core gene alignment looks like it aligns paralogs by accident

[pallevillesen]

Hi, We are analyzing Mycobacteria isolated from cystic fibrosis patients and two patients are known to have infected each other (twins). For these two genomes isolated from these two patients we observe ~760 SNP differences (snpdist run on core_gene_alignment). If I visualize the position of the core genome alignment variation (custom script in R), I see strong vertical lines. Upon close inspection I found that is was very different sequences that were forced to align (with similarity in the beginning of the gene). Outside this single gene, the two genomes are 99.99% identical (11 snps in ~4 megabases).

So it is probably paralogs that are treated as orthologs by accident(?)

I checked more samples and this pattern of variation in the core_gene_alignment is not unusual.

Is there a way for me to force roary to be very stringent when detecting orthologs? I have a feeling that it is a matter of command line options to roary.

I have attached a plot that show many of these "lines" from a roary run of 29 gff files (generated with prokka). It also shows a really weird pattern from codon positions (defined as alignment position %% 3).

[tseemann]

useful paramaters are

         -i        minimum percentage identity for blastp [95]
         -iv STR   Change the MCL inflation value [1.5]

Increasing inflation will increase granularity, that is it will produce smaller clusters.

[pallevillesen]

Thanks a lot for quick reply - will try that.

I developed a simple visualization of the entire alignment to check this.

The SNPs/INDEL columns are scaled with density so the "peaks" are in regions of high density, so it is easy to see some regions of missing paralogue splits.

I don't know if this would of general interest to users of roary(?)

[fconstancias]

Hi @pallevillesen that will be great if you could share your script to generate those nice alignment visualisation.

[pallevillesen]

Hi @fconstancias . I rarely use git (sorry) but actually went all the way to set it up.

https://github.com/pallevillesen/plot_roary