Machine learning–based classification of Fuchs’ endothelial corneal dystrophy reveals disease-associated multigene expression signatures

Background Fuchs’ endothelial corneal dystrophy (FECD) is a genetically and biologically heterogeneous corneal disorder that is clinically diagnosed as a single disease entity. Although bulk RNA sequencing (RNA-seq) studies have provided important insights into FECD-associated transcriptional changes, most analyses have focused on differential expression of individual genes, which may be sensitive to cohort-specific effects and may not fully capture stable disease-associated patterns across heterogeneous samples. Objective To reframe FECD transcriptomic data analysis using a classification-oriented machine learning approach that emphasizes disease–control discrimination at multigene level. Methods Publicly available bulk RNA-seq datasets from FECD and control corneal endothelial samples were integrated and analyzed using supervised machine learning-based models. The gene expression quantification matrix was generated. Further, matrix was provided as an input for logistic regression (LR) and support vector classifier (SVC) for feature selection and classification. Accuracy, precision, and F1-score were calculated along with Receiver operating characteristic- area under the curve (ROC-AUC) to evaluate the performance of classifier. Furthermore, selected features were checked for their biological relevance with FECD pathogenesis using literature mining approach. Results LR and SVC models achieved robust discrimination between FECD and control samples, with classification accuracies of 0.842 (confidence interval (CI) 0.684-1.000) and 0.842 (CI 0.684-1.000), respectively. Moreover, LR provided an AUC of 0.923 (95% CI: 0.767–1.000), SVC achieved an AUC of 0.974 (95% CI: 0.886–1.000). Despite differences in model architecture, 25 gene features overlapped between LR and SVC models, indicating stable transcriptional patterns contributing to disease classification. Biological relevance of the classification-relevant gene signatures revealed convergence on biological processes and signaling pathways previously implicated in corneal endothelial dysfunction, including extracellular matrix remodelling, oxidative stress, and inflammatory signaling. Conclusion FECD patients and control samples were classified based on their gene expression profiles. Feature selection further identified key genes that contributed to distinguishing FECD samples from controls. Functional analysis of these genes and their associated biological processes provided insights into the molecular mechanisms underlying FECD.