Closed FerrenaAlexander closed 1 year ago
I met the samle problem. Don't know how to fix it.
Hi, just wanted to mention that I was able to get past this error by running on an HPC system with 1000GB of RAM. It seems excessive though, I wonder if this can be optimized. But leaving open for now since others have had this issue too. Thanks for all the work on this package!
Hi @FerrenaAlexander @singlece ,
Based on the size of the dataset shown in that log, 128GB of RAM should be way more than needed, especially if only during the plotting. There is a different error that occasionally occurs during plotting but with a different message (saying an outrageous amount of memory in TB cannot be allocated) which appears to be related to the useRaster
option. One thing you could give a try is to set useRaster=FALSE
in run()
or plot_cnv()
. The option controls R rasterization optimization which make the plotting significantly faster, but occasionally trigger weird behavior.
Regards, Christophe.
The code is corrected as following but still gets the same problem mentioned above.
Code:
infercnv_obj2 = infercnv::run(infercnv_obj,
cutoff=0.1,
out_dir= "./output",
cluster_by_groups = TRUE,
hclust_method="ward.D2",
denoise=TRUE,
HMM=FALSE,
useRaster=FALSE,
num_threads = 10)
Hi @singlece ,
I have never seen this specific type of error before, so I am not sure what causes it. To debug a different issue I just plotted a dataset with 160k+ cells on a machine with 180Gb of RAM, so there's no reason why it should ask for 128Gb of RAM for 20k cells.
Could you let me know what the return of sessionInfo()
in your R session is?
Could you also privately share the backup object from the last completed step so I can try to reproduce the issue?
Regards, Christophe.
Hi all, I'm closing this as I do not seem to get this issue when either running with the following parameters:
useRaster = F
in infercnv::run()
I continue to have issues with runtime with some runs but that's different so I'll open a second issue.
Thanks very much for the suggestions!
Hi - thank you for all the work on this package!
I seme to be running into some issue related to memory during plotting:
I believe it is during Step 19 as that is the last STEP that gets grepped out from the log.
The data is 34817 cells by 23160 genes at input, 10X. It is tumor data from 10 samples. I am running to find CNVs in each sample and in fibroblast. 13616 cells are the reference. The "target" cells are split across CAFs and malignant source samples like so:
I ran this on a SLURM system on a linux HPC with 350 GB memory and 7 threads (sbatch cpus-per-task=7) like below:
Is there anything I can do to rectify this?