Executive Summary : | The recent surge in interest in machine learning and deep learning algorithms has led to the analysis of complex omics data from various biological experimentation techniques. This data, encompassing diversified diseases, allows us to understand the systems-level properties of genes, proteins, and metabolites and demystify the dynamics of biological systems. The aim is to develop novel machine learning and deep learning models to uncover the complexities and interaction dynamics of different molecules in biological systems. This will help identify potential biomarkers and gain insights into the formation, pathogenesis, and development of new therapeutics for diseases like cancers. The "curse of dimensionality" problem in omics data is addressed by developing deep-learning models to extract the compressed latent space-encoded representation of molecular features through autoencoders. Additionally, the "class-imbalanced" problem in omics and biomedical data for different classes is addressed by developing models for generating realistically synthetic biological samples through deep generative models like Variational Autoencoders (VAEs) and Generative Adversarial Networks (GANs). These models will help overcome the challenges of class-imbalanced and dimensionality in machine learning models. |