Integrating single-cell datasets produced by multiple omics technologies is essential for defining cellular heterogeneity. Mosaic integration, in which different datasets share only some of the measured modalities, poses major challenges, particularly regarding modality alignment and batch effect removal. Here, we present a deep probabilistic framework for the mosaic integration and knowledge transfer (MIDAS) of single-cell multimodal data. MIDAS simultaneously achieves dimensionality reduction, imputation and batch correction of mosaic data by using self-supervised modality alignment and information-theoretic latent disentanglement. We demonstrate its superiority to 19 other methods and reliability by evaluating its performance in trimodal and mosaic integration tasks. We also constructed a single-cell trimodal atlas of human peripheral blood mononuclear cells and tailored transfer learning and reciprocal reference mapping schemes to enable flexible and accurate knowledge transfer from the atlas to new data. Applications in mosaic integration, pseudotime analysis and cross-tissue knowledge transfer on bone marrow mosaic datasets demonstrate the versatility and superiority of MIDAS. MIDAS is available at https://github.com/labomics/midas.
Motivation: Clustering analysis for single-cell RNA sequencing (scRNA-seq) data is an important step in revealing cellular heterogeneity. Many clustering methods have been proposed to discover heterogenous cell types from scRNA-seq data. However, adaptive clustering with accurate cluster number reflecting intrinsic biology nature from large-scale scRNA-seq data remains quite challenging. Results: Here, we propose a single-cell Deep Adaptive Clustering (scDAC) model by coupling the Autoencoder (AE) and the Dirichlet Process Mixture Model (DPMM). By jointly optimizing the model parameters of AE and DPMM, scDAC achieves adaptive clustering with accurate cluster numbers on scRNA-seq data. We verify the performance of scDAC on five subsampled datasets with different numbers of cell types and compare it with 15 widely used clustering methods across nine scRNA-seq datasets. Our results demonstrate that scDAC can adaptively find accurate numbers of cell types or subtypes and outperforms other methods. Moreover, the performance of scDAC is robust to hyperparameter changes. Availability and implementation: The scDAC is implemented in Python. The source code is available at https://github.com/labomics/scDAC.