Data-driven fine-grained region discovery in the mouse brain with transformers.

Area of Science:

• Spatial transcriptomics
• Computational biology
• Neuroscience

Background:

• Spatial transcriptomics enables detailed mapping of tissue organization, but analyzing large datasets presents computational challenges.
• Existing methods struggle with scalability for organ-scale spatial transcriptomic data, hindering comprehensive analysis.
• Identifying distinct spatial domains and niches is crucial for understanding tissue architecture and function.

Purpose of the Study:

• To establish a scalable workflow for self-supervised spatial domain detection in large-scale spatial transcriptomic datasets.
• To develop a novel computational framework, CellTransformer, for learning latent representations of tissue spatial domains.
• To enable the discovery of both known and uncataloged spatial domains within complex organ structures like the mouse brain.

Main Methods:

• Utilized a self-supervised framework for learning latent representations of tissue spatial domains.
• Developed a novel encoder-decoder architecture, CellTransformer, for hierarchical feature learning from cellular and molecular data.
• Integrated representation learning with minibatched GPU-accelerated clustering for scalability to multi-million cell datasets (e.g., MERFISH, Slide-seqV2).

Main Results:

• CellTransformer successfully identified spatial domains in multi-million cell MERFISH and whole-brain Slide-seqV2 datasets.
• The workflow integrated cells across tissue sections, identified known domains (e.g., Allen Mouse Brain CCF), and discovered hundreds of uncataloged areas.
• Demonstrated high consistency (nearly perfect) of up to 100 spatial domains across multiple mice and tissue sections in large-scale analyses.

Conclusions:

• CellTransformer provides a performant and scalable solution for fine-grained spatial domain detection in large-scale spatial transcriptomics data.
• The workflow facilitates complex multi-animal analyses and aids in the discovery of novel neuroanatomical subregions.
• This advancement in computational analysis is critical for advancing the field of spatial transcriptomics and brain mapping.