DSS-2.40.0 running out of time

[MithilG]

Hello Developers and the Users of DSS,
I am working with several differentially methylated region (DMR) callers, including DSS. I am using the DSS v2.40 in order to maintain the consistency across on-going projects in the lab.

Presently I am calling the DMRs in non-CG context which cannot be collapsed and thus the loci (or sites) are high in number from both strands.

From couple of weeks, I am facing a problem while using the DSS v2.40 as follows:

DSS runs out of time even when I try to run the program in the HPC cluster with its highest resources (shown below) for the following code:

HPC resources used were:

#! /bin/bash
#SBATCH --time=96:00:00
#SBATCH --mem=255536
#SBATCH --cpus-per-task=20
#SBATCH --job-name=DSS_DMRcalling

Code for DMR-calling using the DSS:

mParam = MulticoreParam(workers=madetectCores() - 3, 1), progressbar=TRUE) # which are 29 cores
dmlTest.sm <- DMLtest(bs.filtered,
                             group1=1:6,
                             group2=7:12,
                             smoothing=TRUE,
                 BPPARAM=mParam) # computes parallely
dmrs0<- callDMR(dmlTest.sm, p.threshold=0.05)

The 96hrs (maximum time period) at HPC cluster is less for DSS to finish the DMRcalling step by DSS. Further, I extracted the samples into fractions such as chromosomes 1-6. But still my program cannot complete the "computing test statistic" function. It gives the following text as an output:

Smoothing ...
Estimating dispersion for each CpG site, this will take a while ...
  |======================================================================| 100%

  |======================================================================| 100%

Computing test statistics ...
slurmstepd: error: *** JOB 17623606 ON hpc-n547 CANCELLED AT 2024-11-20T13:40:35 DUE TO TIME LIMIT ***

Previously even when I used chromosomes 1-2 or 1-3 (without using the BPPARAM argument in the DMLTest), I faced the same issue.

Is this expected or there is some crucial parameter that I am missing out?
Did anyone faced this issue earlier?
I will be really grateful if anyone of you provide me your feedback over this issue.

Looking forward to hearing from you.

Have a nice day.

[WangYangSDSZ]

Hi there, I have faced the same problem. Have you found any solutions? Thank you.

[MithilG]

Hi @WangYangSDSZ,
No, I couldn't find any solution in DSS itself.

However, an alternative indirect solution to circumvent the non-CG context issue is that I collapsed the strand and then treated it with DSS dmr-caller. This worked!

Unlike in the CG context, non-CG context methylations are asymmetric on complementary DNA strands. Therefore, collapsing the strand shouldn't lead to the loss of information and eventually shouldn't affect the ultimate result.

I hope it helps.

[WangYangSDSZ]

Thank you for your answer. I have another question. Could you tell me that how much longer it will cost to smooth a data containing ~2000 samples? I only want to focus on the CG context CpG sites. And also, do you recommend other software to analyse WGBS data?

[MithilG]

Hi @WangYangSDSZ again,

(1) When I used DSS for my 12 samples (6 vs 6)—CG contexts, it took me ~13 hours to get the results. However, this speed was a function of the limited computational resources I provided—64 GB memory with one full node.

If you provide sufficient cores/RAM and leverage DSS's parallel computing feature (which I didn't for my analysis), you should get your results in 3-4 days (a naive guess). Since I have never worked with many samples, I can't be precise about the computational resources, even if I want to.

(2) I have used [methylsig] (https://doi.org/10.1093/bioinformatics/btu339) and [dmrseq] (https://doi.org/10.1093/biostatistics/kxy007). They are faster than DSS since they don't define DMRs based on the CG-loci and their smoothing. If you are interested, you may have a look at them.

[WangYangSDSZ]

Thank you sososososo much!!!!!