Deciphering progressive lesion areas in breast cancer spatial transcriptomics via TGR-NMF

Area of Science:

• Oncology
• Bioinformatics
• Computational Biology

Background:

• Understanding breast cancer tissue heterogeneity and tumor progression requires identifying spatial domains.
• Dropout events in spatial transcriptomics pose computational challenges.
• Existing methods like graph neural networks often lack transparency and interpretability.

Purpose of the Study:

• To decipher disease progression-related spatial domains in breast cancer spatial transcriptomics.
• To develop an interpretable method for analyzing spatial transcriptomic data.
• To identify pathogenic genes and pathways associated with breast cancer progression.

Main Methods:

• Development of three graph regularized non-negative matrix factorization (TGR-NMF).
• Implementation of a unitization strategy to address dropout events and utilize complete gene expression data.
• Integration of gene expression and spatial position neighbor topologies for data representation.

Main Results:

• Identification of a progressive lesion area indicative of breast cancer progression through heterogeneity analysis.
• Uncovering of related pathogenic genes and signal pathways within the identified area.
• Development of an interpretable low-dimensional representation of spatial transcriptomic data.

Conclusions:

• TGR-NMF provides an interpretable approach for analyzing breast cancer spatial transcriptomics.
• The method effectively handles dropout events and reveals critical spatial domains related to disease progression.
• This approach facilitates the identification of key genes and pathways for targeted therapeutic strategies.