Detecting anomalous anatomic regions in spatial transcriptomics with STANDS.

Area of Science:

• Computational Biology
• Genomics
• Biotechnology

Background:

• Spatial transcriptomics (ST) data offers insights into disease mechanisms by characterizing anomalous tissue domains (ATDs).
• Current methods lack the ability for de novo detection and dissection of ATDs, especially in multi-sample ST data.
• Understanding pathogenic heterogeneity requires methods that can identify both population-level and individual-specific factors.

Purpose of the Study:

• To introduce STANDS, an innovative framework for de novo detection, alignment, and subtyping of ATDs from multi-sample ST data.
• To address challenges in multi-sample DDATD, including unalignable ATDs, multimodal data integration, and limited normal ST datasets.
• To improve the characterization of anomalous tissue domains for a deeper understanding of disease pathogenesis.

Main Methods:

• Development of STANDS, a framework utilizing Generative Adversarial Networks (GANs).
• Integration of multimodal-learning, transfer-learning, and style-transfer techniques within the STANDS framework.
• Application of STANDS to diverse multi-sample ST datasets for benchmarking.

Main Results:

• STANDS demonstrates superior performance in identifying common and individual-specific ATDs.
• The framework effectively dissects ATDs into biologically distinct subdomains.
• STANDS provides insights into the development of ATDs that are initially indistinguishable from normal tissues.

Conclusions:

• STANDS offers a powerful new approach for the detection and dissection of anomalous tissue domains in multi-sample ST data.
• The framework advances the characterization of pathogenic heterogeneities and individual-specific disease factors.
• STANDS has the potential to reveal early-stage ATDs crucial for understanding disease progression.