Unifying Multimodal Single-Cell Data Using a Mixture of Experts β-Variational Autoencoder-Based Framework

Multimodal single-cell assays measure complementary layers of cell state, but integration is complicated by differences in modality, sparsity, and cohort coverage. We present UniVI (Unified Variational Inference), a scalable mixture-of-experts {beta}-variational autoencoder that learns a shared latent space while preserving modality-specific structure. UniVI uses modality-specific encoders/decoders with a shared latent prior and a symmetric cross-modal alignment objective, enabling consistent integration of paired measurements without curated feature-link graphs or pre-annotated reference atlases; optional supervised heads can be added when labels are available. Across paired RNA-protein and RNA-chromatin datasets, UniVI yields coherent embeddings, improves label transfer, and supports cross-modal reconstruction and denoising. Extending to tri-modal measurements, UniVI maintains robust three-way alignment among RNA, chromatin accessibility, and surface proteins. UniVI also degrades smoothly under severe cell-type imbalance and in the presence of modality-exclusive populations. Finally, in an acute myeloid leukemia mosaic design, a paired RNA-protein bridge co-organizes independent RNA-only and protein+genotype cohorts, revealing genotype-associated neighborhoods that strengthen with mutation-head fine-tuning. Together, UniVI provides a flexible, interpretable framework for multimodal integration across paired, tri-modal, and mosaic study designs and supports practical reference-to-query projection in partially observed studies. The full package is available online at github.com/Ashford-A/UniVI.