Abstract
MOTIVATION: Single-cell multiomics technologies are emerging to characterize different molecular features of cells. This gives rise to an issue of combining various kinds of molecular features to dissect cell heterogeneity. Most single-cell multiomics integration methods focus on shared information among modalities while complementary information specific to each modality is often discarded. RESULTS: To disentangle and combine shared and complementary information across modalities, we develop a dual-modality factor model named scME by using deep factor modeling. Our results demonstrate that scME can generate a better joint representation of multiple modalities than those generated by other single-cell multiomics integration algorithms, which gives a clear elucidation of nuanced differences among cells. We also demonstrate that the joint representation of multiple modalities yielded by scME can provide salient information to improve both single-cell clustering and cell-type classification. Overall, scME will be an efficient method for combining various kinds of molecular features to facilitate the dissection of cell heterogeneity. AVAILABILITY AND IMPLEMENTATION: The code is public for academic use and available on the GitHub site (https://github.com/bucky527/scME).