Improving shape/color assignments in input/alignment visualizations

[nfranz]

This issue is related to the following input file and run: Dropbox/Euler-Runs/BirdPhylogenies/NMF-nmf.mobile.asu.edu/2015-12-10-13:24:50-2015-Galloanserae

A. Regarding the input visualization:
2015-Galloanserae.pdf

1. Notice how 2 input articulations (ncAnas, ncPhasianidae) involving "nc" regions on each side are not connected to their input taxonomies, and are also not correctly shaped/colored. The input concepts also occur in the corresponding trees, but are not shown as articulated there.
2. Wondering if each tree can somehow be made more wide (as opposed to long), resulting in a more quadratic than rectangular "canvas".
3. Related to 2: is it possible to get all or most leaf-level concepts for each tree (separately) more or less into a single vertical region? Right now some leaves are very close to the root and others at the extreme (narrow) tips of the tree spatial arrangement. That makes it harder to keep track of which leaves are and are not (yet) articulated.

B. Regarding the alignment visualization:
2015-Galloanserae_0_mnpw.pdf

1. This use case has (1) very differential leaf-level sampling (both in terms of actual leaf concept identity and numbers of leaves) and (2) many 1 parent/1 child "chains" ("monotypy") in the input taxonomies. As a result, there are a lot of rounded corners/grey color regions in the alignment. However, only 6/20 of the "terminal grey regions" involve concepts from both the 2015 and 2014 input taxonomies. The other 14 terminal grey regions involve only 2015 taxonomy concepts.

Proposal: I think we want to visually differentiate the following 5 classes of concept interactions in an alignment:

1. Single concept in an alignment, Taxonomy 1: yellow, octagon [= current]
2. Single concept in an alignment, Taxonomy 2: green, rectangle with sharp corners [= current]
3. Multiple concepts in an alignment, ALL Taxonomy 1: yellow, rectangle with rounded corners [= new]
4. Multiple concepts in an alignment, ALL Taxonomy 2: green, rectangle with rounded corners [= new]
5. Multiple concepts in an alignment, from BOTH Taxonomy 1 AND 2: grey, rectangle with rounded corners [current]

So this would convey the following:

a. Yellow shapes: unique to Taxonomy 1, occupied by either a single (octagon) or multiple concepts (rectangle with rounded corners)

b. Green shapes: unique to Taxonomy 2, occupied by either a single (rectangle with sharp corners) or multiple concepts (rectangle with rounded corners)

c. Grey: truly congruent, multi-taxonomy regions.

I believe this system is scalable to more than two taxonomies (usually additional shapes and colors, but reserving "rounded corners" and "grey color" for special meanings.

1. The same higher level Euler-Runs folder contains an "mncb" run -- that alignment visualization needs attention as well (but lower priority).

[shizhuoyu]

Problem A:
According to what Mingmin’s implementation, we need to analysis the meaning of nc. “nc” stands for “no coverage”. It means “the coverage does not NEED to be hold” (but it is OK if it holds).

I use an example to explain:
(A B C) means “B isa A” and “C isa A”, which are encoded as “A {>, =} B” and “A {>, =} C”, because by default we have coverage “on”, and non-emptiness (because A, B, C are real concepts), so, finally we have “A {>} B and A {>} C and A = B U C”

(A B nc) means “B isa A”, which is encoded as “A {>,=} B”. However, we have a special keyword “nc” here means the coverage does not NEED to be hold. So, we have exactly “A {>, =} B”, while at the same time, we create an “placeholder” (we called hidden child), named “nc_A”, and we encode it as “A {>} nc_A“. “nc_A” is a placeholder, it is not a concept, so it does not have the constraint “non-emptiness”. So, if the coverage does not hold, we have “A {>} B and A {>} nc_A and A = B U nc_A”, but if the coverage does hold, we have “A {=} B” and the “nc_A” is empty (in encoding: all regions involving nc_A are invalid regions). So, finally we just have “A {>, =} B”.

Regarding to Nico’s issue, we have the following in taxonomy 2015:
(Anas Anas_superciliosa nc)
(Phasianidae Bona_Gall_Clade Rollulus nc)

and in input taxonomy 2014:
(Anas Anas_platyrhynchos nc)
(Phasianidae Gallus Meleagris nc)

Since the “nc” means the coverage does not NEED to be hold, we create the placeholder
2015.ncAnas, 2014.ncAnas, 2015.ncPhasianidae and 2014.ncPhasianidae

These four “hidden children” do not have the constraint “non-emptiness”, so they are not concepts and they can be empty, I am guessing we should not put them in the articulations. Instead, we can specify them in the input taxonomies as something like
(Anas Anas_superciliosa ncAnas)
so that "ncAnas" becomes a “real” concept, which can be used in articulations.

Our program find these four in the articulations, and treat them as new real concepts, that is why they are not connected to their input taxonomies. The input concepts also occur in the corresponding trees, but are not shown as articulated.

The #2 and #3 in ProblemA is due to the limitation of graphviz. Maybe we can tweak the dot file to get it. Still working…

Problem B has been solved for mnpw encoding, it should work with mncb, too. But I have not tested all cases.

[nfranz]

Thanks, Shizhuo. Very helpful. I have worked with "nc" more now, and I think that logically it works very well and is actually essential to the current "Aves 2015/2014" use case. My Issue #6 was about the visualizations only; I have no issues with the input encoding, PWs, or MIR, etc. Let me explain.

taxonomy 2015:
(Anas Anas_superciliosa nc)

taxonomy 2014:
(Anas Anas_platyrhynchos nc)

The reason why I (user) wish to use this encoding is that both 2015/2014 studies can be assumed to work off of a phylogenetically congruent circumscription of the genus-level concept Anas (sec. either). For instance, the cite identical "catalogues" for lower-level classifications. However, due to research constraints/coordination, each source sequenced the entire genome of a different species-level concept (superciliosa sec. 2015 and pltayrhynchos sec. 2014). Very likely this was on purpose, because there is a globally coordinated effort (and budget) to reconstruct the bird tree(s) with genomic data, and so "redundant tips" are purposefully avoided.

Thus, I want to articulate the following:

[2015.Anas equals 2014.Anas] => congruent genus-level circumscriptions.
[2015.Anas_superciliosa disjoint 2014.Anas_platyrhynchos] => disjoint species-level concepts sampled.
[2015.Anas_superciliosa is_included_in 2014.Anas] => see 1st articulation.
[2015.Anas includes 2014.Anas_platyrhynchos] => "inverse" of the preceding articulation.

I could stop now, and would get one or several possible worlds (several, if the example gets more complicated and the leaf-level sampling partially overlaps). As more and more "nc" annotations appear in the input, given the purposefully non-congruent leaf-level sampling approach, it can be helpful to add the following:

[2015.ncAnas disjoint 2014.ncAnas]

which will limit the consistent outcomes to one possible world. Interpretation: For each study (2015, 2014), that part of Anas which is not explicitly included as a child (the nc part) is non-overlapping with other study's corresponding (nc) part. Or, simpler, in each case the nc includes exactly that child of the other source that was not sampled herein. Given the authors' intentions, this is a valid way to represent the alignment.