You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Setting clipping parameters does not seem to trim reads. Here is an example:
I run nextflow run nf-core/rnaseq -r 3.10 -profile test,singularity --outdir output
I check the Sequence Length Distribution in FastQC (trimmed) and the max length is 100.
then I add --clip_r1 17 and run nextflow run nf-core/rnaseq -r 3.10 -profile test,singularity --outdir output --clip_r1 17
the Sequence Length Distribution in FastQC (trimmed) is also 100 (whereas R1 reads are expected to be 17nt shorter). The trimgalore report (trimgalore/RAP1_IAA_30M_REP1_1.fastq.gz_trimming_report.txt) does not mention anything about trimming by 17 bp.
Finally, I run the same as just above but with version 3.9: nextflow run nf-core/rnaseq -r 3.9 -profile test,singularity --outdir output --clip_r1 17
Here, the Sequence Length Distribution in FastQC (trimmed) is 84 for R1 and 100 for R2 as expected. The trimgalore report mentions: "All Read 1 sequences will be trimmed by 17 bp from their 5' end to avoid poor qualities or biases"
Mattias Zepper figured out (see discussion on Slack) that
In pipeline version 3.9, the module still contained the lines:
// Clipping presets have to be evaluated in the context of SE/PE
def c_r1 = params.clip_r1 > 0 ? "--clip_r1 ${params.clip_r1}" : ''
def c_r2 = params.clip_r2 > 0 ? "--clip_r2 ${params.clip_r2}" : ''
def tpc_r1 = params.three_prime_clip_r1 > 0 ? "--three_prime_clip_r1 ${params.three_prime_clip_r1}" : ''
def tpc_r2 = params.three_prime_clip_r2 > 0 ? "--three_prime_clip_r2 ${params.three_prime_clip_r2}" : ''
which went missing in version 3.10 of the pipeline.
This seem to have been removed from the module and not added back in the pipeline.
Command used and terminal output
No response
Relevant files
No response
System information
Nextflow 22.10.4
ran tests locally and tried to trim real data on HPC (slurm) with the same results.
Container: singularity
OS: Linux Ubuntu
nf-core/rnaseq v3.9 and v3.10
The text was updated successfully, but these errors were encountered:
@Bin-Xia: As Ido Tamir pointed out in another issue, one can also work around this. Either use a custom config for the process or trim directly with the aligner --extra_star_align_args "--clip5pNbases 23,4"
@maxulysse: Is there a point in cherry-picking Julies fix and making a fixed release for the 3.10 branch also? No version is entirely free of bugs, but that parameters are simply ignored to me is a rather serious issue.
Description of the bug
Setting clipping parameters does not seem to trim reads. Here is an example:
I run
nextflow run nf-core/rnaseq -r 3.10 -profile test,singularity --outdir output
I check the Sequence Length Distribution in FastQC (trimmed) and the max length is 100.
then I add
--clip_r1 17
and runnextflow run nf-core/rnaseq -r 3.10 -profile test,singularity --outdir output --clip_r1 17
the Sequence Length Distribution in FastQC (trimmed) is also 100 (whereas R1 reads are expected to be 17nt shorter). The trimgalore report (trimgalore/RAP1_IAA_30M_REP1_1.fastq.gz_trimming_report.txt) does not mention anything about trimming by 17 bp.
Finally, I run the same as just above but with version 3.9:
nextflow run nf-core/rnaseq -r 3.9 -profile test,singularity --outdir output --clip_r1 17
Here, the Sequence Length Distribution in FastQC (trimmed) is 84 for R1 and 100 for R2 as expected. The trimgalore report mentions: "All Read 1 sequences will be trimmed by 17 bp from their 5' end to avoid poor qualities or biases"
Mattias Zepper figured out (see discussion on Slack) that
In pipeline version 3.9, the module still contained the lines:
which went missing in version 3.10 of the pipeline.
This seem to have been removed from the module and not added back in the pipeline.
Command used and terminal output
No response
Relevant files
No response
System information
Nextflow 22.10.4
ran tests locally and tried to trim real data on HPC (slurm) with the same results.
Container: singularity
OS: Linux Ubuntu
nf-core/rnaseq v3.9 and v3.10
The text was updated successfully, but these errors were encountered: