Hi Guys! How are you? I managed to put the pipeline to run, and I thank @Niklas Schandry for all the tips and help. However I am currently struggling to run the assembly step due to "lack of memory". Despite my data is really heavy, I have more than enough computational resources and I tried several configurations for FLYE or process high / high memory, but it does not work even when setting --exclusive use of nodes through SLURM inside my HPC system. The assembly runs normally If I try with the same parameters outside nextflow as normal sbatch/slurm script. I am running it on AMD 512GB RAM with 128 cores partitions (I can allocate like 4-5 nodes exclusively, if available). This is an example (of thousands hahaha) of how I tried to set my .nextflow/config:

singularity{
  autoMounts = true
}

process {
  executor = 'slurm' 
  queue = 'amd-512'
  queueSize = 30
  submitRateLimit = '10/1min'
  clusterOptions = '--qos=preempt'
  errorStrategy = 'retry'
  maxRetries = 3
  time = '24h'
  errorStrategy = { task.exitStatus in [125,139] ? 'retry' : 'finish' }
  containerEngine = 'singularity'
  nextflow.enable.virtualThreads = true
}

process {
  withName: FLYE {
    executor = 'slurm'
    queue = 'amd-512'
    clusterOptions = '--qos=preempt --exclusive'
    memory = { 500.GB }
    cpus = { 120 }
  }
}

If anyone has a suggestion, or noticed if I am doing anything wrong, I would appreciate the help! Good sunday to all of you

I dont think you are doing anything wrong, you seem to have a large genome, and maybe you sequenced at a high coverage? "To reduce memory consumption for large genome assemblies, you can use a subset of the longest reads for initial disjointig assembly by specifying

--asm-coverage

and

--genome-size

options. Typically, 40x coverage is enough to produce good disjointigs." Is a suggestion from the flye docs https://github.com/mikolmogorov/Flye/blob/flye/docs/USAGE.md; you can pass this using for example:

process {
  withName: FLYE {
    executor = 'slurm'
    queue = 'amd-512'
    clusterOptions = '--qos=preempt --exclusive'
    memory = { 500.GB }
    cpus = { 120 }
    ext.args = "--asm-coverage 40"
  }
}

Thanks for the Args suggestion! Well, the estimated genome size is 3 billion bp at maximum. Files are indeed big, though.

Yes, this is confusing, but the general approach most people seem to take is to basically to use cpu refer to "threads"

hmmm

the main thing that I was trying to get at; if you set 120 cpus in your nextflow config, the

--threads

for flye should also be 120

I got this same exact error for all runs hahaha But I will make a new run with these modifications later and see what happens

This is a problem with assembling large genomes in general, if you have ONT reads that are >=10.4 you can also try using

hifiasm

in

--ont

mode

Yeah, I am also testing it outside nextflow

If it is fine outside nextflow then it is some configuration problem

Or some limitation inside my HPC system

I strongly recommend taking a look at the

.command.sh

and also

.command.run

files in the work directories of the failed tasks and see if they are as expected

Okay, thanks a lot Niklas! I am very new to Nextflow

Sure, nextflow is also pretty confusing sometimes

Yeah hahaha I even tried multiple configurations suggested by Seqera AI after explaining how my system works

Thanks, you too