Introduction The integration of multi-omics data for disease diagnosis holds transformative potential in the field of computational biology, especially when applied to the intricate and dynamic interactions between microbial communities and their human hosts. Methods This integrative approach enables to capture diverse biological signals across genomic, transcriptomic, proteomic, and metabolomic layers, providing a more comprehensive understanding of disease mechanisms. In alignment with emerging priorities in disease microbiology, our study addresses a critical and timely need for interpretable, scalable, and biologically robust computational models that can extract clinically meaningful diagnostic insights from inherently high-dimensional, heterogeneous, and often incomplete biological datasets. Results and Discussion Traditional image classification approaches in disease contexts—such as those relying solely on histopathological features or genomic imaging—tend to overlook the broader ecological and systemic dimensions that are essential for decoding the mechanisms of microbial pathogenesis. These single-modal methods often suffer from significant limitations, including reduced scalability to diverse clinical settings, poor generalizability across patient populations, and an inability to handle partially observed or biologically variable data. Such constraints diminish their effectiveness in precision diagnostics, disease subtyping, and therapeutic decision-making. By contrast, our approach emphasizes multi-modal integration and model interpretability, aiming to overcome these limitations and advance the development of next-generation diagnostic tools that are both clinically actionable and biologically grounded.
Jinshan Che (Thu,) studied this question.