Parallel computation

[peicai]

Hi Jean and team,

Thanks for your work! I'm wondering if I can use MERINGUE with multicore properly. I'd like to compare the computational effeiciency of MERINGUE with other approaches.

Thanks in advance!

[JEFworks]

Great question.

To identify spatially heterogenous genes, MERINGUE evaluates each gene independently. So one potential approach to multicore processing would be to split genes into groups and evaluate each group in parallel.

Here is an example starting from the MOB analysis tutorial:

dim(mat)

[1] 7365  260

# Identify sigificantly spatially auto-correlated genes
start_time <- Sys.time()
I <- getSpatialPatterns(mat, w, verbose = FALSE)
end_time <- Sys.time()
print(end_time - start_time)

Time difference of 6.233067 secs

## Alternative parallelized version
## split genes into 4 groups
ncore <- 4
groups <- split(rownames(mat), sample(1:ncore,nrow(mat),replace=TRUE))
length(groups)

[1] 4

sapply(groups, length)

   1    2    3    4 
1886 1860 1820 1799 

## parallelize groups 4 cores
start_time <- Sys.time()
I <- do.call(rbind, parallel::mclapply(groups, function(gs) {
+   I <- getSpatialPatterns(mat[gs,], w, verbose=FALSE)
+ }, mc.cores=ncore))
end_time <- Sys.time()
print(end_time - start_time)

Time difference of 2.453482 secs

Hope that helps,
Jean

[peicai]

Hi Jean,

Thanks! It works.