Machine Learning for Glaucoma Identification: A Comparative Study of Algorithms

• Robson Paulino da Silva UFMG
• Janier Arias García UFMG
• Frederico Gualberto Ferreira Coelho UFMG

DOI: https://doi.org/10.5753/eniac.2026.27532

Resumo

Glaucoma is one of the leading causes of irreversible blindness worldwide, and its early detection remains a significant challenge. Although machine learning models have shown promising performance in the classification of transcriptomic data, there is still a gap in understanding the robustness of these models under realistic perturbations and dataset heterogeneity. In this context, this work proposes a comprehensive evaluation of 13 machine learning algorithms for glaucoma identification using gene expression data obtained from public repositories. A unified experimental pipeline was adopted, including preprocessing steps, batch effect correction, and validation through repeated stratified hold-out, with evaluation based on metrics such as accuracy, precision, recall, F1-score, and AUC. Statistical comparisons were performed using Friedman and Nemenyi tests, and model interpretability was analyzed using SHAP. Additionally, robustness was investigated through the introduction of controlled transcriptomic perturbations, including Gaussian noise and gene-level mutations. The results indicate that the best performance was achieved with the combination of datasets GSE27276 and GSE9944, reaching an AUC of 0.99, with Logistic Regression and Stacking models standing out. In contrast, the inclusion of multiple additional datasets resulted in significant performance degradation, with AUC reduced to approximately 0.78, highlighting the impact of transcriptomic heterogeneity on generalization capability. Furthermore, the best-performing models demonstrated high robustness, maintaining stable AUC values even under perturbation scenarios. These findings indicate that regularized linear models can provide competitive performance and greater stability compared to more complex approaches, reinforcing their suitability for high-dimensional biomedical data.

Referências

Cortes, C. and Vapnik, V. (1995). Support-vector networks. Machine Learning, 20(3):273–297.

Dai, M., Hu, Z., Kang, Z., and Zheng, Z. (2022). Based on multiple machine learning to identify the eno2 as diagnosis biomarkers of glaucoma. BMC ophthalmology, 22(1):155.

Demsar, J. (2006). Statistical comparisons of classifiers over multiple data sets. Journal of Machine Learning Research, 7:1–30.

Feng, J. and Xu, J. (2019). Identification of pathogenic genes and transcription factors in glaucoma. Molecular medicine reports, 20(1):216–224.

Friedman, M. (1937). The use of ranks to avoid the assumption of normality implicit in the analysis of variance. Journal of the American Statistical Association, 32(200):675–701.

Garcia, D. and Ramos, J. (2022). Neuroinflammatory pathways and their role in glaucoma. Frontiers in Immunology, 13:889302.

Howell, G. R., Soto, I., Zhu, X., Ryan, M., Macalinao, D. G., Sousa, G. L., Caddle, L. B., MacNicoll, K. H., Barbay, J. M., Porciatti, V., et al. (2012). Radiation treatment inhibits monocyte entry into the optic nerve head and prevents neuronal damage in a mouse model of glaucoma. The Journal of clinical investigation, 122(4):1246–1261.

Kaur, M. et al. (2023). Emerging molecular biomarkers and pathways in glaucoma: A comprehensive review. Molecular Vision, 29:552–570.

Lee, J. and Kim, S. (2020). Augmenting transcriptomic data for deep learning applications. PLOS Computational Biology, 16(12):e1008446.

Liu, M., Hu, F., Lei, C., Fu, M., Li, X., and Yu, L. (2023). Candidate genes identification and rna-seq based pathway analysis associated with primary angle-closure glaucoma with cataract. BMC ophthalmology, 23(1):192.

Lundberg, S. M. and Lee, S.-I. (2017). A unified approach to interpreting model predictions. Advances in Neural Information Processing Systems (NeurIPS), 30:4765–4774.

Nemenyi, P. (1963). Distribution-free multiple comparisons. PhD thesis, Princeton University.

Peng, C., Hu, L., Su, W., and Hu, X. (2025). Identification of diagnostic biomarkers and dissecting immune microenvironment with crosstalk genes in the poag and covid-19 nexus. Scientific Reports, 15(1):25244.

Pulukool, S. K., Bhagavatham, S. K. S., Kannan, V., Sukumar, P., Dandamudi, R. B., Ghaisas, S., Kunchala, H., Saieesh, D., Naik, A. A., Pargaonkar, A., et al. (2021). Elevated dimethylarginine, atp, cytokines, metabolic remodeling involving tryptophan metabolism and potential microglial inflammation characterize primary open angle glaucoma. Scientific Reports, 11(1):9766.

Qiu, H., Zhu, B., and Ni, S. (2018). Identification of genes associated with primary open-angle glaucoma by bioinformatics approach. International Ophthalmology, 38(1):19–28.

Suo, L., Dai, W., Qin, X., Li, G., Zhang, D., Cheng, T., Yao, T., and Zhang, C. (2022). Screening of primary open-angle glaucoma diagnostic markers based on immune-related genes and immune infiltration. BMC Genomic Data, 23(1):67.

Tezel, G. (2011). The molecular regulation of optic nerve degeneration in glaucoma. Progress in Retinal and Eye Research, 30(6):369–380.

Ullah, Z., Tao, Y., and Huang, J. (2024). Integrated bioinformatics-based identification and validation of neuroinflammation-related hub genes in primary open-angle glaucoma. International Journal of Molecular Sciences, 25(15):8193.

Wang, D., Pu, Y., Tan, S., Wang, X., Zeng, L., Lei, J., Gao, X., and Li, H. (2024). Identification of immune-related biomarkers for glaucoma using gene expression profiling. Frontiers in Genetics, 15:1366453.

Wu, J., Lin, C., Yang, C., Pan, L., Liu, H., Zhu, S., Wei, S., Jia, X., Zhang, Q., Yu, Z., et al. (2023). Identification and validation of key biomarkers and potential therapeutic targets for primary open-angle glaucoma. Science China Life Sciences, 66(12):2837–2850.

Xie, Y. and Yu, K. (2024). Identifying hub genes for glaucoma based on bulk rna sequencing data and multi-machine learning models. Current Medicinal Chemistry, 31(42):7059–7071.

Xu, W., Sun, Y., Zhao, S., Zhao, J., and Zhang, J. (2023). Identification and validation of autophagy-related genes in primary open-angle glaucoma. BMC Medical Genomics, 16(1):287.

Xu, W., Zhang, F., and Huang, J. (2025). Identification and validation of oxidative stress signature genes in the trabecular meshwork of glaucoma. Scientific Reports, 15(1):31260.

Zhang, X., Chen, J., Zhou, X., Zhou, D., Liao, L., Zhao, Y., Wu, P., Nie, F., Liao, Z., Cai, Z., et al. (2025). Exploring diagnostic m6a regulators in primary open-angle glaucoma: insight from gene signature and possible mechanisms by which key genes function. BMC Medical Genomics, 18(1):57.

Zhao, Y. and Wang, Y. (2020). Data augmentation for rna-seq using generative neural networks. Bioinformatics, 36(14):4280–4287.

Zhou, Z.-H. (2012). Ensemble Methods: Foundations and Algorithms. Chapman and Hall/CRC.