MAT(2): manifold alignment of single-cell transcriptomes with cell triplets

Motivation: Aligning single-cell transcriptomes is important for the joint analysis of multiple single-cell RNA sequencing datasets, which in turn is vital to establishing a holistic cellular landscape of certain biological processes. Although numbers of approaches have been proposed for this problem, most of which only consider mutual neighbors when aligning the cells without taking into account known cell type annotations. Results: In this work, we present MAT(2) that aligns cells in the manifold space with a deep neural network employing contrastive learning strategy. Compared with other manifold-based approaches, MAT(2) has two-fold advantages. Firstly, with cell triplets defined based on known cell type annotations, the consensus manifold yielded by the alignment procedure is more robust especially for datasets with limited common cell types. Secondly, the batch-effect-free gene expression reconstructed by MAT(2) can better help annotate cell types. Benchmarking results on real scRNA-seq datasets demonstrate that MAT(2) outperforms existing popular methods. Moreover, with MAT(2), the hematopoietic stem cells are found to differentiate at different paces between human and mouse.

Automated summary

MAT(2) aligns cells in the manifold space using a deep neural network with a contrastive learning strategy, defining cell triplets based on known cell type annotations to produce a more robust consensus manifold and reconstructing batch-effect-free gene expression, which helps to annotate cell types more effectively.