Space: reconciling multiple spatial domain identification algorithms via consensus clustering.

Motivation: The rapid development of spatially resolved transcriptomics (SRT) technologies has provided unprecedented opportunities for characterizing and understanding tissue architecture. As this field continues to advance, various methods have been developed to computationally identify spatial domains within tissues. However, the performance of different algorithms on the same dataset is not always consistent. This inconsistency makes it difficult for researchers to select the most reliable results for downstream analysis.

Results: To address this challenge, we propose a domain identification method named Space. Space measures consistency between different methods to select reliable algorithms. It then constructs a consensus matrix to integrate the outputs from multiple algorithms. We introduce similarity loss, spatial loss, and low-rank loss in Space to enhance the accuracy and optimize computational efficiency. This strategy not only resolves the inconsistent issue of clustering labels among different methods but also achieves highly reliable clustering output. Flexible interfaces are also provided for downstream analysis such as visualization, domain-specific gene analysis and trajectory inference. Testing results on multiple publicly available SRT datasets demonstrate that Space performs exceptionally well in deciphering key tissue structures and biological features.

Availability and implementation: The Space package can be easily installed through conda or mamba, and its source code is available at https://honchkrow.github.io/Space.