Sample codes

[whelena]

Hi I am attempting to use the novoSpaRc package in R 3.5.0 using a matrix.mtx file from 10x genomics. However, I am having difficulty setting up the inputs and parameters for gw_assign. i am wondering if there are any sample codes out there (python/R) that could help me out a bit

[lambdamoses]

Sorry for the late reply. I'm no longer working on predicting gene expression in spatial data. First you should normalize the data whichever way you like. Here allen_use is a dgCMatrix for scRNA-seq data with genes in rows and cells in columns. starmap_use is a dgCMatrix for spatial data with genes in rows and cells in columns. locations is a data frame with columns x, y, and if you wish, z, for locations of cells. Calculate graph-based distance:

D_cell <- calc_graph_dist(allen_use, k = 20)
D_loc <- calc_graph_dist(locations, k = 20, transposed = TRUE)

If a reference spatial gene expression atlas is provided, then you should also provide Euclidean distances between each cell in the scRNA-seq dataset and each location in gene expression space. For this example, a randomly selected panel of 50 genes present in both the scRNA-seq and the spatial dataset is used.

set.seed(50)
genes_use <- sample(rownames(allen_use), 50)
allen_sub <- as.matrix(allen_use[genes_use,])
starmap_sub <- as.matrix(starmap_use[genes_use,])
D_cell_loc <- Rfast::dista(t(allen_sub), t(starmap_sub))

Then probabilistically assign cells to locations: This step will be very time consuming if you are not using an optimized BLAS.

p <- rep(1, ncol(allen_sub)) / ncol(allen_sub)
q <- rep(1, nrow(locations)) / nrow(locations)
cell_loc <- gw_assign(D_cell, D_loc, D_cell_loc, alpha = 0.5, p = p, q = q, check_every = 1)

Predict gene expressions at the locations:

gene_pred <- predict_expr_loc(allen_use, cell_loc)

Use the function plot_spatial_expression to plot the predictions and group truth.