Interpretable deep generative ensemble learning for single-cell omics with Hydra

Single-cell omics technologies enable the dissection of cellular heterogeneity, yet the high dimensionality, inherent noise, and sparsity present significant analytical challenges. As the technologies evolve to measure multiple molecular features, such as chromatin accessibility and surface protein expression, they introduce additional complexity to the integrative analysis of multimodal single-cell omics data. Here, we propose Hydra, a deep generative framework based on an ensemble of variational autoencoders for effective utilization of diverse unimodal and multimodal single-cell omics data. Hydra implements interpretable learning modules for capturing cell type-specific molecular signatures. The ensemble of such interpretable modules improves feature selection reproducibility and robust annotation of cell types in query datasets. We benchmarked Hydra on a rich repertoire of 21 datasets, including unimodal and multimodal single-cell omics data spanning multiple tissue types and technologies. Our results demonstrate that Hydra offers comparable to superior performance against several state-of-the-art methods. Finally, we highlight the utility of Hydra in learning and robustly annotating brain cellular subtypes using our previously published dataset that profiles early and late stages of Alzheimer’s disease.