Identification of cell-type-specific spatially variable genes accounting for excess zeros

Area of Science:

• * Computational biology and bioinformatics.
• * Statistical genomics and spatial analysis.
• * Molecular and systems biology.

Background:

• * Spatial transcriptomics enables gene expression profiling with spatial context.
• * Identifying cell-type-specific spatially variable (SV) genes is crucial but challenging.
• * Existing methods lack robust statistical approaches for cell-type-specific SV gene detection, especially when modeling zero-inflation and cell-type proportions.

Purpose of the Study:

• * To develop a novel statistical method, CTSV, for identifying cell-type-specific SV genes.
• * To address limitations in current methods by simultaneously modeling zero-inflation and cell-type proportions.
• * To provide a robust tool for analyzing spatial transcriptomic data and uncovering cell-type-specific spatial gene expression patterns.

Main Methods:

• * Developed CTSV, a statistical approach modeling spatial count data using a zero-inflated negative binomial distribution.
• * Incorporated cell-type proportions and spatial effect functions within a regression framework.
• * Utilized the R package `pscl` for model fitting and a Cauchy combination rule for P-value integration.

Main Results:

• * Simulation studies demonstrated CTSV's superior performance over existing methods at both aggregated and cell-type levels.
• * CTSV achieved higher statistical power for detecting cell-type-specific SV genes.
• * Analysis of pancreatic ductal adenocarcinoma data revealed novel biological insights through CTSV-identified SV genes.

Conclusions:

• * CTSV is an effective statistical method for identifying cell-type-specific spatially variable genes.
• * The method provides enhanced biological insights from spatial transcriptomic data.
• * An R package for CTSV is publicly available for broader scientific application.