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

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell omics data analysis faces challenges due to high dimensionality, noise, and sparsity.
• Accurate annotation of rare cell populations is crucial but difficult, especially in multimodal datasets.

Purpose of the Study:

• To develop a deep generative framework, Hydra, for effective analysis of unimodal and multimodal single-cell omics data.
• To improve the identification and annotation of rare cell populations.

Main Methods:

• Hydra utilizes an ensemble of variational autoencoders for data learning.
• Incorporates interpretable modules for capturing cell-type-specific molecular signatures.
• Ensemble approach enables reproducible feature selection and robust cell-type annotation.

Main Results:

• Hydra demonstrated comparable or superior performance across 21 diverse single-cell omics datasets.
• The framework excels in robustly annotating brain cellular subtypes.
• Hydra effectively preserves disease-relevant signatures, as shown in Alzheimer's disease data.

Conclusions:

• Hydra provides a powerful and robust framework for single-cell omics data integration and analysis.
• The method shows particular promise for advancing the study of rare cell populations in development and disease.
• Hydra facilitates deeper understanding of cellular heterogeneity and disease mechanisms.