MULTI-OMICS INTEGRATION FOR ELUCIDATING COMPLEX DISEASE MECHANISMS: A MACHINE LEARNING PERSPECTIVE

Abstract

Complex diseases like cancer, neurodegenerative diseases, and metabolic syndromes develop on complex interactions involving biological processes through genomic, transcriptomic, proteomic, and metabolomic. Conventional single-omics techniques can tend to miss this complexity and result in incomplete mechanistic insight. Against this background, the multi-omics integration has come as a powerful tool to give an overarching picture of the disease biology, through integration of heterogeneous data at different levels of molecular biology. Nonetheless, multi-omics data are high dimensional, noisy, and heterogeneous, and they pose a big challenge in the analysis process, which requires sophisticated computational models. Machine learning algorithms, including classical algorithms like support-vector machines and random forests, deep learning, and network-based models, have shown great potential in revealing hidden patterns, finding biomarkers, and predicting the outcome of a disease. This review critically analyzes the existing multi-omics integration approaches, the importance of machine learning in enhancing classification and interpretability, and evaluation metrics including accuracy, F1-score, and ROC-AUC used to measure the performance of a model. The main insights are that integrative methods can greatly contribute to the knowledge of disease pathogenesis and allow to identify biomarkers more accurately compared to single-omics. In spite of these developments, there are issues of data standardization, model interpretability and clinical translation. To complete the gap between computational predictions and real-world biomedical applications, future studies should emphasize explainable AI, single-cell and spatial multi-omics, and scalable frameworks.