The above made me suspicious. My latest bug fix in the pipeline code aimed to fix the fact that earlier versions of the pipeline didn't manage to remove a set of S. aureus sequences assigned to E. coli. They're good quality, i.e. long and aligning well with the hmm profile but are most similar to sequences from S. aureus when I megablast against the NCBI database. The S. aureus are not in my r226-2 release, but there are three in the r232-1 release! There are a number of things here: 1. My different pipeline versions are messed up. Sorry again. 2. It's likely that the

5a3b50c7da

commit was the one I ran for r226-2 but not r232-1, and that it does a better job at removing the S. aureus sequences at the cost of also removing some true positives, at least the

GB_GCA_902827205.1~LR792634.1

sequence. (I'm not sure if I can confirm the commit hashes. Will look into that.) 3. I think Sativa gets overwhelmed by the presence of several false positives as it pulls out one sequence at a time to place back in the tree. 4. I can see two ways of dealing with problems like this that would be complementary to the sequence-quality style of filtering I do today: Blast sequences and throw away anything that is way off (wrong class/order?) and cluster sequences and use only the representative sequence in the Sativa step. (The latter would also reduce the number of sequences going into Sativa which would be an advantage in itself.) 5. I need a test case that exposes both the false positive (S. aureus) problem and false negative (

GB_GCA_902827205.1~LR792634.1

) type of problem, and that doesn't take a month to run. 6. The false positive E. coli sequences do not occur in the

n5

files, and I should update #CEA7TBJGJ to use this file for the

addSpecies

method rather than the

n20

.

Hi. Thanks for the info. It is somewhat concerning that SATIVA doesn't seem to catch some clear cut problematic cases. Do you have a list of these problematic S. aureus sequences that are found in E. coli genomes? I'd like to see if my approach is identifying these.

At the moment, I only have a couple:

GB_GCA_009600675.1~SQRB01000048.1

,

RS_GCF_019375135.1~NZ_JAENHD010000101.1

, and

RS_GCF_009600665.1~NZ_SQRC01000101.1

. These are the three that got included among the 20 that were chosen by my pipeline. (There are at least five among the >80,000.)