Thanks @dparks1134 - apologies for the bug, and we'll deal with this as a priority.

Quick notes

• although it stems from the same misunderstanding of/assumptions about mummer's operation, this is not the same issue as Alignment coverage >1.0 #340 and will probably not be fixed by the proposed changes to resolve that issue.
• the resolution currently appears to be to run mummer for the reciprocal pairwise comparisons.

L.

NOTE: Handling under branch issue_373 for version 0.3.

For reference, two sets of output for the genomes provided by @dparks1134, named <reference>_vs_<query>.zip.

genome1_vs_genome2.zip
genome2_vs_genome1.zip

I have pushed changes to issue_373 that remove the assumption that ANIm is symmetric; skip a test that fails if we no longer enforce genome sorting alphabetically as a result; and updated docs, so that they don't claim it is symmetric.

I will see if there are other things I have missed; in particular, I think perhaps the test outputs might need to be updated; (they pass, but this might just mean the differences fall within the designated tolerances).

I think perhaps the test outputs might need to be updated; (they pass, but this might just mean the differences fall within the designated tolerances).

I think your guess is probably correct. It would be useful to have (for the record) a pyani compare output comparing (i) the old, symmetric behaviour and (ii) the new, non-symmetric behaviour so we can see how big the difference is for at least one real-world example we're familiar with.

I tried uploading the archive of inputs I used here, but I think that must be too big; I've instead included the classes.txt file. @widdowquinn, it's a dataset you sent me, so you should already have the sequence files.

classes.txt

Here's the output from running pyani compare on the symmetric and asymmetric versions of anim, using default settings. I have not yet addressed the suggestions made on pyani compare output in #364, so apologies for the less-than-useful colour schemes, et cetera.

summary_run1_vs_run2.md
ref_1_vs_query_2_plots.zip

Update

Here are the updated versions of the summary report and plots; changes include—better colour maps, more useful axes, and more informative titles. (The run numbers differ because I always forget when you need to specify labels, et cetera.)

updated_plots_with_labels.zip

summary_run4_vs_run5.md

It looks like Dickeya didantii NCPPB 898 chromosome, whole genome shotgun sequence specifically sees a lot of differences between the two.

I'm also marking this for closure, because my understanding is that this issue was fixed with 1f10a6c.

Specifically, we have now decided to run two pairwise comparisons (forward and reverse), and calculate ANIm values for query for each comparison.