Integrative Transcriptomic Analysis and Machine Learning Identify Robust Gene Expression Signatures for Cardiovascular Disease Classification

BACKGROUND: Cardiovascular diseases (CVDs) remain the leading cause of mortality worldwide and contribute substantially to the global health burden. Early diagnosis and the identification of reliable molecular biomarkers are critical for improving disease prognosis and guiding therapeutic strategies. Recent advances in high-throughput transcriptomic technologies, along with integrative bioinformatics approaches, have enabled large-scale exploration of gene expression patterns in complex diseases. However, variability across independent datasets often limits the identification of robust and reproducible biomarkers. In this context, the present study aimed to integrate multiple transcriptomic datasets and apply machine learning techniques to identify gene expression signatures capable of accurately distinguishing cardiovascular disease samples from healthy controls. METHODS: Seven publicly available cardiovascular disease transcriptomic datasets were obtained from the Gene Expression Omnibus database maintained by the National Center for Biotechnology Information. Based on predefined inclusion criteria, a total of 1,188 samples (763 cardiovascular disease and 425 healthy controls) were included for analysis. Data preprocessing, normalization, and batch effect correction were performed using the R programming environment. Differential gene expression analysis was carried out using the limma framework, and significantly altered genes were identified using an adjusted p-value threshold of < 0.05. The top 200 most significant genes were selected as predictive features for model development. Multiple machine learning algorithms were implemented, and model performance was assessed using receiver operating characteristic (ROC) curves, precision–recall analysis, and standard classification metrics. RESULTS: Differential expression analysis identified several genes significantly associated with cardiovascular disease. Machine learning models trained on the selected gene expression features demonstrated strong predictive performance in distinguishing disease samples from healthy controls. Among the tested approaches, ensemble models consistently showed improved classification performance compared to individual algorithms. Receiver operating characteristic analysis further indicated high discriminative ability, highlighting the potential of transcriptomic signatures as diagnostic biomarkers. CONCLUSION: This study demonstrates that integrative transcriptomic analysis combined with machine learning can effectively identify gene expression signatures associated with cardiovascular disease. The identified gene set may serve as a potential biomarker panel for early diagnosis and may also contribute to a better understanding of underlying disease mechanisms. Further validation using independent cohorts and functional studies is required to confirm the clinical applicability of these findings.