SCG is implemented in Python but can be accessed from R via the reticulate package. This page demonstrates how to load the pre-computed SCG results from the data application and work with them in R. We also show how to fit the model from scratch in R for users who prefer that workflow.
Replace the path in use_python() with the path to your own Python installation that has scr installed. To find it, run which python or which python3 in your terminal. Please see reticulate for further guidance.
library(reticulate)
library(ggplot2)
library(dplyr)
Attaching package: 'dplyr'The following objects are masked from 'package:stats':
filter, lagThe following objects are masked from 'package:base':
intersect, setdiff, setequal, unionuse_python("~/.pyenv/versions/3.11.14/bin/python")
np <- import("numpy")The results from the data application are saved in ../results/ and can be loaded directly in R without re-fitting the model.
results <- read.csv("../results/results.csv")
spot_info <- read.csv("../data/spot_info.csv")
gene_names <- read.csv("../data/gene_names.csv")
scgs <- results[results$scg == "True", ]
cat(nrow(scgs), "SCGs identified out of", nrow(results), "edges\n")143 SCGs identified out of 1275 edgesCorr_est <- np$load("../results/Corr_est.npy")
dim(Corr_est)[1] 4727 51 51boundary_colors <- c("T" = "#7651A6", "N" = "#2E7D6E", "I" = "#C4952D")
plot_corr_field <- function(
Corr_est,
spot_info,
g,
g_prime,
gene_names = NULL
) {
field <- Corr_est[, g + 1, g_prime + 1] # convert 0-indexed to 1-indexed
df <- data.frame(
x = spot_info$x,
y = spot_info$y,
corr = field,
region = spot_info$region
)
vmax <- max(abs(df$corr))
g_label <- if (!is.null(gene_names)) {
gene_names$gene[g + 1]
} else {
as.character(g)
}
g_prime_label <- if (!is.null(gene_names)) {
gene_names$gene[g_prime + 1]
} else {
as.character(g_prime)
}
ggplot(df, aes(x = x, y = y, color = corr)) +
geom_point(size = 2.0) +
scale_color_gradient2(
low = "#2166ac",
mid = "white",
high = "#d6604d",
midpoint = 0,
limits = c(-vmax, vmax),
name = "Correlation"
) +
coord_equal() +
ggtitle(paste0(g_label, "—", g_prime_label)) +
theme_void() +
theme(
legend.position = "right",
plot.title = element_text(hjust = 0.5, size = 11)
)
}
plot_corr_field(
Corr_est,
spot_info,
g = 2,
g_prime = 4,
gene_names = gene_names
)
plot_corr_field(
Corr_est,
spot_info,
g = 39,
g_prime = 49,
gene_names = gene_names
)
Users who wish to fit SCG from scratch in R can do so via reticulate. The workflow mirrors the Python data application exactly.
scr <- import("scr")
torch <- import("torch")
Y_np <- as.matrix(read.csv("../data/Y.csv"))
Y_np <- t(Y_np)
dims <- dim(Y_np)
N <- dims[1]
P <- dims[2]
K <- L <- as.integer(round(2 * log(P)))
init <- scr$scr_init(
Y = Y_np,
L = L,
K = K,
a_delta = c(2.1, 3.1),
b_delta = 1.0,
a0 = 5.0,
b0 = 0.5,
nu = 5.0,
spca_center = TRUE,
spca_alpha = 0.01,
spca_ridge_alpha = 0.01,
device = scr$DEVICE,
dtype = scr$DTYPE
)
kernel_hypers <- scr$pick_kernel_params(
init = init,
S = as.matrix(spot_info[, c("x", "y")])
)
K_np <- scr$rq_kernel(
as.matrix(spot_info[, c("x", "y")]),
v2 = 1.0,
rho = 0.1 * kernel_hypers$rho,
alpha = kernel_hypers$alpha
)
fit <- scr$cavi(
torch$as_tensor(Y_np, device = scr$DEVICE, dtype = scr$DTYPE),
torch$as_tensor(K_np, device = scr$DEVICE, dtype = scr$DTYPE),
init,
max_iter = 60L,
verbose = FALSE
)
results <- scr$classify_edges(fit$params, M = 500L, alpha = 0.1)When passing data from R to Python via reticulate, NumPy arrays are non-writable by default. This triggers a PyTorch warning during torch$as_tensor that can be safely ignored.
sessionInfo()R version 4.5.2 (2025-10-31)
Platform: aarch64-apple-darwin20
Running under: macOS Tahoe 26.6.2
Matrix products: default
BLAS: /System/Library/Frameworks/Accelerate.framework/Versions/A/Frameworks/vecLib.framework/Versions/A/libBLAS.dylib
LAPACK: /Library/Frameworks/R.framework/Versions/4.5-arm64/Resources/lib/libRlapack.dylib; LAPACK version 3.12.1
locale:
[1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8
time zone: America/Chicago
tzcode source: internal
attached base packages:
[1] stats graphics grDevices utils datasets methods base
other attached packages:
[1] dplyr_1.2.1 ggplot2_4.0.3 reticulate_1.46.0.9000
loaded via a namespace (and not attached):
[1] vctrs_0.7.3 cli_3.6.6 knitr_1.51 rlang_1.3.0
[5] xfun_0.60 otel_0.2.0 png_0.1-9 generics_0.1.4
[9] S7_0.2.2 jsonlite_2.0.0 labeling_0.4.3 glue_1.8.1
[13] htmltools_0.5.9 scales_1.4.0 rmarkdown_2.31 grid_4.5.2
[17] tibble_3.3.1 evaluate_1.0.5 fastmap_1.2.0 yaml_2.3.12
[21] lifecycle_1.0.5 compiler_4.5.2 RColorBrewer_1.1-3 pkgconfig_2.0.3
[25] htmlwidgets_1.6.4 Rcpp_1.1.2 farver_2.1.2 lattice_0.22-9
[29] digest_0.6.39 R6_2.6.1 tidyselect_1.2.1 pillar_1.11.1
[33] magrittr_2.0.5 Matrix_1.7-4 withr_3.0.3 tools_4.5.2
[37] gtable_0.3.6