In today’s World Rare diseases are combinely affect over 300 million people; however, very poor information is available based on these types of conditions, especially about the molecular mechanisms behind rare phenotypes. In this paper, a novel analysis is presented in which multi-omics data—such as genomics, transcriptomics, epigenomics, proteomics, and metabolomics—are integrated using a meta-learning framework. By analyzing existing computational frameworks, several major challenges are detected, such as extreme class imbalance, cross-cohort heterogeneity, and sparse labelling. To overcome these types of problems, a new novel framework called MetaOmics-RPD is proposed for the discovery of rare phenotypes.

In this novel framework, few-shot learning, Model-Agnostic Meta-Learning (MAML), and graph-based omics integration are used within a hierarchical Bayesian structure. Almost 112 case studies published between 2015 and 2025 are reviewed, and their datasets, evaluation methods, and experimental setups are examined and analysed thoroughly. Different kinds of open challenges are also focused on, such as interpretability, compatibility with federated learning, and multi-modal model integration. The main contributions of this work include a specialized meta-learning technique for imbalanced data and a Hierarchical Graph Neural Network (HGNN) for amalgamating features from different data types. The proposed method shows an improvement in AUROC scores of 12% to 27% compared to baseline models. The main purpose of this paper is to provide a basic reference guide for researchers in the field of systems biology, machine learning, and clinical genomics.

Keywords— Multi-omics amalgamation, rare disease, meta-learning, few-shot learning, rare phenotype discovery, graph neural networks, MAML, systems biology, class imbalance, federated genomics.