R crashes when running run_mofa

[elffairy90]

My R just crashes when I run the last step using run_mofa, I have tried reducing the number of features, and also reinstalled basilisk. Could you help me take a look of the issue please?

Align samples based on common IDs

common_samples <- Reduce(intersect, list(
colnames(lipidomics_matrix),
colnames(metabolomics_matrix),
colnames(proteomics_matrix)
))

Subset each matrix to keep only the common samples

lipidomics_matrix <- lipidomics_matrix[, common_samples]
metabolomics_matrix <- metabolomics_matrix[, common_samples]
proteomics_matrix <- proteomics_matrix[, common_samples]

Combine data into a list for MOFA

data_list <- list(
Lipidomics = lipidomics_matrix,
Metabolomics = metabolomics_matrix,
Proteomics = proteomics_matrix
)

Visualize dimensions of input data

lapply(data_list, dim)

Create MOFA object

MOFAobject <- create_mofa(data_list)

Optional: Group information if relevant

For example, grouping participants by a clinical phenotype or intervention

groups <- c(rep("Group1", 100), rep("Group2", 100))

MOFAobject <- create_mofa(data_list, groups = groups)

Visualize data overview

plot_data_overview(MOFAobject)

---- Set model options ------------------------------------------------------

Get and modify default model options

model_opts <- get_default_model_options(MOFAobject)
model_opts$num_factors <- 10 # Adjust based on expected latent factors

---- Set training options ---------------------------------------------------

train_opts <- get_default_training_options(MOFAobject)
train_opts$maxiter <- 1000 # Adjust training iterations if needed

---- Prepare and run MOFA ---------------------------------------------------

MOFAobject <- prepare_mofa(
object = MOFAobject,
model_options = model_opts,
training_options = train_opts
)

Define output file path for the trained model

outfile <- file.path(".", "MOFA_model.hdf5")
MOFAobject.trained <- run_mofa(MOFAobject, outfile, use_basilisk = TRUE)

[halterc]

I got the same problem here. Runs in one project folder but crashes in another. Unfortunately, RStudio does not produce a crash log, so no information on why.

[elasline]

I had the same problem. Every time I run MOFAobject.trained <- run_mofa(MOFAobject, outfile, use_basilisk = TRUE , R crashes.

I was able to resolve it by cleaning the Basilisk cache, which had become corrupted. The steps I followed:

1. Load basilisk -> library(basilisk)
2. Find the external directory used by Basilisk -> basilisk::getExternalDir()
3. Copy the path returned by getExternalDir() and delete the folder in the terminal -> rm -rf /path/to/basilisk/

After this, running run_mofa(..., use_basilisk = TRUE) recreates a clean environment, and MOFA2 runs without crashing.

Even if R still crashes after cleaning the cache, try creating a conda environment manually and point MOFA2 to it using reticulate::use_condaenv().

Hope it could help

[HelenaLC]

Having the same issue. I tried the suggestion re cleaning out basilisk, but it didn't help. Then tried using a conda env, with mofapy2 pre-installed, same issue. When I run in terminal (not RStudio) to see back-log, I only get this:

...
...
Loaded 2 covariate(s) for each sample...


Model options:
- Automatic Relevance Determination prior on the factors: False
- Automatic Relevance Determination prior on the weights: True
- Spike-and-slab prior on the factors: False
- Spike-and-slab prior on the weights: False
Likelihoods:
- View 0 (counts): gaussian


zsh: killed     R

Further above, this also caught me attention, and makes we wonder whether it's a coercion issue, causing the session to run out of memory and crash? I don't know...

Checking inference options for mefisto covariates...
Warning message:
In asMethod(object) :
  sparse->dense coercion: allocating vector of size 3.9 GiB

And suggestions/advice would be grand!
(R v4.5.1 with Bioc v3.22)

*Update: I tried running natively in Python, which also results in zsh: killed python. However, subsetting 1,000 cells runs through. So I am pretty sure it is a memory issue, and that densification (of originally delayed and/or sparse data) causes sessions to execute themselves for protection, or something like that :D