From Single-Modality Sequencing to Multi-Omics Intelligence: A Profound Paradigm Shift in Rare Disease Diagnosis

Abstract

Although high-throughput sequencing technologies have significantly enhanced the molecular diagnostic capabilities for rare diseases, more than half of rare disease patients worldwide still lack a definitive diagnosis. The core bottleneck of this dilemma has shifted from data acquisition to data integration—specifically, the siloed storage and analysis of multi-omics information (including genomics, transcriptomics, and epigenomics), which results in the burial of numerous potential diagnostic clues. This paper systematically reviews the latest advancements in multi-omics data fusion methods within the field of rare disease diagnosis. It focuses on analyzing how deep learning frameworks—exemplified by multi-modal variational autoencoders and attention mechanisms—enable the unified representation and collaborative inference of heterogeneous omics data.