A Zero-Inflated Hierarchical Generalized Transformation Model to Address Non-Normality in Spatially-Informed...

Area of Science:

• Genomics
• Computational Biology
• Oncology

Background:

• Oral squamous cell carcinomas (OSCC) present diagnostic challenges, leading to poor survival rates.
• Spatial transcriptomics is key to understanding the OSCC tumor microenvironment.
• Existing cell-type deconvolution methods struggle with zero-inflated OSCC data.

Purpose of the Study:

• To develop a novel method for accurate cell-type deconvolution in zero-inflated spatial transcriptomics data from OSCC.
• To improve the understanding of the OSCC tumor microenvironment.
• To quantify uncertainty in cell-type proportion estimations.

Main Methods:

• Developed a zero-inflated hierarchical generalized transformation model (ZI-HGT).
• Integrated ZI-HGT with Conditional AutoRegressive Deconvolution (CARD) for enhanced cell-type deconvolution.
• Applied the ZI-HGT + CARD framework to OSCC spatial transcriptomics data.

Main Results:

• The ZI-HGT + CARD framework significantly improved cell-type deconvolution accuracy for OSCC data.
• The method effectively handles high zero-inflation in spatial transcriptomics data.
• Accurate mapping of diverse fibroblast populations within the OSCC tumor microenvironment was achieved.

Conclusions:

• The ZI-HGT + CARD framework offers a robust solution for cell-type deconvolution in challenging spatial transcriptomics datasets.
• This approach enhances the understanding of tumor heterogeneity and immunosuppression in OSCC.
• The findings are critical for advancing OSCC research and therapeutic strategies.