MENDER: fast and scalable tissue structure identification in spatial omics data.

Area of Science:

• Spatial omics
• Computational biology
• Bioinformatics

Background:

• Tissue structure identification is vital for spatial omics data analysis.
• Complex models like Graph Neural Networks are often used, but their performance benefits are questioned.
• Cellular neighborhood structures are consistently observed across spatial technologies.

Purpose of the Study:

• To develop an effective and efficient method for tissue structure identification in spatial omics data.
• To challenge the assumption that increased model complexity directly correlates with improved performance.
• To leverage consistent cellular neighborhood patterns for improved biological insights.

Main Methods:

• Proposed Multi-range Cell Context Decipherer (MENDER), a biology-driven computational model.
• Applied MENDER to diverse spatial omics datasets, including brain regions and a whole-brain atlas.
• Compared MENDER's performance against existing complex models in terms of accuracy and running time.

Main Results:

• MENDER demonstrated substantial performance improvements over modern complex models.
• The method achieved significant speed advantages, being much faster than the second-fastest alternative.
• MENDER successfully identified previously overlooked spatial domains associated with brain aging.
• The model differentiated breast cancer patient subtypes that were previously obscured.

Conclusions:

• MENDER offers a powerful and efficient solution for tissue structure identification in spatial omics.
• The biology-driven design provides superior performance and biological discovery capabilities.
• MENDER's scalability supports analysis of large-scale datasets, like million-level brain atlases.