Powerful and accurate case-control analysis of spatial molecular data

Area of Science:

• Computational biology and bioinformatics
• Spatial transcriptomics and molecular pathology

Background:

• Increasing spatial molecular data necessitates advanced methods for disease-associated structure identification.
• Current methods relying on manual annotations may miss crucial biological signals.

Purpose of the Study:

• Introduce Variational Inference-based Microniche Analysis (VIMA) for flexible and precise discovery of spatial disease features.
• Develop a deep learning and statistical approach to overcome limitations of existing methods.

Main Methods:

• VIMA employs a variational autoencoder to generate numerical 'fingerprints' from tissue patches.
• These fingerprints define 'microniches'—biologically similar tissue groups across samples.
• Rigorous statistics identify microniches correlating with case-control status.

Main Results:

• VIMA demonstrated superior calibration, power, and accuracy in simulations compared to other approaches.
• Applied to Alzheimer's dementia, ulcerative colitis (UC), and rheumatoid arthritis (RA) datasets, VIMA recapitulated known biology.
• Novel spatial disease features were identified across all tested datasets.

Conclusions:

• VIMA offers a powerful, data-driven approach to uncover spatial features of disease.
• The method enhances the analysis of complex spatial molecular data for biological discovery.