great amount memory consumption inscreased

[Alipha-Clisa]

I am not sure whether the problem comes from the new version 0.25.0 or the option --telo-m / -u / -l, but according the 0.19.9 version hifiasm, using the following command would finish complete workflows with 1.5 TB memory:

${pathHifiasm} -t 100 \
     -o "${pathDirOut}/ultimate" \
     -x 0.98 \
     -l 0 \
     --primary \
     -k 41 \
     --ul-rate 0.05 \
     --h1 HIC-1_R1.fq.gz \
     --h2 HIC-1_R2.fq.gz \
     --ul pass.all.fq.gz \
     [[:digit:]].fastq.gz \
     1>${pathStdOut}\
     2>${pathStdErr}

while update hifiasm to version 0.25.0 and additionally switch on the --telo-m TTTAGGG -u -l 1would leading to 1.5TB memory depletion without any file generated, and now hifiasm only finished three round of kmer cleaning:

 ${pathHifiasm} -t 100 \
     -o "${pathDirOut}/k41_purge" \
     -x 0.98 \
     -l 1 \
     -k 41 \
     --primary \
     --telo-m TTTAGGG \
     --ul-rate 0.05 \
     -u \
     --ul pass.all.fq.gz \
     --h1 HIC-1_R1.fq.gz \
     --h2 HIC-1_R2.fq.gz \
     [[:digit:]].fastq.gz \
     1>${pathStdOut}\
     2>${pathStdErr}

I would try the older version hifiasm with same option, and I would wonder whether the option --telo-m / -u / -l are designed as memory expensive operation or not ?

[chhylp123]

--telo-m TTTAGGG -u -l 1 would only affect the post assembly steps, instead of the error correction. I am not so much sure about the performance. It really depends on how much HiFi data you are using. In addition, -k41 and large number CPUs used will also increase the memory requirement.

[Alipha-Clisa]

After resorting to another server the hifiasm managed to produce the ec.bin, ovlp.source.bin and ovlp.reverse.bin and so on the contig gfas, but subjecting to limited 1.5T memory, hifiasm failed to finish the complete pipeline. I would wonder whether resuming based on the *.bins generated would reduce the memory requirement?

[chhylp123]

It should be. You may also consider to reduce the number of CPUs.