Multimodal spatial alignment and morphology mapping with MOSAICField.

Area of Science:

• Spatial biology
• Computational pathology
• Bioinformatics

Background:

• Comprehensive tissue and tumor atlases require integrating diverse spatial data from multiple experimental modalities.
• Challenges exist in aligning and integrating data across spatial technologies with varying molecular features, resolutions, and tissue morphologies.

Purpose of the Study:

• To introduce MOSAICField, a unified framework for aligning spatial slices across arbitrary combinations of experimental modalities.
• To enable the reconstruction of 3D tissue models and the mapping of anatomical structures from multi-slice spatial data.

Main Methods:

• MOSAICField utilizes a deep neural network to compute physical and morphological alignments across multiple tissue slices.
• The framework estimates nonlinear deformation fields using a multimodal feature loss.
• It was evaluated on simulated data and a prostate cancer sample from the Human Tumor Atlas Network (HTAN).

Main Results:

• MOSAICField successfully constructs accurate 3D tumor models and tracks the architecture of the prostatic ductal system.
• The framework improves the analysis of features within and across different modalities.
• MOSAICField demonstrates superior performance compared to existing methods.

Conclusions:

• MOSAICField provides a unified solution for integrating multimodal spatial data from diverse experimental techniques.
• The framework facilitates the creation of comprehensive 3D tissue atlases and enhances biological insights.
• Accurate 3D reconstruction and morphological alignment are crucial for advancing spatial biology research.