MORPHE: Bridging Image Generation and Spatial Omics for Tissue Synthesis.

Area of Science:

• Computational Biology
• Bioinformatics
• Artificial Intelligence in Biology

Background:

• Spatially resolved omics technologies offer single-cell resolution but face limitations like cost, incomplete coverage, and 2D imaging.
• Existing methods struggle to reconstruct or extend tissue context beyond experimental constraints.

Purpose of the Study:

• To develop an AI framework, MORPHE, for synthesizing biologically faithful tissue architecture from spatial omics data.
• To enable reconstruction and extension of tissue context computationally, addressing current technological limitations.

Main Methods:

• MORPHE utilizes a graph-informed probabilistic embedding to map cell identities and spatial relationships into a continuous latent space.
• This latent space is compatible with diffusion modeling, leveraging pre-trained image-generative models for synthesis.
• Cells are modeled as fundamental units, learning collective relationships to generate large-scale tissue structures.

Main Results:

• MORPHE successfully synthesized tissue architecture at single-cell resolution across large-scale datasets (millions of cells) from the intestine and brain.
• The framework demonstrated computational scalability and enabled outpainting, inpainting, and cross-tissue imputation in 2D and 3D.
• MORPHE generated biologically faithful tissue reconstructions, extending beyond experimental limitations.

Conclusions:

• MORPHE represents a novel class of tissue generation algorithms for spatial omics data.
• This AI framework effectively addresses current limitations in spatial omics, enhancing data reconstruction and extension capabilities.
• MORPHE facilitates a deeper understanding of tissue organization by computationally reconstructing complex spatial cellular maps.