Graph Neural Networks based Model for Multi-Omics Data Integration in Disease Classification and Biomarker Discovery

In recent years, advanced deep learning tools are used to improve disease classification and discovery of biomarker by combing multi-omics data. However, many challenges are faced by using these tools include heterogeneity of data which means different technologies are used to create omics data which cause some variations in data formats and also faced high dimensionality issues. Therefore, this research proposes a Multi Modal Neural Network (MMNN) for disease classification and biomarkers discovery for which the data is collected from multiple biomedical datasets for multi-omics data integration and classification tasks. The, the collected data is preprocessed with Analysis of Variance (ANOVA), Principal Component Analysis (PCA) and normalization for reducing noise and ensuring consistency among the omics dataset. Further, Multiple Kernel Learning (MKL) is performed which fuses omics datasets effectively for enhancing classification. Finally, MMNN are incorporated to train the optimal kernel fusion automatically and ensures accurate disease classification and bio marker discovery which is completed through Shapley Additive exPlanations (SHAP). The proposed MMNN model obtained better results in terms of accuracy and precision with 95% and 88% respectively when compared to existing Support Vector Machine (SVM).