Whole brain alignment of spatial transcriptomics between humans and mice with BrainAlign.

Area of Science:

• Neuroscience
• Computational Biology
• Genomics

Background:

• Mouse models are crucial for human neuroscience research, necessitating advanced computational tools for cross-species data translation.
• Spatial transcriptomics (ST) offers high-resolution gene expression data, but aligning human and mouse ST data remains challenging.

Purpose of the Study:

• To develop a computational method, BrainAlign, for whole-brain alignment of spatial transcriptomics data between humans and mice.
• To leverage self-supervised learning to identify conserved brain regions and homologous gene expression patterns across species.

Main Methods:

• Developed BrainAlign, a self-supervised learning approach utilizing a heterogeneous graph neural network.
• BrainAlign encodes spatial transcriptomics spots and genes into shared embedding spaces for cross-species integration.
• Employed genomic analysis to investigate gene expression connections and patterns.

Main Results:

• BrainAlign successfully integrated cross-species ST data, revealing conserved brain regions.
• The method facilitated the detection of homologous regions between human and mouse brains.
• Genomic analysis confirmed similar expression patterns for marker genes and identified cross-species gene expression connections.
• BrainAlign accurately mapped homologous regions into a unified spatial domain while preserving relative positions.

Conclusions:

• BrainAlign provides a robust computational framework for cross-species brain alignment using spatial transcriptomics.
• This approach enhances the translational potential of mouse models in human neuroscience research.
• The findings highlight conserved organizational principles and gene expression patterns in mammalian brains.