stClinic dissects clinically relevant niches by integrating spatial multi-slice multi-omics data in dynamic graphs

Area of Science:

• Computational biology
• Cancer research
• Bioinformatics

Background:

• Spatial multi-omics (SMSMO) integration advances cellular niche understanding in tumors.
• Challenges include data scale, diversity, heterogeneity, and small sample sizes, limiting clinical insights.

Purpose of the Study:

• To introduce stClinic, a dynamic graph model for integrating SMSMO and phenotype data.
• To uncover clinically relevant cellular niches and link them to disease manifestations.
• To enhance understanding of tumor microenvironments and their clinical impact.

Main Methods:

• stClinic aggregates information from neighboring nodes across slices using a Mixture-of-Gaussians prior.
• It employs attention-based geometric statistical measures for slice characterization relative to the population.
• Zero-shot learning is utilized for label annotation and multi-omics integration.

Main Results:

• stClinic identifies aggressive tumor niches enriched with tumor-associated macrophages.
• Favorable prognostic niches abundant in B and plasma cells were discovered.
• A specific niche involving myeloid cells and fibroblasts driving colorectal cancer invasion was identified.

Conclusions:

• stClinic effectively integrates multi-omics data to reveal clinically significant cellular niches.
• The model aids in predicting cancer malignancy and patient prognosis.
• Findings provide novel insights into tumor heterogeneity and therapeutic targets.