Introduction

Chronic obstructive pulmonary disease (COPD) has become a public health challenge due to its high prevalence worldwide and the associated disability, morbidity, mortality and socioeconomic burden.1–3 Rehman et al reported a prevalence of COPD of 3.4–13.4%3 in Europe and the United States and 3.5–19.1% in Asia due to urbanisation, industrial pollution, tanneries and high household use of biofuels.4,5 The number of COPD deaths in China exceeded 900,000 in 2013 and COPD is now the third leading cause of death in China. Typical symptoms of COPD include dyspnoea, chronic cough and sputum production, and spirometry is considered the gold standard in the diagnosis of COPD,6 however, early-stage COPD often goes undetected, resulting in patients with early-stage COPD being under-diagnosed and under-treated. Therefore, there is a need to develop a reliable early warning method for COPD. This will lead to early intervention and treatment of COPD.

A single nucleotide polymorphism (SNP) is a type of DNA polymorphism that refers to a change in a single nucleotide that result in different DNA sequences that, after transcription and translation, result in functional differences in the final expression of the protein.7 SNPs are the most common genetic variation in the human genome and the most common form of DNA sequence variation that reflects individual differences. On average, there are about 1 SNPs per 1000 bases, and only a fraction of these specific SNPs are associated with disease.8 They are known as the third generation of genetic markers because of their widespread use, large numbers, stable genetic properties and ease of automated batch detection. Currently, with the development of SNPs detection technology, it is widely used in the study of common and complex diseases, medical diagnosis, drug development and the exploration of disease susceptibility genes.9

In a genome-wide association study (GWAS), one study analysed a large cohort of patients and found that as many as 3 ~ 1 million SNPs in cases were COPD disease-associated loci.10 In 2017, Wain LV et al found that 95 loci in FEVI, FVC and FEV1/FVC were associated with COPD risk, and enrichment analysis showed that these loci were associated with lung development, elastic fibres and epigenetic regulatory pathways.11 In 2019, a study found 82 loci associated with COPD, with a total of 156 risk genes located in these loci.12 Recently, Shrine N et al identified 257 loci associated with lung function phenotypes, of which 107 were identified as risk genes for COPD.13 Currently, for the set of SNPs genes that are significantly associated with COPD susceptibility, it is crucial that gene targeting and identification of individual disease-causing variants is carried out in subsequent studies.14

Least Absolute Shrinkage and Selection Operator (LASSO) method is a statistical approach that integrates feature selection with regularization, which improves the predictive power of models by applying a penalty to the magnitude of the coefficients, thus reducing the complexity and preventing overfitting.15 Recursive Feature Elimination (RFE) is an efficient machine learning technique suitable for both classification and regression. It works by determining the optimal dividing hyperplane in the feature space to distinguish between classes or to minimize errors in fitting the regression function.16 Random Forest algorithm is a form of ensemble learning that operates by generating an ensemble of decision trees, and it enhances predictive accuracy and reliability by considering the majority vote among the trees for classification tasks or by averaging their predictions in the case of regression.17 Cross-validation with Random Forest, LASSO and RFE algorithms was performed to mitigate the risk of overfitting.

Therefore, in this study, we used a variety of statistical algorithms to construct models by one-way logistic regression analysis, LASSO regression, RFE Algorithm and Random Forest with feature selection and screening of key SNPs, plotted column-line plots based on the SNPs screened by the best model, and assessed the discriminative power of the model in the original dataset using calibration curves and receiver operating characteristic (ROC) curves. To our knowledge, this is the first study to investigate the contribution of SNPs to COPD risk using LASSO regression, the RFE algorithm and random forests.

Materials and Methods Study Population

A total of 233 people with COPD and 290 healthy controls were included in the study for a case control study. Based on the Global Initiative for Chronic Obstructive Lung Disease criteria, individuals were diagnosed with COPD with the ratio of forced expiratory volume in 1 second (FEV1) /forced vital capacity (FVC) < 70% and FEV1<80% predicted. COPD patients with a history of serious illnesses such as bronchial asthma, tuberculosis and lung cancer were not included in this study. The control group consisted of healthy people without pulmonary dysfunction, lung-related diseases, other chronic diseases and disorders, and severe endocrine, metabolic and nutritional disorders, who underwent a health check-up at the same hospital during the same period. Clinical characteristics of the subjects, including smoking, body mass index (BMI), complications, wheezing, gasping, chest pain and respiratory infections, were collected from medical records and questionnaires. The study protocol was approved by the Ethics Committee of Hainan Provincial People’s Hospital in accordance with the Declaration of Helsinki. All subjects signed an informed consent form.

Selection of SNPs

We identified SNPs associated with COPD based on the literature in PubMed (https://pubmed.ncbi.nlm.nih.gov/) and our case-control study of COPD. We then screened SNPs located on these genes from the Chinese Han Beijing (CHB) dataset of the Thousand Genomes Project (https://www.internationalgenome.org/) and the Ensembl website (http://www.ensembl.org), considering only the minimum allele frequency (MAF) ≥ 0.05 for SNPs. Haploview v4.2 software (Broad Institute of MIT and Harvard) was used to predict marker SNPs for each gene.

Genomic DNA Extraction and SNPs Genotyping

Peripheral blood samples were collected from all subjects and genome extraction kits were purchased from Xi’an Gold & Magnesium Co. Amplification primers were designed using the MassARRAY Assay Design software and genotyping was performed using the MassARRAY platform (Agena, San Diego, CA, USA). The generated assay data was analysed using AgenaTyper v4.0 software, which requires a call rate of ≥95% for candidate SNPs.

Definition of Data Characteristics

The total study population in this study was 523 individuals, with the minimum allele being the risk allele in the healthy control population, and 0, 1 and 2 denoting the number of risk alleles carried by an individual, being 2 carried by AA, 1 carried by AB and 0 carried by BB (the minimal allele was A). In addition, we specified the number of COPD patients and healthy controls as the dependent variable and the number of SNPs carrying risk alleles in each sample as the explanatory variable. These data were finally screened as data features for machine learning by one-way logistic regression and LASSO model, RFE-Caret, RFE-Lda, RFE-lr, RFE-nb, RFE-rf, RFE-treebag algorithms and random forest model.

Annotation Analysis of SNPs

Expression quantitative trait locus (eQTL) analysis can identify possible causative genes within COPD susceptibility loci.18 Motifs are a class of gene loci that can influence gene expression, and most of these loci are SNPs. In this study, we used the online tool HaploReg v4.1 (https://pubs.broadinstitute.org/mammals/haploreg/haploreg.php) to perform functional annotation analyses of the screened SNPs, including eQTL analysis, motif change regulation analysis and SNPs mapping.

Data Analysis

In this study, we used R v4.2.1 to perform batch one-way logistic regression analysis on 146 SNPs loci from 523 samples, and screened the SNPs obtained by screening in LASSO regression, RFE algorithm and randomforest algorithm, respectively, to construct the models associated with COPD risk, and plotted ROC curves to evaluate the model classification performance was selected, and the model with the best performance was selected to construct the column line graph of SNPs loci associated with COPD risk. The Hosmer-Lemeshow test was used to assess the goodness of fit of the column line plots and visualised by calibration curves. SPSS 22.0 statistical software was used and comparisons of normally distributed measures were analysed by ANOVA, with measures expressed as mean ± s of x and non-normally distributed measures expressed as median (interquartile range) using the rank sum test. Count data were analysed using the χ2 test. Logistic regression analysis was performed using the Wald test with p<0.05 as a statistically significant difference.

LASSO regression using the R package “glmnet” and 10-fold cross-validation using the “cv” function. Use the Glmnet package to obtain the most appropriate penalty factor λ. The importance of each SNPs was assessed using the R package “randomforest” and the Lda, lr, nbFuncs, rf and treebag parameters in “caret”, followed by plotting the ROC curves using the functions in the “pROC” package and performing the Hosmer-Lemeshow test using the R package “ResourceSelection”, where a significant p-value indicates a poorly fitted model.

Results Genotyping results of SNPs

Based on the screening criteria, 146 SNPs from 43 genes were screened and genotyped among 233 COPD patients and 290 healthy controls using the Agena MassARRAY technique, and all SNPs met the typing success rate of ≥95% and Hardy-Weinberg equilibrium p>0.05 after chi-square test. The information corresponding to 146 of these SNPs and 43 genes is shown in Table S1. The results of 146 SNPs genotyping were displayed in Table 1.

Table 1 The Results of 146 SNPs Genotyping Using the MassARRAY Platform

One-Way Logistic Regression Analyses

The results of univariate analysis showed that among the successfully typed loci, 44 SNPs had statistically significant effects on the risk of COPD (p< 0.05) (Table 2).

Table 2 Univariate Logistic Regression Results (Only Significant SNPs are Included in the Table)

LASSO Regression Analysis

Based on the results of the 10-fold cross-validation, we obtained the value of λ at the minimum of the mean square error (MSE) (lambda.min) and the value of λ one standard error away from the minimum of the MSE (lambda.1se), with the corresponding number of SNPs varying with the value of the penalty coefficient λ (Figure 1A). In this study, we chose λ=0.033, which had the highest penalty value, as the optimal λ. Figure 1B shows a total of 25 significant SNPs observed at λ=0.033, of which 13 SNPs were positively correlated with the risk of COPD, namely rs12479210 (β=0. 411), rs1420101 (β=0.0000572), rs9320913 (β=0.128), rs4646437 (β=0.0611), rs298207 (β=0.0207), rs16907751 (β=0.377), rs759648 (β=0.126), rs2420915 (β=0.0520), rs78750958 (β=0. 0520), rs1484215 (β=0.0846), rs3024622 (β=0.165), rs1038376 (β=0.511) and rs2853676 (β=0.209), and 12 SNPs were negatively correlated with the risk of COPD, namely rs13097407 (β=−0.152), rs352140 (β=−0.0769), rs911186 (β=−1. 42), rs2505059 (β=−0.141), rs10245353 (β=−0.128), rs4719841 (β=−0.231), rs13271489 (β=−0.294), rs7934083 (β=−0.441), rs9525927 (β=−0.197), rs3093193 (β=−0.233), rs3093110 (β=−0.250), rs4803420 (β=−0.115) (Table 3). The area under the curve (AUC) of the ROC curve was 0.809, an indication that the model had good classification results (Figure 1C).

Table 3 Significant SNPs and Their Coefficients After LASSO Regression

Figure 1 LASSO regression analysis. (A) 10-fold cross-validation of the results. The value in the middle of the two dotted lines is the range of the positive and negative standard deviations of log(λ). The dotted line on the left indicated the value of the harmonic parameter log(λ) when the error of the model is minimized. 25 variables were selected when log(λ) = 0.033. (B) LASSO coefficient profiles of 25 significant SNPs. A vertical line was drawn at the value chosen by 10-fold cross-validation. As the value of λ decreased, the degree of model compression increased and the function of the model to select important variables increased. (C) Receiver operating characteristic (ROC) curves of 25 SNPs in LASSO regression analysis. AUC = 0.809.

RFE Algorithm

Based on the RFE algorithm analysis, a total of 38 significant SNPs were screened in the caret model, 42 significant SNPs in the Lda model, 42 significant SNPs in the lr model, 4 significant SNPs in the nb model, 42 significant SNPs in the rf model, and 44 significant SNPs in the treebag model (Table 4). In addition, the AUC of the ROC curve of the caret model is 0.769, the AUC of the ROC curve of the Lda model is 0.798, the AUC of the ROC curve of the lr model is 0.743, the AUC of the ROC curve of the nb model is 0.686, the AUC of the ROC curve of the rf model is 0.766, the AUC of the ROC curve of the treebag model is 0.743, and all these AUC values have AUC values of 0.769. 686, the AUC of the ROC curve of the rf model is 0.766, the AUC of the ROC curve of the treebag model is 0.743, and all these AUC values have AUC > 0.5, so all six models are considered to have good classification performance (Figure 2).

Table 4 SNPs Selected by Caret, Lda, Lr, Nb, Rf and Treebag Models Constructed Based on Recursive Feature Elimination (RFE) Algorithm

Figure 2 ROC curves for the six models of Recursive Feature Elimination (RFE). (A) ROC curves of 38 SNPs in caret model. AUC = 0.769. (B) ROC curves of 42 SNPs in Lda model. AUC = 0.798. (C) ROC curves of 42 SNPs in lr model. AUC = 0.734. (D) ROC curves of 4 SNPs in nb model. AUC = 0.686. (E) ROC curves of 42 SNPs in rf model. AUC = 0.766. (F) ROC curves of 44 SNPs in treebag model. AUC = 0.734.

Random Forest(RF) Assessment

To assess the significance of the contribution of SNPs obtained from genotyping to COPD risk, we made a random forest decision based on the characteristics of the sample data described above. In the random forest model, the relative importance of a variable is the total reduction in node impurity when that variable is equally distributed across all trees, and node impurity is defined by the Gini coefficient. Therefore, we ranked the variables according to the size of the average decreasing Gini coefficient of the Random Forest output and ranked the 44 SNPs in order of importance from largest to smallest (Table 5). The AUC of the ROC curve is 0.719, which is an indication that the model has a good classification performance (Figure 3).

Table 5 Random Forest Decision Results for 44 SNPs (MeanDecreaseGini Coefficients Represent the Importance of SNPs, Ranked from Most to Least)

Figure 3 ROC curves of 44 SNPs for random forest model. AUC = 0.719.

Identification and Validation of Personalized Predictive Models

Based on the above AUC values, the performance of these eight classifiers was evaluated, and Lasso (0.809) > lda (0.798) > caret (0.769) > rf (1.766) > lr (0.743) = treebag (0.743) > RF (0.719) > nb (0.686), the 25 SNPs screened by the best LASSO model were selected as independent predictors of COPD risk. Based on HaploReg v4.1 database, the potential functions of these SNPs were displayed in Table S2. Nomogram for predictive models were constructed based on 25 SNPs screened by the best LASSO model (Figure 4A). Nomogram results showed that rs1038376 and rs12479210 polymorphic loci contributed most to the increased risk of developing COPD, whereas rs13097407, rs352140, rs911186, rs2505059, rs1024535, rs471984, rs1327148, rs7934083, rs952592, rs3093193, rs3093110 and rs4803420 risk alleles were the protective factors for COPD risk. Figure 4B shows the calibration curves for the Nomogram we constructed, and the actual curves are closer to the ideal curves, indicating that the model is well calibrated in the dataset.

Figure 4 LASSO model Performance validation. (A) Nomogram Nomogram model predicting COPD risk. The nomogram is used by summing all points identified on the scale for each variable. (B) Curve of calibration for predicting COPD risk. The predicted Probability by the nomogram model is plotted on the x-axis, and the Observed Probability is plotted on the y-axis.

Discussion

COPD is an irreversible and progressive disease, so there is an urgent need to diagnose COPD in its early stages.19 A combination of genome-wide association studies and candidate gene analysis can help identify genetic variants that contribute to an individual’s predisposition to COPD.10 Although various types of risk prediction models have been developed in abundance in recent years, most are based on individual models or algorithms for prediction, eg Jin et al identified race SNPs by filtering through best linear unbiased prediction (BLUP) in a linear mixed model,20 correlation between IL95R SNPs and the risk of COPD as calculated by logistic regression analysis according to Zhou et al,21 although the overall predictive ability of KNN, LR and XGboost models has been reported,19 the most effective model for predicting genetic polymorphisms has not been reported in individual prediction models.

Previous studies assessed the heritability of COPD and related phenotypes in smokers among the non-Hispanic whites.22 Matthew Moll constructed a polygenic risk score using a genome-wide association study of lung function for COPD from the UK Biobank and SpiroMeta.23 A multi-ancestry genome-wide association analyses and systematic variant-to-gene mapping strategies implicate new genes and pathways influencing lung function and COPD risk.24 Jingzhou Zhang reported that a polygenic risk score is associated with earlier age of diagnosis of COPD and retains predictive value when added to known early-life risk factors in 6647 non-Hispanic White (NHW) and 2464 African American (AA) participants.25 Moreover, in 400,102 individuals of European ancestry, a new genetic signals for lung function highlight pathways and COPD associations across multiple ancestries.13 Despite the advancements in COPD risk modeling, the majority of these studies have been centered on European populations. There are few studies on COPD risk models in Chinese Han population.

In this study, we included SNPs that have been published as significant in association analyses for COPD. In total, we included 146 significant loci. On this basis, 233 patients diagnosed at Hainan Provincial People’s Hospital and 290 healthy controls who underwent medical check-ups during the same period were screened using the Agena MassARRAY technique in a case-control study method, and 44 SNPs were significantly associated with COPD susceptibility using one-way logistic regression analysis. The contribution of these 44 SNPs to the risk of COPD was then assessed using models constructed by LASSO, Caret, LDA, LR, NB, Rf and Treebag and the Random Forest model, comparing the classification performance of the different models and working to find a predictive model with higher performance.

LASSO is a regression analysis method that performs both variable selection and regularisation to improve the predictive accuracy and interpretability of statistical models.26 An attractive feature for SNPs selection is the sparsity of the LASSO model and the shrinking of the regression coefficients, which can be effective in selecting SNPs that predict quantitative traits but are limited by certain conditions.27 Jeremy Sabourin’s study shows that the performance of LASSO-based RMA methods in distinguishing between multiple real signals and highly correlated SNPs can be continuously improved by randomising the penalty parameter.28 In genomic studies, the ability to identify SNPs that affect a target trait is important for understanding the genetic basis of the trait.29 Caret (Classification And REgression Training) is a powerful package for building, evaluating and comparing predictive models in the R language.30 `Caret` provides a unified interface that makes it much easier to switch between algorithms.31 On this basis, we used the RFE-Caret, RFE-Lda, RFE-lr, RFE-nb, RFE-rf and RFE-treebag algorithms to assess the risk of SNPs for COPD. In previous studies of SNPs, Caret has tuned models to select appropriate parameters to improve model accuracy. In the diabetes study, Quincy A Hathaway performed 10-fold cross-validation of the results using LDA, NB, Support Vector Machine (SVM) and Classification and Regression Tree (CART) models. The ultimate goal is to select the optimal model to determine the biomarkers of the disease.32 Random forest models make predictions by constructing multiple decision trees and combining them together.33 SNP data usually contains a large number of features, and Random Forest can effectively deal with high-dimensional data to predict the most important SNPs in the dataset.34 One study used random forest modelling to distinguish the ability of Parkinson’s patients from controls.35 The RF algorithm is trained on relevant data and the discriminative importance of individual SNPs is assessed by a technical construct known as graph depth.36 As in our preliminary study, the predictive power of the tested SNPs was visualised and quantified using ROC curves and AUC, respectively.37

In addition to screening the best predictive models, we performed a column-line graphical model of the risk of incident COPD for the 25 independent predictors screened by the best model, lasso, and found that among the 25 high-risk SNPs, the rs1038376 and rs12479210 polymorphic loci contributed most to the increased risk of incident COPD. This result was crudely demonstrated in previous studies, where rs1038376 A/T and A/T-T/T/T were associated with an increased risk of COPD in co-dominant and dominant models, respectively, compared to the AA genotype.38 Notably, rs12479210 was screened and strongly correlated with COPD in all of the above models, but no other study has yet clarified its association with COPD. Studies have shown that rs12479210, a candidate SNP for the IL-1RL1 gene, is significantly associated with lung cancer risk,39 that IL-1RL1 is considered a targeted biomarker or target for pharmacological intervention in asthma,40 and that people with COPD have a higher risk of lung cancer.41 In conclusion, combining previous studies and our prediction results, we speculate that rs12479210 may be a potential risk locus for COPD.

However, our studies invariably have some limitations. On the one hand, although this was a case-control study, the study population was mostly from Hainan Province, China, so it would be cautious to generalise the conclusions or findings of this study to the general population. On the other hand, and we did not have external data to validate it, so we need to obtain more external data to further evaluate the nomogram constructed in this study.

Conclusions

In conclusion, based on the combination of single-factor analysis, LASSO regression, RFE algorithm and random forest model, 25 SNPs were screened to construct a simple prediction model with high predictive performance for COPD risk in the Chinese Han population.

Data Sharing Statement

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Ethical Approval

This study was conducted under the standards approved by the Ethics Committee of Hainan Provincial People’s Hospital and was in accordance with the ethical principles of the World Medical Association Declaration of Helsinki for medical research involving humans. Informed consent was obtained from all individual participants included in this study.

Consent for Publication

Consent to publish statements must confirm that the details of any images, videos, recordings, etc can be published, and that the person(s) providing consent have been shown the article contents to be published.

Acknowledgments

We thank all members of our research team for their contributions to this study, as well as Hainan Provincial People’s Hospital and all participants for their support to this study. We also thank National Natural Science Foundation of China for funding this study.

Author Contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Funding

This study was supported by Hainan Province Science and Technology Special Fund (No. ZDYF2024SHFZ094), the research project of Innovation Platform for Academicians of Hainan Province (YSPTZX202312), the Innovation Platform for Academicians of Hainan Province and the Key Research and Development Program of National Natural Science Foundation of China (No. 81860015).

Disclosure

The authors declare no conflicts of interest in this work.

References

1. Global. regional, and national incidence, prevalence, and years lived with disability for 310 diseases and injuries, 1990-2015: a systematic analysis for the global burden of disease study 2015. Lancet. 2016;388(10053):1545–1602. doi:10.1016/S0140-6736(16)31678-6

2. Adeloye D, Chua S, Lee C, et al. Global and regional estimates of COPD prevalence: systematic review and meta-analysis. J Global Health. 2015;5(2):020415. doi:10.7189/jogh.05.020415

3. Anees Ur R, Ahmad Hassali MA, Muhammad SA, et al. The economic burden of chronic obstructive pulmonary disease (COPD) in the USA, Europe, and Asia: results from a systematic review of the literature. Expert Rev Pharmacoecon Outcomes Res. 2020;20(6):661–672. doi:10.1080/14737167.2020.1678385

4. AS B, MA M, WM V, et al. International variation in the prevalence of COPD (the BOLD Study): a population-based prevalence study. Lancet. 2007; 370(9589):741–750.

5. Mannino DM, Buist AS. Global burden of COPD: risk factors, prevalence, and future trends. Lancet. 2007; 370(9589):765–773.

6. KF R, Hurd S, Anzueto A, et al. Global strategy for the diagnosis, management, and prevention of chronic obstructive pulmonary disease: GOLD executive summary. Am J Respir Crit Care Med. 2007;176(6):532–555. doi:10.1164/rccm.200703-456SO

7. Wang CD, Chen N, Huang L, et al. Impact of CYP1A1 polymorphisms on susceptibility to chronic obstructive pulmonary disease: a meta-analysis. Biomed Res Int. 2015;2015:942958. doi:10.1155/2015/942958

8. Humbert M, Montani D, Perros F, Dorfmüller P, Adnot S, Eddahibi S. Endothelial cell dysfunction and cross talk between endothelium and smooth muscle cells in pulmonary arterial hypertension. Vasc Pharmacol. 2008;49(4–6):113–118. doi:10.1016/j.vph.2008.06.003

9. Yuksel H, Yilmaz O, Karaman M, et al. Role of vascular endothelial growth factor antagonism on airway remodeling in asthma. Annals Allergy Asthma Immunol. 2013;110(3):150–155. doi:10.1016/j.anai.2012.12.015

10. Marciniak SJ, Lomas DA. Genetic susceptibility. Clinics Chest Med. 2014;35(1):29–38. doi:10.1016/j.ccm.2013.10.008

11. BD H, de Jong K, Lamontagne M, et al. Genetic loci associated with chronic obstructive pulmonary disease overlap with loci for lung function and pulmonary fibrosis. Nature Genet. 2017;49(3):426–432. doi:10.1038/ng.3752

12. Sakornsakolpat P, Prokopenko D, Lamontagne M, et al. Genetic landscape of chronic obstructive pulmonary disease identifies heterogeneous cell-type and phenotype associations. Nature Genet. 2019;51(3):494–505. doi:10.1038/s41588-018-0342-2

13. Shrine N, AL G, AM E, et al. New genetic signals for lung function highlight pathways and chronic obstructive pulmonary disease associations across multiple ancestries. Nat Genet. 2019;51(3):481–493. doi:10.1038/s41588-018-0321-7

14. Wu MC, Kraft P, Epstein MP, et al.:Powerful SNP-set analysis for case-control genome-wide association studies. Am J Hum Genet. 2010;86(6):929–942. doi:10.1016/j.ajhg.2010.05.002

15. Omranian N, Eloundou-Mbebi JM, Mueller-Roeber B, Nikoloski Z. Gene regulatory network inference using fused LASSO on multiple data sets. Sci Rep. 2016;6:20533. doi:10.1038/srep20533

16. Huang S, Cai N, Pacheco PP, Narrandes S, Wang Y, Xu W. Applications of support vector machine (SVM) learning in cancer genomics. Cancer Genomics Proteomics. 2018;15(1):41–51.

17. Yu X, Zeng Q. Random forest algorithm-based classification model of pesticide aquatic toxicity to fishes. Aquatic Toxicol. 2022;251:106265. doi:10.1016/j.aquatox.2022.106265

18. Lamontagne M, JC B, Obeidat M, et al. Leveraging lung tissue transcriptome to uncover candidate causal genes in COPD genetic associations. Human Mol Gene. 2018;27(10):1819–1829. doi:10.1093/hmg/ddy091

19. Ma X, Wu Y, Zhang L, et al. Comparison and development of machine learning tools for the prediction of chronic obstructive pulmonary disease in the Chinese population. J Transl Med. 2020;18(1):146. doi:10.1186/s12967-020-02312-0

20. Gim J, An J, Sung J, Silverman EK, Cho MH, Won S. A between ethnicities comparison of chronic obstructive pulmonary disease genetic risk. Front Genet. 2020;11:329. doi:10.3389/fgene.2020.00329

21. Zhou Y, Chen J, Bai F, et al. Suggestive evidence of genetic association of IL23R polymorphisms with chronic obstructive pulmonary disease risk in the Chinese population. J Gene Med. 2023;25(5):e3479. doi:10.1002/jgm.3479

22. Zhou JJ, Cho MH, Castaldi PJ, Hersh CP, Silverman EK, Laird NM. Heritability of chronic obstructive pulmonary disease and related phenotypes in smokers. Am J Respir Crit Care Med. 2013;188(8):941–947. doi:10.1164/rccm.201302-0263OC

23. Moll M, Sakornsakolpat P, Shrine N, et al. Chronic obstructive pulmonary disease and related phenotypes: polygenic risk scores in population-based and case-control cohorts. Lancet Respir Med. 2020;8(7):696–708. doi:10.1016/S2213-2600(20)30101-6

24. Shrine N, AG I, Chen J, et al. Multi-ancestry genome-wide association analyses improve resolution of genes and pathways influencing lung function and chronic obstructive pulmonary disease risk. Nat Genet. 2023;55(3):410–422. doi:10.1038/s41588-023-01314-0

25. Zhang J, Xu H, Qiao D, et al. A polygenic risk score and age of diagnosis of COPD. Eur Respir J. 2022;60(3):2101954. doi:10.1183/13993003.01954-2021

26. Kang J, Choi YJ, Kim IK, et al. LASSO-based machine learning algorithm for prediction of lymph node metastasis in T1 colorectal cancer. Cancer Res Treat. 2021;53(3):773–783. doi:10.4143/crt.2020.974

27. Feng ZZ, Yang X, Subedi S, McNicholas PD. The LASSO and sparse least square regression methods for SNP selection in predicting quantitative traits. IEEE/ACM trans comput biol bioinfo. 2012;9(2):629–636. doi:10.1109/TCBB.2011.139

28. Yang C, Wan X, Yang Q, Xue H, Yu W. Identifying main effects and epistatic interactions from large-scale SNP data via adaptive group Lasso. BMC Bioinf. 2010;1(1 Suppl):S18. doi:10.1186/1471-2105-11-S1-S18

29. Wang H, Zhang Y, Chen L, et al. Identification of clinical prognostic features of esophageal cancer based on m6A regulators. Front Immunol. 2022;13:950365. doi:10.3389/fimmu.2022.950365

30. Beck MW. NeuralNetTools: visualization and analysis tools for neural networks. J Stat Software. 2018;85(11):1–20. doi:10.18637/jss.v085.i11

31. TM D, Dankers F, Valdes G, et al. Machine learning algorithms for outcome prediction in (chemo)radiotherapy: an empirical comparison of classifiers. Med Phys. 2018;45(7):3449–3459. doi:10.1002/mp.12967

32. QA H, SM R, MV P, et al. Machine-learning to stratify diabetic patients using novel cardiac biomarkers and integrative genomics. Cardiovasc diabetol. 2019;18(1):78. doi:10.1186/s12933-019-0879-0

33. Elbeltagi A, Pande CB, Kumar M, et al. Prediction of meteorological drought and standardized precipitation index based on the random forest (RF), random tree (RT), and Gaussian process regression (GPR) models. Environ Sci Pollut Res Int. 2023;30(15):43183–43202. doi:10.1007/s11356-023-25221-3

34. Botta V, Louppe G, Geurts P, Wehenkel L. Exploiting SNP correlations within random forest for genome-wide association studies. PLoS One. 2014;9(4):e93379. doi:10.1371/journal.pone.0093379

35. Cibulka M, Brodnanova M, Grendar M, et al. Alzheimer’s disease-associated SNP rs708727 in SLC41A1 may increase risk for parkinson’s disease: report from enlarged Slovak study. Int J Mol Sci. 2022;23(3):1604. doi:10.3390/ijms23031604

36. Deo RC. Machine learning in medicine. Circulation. 2015;132(20):1920–1930. doi:10.1161/CIRCULATIONAHA.115.001593

37. Cibulka M, Brodnanova M, Grendar M, et al. SNPs rs11240569, rs708727, and rs823156 in SLC41A1 do not discriminate between Slovak patients with idiopathic parkinson’s disease and healthy controls: statistics and machine-learning evidence. Int J Mol Sci. 2019;20(19):4688. doi:10.3390/ijms20194688

38. Ding Y, Li Q, Feng Q, et al. CYP2B6 genetic polymorphisms influence chronic obstructive pulmonary disease susceptibility in the Hainan population. Int J Chronic Obstr. 2019;14:2103–2115. doi:10.2147/COPD.S214961

39. Li Q, Zhang C, Cheng Y, et al. IL1RL1 polymorphisms rs12479210 and rs1420101 are associated with increased lung cancer risk in the Chinese Han population. Front Genetics. 2023;14:1183528. doi:10.3389/fgene.2023.1183528

40. Saikumar Jayalatha AK, Hesse L, Ketelaar ME, Koppelman GH, Nawijn MC. The central role of IL-33/IL-1RL1 pathway in asthma: from pathogenesis to intervention. Pharmacol Ther. 2021;225:107847. doi:10.1016/j.pharmthera.2021.107847

41. Forder A, Zhuang R, VGP S, et al. Mechanisms contributing to the comorbidity of COPD and lung cancer. Int J Mol Sci. 2023;24(3):2859. doi:10.3390/ijms24032859