Release: v5.3.0 Rainbow Rattlesnake

[dialvarezs]

Added

• #905 - Add nf-test snapshot for test_assembly_input profile (by @dialvarezs)
• #930 - Add binner SemiBin2 (by @d4straub)
• #861 - Added --generate_bigmag_file to execute the bigmag workflow that generates the file to be used as input for BIgMAG (added by @jeffe107)

Changed

• #932 - Replaced usages of deprecated Channel() with channel() and fix other LSP warnings (by @dialvarezs)
• #937 - Updated to nf-core 3.5.1 TEMPLATE (by @dialvarezs)
• #938 - Update nf-core modules (by @dialvarezs)

Fixed

• #894 - Fix read order in metaSPAdes to allow co-assembly of paired-end data of multiple samples (reported by @maartenciers, fix by @jfy133 with contributions from @prototaxites, @d4straub and @dialvarezs)
• #927 - MetaBinner now succeeds when no contigs are too short or all are binned (reported by @MicroSeq, fix by @d4straub)
• #929 - Allow the domain_classification.R script to run with any assembler, not just Megahit or Spades (reported by @MicroSeq, fix by @prototaxites)
• #943 - Fixed concatenation of BUSCO summaries with uneven columns by changing from csvtk to qsv (reported by @jfy133 and @julianu, fix by @dialvarezs)
• #943 - Fixed creation of the Tiara report channel used for concatenation (by @dialvarezs)
• #945 - Skip mixing of GTDB-Tk MultiQC files when binning is skipped (reported by @amizeranschi, fix by @dialvarezs)
• #953 - metaSPAdes retries upon error 250 (out of memory), rather than finishing the pipeline.
• #954 - Skip GTDB-Tk when no bin QC tool is enabled and add warning messages (fix by @dialvarezs)
• #956 - Support long reads assemblers in assembly input (fix by @dialvarezs)

Dependencies

Tool	Previous version	New version
bcftools	1.21	1.22
csvtk	0.31.0
fastp	0.24.0	1.0.1
geNomad	1.11.1	1.11.2
metamdbg	1.1	1.2
mmseqs	17.b804f	18.8cc5c
nf-core		3.5.1
qsv		5.1.0
samtools	1.21	1.22.1
SemiBin2		2.2.0

Deprecated

• #943 - Remove csvtk/concat module (by @dialvarezs).

[dialvarezs]

When using conda, Semibin2 seems to be trying to train a new model... Any ideas why, @d4straub?

[d4straub]

When using conda, Semibin2 seems to be trying to train a new model.

Yes that is intentionally. Semibin2 should train a model with not only conda but also singularity & docker. Because pre-trained models are only available for single-bam input and CI test 1 & 3 have multiple bam input. However, I just checked the error report, and the model is trained in just a few seconds. The last message by Semibin2 is SemiBin2[11530] INFO Start binning.. I am not sure though why binning should take longer with conda than singularity/docker.
I re-started the failed tests just to make sure that wasnt some weird coincidence.

Edit:

• no, timeout seems reproducible for tests 1 & 3
• curious is though that SemiBin2 works for test 4 ('-profile hybrid' ) with conda

Edit2: CI test 4 ('-profile hybrid' ) has a single sample, i.e. model is pre-trained, input only one bam file. Still a mystery why SemiBin2 fails with conda on multiple samples in CI tests.

[FriederikeHanssen]

Got no additional things. All my comments are merely suggestions. Nice work

[dialvarezs]

no, timeout seems reproducible for tests 1 & 3

I can reproduce it locally with conda. It's strange, because when it reaches that point of "Start binning.", CPU usage drops to 0% and stays like that.

I'm going to check the conda vs the container dependencies if I this is because some bad pin.

EDIT: There is an issue that seems to resemble this behaviour: BigDataBiology/SemiBin#208

EDIT 2: I found the issue! Seems to be the igraph package that recently got v1.0.0