MACHINE LEARNING FOR EARLY DIAGNOSIS OF SMALL CELL LUNG CANCER: MULTI-OMICS BIOMARKER STRATEGIES

Abstract

Small cell lung cancer (SCLC) is an aggressive malignancy characterized by late diagnosis and five-year survival rates below 7%. Existing serological markers, including neuron-specific enolase and progastrin-releasing peptide, lack the sensitivity and specificity required for reliable early detection. Machine learning applied to multi-omics data has emerged as a powerful strategy to overcome these limitations. In this review, we systematically evaluate recent progress in machine-learning-based screening of diagnostic biomarkers for SCLC across transcriptomic, exosomal RNA, proteomic, metabolomic, and DNA methylation datasets. We compare the performance and methodological characteristics of leading feature-selection algorithms (LASSO, random forest, SVM-RFE, and XGBoost) and critically appraise the diagnostic accuracy, validation rigor, and reproducibility of published signatures. While several multi-omics panels, particularly those derived from exosomal RNA and DNA methylation, have achieved AUCs exceeding 0.90, widespread clinical translation remains hindered by single-center retrospective designs, inadequate external validation, heterogeneous preprocessing pipelines, and data leakage. We outline standardized analytical and validation frameworks required to advance these signatures toward clinical utility and provide practical recommendations for future high-quality biomarker discovery. This review offers a comprehensive and critical synthesis to guide the development of robust, clinically deployable machine-learning models for early SCLC diagnosis.