ISPAT: Informed Spatially aware Patterns for Multiplexed Immunofluorescence Data

An R package implementing ISPAT, a Bayesian framework that identifies shared and region-specific cell-type interaction patterns in multiplexed immunofluorescence (mIF) tissue imaging data, with multi-modal integration for Poisson and Gaussian spatial mixed-effects models.

Methods and application: Bhadury, S. et al., ISPAT: Informed Spatially Aware Patterns for Multiplexed Immunofluorescence data, Nature Scientific Reports.

Installation

# install.packages("remotes")
remotes::install_github("sagnikbhadury/ISPAT")

ISPAT depends on rstan; make sure a working C++ toolchain and rstan are installed first (see the RStan getting-started guide).

Usage

library(ISPAT)

mydata <- readRDS(file = paste0(savepath, "data/",
                   "Data - Analysis ready Marginally obtained Spatial intensities ",
                   "of Cells on Tissue Slide_", diseasetype, "_", mypatID, ".rds"),
                   refhook = NULL)
rawdata <- mydata %>% tidyr::pivot_wider(names_from = Cells, values_from = Intensities)
S <- rawdata[, c("X", "Y", "area_category")] %>% as.matrix()
Y <- rawdata %>% dplyr::select("APC", "CTL", "Epi", "THelper", "Treg") %>% t() %>% as.matrix()

saved <- ISPAT(Y, S, ncores = numcores, RefPrior = diag(1, 5, 5), use_ref = FALSE,
               Kernel = "Matern", sGLM_method = "MLE", VB_MSFA = TRUE, MSFA_method = "CAVI")

ISPAT() returns a shared cell-type interaction network common across all tissue regions/clusters, plus a region-specific network for each cluster.

Package layout

• R/ISPAT.R — main model-fitting function, ISPAT().
• R/msfa.R — the multi-study factor analysis step, cavi_msfa() and svi_msfa().
• R/ext-loop.R — internal per-marker spatial GLM worker.
• inst/stan/ — bundled Stan programs for the per-marker spatial fit (Gaussian and Poisson spatial mixed models, RBF and Matern kernels).

See ?ISPAT for full documentation of the main function.