Tera-MIND: Tera-scale mouse brain simulation via spatial mRNA-guided diffusion

Area of Science:

• Neuroscience
• Computational Biology
• Bioinformatics

Background:

• 3D modeling of molecularly defined brain structures is essential for understanding brain functions.
• Existing tera-scale volumetric atlases present computational challenges for modeling complex structures.
• Spatially resolved transcriptomic data offers sub-cellular resolution but requires advanced computational tools.

Purpose of the Study:

• To propose Tera-MIND, a novel generative framework for simulating teravoxel-scale 3D mouse brains.
• To address computational challenges in modeling intricate brain structures within their native spatial context.
• To enable detailed 3D modeling of brain structures using spatial gene expression data.

Main Methods:

• Developed Tera-MIND, a patch-based and boundary-aware diffusion model for 3D brain simulation.
• Utilized spatial gene expression as conditional input for generating virtual mouse brains.
• Employed 3D gene-gene self-attention to identify spatial molecular interactions.

Main Results:

• Generated teravoxel-scale virtual mouse brains with comprehensive cellular morphological detail.
• Identified spatial molecular interactions within key transcriptomic pathways, including glutamatergic and dopaminergic systems.
• Demonstrated the translational applicability of Tera-MIND on human brain samples.

Conclusions:

• Tera-MIND provides an efficient generative modeling framework for whole virtual organisms.
• The framework facilitates the study of complex brain structures and molecular interactions.
• Tera-MIND paves the way for integrative applications in biomedical research, including comparative studies.