[Bug]: SeqSero2 sub-species misidentification

[kapsakcj]

What container were you trying to use, and how were you attempting to use it?

The docker image for most current version of SeqSero2 v1.3.1 contains SalmID 0.11

Image name staphb/seqsero2:1.3.1 and the :latest tag.

I have found multiple occurrences/samples where this docker image has misclassified the subspecies, resulting in discrepancies between results produced from this docker image and the results from the PulseNet 2.0 platform (which I think uses SeqSero2S and some kind of look up table)

For example, with one sample I expected subspecies "diarizonae IIIb", but this docker image predicts "arizonae IIIa"

And vice-versa: with another sample I expected subspecies "arizonae IIIa", but this docker image predicts "diarizonae IIIb"

Another example, I expected subspecies "houtenae IV", but this docker image did not predicted a subspecies at all, just -

So, when running the same data through the biocontainers docker image quay.io/biocontainers/seqsero2:1.3.1--pyhdfd78af_1, all predictions are correct and as expected.

I've concluded that the reason the biocontainers docker image runs successfully, is that SalmID is on v0.1.23 (the most recent version) instead of the older 0.11 version

SO. I'm working on re-vamping the SeqSero2 v1.3.1 dockerfile to install via the bioconda recipe instead of installing dependencies manually. SeqSero2 is old & unmaintained python code and there's so many dependencies which leads to mess when attempting to install everything independently. So many issues with installing the various python modules and external deps like SalmID which require poetry (yet another python package manager????) for manual installation.

I'm also planning on adding tests for the various subspecies to ensure it can predict them accurately.

I will open a draft PR soon.

Relevant log output

[kapsakcj]

branch is called cjk-seqsero2-bugfix. Needs more work before I open a draft PR

[kapsakcj]

For example, with a sample that is known to be Salmonella enterica subsp. arizonae (IIIa), this docker image incorrectly predicts subspecies diarizonae IIIb: https://www.ncbi.nlm.nih.gov/biosample/SAMN02568554/

# get illumina reads
wget -q https://ftp.sra.ebi.ac.uk/vol1/fastq/SRR429/000/SRR4293100/SRR4293100_1.fastq.gz
wget -q https://ftp.sra.ebi.ac.uk/vol1/fastq/SRR429/000/SRR4293100/SRR4293100_2.fastq.gz

# run staphb/seqsero2:1.3.1 docker image on reads, in SeqSero2's allele mode
$ docker_run -ti staphb/seqsero2:1.3.1 SeqSero2_package.py -i SRR4293100_1.fastq.gz SRR4293100_2.fastq.gz -t 2 -m a -d SRR4293100-out -n SRR2493100 -p 2
building database...
mapping...
check samtools version: 1.8
[bam_sort_core] merging from 0 files and 2 in-memory blocks...
assembling...
blasting... 

Sample name:    SRR2493100
Output directory:       /data/SRR4293100-out
Input files:    /data/SRR4293100_1.fastq.gz     /data/SRR4293100_2.fastq.gz
O antigen prediction:   41
H1 antigen prediction(fliC):    z4,z23
H2 antigen prediction(fljB):    -
Predicted identification:       Salmonella enterica subspecies diarizonae (subspecies IIIb)
Predicted antigenic profile:    41:z4,z23:-
Predicted serotype:     IIIb 41:z4,z23:-
Note:   This predicted serotype is not in the Kauffman-White scheme. This antigenic profile has been associated with serotype 'Waycross or IV 41:z4,z23:- or IIIa 41:z4,z23:-' in the Kauffman-White scheme. The existence of the same antigenic formula in multiple species or subspecies is well documented in the Kauffman-White Scheme.

Another example, expected subspecies IIIb and docker image predicted IIIa: https://www.ncbi.nlm.nih.gov/biosample/SAMN03371464/

wget -q https://ftp.sra.ebi.ac.uk/vol1/fastq/SRR195/009/SRR1955549/SRR1955549_1.fastq.gz
wget -q https://ftp.sra.ebi.ac.uk/vol1/fastq/SRR195/009/SRR1955549/SRR1955549_2.fastq.gz

docker_run -ti staphb/seqsero2:1.3.1 SeqSero2_package.py -i SRR1955549_1.fastq.gz SRR1955549_2.fas
tq.gz -t 2 -m a -d SRR1955549-out -n SRR1955549 -p 2
building database...
mapping...
check samtools version: 1.8
[bam_sort_core] merging from 0 files and 2 in-memory blocks...
assembling...
blasting... 

Sample name:    SRR1955549
Output directory:       /data/SRR1955549-out
Input files:    /data/SRR1955549_1.fastq.gz     /data/SRR1955549_2.fastq.gz
O antigen prediction:   60
H1 antigen prediction(fliC):    r
H2 antigen prediction(fljB):    z
Predicted identification:       Salmonella enterica subspecies arizonae (subspecies IIIa)
Predicted antigenic profile:    60:r:z
Predicted serotype:     IIIa 60:r:z
Note:   This predicted serotype is not in the Kauffman-White scheme. This antigenic profile has been associated with serotype 'IIIb 60:r:z' in the Kauffman-White scheme. The existence of the same antigenic formula in multiple species or subspecies is well documented in the Kauffman-White Scheme.

I have added these samples (among a few more) to test test stage of the dockerfile I'm working on.