence of LUSC 5 . Previous studies have shown that most LUSC patients are already in the highly advanced cancer stage at diagnosis, and the clinical treatment is relatively difficult and the prognosis is relatively poor 6 . With the continuous advancement of modern medical technology, various novel therapeutic strategies have been actively explored and applied to improve the therapeutic effects of LUSC 7 . These therapeutic strategies aim to precisely intervene in abnormal cancer cells, inhibit cancer cell progression, and stimulate the body's immunity, such as targeted therapy, immunotherapy, and gene therapy. And targeted therapy is a therapeutic strategy that can accurately intervene in the special signaling pathways of cancer cells. It has greatly improved the therapeutic outcome of LUSC patients, and significantly prolonged the survival time of LUSC patients with genetic mutations 8 . Although the application of these advanced therapeutic approaches has provided new promises and opportunities for patients with LUSC, the great heterogeneity of its pathogenesis and biological characteristics has led to great variability in the treatment outcomes of different patients 9 . Therefore, how to fully recognise the heterogeneity of LUSC and improve the precision and effectiveness of treatment is an issue that we need to continuously focus on and explore in LUSC treatment research. Gene expression is regulated by a range of genetic and epigenetic changes 10 . Multi-omics data analysis is an approach that integrates multiple biological information and aims to probe deeply into the regulatory mechanisms and functional networks of complex biological systems 11 . Through combined integration and analysis of biological data at different levels, we could obtain more comprehensive and accurate information, reveal the association between genes and phenotypes, identify key biological processes and signalling pathways, and further understand the functions and regulatory mechanisms of biological systems 12 . Nevertheless, most of the current studies are still mainly focused on exploring the high-throughput data and biological functions of individual-omics, limiting in-depth understanding of the complex interactions between different biological pathways 13 . Hence, we developed novel LUSC molecular subtypes through integrated multi-omics analysis to further search for new biomarkers, predict LUSC patients’ survival and prognosis, and explore potential therapeutic targets, thus providing a reference for individualised therapy. In the study, multi-omics data of LUSC were gained from the TCGA-LUSC databases, such as mRNA, lncRNA, miRNA, DNA methylation, somatic mutations and clinical data. And we construct an comprehensive consensus subtype of LUSC via ten integration clustering algorithms. Then, 108 prognosis-related genes (PRGs) were selected and adopted ten machine learning methods to build the consensus machine learning-driven signature (CMLS) for LUSC. CMLS exhibited remarkable biological significance for prognosis and clinical application in LUSC. CMLS also proved robust capability in predicting immunotherapy response and targeted drugs therapeutic efficacy. Overall, our results provide an important theoretical basis for the precise differentiation of LUSC biological subtypes, in-depth understanding of tumour heterogeneity and exploration of personalised clinical therapeutic strategy. Materials and methods Preparation and pre-processing of LUSC data We obtained multi-omics data on LUSC from the TCGA-LUSC set, such as mRNA, lncRNA, miRNA, DNA methylation, somatic mutations and clinical data. Furthermore, we obtained complete information on three LUSC sets from GEO (GSE73403, GSE74777 and GSE157010) 14 , 15 . The different datasets were merged using the “sva” package to remove the batch effect of each dataset. Finally, PCA analyses were used to further validate the validity and stability of the merging of different data sets. Multiomics consensus ensemble analysis Firstly, we use the "getElite" function to filter the features 16 . For continuous variables, we selected the “mad” parameter to select the top 3,000 genes with the highest variation. We set the “cox” parameter to recognize prognostic genes in each data dimension (p < 0.05). We recognized frequently mutated genes by setting the "freq" parameter based on mutation frequency (elite.pct = 0.1) 17 . Then, we used the "getClustNum" function to get the optimal number of clusters and used the "getMOIC" function to perform the clustering analysis, which contained ten clustering algorithms. Specific molecular characteristics Firstly, we computed the enrichment scores of multiple therapeutically relevant signatures by the GSVA algorithm and constructed a transcriptional regulatory network by the "RTN" package 18 , 19 . Secondly, immune checkpoint and immune/stromal score were also assessed 20 . Finally, we calculated DNA MeTIL scores and assessed the enrichment of 24 immune cells by GSVA. The clustering results were verified in the LUSC-META set using markers and the consistency of the consensus clustering was compared with the NTP and PAM classifiers. Development of CMLS Given that some sets had fewer samples, we merged the three sets into the META-LUSC set and used the "sva" package to remove batch effects. To establish CMLS, ten machine learning algorithms are used. The TCGA-LUSC set was used as the training set and the META-LUSC set was used as the validation set. Prognostic value and clinical applications We scored each sample based on the obtained model and categorised the patients into different groups. Survival analyses were performed on the different groups to assess prognostic value. Furthermore, we systematically searched for 12 prognostic signatures related to LUSC. To improve the clinical application, we established a nomogram containing multiple clinical features. Exploration of immune landscape On the basis of the "IOBR" package, several dozen previous published signatures were obtained with respect to Tumour microenvironment (TME) cell type, immunotherapy response, immunosuppression, and immune exclusion and using a uniform methodology to compute concentration scores for each sample 20 . We compared the differences in TMB and M1 macrophage distribution between the two groups and regrouped the patients in conjunction with CMLS 21 . For immunotherapy response, we assessed patients' survival in response to immunotherapy and combined subclass mapping to assess immunotherapy response 22 . This was further verified in IMvigor210, GSE78220, GSE91061, and GSE135222 23–27 . Identification of potential therapeutic agents The activation status of the oncogenic pathway was analysed by the GSEA algorithm between patients with different CMLS groups. Expression data of human CCLs were obtained from CCLE 28 . Drug sensitivity data for CCLs were obtained from the CTRP v.2.0 and PRISM databases 29 , 30 . Results Identification of multi-omics consensus molecular subtypes Figure 1 illustrated the research process of the study. The PCA analysis demonstrated the distribution of samples before and after batch effect processing ( Figs S1 A and S1B ). Two subtypes were recognised from ten clustering algorithms ( Figs. 2A and 2B ). Unique molecular expression patterns across transcriptomes (mRNAs, lncRNAs, and miRNAs), epigenetic methylation, and somatic mutations were then further combined by a consensus-integrated approach ( Fig. 2C-2E ). Our classification system correlated strongly with OS (p < 0.001; Fig. 2F ). CS2 showed the most favourable survival outcome. Specific molecular characteristics Intriguingly, we discovered that pathways including saccule, epithelium, and vasculature development were remarkably enriched in CS1, whereas pathways including mesenchyme development and mesenchymal cell proliferation were remarkably enriched in CS2. In addition, these pathways, including immunosuppressive oncogenic pathways, are markedly enriched in CS1, whereas CS2 is potentially able to profit from treatments, including radiotherapy or targeted therapies ( Fig. 3A ). We analysed potential regulators related to cancer chromatin remodelling and 23 LUSC transcription factors in order to investigate transcriptome differences. GATA3, RARA, AR, STAT3, GATA6, ESR1, and PGR regulators were significantly activated in CS1, whereas RARG, RXRA, FGFR3, EGFR, HIF1A, FOXM1, ESR2, KLF4, PPARG, and TP63 were particularly enriched in LUSCs ( Fig. 3B ). Considering the critical role of tumour immunity in tumourigenesis and development, we found a significant increase in immune cell infiltration in CS1 ( Fig. 3C ). We selected 2000 genes up-regulated in each subtype as markers. Similar results were obtained in the META-LUSC set ( Fig. 4A ). The consistency of the CSs with the NTP and PAM algorithms was assessed (p < 0.001; Figs. 4B ). Establishment of the CMLS We used univariate Cox analysis to filter 108 PRGs from the TCGA-LUSC and META-LUSC whose expression was significantly correlated with OS. PRGs were incorporated into the integration framework to establish CMLS. We developed signatures based on ten algorithmic combinations and computed all the sets in the mean C-index for each signature to evaluate the predictive power ( Fig. 5A ). The algorithm consisting of StepCox (direction = backward) + CoxBoost retained the highest average C-index to construct the final signature. The StepCox (direction = backward) method recognized the most valuable PRGs, and the CoxBoost method filtered out the most valuable model, established from the 24 hub genes ( Figs. 5B and 5C ). We computed CMLS scores for each sample across all sets. In the TCGA-LUSC and META-LUSC sets, patients with high-CMLS group had a poorer clinical prognosis ( Figs. 5D and 5E ). We validated the prognostic value of these genes by survival analysis, and the results derived were in general agreement with those of Cox analysis ( Figs S2 and S3 ). These genes were significantly related to DSS and PFS in LUSC, emphasising their close prognostic correlation with patients ( Fig S4 and S5 ). Subsequently, we systematically examined the multi-omics phenotypes of CMLS in 33 different cancer types using the GSCALite database. CHEK2, KIF22, BCAT1, RPIA, ITGA3, GRHL3 and FN1 genes were highly expressed in some cancer tissues, while GPX3, MMRN1, AKAP12, KANK4, PINK1, F13A1, ALDH7A1, STAB1 and VSIG4 genes were lowly expressed in some cancer tissues ( Fig S6 A ). The mRNA expression levels of CMLS genes were positively related to CNVs in most cancer types, especially PINK1, BAG4, RPIA, CHEK2, ALDH7A1, ITGB1 and KIF22 ( Fig S6 B ). By analysing the CNV frequency changes, the CNVs of these gene differed significantly across cancer types, with NLRP3, SOX2, BCAT1 and GPR160 having the highest CNV frequencies, predominantly copy number heterozygous amplifications ( Fig S6 C and S6D ). Furthermore, methylation levels of these genes in most tumour types differed remarkably between tumour and normal samples ( Fig S7 A ). In most cancers, the methylation levels were negatively related to the mRNA levels ( Fig S7 B ). These genes activated the pan-cancer EMT pathway and had a remarkable inhibitory effect on the RAS/MAPK pathway ( Fig S7 C and S7D ). Comparison with established signatures These signatures are related to different biological processes, including cellular pyroptosis, glycolysis, immunity, senescence, inflammation, T cells, ferroptosis, autophagy, hypoxia, etc. Remarkably, in both the TCGA-LUSC and META-LUSC datasets, CMLS exhibited better C-index performance than both constructed models ( Fig. 6A and 6B ). We constructed a nomogram containing features and clinical characteristics that were used to accurately forecast the survival of LUSC patients ( Fig. 6C ). The calibration curve demonstrated that the accuracy of the nomogram was accordance with reality ( Fig. 6D ). DCA showed a significantly higher clinical benefit for nomogram than for CMLS alone ( Fig. 6E ). The C-index demonstrated the better predictive performance of the nomogram ( Fig. 6F ). Univariate and multivariate Cox analyses demonstrated that RS was an independent prognostic factor for LUSC ( Figs. 6G and 6H ). Evaluation of the immunological landscape Using the IOBR package, we analysed the TME of LUSC and found that low-CMLS group had remarkably higher levels of immune cell infiltration, such as NK cells, T cells and B cells, suggesting the presence of an immune activation state ( Fig. 7A ). These results indicated that LUSCs with low-CMLS group may be classified as "hot tumors". In contrast, fibroblasts and neutrophils were observed to be predominantly enriched in high-CMLS group, along with molecular markers associated with immunosuppression and exclusion, including the EMT pathway which indicated an immunosuppressive state ( Figs. 7B and 7C ). Therefore, high-CMLS LUSC tended to be classified as a "cold tumor". Furthermore, previously signatures related to better immunotherapy were remarkably enriched in the low-CMLS group ( Fig. 7D ), which also exhibited higher TMB and M1 macrophage infiltration indicating greater immunogenicity ( Figs. 7E and 7F ). Survival analysis revealed that CMLS could effectively complement M1 macrophages as a differentiating factor for patient outcomes ( Fig. 7G ). While low-CMLS group combined with high TMB or M1 macrophage infiltration was associated with better survival outcomes for LUSC patients. Prediction of immunotherapy response The lower CMLS group exhibited significantly improved prognostic outcomes in the IMvigor210 set, indicating a greater benefit from immunotherapy (p < 0.0001; Fig. 8A ). Furthermore, the results revealed a significant decrease in CMLS score for the responder group (CR/PR) compared to the non-responder group (PD/SD) (p < 0.001; Fig. 8B ). These findings were further validated in several immunotherapy validation sets with prognostic information. In the post-immunotherapy population, low-CMLS group was consistently associated with better prognostic outcomes (p < 0.0001, Fig. 8C ; p < 0.0001, Fig. 8D ), and low-CMLS group was also indicative of improved immunotherapy outcomes (GSE91061, p < 0.001; Fig. 8E ). Selection of potential drugs Significant differences were observed in prognosis between patients with different CMLS group, and the GSEA analysis revealed that several pathways, including epithelial mesenchymal transition, coagulation, KRAS signaling, and TNF-α signaling by NF-κB, were remarkably activated in high-CMLS group ( Fig. 9A ). To ensure the validity of our approach, we validated our findings using cisplatin, a well-established treatment for LUSC. Our algorithm demonstrated that low levels of ERCC1 expression were related to a better response to cisplatin therapy, suggesting the potential benefits of chemotherapy for these patients ( Fig. 9B ). Subsequently, we identified five CTRP-derived drugs (canertinib, dasatinib, neratinib, PD318088, and selumetinib; Fig. 9C ) and six PRISM-derived drugs (homoquinolinic-acid, kifunensine, ornithine, Ro-04-5595, romidepsin, and ZLN005; Fig. 9D ). Furthermore, we compared the expression levels of target genes for drug candidates in tumor and normal tissues ( Figs. 9E and 9F ). Discussion The metabolome, genome, transcriptome and proteome are core components of systems biology, and their complicated interactions crucially influence cancer development and metastasis 31 . In the field of biology, machine learning algorithms have been widely used to analyse multiple high-throughput data, providing new methods to solve complex biological problems 31 . Immunotherapy has been considered to be a prevailing therapeutic approach for LUSC, however, the knowledge of biomarkers that predict immune response remains relatively lacking, which greatly limits the development of immunotherapies for LUSC. Li et al. analysed the SEER database data on demographic characteristics, diagnosis time, and treatments of LUSC patients via six machine-learning algorithms to forecast the prognostic status and treatment response of LUSC, further promoting the clinical management and improvement of treatment outcomes for LUSC 34 . Based on the TCGA database, Chen et al. used four different machine learning methods to screen five hub genes to construct prediction models for clinical applications of hepatocellular carcinoma 35 . Besides, machine learning algorithms are an effective tool in analyzing the potential connections and biological significance of multi-omics data 36 . Chu et al. offered research basis for early diagnosis and timely treatment of muscle-invasive urothelial cancer patients by combining multi-omics data and multiple machine learning algorithms 17 . However, there are still fewer multi-omics studies on LUSC. Hence, we applied machine learning analysis to the integrated multi-omics data of LUSC to further filter biomarkers and drug therapy targets, providing new possibilities for delivering personalised treatments. TME is a highly complex local environment including immune cells, tumour cells, epithelial cells, and stromal cells 37 . Complex interactions between tumour cells and multiple constituents in the TME become central drivers of tumour metastasis and accelerated cancer progression 38 . Within the same tumour, enormous variations may exist in terms of category, intrinsic characteristics, stage, and the TME status. It has been observed that the heterogeneity of tumours leads to remarkable differences in the treatment response of patients with the same type of tumour, such as LC 39 . There are marked differences at the molecular level among different individuals of the same tumour or among different regions of the same tumour 39 . Therefore, in-depth investigation of the TME can help to reveal the dynamic evolution of tumours, and unravelling the underlying molecular mechanisms of tumour heterogeneity is crucial for the realisation of precision medicine and the improvement of patient prognosis. In recent years, owing to the rapid advances in molecular biology and artificial intelligence, our recognition of tumour heterogeneity has gradually deepened. We analyzed the integrated sequencing data through multi-omics methods to deeply explore the mechanism of heterogeneity in the process of tumor development, aiming to offer novel insights into the individualised tumours treatment and to enhance the quality of patient's survival. In the study, we gained multi-omics data on LUSC from the TCGA and GEO databases. First, we performed the "ComBat" function to merge diverse datasets to remove the batch effect of each datasets. We utilized PCA analysis to verify the efficacy and robustness of merging the diverse datasets. Besides, we identified two subtypes (CS1 and CS2) from ten multi-omics integrated clustering algorithms. We adopted the "getElite" function to filter the gene features, and we found that CS2 showed the most favourable survival outcome among all subtypes. Immune infiltration analysis revealed that immune cell infiltration was remarkably increased in the CS1 group, whereas it was relatively low in the CS2 group. Subsequent analysis has revealed that CS2 patients have a heightened potential for deriving therapeutic benefits from treatment modalities such as radiation therapy and targeted therapy. Since some sets had smaller samples, we integrated them into the META-LUSC set for the next analyses. We screened 108 PRGs for the construction of the CMLS. With StepCox and CoxBoost algorithms, we filtered out 24 hub genes and performed the final model construction. We divided the sample into different CMLS groups and performed survival analyses to assess their prognostic value.We found that high-CMLS group had a worse clinical prognosis. By comparing CMLS with 12 other signatures, we observed that CMLS has better C-index performance, again demonstrating the robustness and reliability of CMLS. Moreover, we evaluated the immunological landscape of LUSC using "IOBR" package. Low-CMLS group exhibited significantly higher levels of immune cell infiltration, including NK cells, T cells and B cells, suggesting that they may have better survival outcomes. Further prediction of immune response outcomes also demonstrated that the low-CMLS group displayed dramatically improved prognosis and more sensitive immune responses. Finally, we screened potential therapeutic drugs to inform targeted drug treatment strategies for LUSC. Numerous studies have found that during the construction of predictive models, the model behaved well in the training set but may behave poorly in validation sets, and further analyses have revealed that this could be related to overfitting of the model 41 . To effectively mitigate the bias in the assessment of truthfulness performance due to model overfitting, we adopted an integrated modelling strategy, whereby the constructed model is placed in multiple sets to enable more objective evaluation of its truthfulness performance with respect to both the training and validation sets. Referring to previous studies 17 , the mean value of the C-index was used as a model evaluation criterion to synthesise the evaluation results of the models built in each set to construct a highly credible prediction model, thus achieving more accurate prediction results for LUSC. The infiltrating growth of immune cells in TME performs various roles in the process of tumourigenesis, thus influencing tumour progression and the outcome of cancer treatment 42 . Whereas the composition of immune cells varies from one tumour to another, we need to analyse the proportions of immune cells that are responsive to different tumour types in order to explore the underlying mechanisms of tumour development. At present, there are many methods and tools for investigating immune infiltration, however, none of their tools are universally applicable and cannot produce relatively reliable results, thus affecting our understanding of the tumour. The emergence of the IOBR package provides a reliable analytical tool to explore integrated multi-omics data on tumour immune interactions and characterisation of TME 43 . Here, we adopted IOBR package to comprehensively explore and analyse TME, and correlated the analysis results with clinical information to provide reference value for the clinical treatment of LUSC. The low-CMLS group was more sensitive to immunotherapy and had more abundant immune cell types, indicating that CMLS might forecast the prognosis and immune response of LUSC in some degree. Nevertheless, our study remains certain deficiencies. First, we included data from multiple sequencing platforms in our analyses, and although we did our best to minimise the differences between them, differences between the data were still inevitable. Second, the analyse results were not subjected to further experimental validation and mechanistic investigation in this study. Conclusion We identified two subtypes from multi-omics integrated clustering algorithms that were closely related to the prognosis of LUSC. 108 PRGs were selected and adopted ten machine learning methods to build the CMLS for LUSC. Low-CMLS group was more sensitive to immunotherapy and had more abundant immune cell types, indicating that CMLS might forecast the prognosis and immune response of LUSC in some degree. In conclusion, our study provided novel ways to optimise the clinical diagnosis and therapeutic approaches of LUSC. Declarations Competing interests The authors declare no competing interests. Funding This work was supported by the 2023 Natural Science Foundation of Sichuan Province (grant No.2023NSFSC1490), the 2021 National Innovation and Entrepreneurship Training Program for College Students in China (grant No. 202110632050), the 2023 Key research and development project of Sichuan Province (grant No.2023YFS0504), and the 2023 Luzhou-Southwest Medical University cooperative application project (grant No.2023LZXNYDJ007). Author Contribution YD: Writing-original draft, Software, Formal analysis, Conceptualization. XZ: Writing-original draft, Visualization.YW: Writing-review & editing, Supervision. TX: Writing-review & editing, Funding acquisition, Investigation, Supervision. All authors read and approved the final manuscript. Acknowledgement All authors express their gratitude to the online databases TCGA and GEO for their provision of the data. Data Availability The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request. References Sung, H. et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin. 71, 209–249 (2021). Dong, M. et al. The Analysis of Surgical Prognostic Factors and Molecular Typing of Locally Advanced Lung Squamous Cell Carcinomas. Asia Pac J Clin Oncol. 17, e226-e237 (2021). Socinski, M. A. et al. Current and Emergent Therapy Options for Advanced Squamous Cell Lung Cancer. J. Thorac. Oncol. 13, 165–183 (2018). Li, C. et al. Global Burden and Trends of Lung Cancer Incidence and Mortality. Chin Med J (Engl). 136, 1583–1590 (2023). Lau, S., Pan, Y., Velcheti, V. & Wong, K. K. Squamous Cell Lung Cancer: Current Landscape and Future Therapeutic Options. Cancer Cell. 40, 1279–1293 (2022). Herbst, R. S., Morgensztern, D. & Boshoff, C. The Biology and Management of Non-Small Cell Lung Cancer. Nature. 553, 446–454 (2018). Malik, P., Rani, R., Solanki, R., Patel, V. H. & Mukherjee, T. K. Understanding the Feasibility of Chemotherapeutic and Immunotherapeutic Targets Against Non-Small Cell Lung Cancers: An Update of Resistant Responses and Recent Combinatorial Therapies. Exploration of targeted anti-tumor therapy. 4, 850–895 (2023). Camidge, D. R., Doebele, R. C. & Kerr, K. M. Comparing and Contrasting Predictive Biomarkers for Immunotherapy and Targeted Therapy of NSCLC. Nat. Rev. Clin. Oncol. 16, 341–355 (2019). Niederst, M. J. et al. RB Loss in Resistant EGFR Mutant Lung Adenocarcinomas that Transform to Small-Cell Lung Cancer. Nat. Commun. 6, 6377 (2015). Jung, G., Hernandez-Illan, E., Moreira, L., Balaguer, F. & Goel, A. Epigenetics of Colorectal Cancer: Biomarker and Therapeutic Potential. Nat Rev Gastroenterol Hepatol. 17, 111–130 (2020). Picard, M., Scott-Boyer, M. P., Bodein, A., Perin, O. & Droit, A. Integration Strategies of Multi-Omics Data for Machine Learning Analysis. Comput Struct Biotechnol J. 19, 3735–3746 (2021). Ge, Q. et al. Molecular Classifications of Prostate Cancer: Basis for Individualized Risk Stratification and Precision Therapy. Ann. Med. 55, 2279235 (2023). Hasin, Y., Seldin, M. & Lusis, A. Multi-Omics Approaches to Disease. Genome Biol. 18, 83 (2017). Noro, R. et al. A Two-Gene Prognostic Classifier for Early-Stage Lung Squamous Cell Carcinoma in Multiple Large-Scale and Geographically Diverse Cohorts. J. Thorac. Oncol. 12, 65–76 (2017). Bueno, R. et al. Multi-Institutional Prospective Validation of Prognostic mRNA Signatures in Early Stage Squamous Lung Cancer (Alliance). J. Thorac. Oncol. 15, 1748–1757 (2020). Lu, X., Meng, J., Zhou, Y., Jiang, L. & Yan, F. MOVICS: An R Package for Multi-Omics Integration and Visualization in Cancer Subtyping. Bioinformatics. 36, 5539–5541 (2021). Chu, G., Ji, X., Wang, Y. & Niu, H. Integrated Multiomics Analysis and Machine Learning Refine Molecular Subtypes and Prognosis for Muscle-Invasive Urothelial Cancer. Mol Ther Nucleic Acids. 33, 110–126 (2023). Hanzelmann, S., Castelo, R. & Guinney, J. GSVA: Gene Set Variation Analysis for Microarray and RNA-seq Data. BMC Bioinformatics. 14, 7 (2013). Castro, M. A. et al. Regulators of Genetic Risk of Breast Cancer Identified by Integrative Network Analysis. Nat. Genet. 48, 12–21 (2016). Zeng, D. et al. IOBR: Multi-Omics Immuno-Oncology Biological Research to Decode Tumor Microenvironment and Signatures. Front Immunol. 12, 687975 (2021). Sha, D. et al. Tumor Mutational Burden as a Predictive Biomarker in Solid Tumors. Cancer Discov. 10, 1808–1825 (2020). Hoshida, Y., Brunet, J. P., Tamayo, P., Golub, T. R. & Mesirov, J. P. Subclass Mapping: Identifying Common Subtypes in Independent Disease Data Sets. PLoS One. 2, e1195 (2007). Balar, A. V. et al. Atezolizumab as First-Line Treatment in Cisplatin-Ineligible Patients with Locally Advanced and Metastatic Urothelial Carcinoma: A Single-Arm, Multicentre, Phase 2 Trial. Lancet. 389, 67–76 (2017). Hugo, W. et al. Genomic and Transcriptomic Features of Response to Anti-PD-1 Therapy in Metastatic Melanoma. Cell. 168, 542 (2017). Riaz, N. et al. Tumor and Microenvironment Evolution During Immunotherapy with Nivolumab. Cell. 171, 934–949 (2017). Jung, H. et al. DNA Methylation Loss Promotes Immune Evasion of Tumours with High Mutation and Copy Number Load. Nat. Commun. 10, 4278 (2019). Kim, J. Y., Choi, J. K. & Jung, H. Genome-Wide Methylation Patterns Predict Clinical Benefit of Immunotherapy in Lung Cancer. Clin. Epigenetics. 12, 119 (2020). Garraway, L. A. et al. Integrative Genomic Analyses Identify MITF as a Lineage Survival Oncogene Amplified in Malignant Melanoma. Nature. 436, 117–122 (2005). Rees, M. G. et al. Correlating Chemical Sensitivity and Basal Gene Expression Reveals Mechanism of Action. Nat. Chem. Biol. 12, 109–116 (2016). Yu, C. et al. High-Throughput Identification of Genotype-Specific Cancer Vulnerabilities in Mixtures of Barcoded Tumor Cell Lines. Nat. Biotechnol. 34, 419–423 (2016). Hu, Y., Sun, L., Zhang, Y., Lang, J. & Rao, J. Phosphoproteomics Reveals Key Regulatory Kinases and Modulated Pathways Associated with Ovarian Cancer Tumors. Onco Targets Ther. 13, 3595–3605 (2020). Guo, R. et al. Omics Strategies Decipher Therapeutic Discoveries of Traditional Chinese Medicine Against Different Diseases at Multiple Layers Molecular-Level. Pharmacol. Res. 152, 104627 (2020). Brahmer, J. et al. Nivolumab Versus Docetaxel in Advanced Squamous-Cell Non-Small-Cell Lung Cancer. N Engl J Med. 373, 123–135 (2015). Li, G. et al. Prediction of Lymph Node Metastasis of Lung Squamous Cell Carcinoma by Machine Learning Algorithm Classifiers. J. Cancer Res. Ther. 19, 1533–1543 (2023). Chen, D. et al. Integrated Machine Learning and Bioinformatic Analyses Constructed a Novel Stemness-Related Classifier to Predict Prognosis and Immunotherapy Responses for Hepatocellular Carcinoma Patients. Int. J. Biol. Sci. 18, 360–373 (2022). Swanson, K., Wu, E., Zhang, A., Alizadeh, A. A. & Zou, J. From Patterns to Patients: Advances in Clinical Machine Learning for Cancer Diagnosis, Prognosis, and Treatment. Cell. 186, 1772–1791 (2023). Chen, X. & Song, E. The Theory of Tumor Ecosystem. Cancer Commun (Lond). 42, 587–608 (2022). de Visser, K. E. & Joyce, J. A. The Evolving Tumor Microenvironment: From Cancer Initiation to Metastatic Outgrowth. Cancer Cell. 41, 374–403 (2023). de Sousa, V. & Carvalho, L. Heterogeneity in Lung Cancer. Pathobiology: journal of immunopathology, molecular and cellular biology. 85, 96–107 (2018). Dagogo-Jack, I. & Shaw, A. T. Tumour Heterogeneity and Resistance to Cancer Therapies. Nature reviews. Clinical oncology. 15, 81–94 (2018). Ching, T. et al. Opportunities and Obstacles for Deep Learning in Biology and Medicine. J. R. Soc. Interface. 15, (2018). Gentles, A. J. et al. The Prognostic Landscape of Genes and Infiltrating Immune Cells Across Human Cancers. Nat. Med. 21, 938–945 (2015). Zeng, D. et al. IOBR: Multi-Omics Immuno-Oncology Biological Research to Decode Tumor Microenvironment and Signatures. Front Immunol. 12, 687975 (2021). Additional Declarations No competing interests reported. Supplementary Files S1.1.tif Fig S1: The PCA analysis demonstrated the distribution of samples before (A) and after (B) batch effect processing. S2.1.tif Fig S2: Survival analysis between 24 hub genes and OS in TCGA-LUSC cohort. S3.1.tif Fig S3: Survival analysis between 24 hub genes and OS in META-LUSC cohort. S4.1.tif Fig S4: Survival analysis between 24 hub genes and DSS in TCGA-LUSC cohort. S5.1.tif Fig S5: Survival analysis between 24 hub genes and PFI in TCGA-LUSC cohort. S6.1.tif Fig S6: Differential expression analysis and CNV-related analysis of hub genes in CMLS. (A) Differential expression analysis of hub genes in CMLS. (B) Correlation analysis between CNV and mRNA expression level of hub genes in CMLS. (C) The CNV pie distribution indicates the constitution of Heterozygous/Homozygous CNV of each hub genes in CMLS in each cancer. (D) The distribution of heterozygous and homozygous CNV of hub genes in CMLS in each cancer. S7.1.tif Fig S7: Fig S8: Methylation-related analysis and pathway analysis of hub genes in CMLS. (A) Differential methylation expression analysis of hub genes in CMLS. (B) The Heatmap showed the activation and repression status of hub genes in CMLS on related pathways. (C) Correlation analysis between methylation and mRNA expression level of hub genes in CMLS. (D) The pie chart showed the activation and repression status of hub genes in CMLS on related pathways. Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 19 Jun, 2024 Reviews received at journal 17 Jun, 2024 Reviewers agreed at journal 05 Jun, 2024 Reviewers agreed at journal 03 Jun, 2024 Reviewers invited by journal 24 May, 2024 Editor assigned by journal 24 May, 2024 Editor invited by journal 21 May, 2024 Submission checks completed at journal 21 May, 2024 First submitted to journal 16 May, 2024 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-4432088","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":308215786,"identity":"c9d80be6-e951-48fb-b67d-8886b43f3ba8","order_by":0,"name":"Ya Dong","email":"","orcid":"","institution":"the Affiliated Traditional Chinese Medicine Hospital, Southwest Medical University","correspondingAuthor":false,"prefix":"","firstName":"Ya","middleName":"","lastName":"Dong","suffix":""},{"id":308215787,"identity":"841edc5d-7b6e-4a4d-8f09-de99487f1316","order_by":1,"name":"Xiang Zhang","email":"","orcid":"","institution":"west China hospital","correspondingAuthor":false,"prefix":"","firstName":"Xiang","middleName":"","lastName":"Zhang","suffix":""},{"id":308215788,"identity":"b3874722-ad5d-421b-af97-d12b86e4d6b1","order_by":2,"name":"Yuhan Wang","email":"","orcid":"","institution":"Luzhou Longmatan District People's Hospital","correspondingAuthor":false,"prefix":"","firstName":"Yuhan","middleName":"","lastName":"Wang","suffix":""},{"id":308215789,"identity":"8df1eb4a-6e0d-42a4-8cdd-1fb42cc7c2c3","order_by":3,"name":"Tao Xu","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA3UlEQVRIiWNgGAWjYNACAwk5Nvbmgw8SKmqIUc4MxBUWxvw8x5INHpw5RqyWMxWJkjN81CQftjAT4aQb+Qc/fGyTSDC4wcNWkdjAxsDf3p2AV4vkjGRmyZltEnkGt3uP3UjcIcMgcebsBrxa+CWSGaR52ySKDe6cS7uReIYNGBS5+LWwSSQz/wZqSdxwI8esILGNmbAWoC1s0jxnJBJnzsgxYyBKi2TPYzPLGRUS4ECWSDhzjIegXwyOJz6+8cGgDhyVH39U1Mjxt/fi14IBeEhTPgpGwSgYBaMAKwAA+ihIwrXvfdIAAAAASUVORK5CYII=","orcid":"","institution":"the Affiliated Hospital of Southwest Medical University","correspondingAuthor":true,"prefix":"","firstName":"Tao","middleName":"","lastName":"Xu","suffix":""}],"badges":[],"createdAt":"2024-05-16 15:29:13","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-4432088/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-4432088/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":57941255,"identity":"89339502-d29b-475f-b71c-31909307fcb0","added_by":"auto","created_at":"2024-06-07 18:56:19","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":174847,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of the study.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-4432088/v1/e1e8822553c3ff0ec7bd528a.png"},{"id":57941258,"identity":"77eec833-19db-4ddd-b01c-e361eea74a17","added_by":"auto","created_at":"2024-06-07 18:56:19","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":5009552,"visible":true,"origin":"","legend":"\u003cp\u003eThe multiomics integrative consensus subtypes of LUSC. (A) Calculation of the CPI and gap statistic to identify the optimal clustering number for LUSC. (B) The sample similarity of each subgroup was evaluated by the Silhoutte score. (C) A comprehensive heatmap of consensus ensemble subtypes. (D) Consensus clustering heatmap for two prognostic subtypes based on the 10 algorithms. (E) Clustering of LUSC\u003c/p\u003e\n\u003cp\u003epatients by 10 algorithms. (F) The survival outcome of the two subtypes via Kaplan‒Meier analysis.\u003c/p\u003e","description":"","filename":"2.1.png","url":"https://assets-eu.researchsquare.com/files/rs-4432088/v1/6a48ec0bc3ad1fa17befadcf.png"},{"id":57941256,"identity":"c400dc39-7626-46f0-8b76-34c6a4fc1ee8","added_by":"auto","created_at":"2024-06-07 18:56:19","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":2753374,"visible":true,"origin":"","legend":"\u003cp\u003eDifferences in genomic features of the two CS subtypes in LUSC. (A) The different enrichment pathways of two subtypes for LUSC. (B) The heatmap illustrated the regulatory activity profiles of 23 transcription factors (top panel) and potential regulators related to chromatin remodelling (bottom panel). (C) Immune landscape in the TCGA-LUSC cohort. The top annotation of the heatmap displayed the immune enrichment score, stromal enrichment score, and MeTIL. The middle panel revealed the expression of immune checkpoint genes, and the bottom panel showed the enrichment score of 22 TME immune cells.\u003c/p\u003e","description":"","filename":"3.1.png","url":"https://assets-eu.researchsquare.com/files/rs-4432088/v1/9018a6d44e8e5d4578322717.png"},{"id":57941265,"identity":"22b94f1d-b45d-4667-b55a-7f7a8d07f613","added_by":"auto","created_at":"2024-06-07 18:56:19","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":2201393,"visible":true,"origin":"","legend":"\u003cp\u003eMolecular characteristics of LUSC. (A) Verification of LUSC CSs in the META-LUSC datasets. (B) The consistency of CSs with NTP in the TCGA-LUSC cohort.\u003c/p\u003e","description":"","filename":"4.1.png","url":"https://assets-eu.researchsquare.com/files/rs-4432088/v1/1b9a673b9e2133334b1c0383.png"},{"id":57941268,"identity":"e54495a2-df4f-457b-ad37-0f6c93133d93","added_by":"auto","created_at":"2024-06-07 18:56:19","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":3371162,"visible":true,"origin":"","legend":"\u003cp\u003eConstruction of CMLS by integrated machine learning. (A) A total of 99 prediction models were fitted in the TCGA-LUSC and META-LUSC cohorts, and the c-index was further calculated for each cohort model. (B) The CoxBoost method filtered out the most valuable model, established from the 24 hub genes. (C) The univariate Cox regression analysis results of 24 hub genes in TCGA-LUSC and META-LUSC cohorts. (D–E) Survival outcome of LUSC patients with different CMLS groups in the TCGA-LUSC and META-LUSC datasets.\u003c/p\u003e","description":"","filename":"5.1.png","url":"https://assets-eu.researchsquare.com/files/rs-4432088/v1/9f5dec129002c659e233d183.png"},{"id":57941266,"identity":"f78216fa-9d96-45de-ac98-4a6f0d51a0d4","added_by":"auto","created_at":"2024-06-07 18:56:19","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":1090238,"visible":true,"origin":"","legend":"\u003cp\u003eClinical significance \u0026nbsp;of CMLS. (A–B) The CMLS compared with and 12 other different biological models published in the TCGA-LUSC and META-LUSC cohorts. (C) Constructed a nomogram based on the CMLS. (D) Evaluating the predictive accuracy of the nomogram for the OS with calibration curves. (E) DCA analysis showed that the clinical efficiency of the nomogram was markedly higher than that of CMLS alone. (F) Time-dependent C-index chart between the nomogram and CMLS. (G)The univariate Cox analysis model of LUSC. (H) The multivariate Cox analyses model of LUSC.\u003c/p\u003e","description":"","filename":"6.1.png","url":"https://assets-eu.researchsquare.com/files/rs-4432088/v1/ccc5126ce6f4aac9ad8bdb49.png"},{"id":57941264,"identity":"e58cb914-fd85-42b9-93ad-2acc946becd8","added_by":"auto","created_at":"2024-06-07 18:56:19","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":1764577,"visible":true,"origin":"","legend":"\u003cp\u003eThe immune landscape. (A) The different distribution of TME cell type. (B-D) The different distribution of immune suppression, immune exclusion, and immunotherapy biomarkers. (E) The distribution of TMB. (F) Correlation between M1 macrophages and two CMLS groups. (G) Survival analysis with different CMLS and M1 macrophages by Kaplan-Meier curves.\u003c/p\u003e","description":"","filename":"7.1.png","url":"https://assets-eu.researchsquare.com/files/rs-4432088/v1/829ee100facc9df166a9891c.png"},{"id":57941261,"identity":"cdaeb750-001c-460f-9613-75435f1731c3","added_by":"auto","created_at":"2024-06-07 18:56:19","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":637862,"visible":true,"origin":"","legend":"\u003cp\u003ePrediction of immunotherapy response and survival probability in LUSC patients. (A) Kaplan-Meier curve showed the survival probability. (B) The distribution of CMLS in four immunotherapy response groups. (C and D) The low-CMLS group was associated with better prognostic outcomes in GSE78220 and GSE135222 sets. (E) The low-CMLS group was indicative of improved immunotherapy outcomes in GSE91061 set.\u003c/p\u003e","description":"","filename":"8.1.png","url":"https://assets-eu.researchsquare.com/files/rs-4432088/v1/3ca847dc7b99c93845d6b815.png"},{"id":57941260,"identity":"3d610655-8c33-4a31-b325-27cc0602eae9","added_by":"auto","created_at":"2024-06-07 18:56:19","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":1514400,"visible":true,"origin":"","legend":"\u003cp\u003eSelection of potential drugs for LUSC patients. (A) Significant prognosis differences by GSEA analysis. (B) Predicting the sensitivity of cisplatin. (C) Identify five CTRP-derived drugs (canertinib, dasatinib, neratinib, PD318088, and selumetinib) for LUSC. (D) Identify six PRISM-derived drugs (homoquinolinic-acid, kifunensine, ornithine, Ro-04-5595, romidepsin, and ZLN005) for LUSC. (E-F) Comparison the expression levels of target genes for drug candidates in tumor and normal tissues.\u003c/p\u003e","description":"","filename":"9.1.png","url":"https://assets-eu.researchsquare.com/files/rs-4432088/v1/dcc31a9105881a78443752df.png"},{"id":57945344,"identity":"474a04ce-c964-4df8-8547-82fd13106afd","added_by":"auto","created_at":"2024-06-07 19:36:34","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":21810013,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-4432088/v1/e0fbf50a-8019-4f66-abdb-13f101fafa46.pdf"},{"id":57943209,"identity":"5448f934-b80f-4d23-bfcd-bdcf513d72e5","added_by":"auto","created_at":"2024-06-07 19:04:19","extension":"tif","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":282336,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFig S1\u003c/strong\u003e: The PCA analysis demonstrated the distribution of samples before (A) and after (B) batch effect processing.\u003c/p\u003e","description":"","filename":"S1.1.tif","url":"https://assets-eu.researchsquare.com/files/rs-4432088/v1/0bac31baee3a0cc49eebdefe.tif"},{"id":57941259,"identity":"b76d9e9c-65fc-43be-bf17-113d6e8e6917","added_by":"auto","created_at":"2024-06-07 18:56:19","extension":"tif","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":1467692,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFig S2\u003c/strong\u003e: Survival analysis between 24 hub genes and OS in TCGA-LUSC cohort.\u003c/p\u003e","description":"","filename":"S2.1.tif","url":"https://assets-eu.researchsquare.com/files/rs-4432088/v1/a1c25d8e9f8c51f37759295e.tif"},{"id":57944663,"identity":"8a128942-83fd-40b5-8d76-cb06685d396e","added_by":"auto","created_at":"2024-06-07 19:12:19","extension":"tif","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":1454380,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFig S3\u003c/strong\u003e: Survival analysis between 24 hub genes and OS in META-LUSC cohort.\u003c/p\u003e","description":"","filename":"S3.1.tif","url":"https://assets-eu.researchsquare.com/files/rs-4432088/v1/804cb1b0fe17cd2c04f5eb1b.tif"},{"id":57941267,"identity":"c4a14852-607e-457b-87d6-32abaeb60cb5","added_by":"auto","created_at":"2024-06-07 18:56:19","extension":"tif","order_by":4,"title":"","display":"","copyAsset":false,"role":"supplement","size":1511872,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFig S4\u003c/strong\u003e: Survival analysis between 24 hub genes and DSS in TCGA-LUSC cohort.\u003c/p\u003e","description":"","filename":"S4.1.tif","url":"https://assets-eu.researchsquare.com/files/rs-4432088/v1/944137db330e0b023ad1c673.tif"},{"id":57943210,"identity":"3ade9017-c941-4d44-a525-51c90af052a3","added_by":"auto","created_at":"2024-06-07 19:04:19","extension":"tif","order_by":5,"title":"","display":"","copyAsset":false,"role":"supplement","size":1414240,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFig S5\u003c/strong\u003e: Survival analysis between 24 hub genes and PFI in TCGA-LUSC cohort.\u003c/p\u003e","description":"","filename":"S5.1.tif","url":"https://assets-eu.researchsquare.com/files/rs-4432088/v1/d8e08c14a74cdc7ec9de33b0.tif"},{"id":57941269,"identity":"e792d706-57af-4d2d-b313-927e08644aa3","added_by":"auto","created_at":"2024-06-07 18:56:20","extension":"tif","order_by":6,"title":"","display":"","copyAsset":false,"role":"supplement","size":2372820,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFig S6\u003c/strong\u003e: Differential expression analysis and CNV-related analysis of hub genes in CMLS. (A) Differential expression analysis of hub genes in CMLS. (B) Correlation analysis between CNV and mRNA expression level of hub genes in CMLS. (C) The CNV pie distribution indicates the constitution of Heterozygous/Homozygous CNV of each hub genes in CMLS in each cancer. (D) The distribution of heterozygous and homozygous CNV of hub genes in CMLS in each cancer.\u003c/p\u003e","description":"","filename":"S6.1.tif","url":"https://assets-eu.researchsquare.com/files/rs-4432088/v1/ef1fcef1af63db1231c5bb70.tif"},{"id":57941270,"identity":"0439e1b1-706e-4d1b-80c9-42543e6f7fa6","added_by":"auto","created_at":"2024-06-07 18:56:20","extension":"tif","order_by":7,"title":"","display":"","copyAsset":false,"role":"supplement","size":1469024,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFig S7\u003c/strong\u003e: Fig S8: Methylation-related analysis and pathway analysis of hub genes in CMLS. (A) Differential methylation expression analysis of hub genes in CMLS. (B) The Heatmap showed the activation and repression status of hub genes in CMLS on related pathways. (C) Correlation analysis between methylation and mRNA expression level of hub genes in CMLS. (D) The pie chart showed the activation and repression status of hub genes in CMLS on related pathways.\u003c/p\u003e","description":"","filename":"S7.1.tif","url":"https://assets-eu.researchsquare.com/files/rs-4432088/v1/e5d797bc64c3b535837fe43e.tif"}],"financialInterests":"No competing interests reported.","formattedTitle":"Integrating multi-omics analysis and machine learning to identify molecular subtypes and construct prognostic models for lung squamous cell carcinoma","fulltext":[{"header":"Introduction","content":"\u003cp\u003eLUSC had a high morbidity and mortality rate in China, resulting in high social burdens. Most LUSC patients are already in the highly advanced cancer stage at diagnosis, and the clinical treatment is relatively difficult and the prognosis is relatively poor. The progress and application of sequencing technologies and machine learning algorithms offer new therapeutic perspectives and survival opportunities for LUSC patients. First, we gained multi-omics data on LUSC from the TCGA and GEO databases and performed batch effect. A total of ten different clustering methods were adopted to conduct multiomics consensus ensemble analysis. Then, we combined the integration analysis with ten machine learning algorithms to develop a CMLS. Besides, we explored the immune landscape and immunotherapeutic response of LUSC. Lastly, we identified potential therapeutic agents in LUSC. We independently identified two subtypes (CS1 and CS2) from ten multi-omics integrated clustering algorithms and CS2 showed the most favourable survival outcome among all subtypes. Subsequently, we identified 24 PRGs based on markers between subtypes and constructed CMLS using ten machine learning algorithms. In the TCGA-LUSC and META-LUSC sets, patients with high-CMLS group had a poorer clinical prognosis. Moreover, we evaluated the immunological landscape of LUSC using \"IOBR\" package. Low-CMLS group exhibited significantly higher levels of immune cell infiltration, including NK cells, T cells and B cells, suggesting that they may have better survival outcomes. CMLS might predict the prognosis and immune response of LUSC patients in some degree. In conclusion, our study provided novel ways to optimise the clinical diagnosis and therapeutic approaches of LUSC.\u003c/p\u003e\n\u003cp\u003eLung cancer (LC) is the most harmful malignant tumour in the world, and the incidence and mortality rates of LC in China account for about one third of those in the world \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e.As the major histological subtype of NSCLC, LUSC accounts for about 20\u0026ndash;35% of the incidence of LC \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. LUSC is mainly derived from the proximal bronchus and has a greater risk of invading large vessels \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e. The risk of LUSC is much higher in men than in women and the tumour mutation burden is positively correlated with increasing age \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. Smoking, a recognised risk factor, has a strong association with the incidence of LUSC \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. Previous studies have shown that most LUSC patients are already in the highly advanced cancer stage at diagnosis, and the clinical treatment is relatively difficult and the prognosis is relatively poor \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eWith the continuous advancement of modern medical technology, various novel therapeutic strategies have been actively explored and applied to improve the therapeutic effects of LUSC \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. These therapeutic strategies aim to precisely intervene in abnormal cancer cells, inhibit cancer cell progression, and stimulate the body's immunity, such as targeted therapy, immunotherapy, and gene therapy. And targeted therapy is a therapeutic strategy that can accurately intervene in the special signaling pathways of cancer cells. It has greatly improved the therapeutic outcome of LUSC patients, and significantly prolonged the survival time of LUSC patients with genetic mutations \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. Although the application of these advanced therapeutic approaches has provided new promises and opportunities for patients with LUSC, the great heterogeneity of its pathogenesis and biological characteristics has led to great variability in the treatment outcomes of different patients \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. Therefore, how to fully recognise the heterogeneity of LUSC and improve the precision and effectiveness of treatment is an issue that we need to continuously focus on and explore in LUSC treatment research.\u003c/p\u003e\n\u003cp\u003eGene expression is regulated by a range of genetic and epigenetic changes \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e. Multi-omics data analysis is an approach that integrates multiple biological information and aims to probe deeply into the regulatory mechanisms and functional networks of complex biological systems \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e. Through combined integration and analysis of biological data at different levels, we could obtain more comprehensive and accurate information, reveal the association between genes and phenotypes, identify key biological processes and signalling pathways, and further understand the functions and regulatory mechanisms of biological systems \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. Nevertheless, most of the current studies are still mainly focused on exploring the high-throughput data and biological functions of individual-omics, limiting in-depth understanding of the complex interactions between different biological pathways \u003csup\u003e\u003cspan class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. Hence, we developed novel LUSC molecular subtypes through integrated multi-omics analysis to further search for new biomarkers, predict LUSC patients\u0026rsquo; survival and prognosis, and explore potential therapeutic targets, thus providing a reference for individualised therapy.\u003c/p\u003e\n\u003cp\u003eIn the study, multi-omics data of LUSC were gained from the TCGA-LUSC databases, such as mRNA, lncRNA, miRNA, DNA methylation, somatic mutations and clinical data. And we construct an comprehensive consensus subtype of LUSC via ten integration clustering algorithms. Then, 108 prognosis-related genes (PRGs) were selected and adopted ten machine learning methods to build the consensus machine learning-driven signature (CMLS) for LUSC. CMLS exhibited remarkable biological significance for prognosis and clinical application in LUSC. CMLS also proved robust capability in predicting immunotherapy response and targeted drugs therapeutic efficacy. Overall, our results provide an important theoretical basis for the precise differentiation of LUSC biological subtypes, in-depth understanding of tumour heterogeneity and exploration of personalised clinical therapeutic strategy.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e"},{"header":"Materials and methods","content":"\u003cdiv id=\"Sec2\" class=\"Section2\"\u003e \u003ch2\u003ePreparation and pre-processing of LUSC data\u003c/h2\u003e \u003cp\u003eWe obtained multi-omics data on LUSC from the TCGA-LUSC set, such as mRNA, lncRNA, miRNA, DNA methylation, somatic mutations and clinical data. Furthermore, we obtained complete information on three LUSC sets from GEO (GSE73403, GSE74777 and GSE157010) \u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e,\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e. The different datasets were merged using the \u0026ldquo;sva\u0026rdquo; package to remove the batch effect of each dataset. Finally, PCA analyses were used to further validate the validity and stability of the merging of different data sets.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eMultiomics consensus ensemble analysis\u003c/h2\u003e \u003cp\u003eFirstly, we use the \"getElite\" function to filter the features \u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. For continuous variables, we selected the \u0026ldquo;mad\u0026rdquo; parameter to select the top 3,000 genes with the highest variation. We set the \u0026ldquo;cox\u0026rdquo; parameter to recognize prognostic genes in each data dimension (p\u0026thinsp;\u0026lt;\u0026thinsp;0.05). We recognized frequently mutated genes by setting the \"freq\" parameter based on mutation frequency (elite.pct\u0026thinsp;=\u0026thinsp;0.1) \u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e. Then, we used the \"getClustNum\" function to get the optimal number of clusters and used the \"getMOIC\" function to perform the clustering analysis, which contained ten clustering algorithms.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003eSpecific molecular characteristics\u003c/h2\u003e \u003cp\u003eFirstly, we computed the enrichment scores of multiple therapeutically relevant signatures by the GSVA algorithm and constructed a transcriptional regulatory network by the \"RTN\" package \u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e,\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e. Secondly, immune checkpoint and immune/stromal score were also assessed \u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e. Finally, we calculated DNA MeTIL scores and assessed the enrichment of 24 immune cells by GSVA. The clustering results were verified in the LUSC-META set using markers and the consistency of the consensus clustering was compared with the NTP and PAM classifiers.\u003c/p\u003e \u003cdiv id=\"Sec5\" class=\"Section3\"\u003e \u003ch2\u003eDevelopment of CMLS\u003c/h2\u003e \u003cp\u003eGiven that some sets had fewer samples, we merged the three sets into the META-LUSC set and used the \"sva\" package to remove batch effects. To establish CMLS, ten machine learning algorithms are used. The TCGA-LUSC set was used as the training set and the META-LUSC set was used as the validation set.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section3\"\u003e \u003ch2\u003ePrognostic value and clinical applications\u003c/h2\u003e \u003cp\u003e We scored each sample based on the obtained model and categorised the patients into different groups. Survival analyses were performed on the different groups to assess prognostic value. Furthermore, we systematically searched for 12 prognostic signatures related to LUSC. To improve the clinical application, we established a nomogram containing multiple clinical features.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003eExploration of immune landscape\u003c/h2\u003e \u003cp\u003eOn the basis of the \"IOBR\" package, several dozen previous published signatures were obtained with respect to Tumour microenvironment (TME) cell type, immunotherapy response, immunosuppression, and immune exclusion and using a uniform methodology to compute concentration scores for each sample \u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e. We compared the differences in TMB and M1 macrophage distribution between the two groups and regrouped the patients in conjunction with CMLS \u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. For immunotherapy response, we assessed patients' survival in response to immunotherapy and combined subclass mapping to assess immunotherapy response \u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e\u003c/sup\u003e. This was further verified in IMvigor210, GSE78220, GSE91061, and GSE135222 \u003csup\u003e23\u0026ndash;27\u003c/sup\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eIdentification of potential therapeutic agents\u003c/h2\u003e \u003cp\u003eThe activation status of the oncogenic pathway was analysed by the GSEA algorithm between patients with different CMLS groups. Expression data of human CCLs were obtained from CCLE \u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e. Drug sensitivity data for CCLs were obtained from the CTRP v.2.0 and PRISM databases \u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e,\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eIdentification of multi-omics consensus molecular subtypes\u003c/h2\u003e \u003cp\u003e \u003cb\u003eFigure\u0026nbsp;1\u003c/b\u003e illustrated the research process of the study. The PCA analysis demonstrated the distribution of samples before and after batch effect processing (\u003cb\u003eFigs \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003eA and S1B\u003c/b\u003e). Two subtypes were recognised from ten clustering algorithms (\u003cb\u003eFigs.\u0026nbsp;2A and 2B\u003c/b\u003e). Unique molecular expression patterns across transcriptomes (mRNAs, lncRNAs, and miRNAs), epigenetic methylation, and somatic mutations were then further combined by a consensus-integrated approach (\u003cb\u003eFig.\u0026nbsp;2C-2E\u003c/b\u003e). Our classification system correlated strongly with OS (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001; \u003cb\u003eFig.\u0026nbsp;2F\u003c/b\u003e). CS2 showed the most favourable survival outcome.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eSpecific molecular characteristics\u003c/h2\u003e \u003cp\u003eIntriguingly, we discovered that pathways including saccule, epithelium, and vasculature development were remarkably enriched in CS1, whereas pathways including mesenchyme development and mesenchymal cell proliferation were remarkably enriched in CS2. In addition, these pathways, including immunosuppressive oncogenic pathways, are markedly enriched in CS1, whereas CS2 is potentially able to profit from treatments, including radiotherapy or targeted therapies (\u003cb\u003eFig.\u0026nbsp;3A\u003c/b\u003e). We analysed potential regulators related to cancer chromatin remodelling and 23 LUSC transcription factors in order to investigate transcriptome differences. GATA3, RARA, AR, STAT3, GATA6, ESR1, and PGR regulators were significantly activated in CS1, whereas RARG, RXRA, FGFR3, EGFR, HIF1A, FOXM1, ESR2, KLF4, PPARG, and TP63 were particularly enriched in LUSCs (\u003cb\u003eFig.\u0026nbsp;3B\u003c/b\u003e). Considering the critical role of tumour immunity in tumourigenesis and development, we found a significant increase in immune cell infiltration in CS1 (\u003cb\u003eFig.\u0026nbsp;3C\u003c/b\u003e). We selected 2000 genes up-regulated in each subtype as markers. Similar results were obtained in the META-LUSC set (\u003cb\u003eFig.\u0026nbsp;4A\u003c/b\u003e). The consistency of the CSs with the NTP and PAM algorithms was assessed (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001; \u003cb\u003eFigs.\u0026nbsp;4B\u003c/b\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eEstablishment of the CMLS\u003c/h2\u003e \u003cp\u003eWe used univariate Cox analysis to filter 108 PRGs from the TCGA-LUSC and META-LUSC whose expression was significantly correlated with OS. PRGs were incorporated into the integration framework to establish CMLS. We developed signatures based on ten algorithmic combinations and computed all the sets in the mean C-index for each signature to evaluate the predictive power (\u003cb\u003eFig.\u0026nbsp;5A\u003c/b\u003e). The algorithm consisting of StepCox (direction\u0026thinsp;=\u0026thinsp;backward)\u0026thinsp;+\u0026thinsp;CoxBoost retained the highest average C-index to construct the final signature. The StepCox (direction\u0026thinsp;=\u0026thinsp;backward) method recognized the most valuable PRGs, and the CoxBoost method filtered out the most valuable model, established from the 24 hub genes (\u003cb\u003eFigs.\u0026nbsp;5B and 5C\u003c/b\u003e). We computed CMLS scores for each sample across all sets. In the TCGA-LUSC and META-LUSC sets, patients with high-CMLS group had a poorer clinical prognosis (\u003cb\u003eFigs.\u0026nbsp;5D and 5E\u003c/b\u003e). We validated the prognostic value of these genes by survival analysis, and the results derived were in general agreement with those of Cox analysis (\u003cb\u003eFigs \u003cspan refid=\"MOESM2\" class=\"InternalRef\"\u003eS2\u003c/span\u003e and S3\u003c/b\u003e). These genes were significantly related to DSS and PFS in LUSC, emphasising their close prognostic correlation with patients (\u003cb\u003eFig \u003cspan refid=\"MOESM4\" class=\"InternalRef\"\u003eS4\u003c/span\u003e and S5\u003c/b\u003e).\u003c/p\u003e \u003cp\u003eSubsequently, we systematically examined the multi-omics phenotypes of CMLS in 33 different cancer types using the GSCALite database. CHEK2, KIF22, BCAT1, RPIA, ITGA3, GRHL3 and FN1 genes were highly expressed in some cancer tissues, while GPX3, MMRN1, AKAP12, KANK4, PINK1, F13A1, ALDH7A1, STAB1 and VSIG4 genes were lowly expressed in some cancer tissues (\u003cb\u003eFig \u003cspan refid=\"MOESM6\" class=\"InternalRef\"\u003eS6\u003c/span\u003eA\u003c/b\u003e). The mRNA expression levels of CMLS genes were positively related to CNVs in most cancer types, especially PINK1, BAG4, RPIA, CHEK2, ALDH7A1, ITGB1 and KIF22 (\u003cb\u003eFig \u003cspan refid=\"MOESM6\" class=\"InternalRef\"\u003eS6\u003c/span\u003eB\u003c/b\u003e).\u003c/p\u003e \u003cp\u003eBy analysing the CNV frequency changes, the CNVs of these gene differed significantly across cancer types, with NLRP3, SOX2, BCAT1 and GPR160 having the highest CNV frequencies, predominantly copy number heterozygous amplifications (\u003cb\u003eFig \u003cspan refid=\"MOESM6\" class=\"InternalRef\"\u003eS6\u003c/span\u003eC and S6D\u003c/b\u003e). Furthermore, methylation levels of these genes in most tumour types differed remarkably between tumour and normal samples (\u003cb\u003eFig \u003cspan refid=\"MOESM7\" class=\"InternalRef\"\u003eS7\u003c/span\u003eA\u003c/b\u003e). In most cancers, the methylation levels were negatively related to the mRNA levels (\u003cb\u003eFig \u003cspan refid=\"MOESM7\" class=\"InternalRef\"\u003eS7\u003c/span\u003eB\u003c/b\u003e). These genes activated the pan-cancer EMT pathway and had a remarkable inhibitory effect on the RAS/MAPK pathway (\u003cb\u003eFig \u003cspan refid=\"MOESM7\" class=\"InternalRef\"\u003eS7\u003c/span\u003eC and S7D\u003c/b\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eComparison with established signatures\u003c/h2\u003e \u003cp\u003eThese signatures are related to different biological processes, including cellular pyroptosis, glycolysis, immunity, senescence, inflammation, T cells, ferroptosis, autophagy, hypoxia, etc. Remarkably, in both the TCGA-LUSC and META-LUSC datasets, CMLS exhibited better C-index performance than both constructed models (\u003cb\u003eFig.\u0026nbsp;6A and 6B\u003c/b\u003e). We constructed a nomogram containing features and clinical characteristics that were used to accurately forecast the survival of LUSC patients (\u003cb\u003eFig.\u0026nbsp;6C\u003c/b\u003e). The calibration curve demonstrated that the accuracy of the nomogram was accordance with reality (\u003cb\u003eFig.\u0026nbsp;6D\u003c/b\u003e). DCA showed a significantly higher clinical benefit for nomogram than for CMLS alone (\u003cb\u003eFig.\u0026nbsp;6E\u003c/b\u003e). The C-index demonstrated the better predictive performance of the nomogram (\u003cb\u003eFig.\u0026nbsp;6F\u003c/b\u003e). Univariate and multivariate Cox analyses demonstrated that RS was an independent prognostic factor for LUSC (\u003cb\u003eFigs.\u0026nbsp;6G and 6H\u003c/b\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eEvaluation of the immunological landscape\u003c/h2\u003e \u003cp\u003eUsing the IOBR package, we analysed the TME of LUSC and found that low-CMLS group had remarkably higher levels of immune cell infiltration, such as NK cells, T cells and B cells, suggesting the presence of an immune activation state (\u003cb\u003eFig.\u0026nbsp;7A\u003c/b\u003e). These results indicated that LUSCs with low-CMLS group may be classified as \"hot tumors\". In contrast, fibroblasts and neutrophils were observed to be predominantly enriched in high-CMLS group, along with molecular markers associated with immunosuppression and exclusion, including the EMT pathway which indicated an immunosuppressive state (\u003cb\u003eFigs.\u0026nbsp;7B and 7C\u003c/b\u003e). Therefore, high-CMLS LUSC tended to be classified as a \"cold tumor\". Furthermore, previously signatures related to better immunotherapy were remarkably enriched in the low-CMLS group (\u003cb\u003eFig.\u0026nbsp;7D\u003c/b\u003e), which also exhibited higher TMB and M1 macrophage infiltration indicating greater immunogenicity (\u003cb\u003eFigs.\u0026nbsp;7E and 7F\u003c/b\u003e). Survival analysis revealed that CMLS could effectively complement M1 macrophages as a differentiating factor for patient outcomes (\u003cb\u003eFig.\u0026nbsp;7G\u003c/b\u003e). While low-CMLS group combined with high TMB or M1 macrophage infiltration was associated with better survival outcomes for LUSC patients.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003ePrediction of immunotherapy response\u003c/h2\u003e \u003cp\u003eThe lower CMLS group exhibited significantly improved prognostic outcomes in the IMvigor210 set, indicating a greater benefit from immunotherapy (p\u0026thinsp;\u0026lt;\u0026thinsp;0.0001; \u003cb\u003eFig.\u0026nbsp;8A\u003c/b\u003e). Furthermore, the results revealed a significant decrease in CMLS score for the responder group (CR/PR) compared to the non-responder group (PD/SD) (p\u0026thinsp;\u0026lt;\u0026thinsp;0.001; \u003cb\u003eFig.\u0026nbsp;8B\u003c/b\u003e). These findings were further validated in several immunotherapy validation sets with prognostic information. In the post-immunotherapy population, low-CMLS group was consistently associated with better prognostic outcomes (p\u0026thinsp;\u0026lt;\u0026thinsp;0.0001, \u003cb\u003eFig.\u0026nbsp;8C\u003c/b\u003e; p\u0026thinsp;\u0026lt;\u0026thinsp;0.0001, \u003cb\u003eFig.\u0026nbsp;8D\u003c/b\u003e), and low-CMLS group was also indicative of improved immunotherapy outcomes (GSE91061, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001; \u003cb\u003eFig.\u0026nbsp;8E\u003c/b\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eSelection of potential drugs\u003c/h2\u003e \u003cp\u003eSignificant differences were observed in prognosis between patients with different CMLS group, and the GSEA analysis revealed that several pathways, including epithelial mesenchymal transition, coagulation, KRAS signaling, and TNF-α signaling by NF-κB, were remarkably activated in high-CMLS group (\u003cb\u003eFig.\u0026nbsp;9A\u003c/b\u003e). To ensure the validity of our approach, we validated our findings using cisplatin, a well-established treatment for LUSC. Our algorithm demonstrated that low levels of ERCC1 expression were related to a better response to cisplatin therapy, suggesting the potential benefits of chemotherapy for these patients (\u003cb\u003eFig.\u0026nbsp;9B\u003c/b\u003e). Subsequently, we identified five CTRP-derived drugs (canertinib, dasatinib, neratinib, PD318088, and selumetinib; \u003cb\u003eFig.\u0026nbsp;9C\u003c/b\u003e) and six PRISM-derived drugs (homoquinolinic-acid, kifunensine, ornithine, Ro-04-5595, romidepsin, and ZLN005; \u003cb\u003eFig.\u0026nbsp;9D\u003c/b\u003e). Furthermore, we compared the expression levels of target genes for drug candidates in tumor and normal tissues (\u003cb\u003eFigs.\u0026nbsp;9E and 9F\u003c/b\u003e).\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe metabolome, genome, transcriptome and proteome are core components of systems biology, and their complicated interactions crucially influence cancer development and metastasis \u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e. In the field of biology, machine learning algorithms have been widely used to analyse multiple high-throughput data, providing new methods to solve complex biological problems \u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e. Immunotherapy has been considered to be a prevailing therapeutic approach for LUSC, however, the knowledge of biomarkers that predict immune response remains relatively lacking, which greatly limits the development of immunotherapies for LUSC. Li et al. analysed the SEER database data on demographic characteristics, diagnosis time, and treatments of LUSC patients via six machine-learning algorithms to forecast the prognostic status and treatment response of LUSC, further promoting the clinical management and improvement of treatment outcomes for LUSC \u003csup\u003e\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u003c/sup\u003e. Based on the TCGA database, Chen et al. used four different machine learning methods to screen five hub genes to construct prediction models for clinical applications of hepatocellular carcinoma \u003csup\u003e\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u003c/sup\u003e. Besides, machine learning algorithms are an effective tool in analyzing the potential connections and biological significance of multi-omics data \u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e. Chu et al. offered research basis for early diagnosis and timely treatment of muscle-invasive urothelial cancer patients by combining multi-omics data and multiple machine learning algorithms \u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e. However, there are still fewer multi-omics studies on LUSC. Hence, we applied machine learning analysis to the integrated multi-omics data of LUSC to further filter biomarkers and drug therapy targets, providing new possibilities for delivering personalised treatments.\u003c/p\u003e \u003cp\u003eTME is a highly complex local environment including immune cells, tumour cells, epithelial cells, and stromal cells \u003csup\u003e\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u003c/sup\u003e. Complex interactions between tumour cells and multiple constituents in the TME become central drivers of tumour metastasis and accelerated cancer progression \u003csup\u003e\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e\u003c/sup\u003e. Within the same tumour, enormous variations may exist in terms of category, intrinsic characteristics, stage, and the TME status. It has been observed that the heterogeneity of tumours leads to remarkable differences in the treatment response of patients with the same type of tumour, such as LC \u003csup\u003e\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e. There are marked differences at the molecular level among different individuals of the same tumour or among different regions of the same tumour \u003csup\u003e\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u003c/sup\u003e. Therefore, in-depth investigation of the TME can help to reveal the dynamic evolution of tumours, and unravelling the underlying molecular mechanisms of tumour heterogeneity is crucial for the realisation of precision medicine and the improvement of patient prognosis. In recent years, owing to the rapid advances in molecular biology and artificial intelligence, our recognition of tumour heterogeneity has gradually deepened. We analyzed the integrated sequencing data through multi-omics methods to deeply explore the mechanism of heterogeneity in the process of tumor development, aiming to offer novel insights into the individualised tumours treatment and to enhance the quality of patient's survival.\u003c/p\u003e \u003cp\u003eIn the study, we gained multi-omics data on LUSC from the TCGA and GEO databases. First, we performed the \"ComBat\" function to merge diverse datasets to remove the batch effect of each datasets. We utilized PCA analysis to verify the efficacy and robustness of merging the diverse datasets. Besides, we identified two subtypes (CS1 and CS2) from ten multi-omics integrated clustering algorithms.\u003c/p\u003e \u003cp\u003eWe adopted the \"getElite\" function to filter the gene features, and we found that CS2 showed the most favourable survival outcome among all subtypes. Immune infiltration analysis revealed that immune cell infiltration was remarkably increased in the CS1 group, whereas it was relatively low in the CS2 group. Subsequent analysis has revealed that CS2 patients have a heightened potential for deriving therapeutic benefits from treatment modalities such as radiation therapy and targeted therapy. Since some sets had smaller samples, we integrated them into the META-LUSC set for the next analyses. We screened 108 PRGs for the construction of the CMLS. With StepCox and CoxBoost algorithms, we filtered out 24 hub genes and performed the final model construction. We divided the sample into different CMLS groups and performed survival analyses to assess their prognostic value.We found that high-CMLS group had a worse clinical prognosis. By comparing CMLS with 12 other signatures, we observed that CMLS has better C-index performance, again demonstrating the robustness and reliability of CMLS. Moreover, we evaluated the immunological landscape of LUSC using \"IOBR\" package. Low-CMLS group exhibited significantly higher levels of immune cell infiltration, including NK cells, T cells and B cells, suggesting that they may have better survival outcomes. Further prediction of immune response outcomes also demonstrated that the low-CMLS group displayed dramatically improved prognosis and more sensitive immune responses. Finally, we screened potential therapeutic drugs to inform targeted drug treatment strategies for LUSC.\u003c/p\u003e \u003cp\u003eNumerous studies have found that during the construction of predictive models, the model behaved well in the training set but may behave poorly in validation sets, and further analyses have revealed that this could be related to overfitting of the model \u003csup\u003e\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e\u003c/sup\u003e. To effectively mitigate the bias in the assessment of truthfulness performance due to model overfitting, we adopted an integrated modelling strategy, whereby the constructed model is placed in multiple sets to enable more objective evaluation of its truthfulness performance with respect to both the training and validation sets. Referring to previous studies \u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e, the mean value of the C-index was used as a model evaluation criterion to synthesise the evaluation results of the models built in each set to construct a highly credible prediction model, thus achieving more accurate prediction results for LUSC. The infiltrating growth of immune cells in TME performs various roles in the process of tumourigenesis, thus influencing tumour progression and the outcome of cancer treatment \u003csup\u003e\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u003c/sup\u003e. Whereas the composition of immune cells varies from one tumour to another, we need to analyse the proportions of immune cells that are responsive to different tumour types in order to explore the underlying mechanisms of tumour development. At present, there are many methods and tools for investigating immune infiltration, however, none of their tools are universally applicable and cannot produce relatively reliable results, thus affecting our understanding of the tumour. The emergence of the IOBR package provides a reliable analytical tool to explore integrated multi-omics data on tumour immune interactions and characterisation of TME \u003csup\u003e\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e\u003c/sup\u003e. Here, we adopted IOBR package to comprehensively explore and analyse TME, and correlated the analysis results with clinical information to provide reference value for the clinical treatment of LUSC. The low-CMLS group was more sensitive to immunotherapy and had more abundant immune cell types, indicating that CMLS might forecast the prognosis and immune response of LUSC in some degree.\u003c/p\u003e \u003cp\u003eNevertheless, our study remains certain deficiencies. First, we included data from multiple sequencing platforms in our analyses, and although we did our best to minimise the differences between them, differences between the data were still inevitable. Second, the analyse results were not subjected to further experimental validation and mechanistic investigation in this study.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eWe identified two subtypes from multi-omics integrated clustering algorithms that were closely related to the prognosis of LUSC. 108 PRGs were selected and adopted ten machine learning methods to build the CMLS for LUSC. Low-CMLS group was more sensitive to immunotherapy and had more abundant immune cell types, indicating that CMLS might forecast the prognosis and immune response of LUSC in some degree. In conclusion, our study provided novel ways to optimise the clinical diagnosis and therapeutic approaches of LUSC.\u003c/p\u003e "},{"header":"Declarations","content":"\u003ch2\u003eCompeting interests\u003c/h2\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003ch2\u003e\u0026nbsp;\u003c/h2\u003e\n\u003ch2\u003eFunding\u003c/h2\u003e\n\u003cp\u003eThis work was supported by the 2023 Natural Science Foundation of Sichuan Province (grant No.2023NSFSC1490), the 2021 National Innovation and Entrepreneurship Training Program for College Students in China (grant No. 202110632050), the 2023 Key research and development project of Sichuan Province (grant No.2023YFS0504), and the 2023 Luzhou-Southwest Medical University cooperative application project (grant No.2023LZXNYDJ007).\u003c/p\u003e\n\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\n\u003cp\u003eYD: Writing-original draft, Software, Formal analysis, Conceptualization. XZ: Writing-original draft, Visualization.YW: Writing-review \u0026amp; editing, Supervision. TX: Writing-review \u0026amp; editing, Funding acquisition, Investigation, Supervision. All authors read and approved the final manuscript.\u003c/p\u003e\n\u003ch2\u003eAcknowledgement\u003c/h2\u003e\n\u003cp\u003eAll authors express their gratitude to the online databases TCGA and GEO for their provision of the data.\u003c/p\u003e\n\u003ch2\u003eData Availability\u003c/h2\u003e\n\u003cp\u003eThe datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eSung, H. et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin. 71, 209\u0026ndash;249 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDong, M. et al. The Analysis of Surgical Prognostic Factors and Molecular Typing of Locally Advanced Lung Squamous Cell Carcinomas. Asia Pac J Clin Oncol. 17, e226-e237 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSocinski, M. A. et al. Current and Emergent Therapy Options for Advanced Squamous Cell Lung Cancer. J. Thorac. Oncol. 13, 165\u0026ndash;183 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, C. et al. Global Burden and Trends of Lung Cancer Incidence and Mortality. Chin Med J (Engl). 136, 1583\u0026ndash;1590 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLau, S., Pan, Y., Velcheti, V. \u0026amp; Wong, K. K. Squamous Cell Lung Cancer: Current Landscape and Future Therapeutic Options. Cancer Cell. 40, 1279\u0026ndash;1293 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHerbst, R. S., Morgensztern, D. \u0026amp; Boshoff, C. The Biology and Management of Non-Small Cell Lung Cancer. Nature. 553, 446\u0026ndash;454 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMalik, P., Rani, R., Solanki, R., Patel, V. H. \u0026amp; Mukherjee, T. K. Understanding the Feasibility of Chemotherapeutic and Immunotherapeutic Targets Against Non-Small Cell Lung Cancers: An Update of Resistant Responses and Recent Combinatorial Therapies. Exploration of targeted anti-tumor therapy. 4, 850\u0026ndash;895 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCamidge, D. R., Doebele, R. C. \u0026amp; Kerr, K. M. Comparing and Contrasting Predictive Biomarkers for Immunotherapy and Targeted Therapy of NSCLC. Nat. Rev. Clin. Oncol. 16, 341\u0026ndash;355 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNiederst, M. J. et al. RB Loss in Resistant EGFR Mutant Lung Adenocarcinomas that Transform to Small-Cell Lung Cancer. Nat. Commun. 6, 6377 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJung, G., Hernandez-Illan, E., Moreira, L., Balaguer, F. \u0026amp; Goel, A. Epigenetics of Colorectal Cancer: Biomarker and Therapeutic Potential. Nat Rev Gastroenterol Hepatol. 17, 111\u0026ndash;130 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePicard, M., Scott-Boyer, M. P., Bodein, A., Perin, O. \u0026amp; Droit, A. Integration Strategies of Multi-Omics Data for Machine Learning Analysis. Comput Struct Biotechnol J. 19, 3735\u0026ndash;3746 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGe, Q. et al. Molecular Classifications of Prostate Cancer: Basis for Individualized Risk Stratification and Precision Therapy. Ann. Med. 55, 2279235 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHasin, Y., Seldin, M. \u0026amp; Lusis, A. Multi-Omics Approaches to Disease. Genome Biol. 18, 83 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNoro, R. et al. A Two-Gene Prognostic Classifier for Early-Stage Lung Squamous Cell Carcinoma in Multiple Large-Scale and Geographically Diverse Cohorts. J. Thorac. Oncol. 12, 65\u0026ndash;76 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBueno, R. et al. Multi-Institutional Prospective Validation of Prognostic mRNA Signatures in Early Stage Squamous Lung Cancer (Alliance). J. Thorac. Oncol. 15, 1748\u0026ndash;1757 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLu, X., Meng, J., Zhou, Y., Jiang, L. \u0026amp; Yan, F. MOVICS: An R Package for Multi-Omics Integration and Visualization in Cancer Subtyping. Bioinformatics. 36, 5539\u0026ndash;5541 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChu, G., Ji, X., Wang, Y. \u0026amp; Niu, H. Integrated Multiomics Analysis and Machine Learning Refine Molecular Subtypes and Prognosis for Muscle-Invasive Urothelial Cancer. Mol Ther Nucleic Acids. 33, 110\u0026ndash;126 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHanzelmann, S., Castelo, R. \u0026amp; Guinney, J. GSVA: Gene Set Variation Analysis for Microarray and RNA-seq Data. BMC Bioinformatics. 14, 7 (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCastro, M. A. et al. Regulators of Genetic Risk of Breast Cancer Identified by Integrative Network Analysis. Nat. Genet. 48, 12\u0026ndash;21 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZeng, D. et al. IOBR: Multi-Omics Immuno-Oncology Biological Research to Decode Tumor Microenvironment and Signatures. Front Immunol. 12, 687975 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSha, D. et al. Tumor Mutational Burden as a Predictive Biomarker in Solid Tumors. Cancer Discov. 10, 1808\u0026ndash;1825 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHoshida, Y., Brunet, J. P., Tamayo, P., Golub, T. R. \u0026amp; Mesirov, J. P. Subclass Mapping: Identifying Common Subtypes in Independent Disease Data Sets. PLoS One. 2, e1195 (2007).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBalar, A. V. et al. Atezolizumab as First-Line Treatment in Cisplatin-Ineligible Patients with Locally Advanced and Metastatic Urothelial Carcinoma: A Single-Arm, Multicentre, Phase 2 Trial. Lancet. 389, 67\u0026ndash;76 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHugo, W. et al. Genomic and Transcriptomic Features of Response to Anti-PD-1 Therapy in Metastatic Melanoma. Cell. 168, 542 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRiaz, N. et al. Tumor and Microenvironment Evolution During Immunotherapy with Nivolumab. Cell. 171, 934\u0026ndash;949 (2017).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJung, H. et al. DNA Methylation Loss Promotes Immune Evasion of Tumours with High Mutation and Copy Number Load. Nat. Commun. 10, 4278 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, J. Y., Choi, J. K. \u0026amp; Jung, H. Genome-Wide Methylation Patterns Predict Clinical Benefit of Immunotherapy in Lung Cancer. Clin. Epigenetics. 12, 119 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGarraway, L. A. et al. Integrative Genomic Analyses Identify MITF as a Lineage Survival Oncogene Amplified in Malignant Melanoma. Nature. 436, 117\u0026ndash;122 (2005).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRees, M. G. et al. Correlating Chemical Sensitivity and Basal Gene Expression Reveals Mechanism of Action. Nat. Chem. Biol. 12, 109\u0026ndash;116 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYu, C. et al. High-Throughput Identification of Genotype-Specific Cancer Vulnerabilities in Mixtures of Barcoded Tumor Cell Lines. Nat. Biotechnol. 34, 419\u0026ndash;423 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHu, Y., Sun, L., Zhang, Y., Lang, J. \u0026amp; Rao, J. Phosphoproteomics Reveals Key Regulatory Kinases and Modulated Pathways Associated with Ovarian Cancer Tumors. Onco Targets Ther. 13, 3595\u0026ndash;3605 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuo, R. et al. Omics Strategies Decipher Therapeutic Discoveries of Traditional Chinese Medicine Against Different Diseases at Multiple Layers Molecular-Level. Pharmacol. Res. 152, 104627 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBrahmer, J. et al. Nivolumab Versus Docetaxel in Advanced Squamous-Cell Non-Small-Cell Lung Cancer. N Engl J Med. 373, 123\u0026ndash;135 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, G. et al. Prediction of Lymph Node Metastasis of Lung Squamous Cell Carcinoma by Machine Learning Algorithm Classifiers. J. Cancer Res. Ther. 19, 1533\u0026ndash;1543 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, D. et al. Integrated Machine Learning and Bioinformatic Analyses Constructed a Novel Stemness-Related Classifier to Predict Prognosis and Immunotherapy Responses for Hepatocellular Carcinoma Patients. Int. J. Biol. Sci. 18, 360\u0026ndash;373 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSwanson, K., Wu, E., Zhang, A., Alizadeh, A. A. \u0026amp; Zou, J. From Patterns to Patients: Advances in Clinical Machine Learning for Cancer Diagnosis, Prognosis, and Treatment. Cell. 186, 1772\u0026ndash;1791 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, X. \u0026amp; Song, E. The Theory of Tumor Ecosystem. Cancer Commun (Lond). 42, 587\u0026ndash;608 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ede Visser, K. E. \u0026amp; Joyce, J. A. The Evolving Tumor Microenvironment: From Cancer Initiation to Metastatic Outgrowth. Cancer Cell. 41, 374\u0026ndash;403 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ede Sousa, V. \u0026amp; Carvalho, L. Heterogeneity in Lung Cancer. Pathobiology: journal of immunopathology, molecular and cellular biology. 85, 96\u0026ndash;107 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDagogo-Jack, I. \u0026amp; Shaw, A. T. Tumour Heterogeneity and Resistance to Cancer Therapies. Nature reviews. Clinical oncology. 15, 81\u0026ndash;94 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChing, T. et al. Opportunities and Obstacles for Deep Learning in Biology and Medicine. J. R. Soc. Interface. 15, (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGentles, A. J. et al. The Prognostic Landscape of Genes and Infiltrating Immune Cells Across Human Cancers. Nat. Med. 21, 938\u0026ndash;945 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZeng, D. et al. IOBR: Multi-Omics Immuno-Oncology Biological Research to Decode Tumor Microenvironment and Signatures. Front Immunol. 12, 687975 (2021).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"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":"Lung squamous cell carcinoma, Multiomics analysis, Machine learning algorithm, Consensus machine learning-driven signature, Prognosis","lastPublishedDoi":"10.21203/rs.3.rs-4432088/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-4432088/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"LUSC had a high morbidity and mortality rate in China, resulting in high social burdens. Most LUSC patients are already in the highly advanced cancer stage at diagnosis, and the clinical treatment is relatively difficult and the prognosis is relatively poor. The progress and application of sequencing technologies and machine learning algorithms offer new therapeutic perspectives and survival opportunities for LUSC patients. First, we gained multi-omics data on LUSC from the TCGA and GEO databases and performed batch effect. A total of ten different clustering methods were adopted to conduct multiomics consensus ensemble analysis. Then, we combined the integration analysis with ten machine learning algorithms to develop a CMLS. Besides, we explored the immune landscape and immunotherapeutic response of LUSC. Lastly, we identified potential therapeutic agents in LUSC. We independently identified two subtypes (CS1 and CS2) from ten multi-omics integrated clustering algorithms and CS2 showed the most favourable survival outcome among all subtypes. Subsequently, we identified 24 PRGs based on markers between subtypes and constructed CMLS using ten machine learning algorithms. In the TCGA-LUSC and META-LUSC sets, patients with high-CMLS group had a poorer clinical prognosis. Moreover, we evaluated the immunological landscape of LUSC using \"IOBR\" package. Low-CMLS group exhibited significantly higher levels of immune cell infiltration, including NK cells, T cells and B cells, suggesting that they may have better survival outcomes. CMLS might predict the prognosis and immune response of LUSC patients in some degree. In conclusion, our study provided novel ways to optimise the clinical diagnosis and therapeutic approaches of LUSC.","manuscriptTitle":"Integrating multi-omics analysis and machine learning to identify molecular subtypes and construct prognostic models for lung squamous cell carcinoma","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-06-07 18:56:14","doi":"10.21203/rs.3.rs-4432088/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2024-06-19T10:16:52+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-06-17T14:47:05+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"284081152842318742480294157070816615988","date":"2024-06-05T18:00:33+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"303889766121793336898385389128388371687","date":"2024-06-03T08:13:09+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-05-24T09:46:38+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-05-24T09:31:39+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2024-05-21T14:19:40+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-05-21T14:15:59+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2024-05-16T15:27:02+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":"68fbcc28-e210-497a-b480-aa6fd490a20d","owner":[],"postedDate":"June 7th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":32558708,"name":"Biological sciences/Cancer"},{"id":32558709,"name":"Health sciences/Oncology"}],"tags":[],"updatedAt":"2024-06-07T18:56:14+00:00","versionOfRecord":[],"versionCreatedAt":"2024-06-07 18:56:14","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-4432088","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-4432088","identity":"rs-4432088","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}