Aurash A Mohaimani <aurash@uwm.edu>
Jan 16

to me
Hi Colin,

In addition to the small issue reported earlier, I'd like to add a [much] more major one that's been forcing my group's installation of JBrowse to disable track histograms entirely in certain cases: the histograms simply aren't being generated for certain tracks on given chromosomes.

An example: http://i.imgur.com/UkdBSBn.png

Chromosome 1 has histograms spanning its entire length, while Chromosome 5 has none at all. Upon attempting to view the entirety of Chromosome 5, we arrive at an "infinite" loading screen for non-existent histograms.

For reference, this is a GFF3 track with long features that can range from kilobases to many megabases in length; it has roughly 800 features in total spread over 20 Rat chromosomes, averaging about 40 features per chromosome.

When attempting to view certain chromosomes in JBrowse, it shows a loading page for the [non-existent] histograms, and then unsurprisingly never loads anything.

It's terribly vexing, and as a short-term workaround, I've had to disable "featureScale" to disable all display of histograms for these kinds of tracks.

If it helps at all, I've generally observed that this phenomenon only appears to occur on GFF3 tracks with a low total feature count.

Aurash

From: Aurash A Mohaimani
Sent: Friday, January 16, 2015 4:24 PM
To: Colin
Subject: JBrowse not Italicizing GFF3 Feature Names

Hi Colin,

I'm writing to inquire about a certain [small, but pestering] issue users have reported to me regarding JBrowse.

For GFF3 features with a "Name" attribute that includes the italicization HTML tags, the JBrowse genome browser shows the literal text of the HTML italics tags, but the mouse-over text in the genome viewer interprets the HTML correctly.

An example can be found here: http://i.imgur.com/mJeg7Kw.png

Might you know of any method to resolve this, or has it already been addressed in a recent JBrowse release?

Warm Regards,
Aurash



Colin <colin.diesh@gmail.com>
Jan 19

to Aurash
Thanks those are certainly interesting issues. The italics seems to only be an issue with CanvasFeatures, but it can be used on HTMLFeatures. I guess you could try that?

With the other issue, that definitely sounds weird, do you have an example file I could look at?


Thanks


Aurash A Mohaimani <aurash@uwm.edu>
AttachmentsJan 19

to me
Hi Colin,

I've attached the sample GFF3 track, which I discussed as the example track regarding the histogram generation issue, to this email.

Would you please try to load it in the current JBrowse version to determine if histograms are being generated on every chromosome?

If you need a link to the relevant Rat5 genome, please don't hesitate to let me know!

Also, I've run some [very] basic stats on this file, just to be sure that there are features covering every chromosome, and here's what I've found:

Feature count for Chr1: 429
Feature count for Chr10: 123
Feature count for Chr11: 8
Feature count for Chr12: 10
Feature count for Chr13: 37
Feature count for Chr14: 21
Feature count for Chr15: 11
Feature count for Chr16: 24
Feature count for Chr17: 9
Feature count for Chr18: 21
Feature count for Chr19: 4
Feature count for Chr2: 80
Feature count for Chr20: 11
Feature count for Chr3: 38
Feature count for Chr4: 67
Feature count for Chr5: 60
Feature count for Chr6: 13
Feature count for Chr7: 36
Feature count for Chr8: 25
Feature count for Chr9: 49

Aurash

From: Colin <colin.diesh@gmail.com>
Sent: Monday, January 19, 2015 2:47 PM
To: Aurash A Mohaimani
Subject: Re: JBrowse not Generating GFF3 Track Histograms

Attachments area
Preview attachment CongenicStrains.gff3


CongenicStrains.gff3

Colin <colin.diesh@gmail.com>
Jan 19

to Aurash
Hi thanks for the data
I tried loading it but didn't see any problems (it actually didnt show histograms by default at all, didn't appear dense enough)

I'm trying to think what other reason the problem is, and I realized that sometimes  "infinite loading" can indicate that javascript crashed, so i might check the javascript console for errors.

Also, I don't suspect this would cause the infinite loading, but there was a important patch added to JBrowse 1.11.4 for canvasfeatures histograms https://github.com/GMOD/jbrowse/issues/475

That said, if you do have the html markup in the feature names, I might recommend using htmlfeatures anyways

Hope that helps


Aurash A Mohaimani <aurash@uwm.edu>
Jan 20

to Marek, Jeffrey, jrsmith, me
Hi Colin,

By "didn't see any problems", do you mean that the latest version of JBrowse was able to generate histograms on every chromosome for the given dataset (Chromosomes 1 through 20, plus X)?

I've inspected the JavaScript console of our JBrowse installation, and I'm receiving the following:


From the console
Warning: "Unable to determine an appropriate data store to use with track 'undefined', please explicitly specify a storeClass in the configuration." (dojo.js:1050)

From the JavaScript console
Error: "TypeError: a is not a constructor" (dojo.js:32)
Warning: "Use of getAttributeNode() is deprecated. Use getAttribute() instead." (dojo.js:248)


I'm unsure of how much any of this may relate to the histogram generation problem, but I checked the number of defined storeClasses in my "trackList.json" and compared it to the number of uniquely defined tracks, and the numbers matched: 87 in total.

The patch you're referring to in the JBrowse 1.11.4 release is something we've actually discussed previously! I do have "binsPerBlock" set to explicitly equal 25 in the "histogram" clauses for each of my Canvas tracks, and that fixed an earlier issue related to histogram displays. However, this issue concerns histogram generation during the process of loading a track into JBrowse using the "flatfile-to-json.pl" script.

I tried to migrate several of our long-feature-length tracks into the older HTMLFeatures track type over the past day, and I've been surprised at this trackType: it fixes another, separate major issue while raising its own independent issues! It provides the new capacity to browse across really long features while moving the feature name (which is very handy), but the histogram display of HTMLFeatures tracks is somewhat wonky--there's fragmentation between histogram bars when zoomed out, and the histogram scale is not easily viewable when using slightly darker colors for the background of the histogram bars. (Also, the JBrowse config wiki's documentation for HTMLFeatures is imperfect.)

Also, it seems that there is some conflict between "featureScale" and "histScale" in HTMLFeatures tracks: "featureScale" will display the "error box" requiring a user to zoom in further in order to view the data, while "histScale" in HTMLFeatures functions more like "featureScale" from the CanvasFeatures tracks--allowing for the zoom threshold at which histograms will be displayed.

Even when I disable "featureScale" and "histScale" on HTMLFeatures tracks, at a sufficiently large viewing window (say, over 250 Megabases on a given chromosome), JBrowse overrides itself into displaying the pre-generated histograms, even though I seemed to not specify for them. If I then remove the actual histogram files for that particular track, JBrowse displays nothing--not even the infinite loading screen from CanvasFeatures tracks.

So, ultimately, HTMLFeatures tracks fix another separate issue, and they also resolve the feature name italicization issue I wrote to you about earlier, but they suffer from their own [significant] problems.

Out of curiosity, why is there a lack of "overlap" between core features of HTML and Canvas feature tracks? The capacity to move feature names along with a user's browsing experience in HTMLFeatures is fantastic, and I don't understand why that would have been neglected when porting the functionality over to the new CanvasFeatures tracks.

My initial experience with HTMLFeatures histograms' display has been a little jarring, and HTMLFeatures tracks also lack the individual feature "hover-over" functionality that's present in CanvasFeatures (when mousing over an individual feature, JBrowse shades the feature, along with creating a small pop-up window containing the given feature's name).

Each track type appears to [independently] solve its own set of issues, without sufficient coverage between the two. And, perhaps worst of all: neither resolves the missing histogram generation.

Aurash

From: Colin <colin.diesh@gmail.com>
Sent: Monday, January 19, 2015 4:51 PM


Colin <colin.diesh@gmail.com>
Jan 20

to Aurash, Marek, Jeffrey, jrsmith
Hi Aurash,
Good assessment of the issues. I'm not an official jbrowse dev but from what I know:

a) The histogram generation uses heuristics and may not output certain histogram resolutions or any histograms files at all if there are not enough features in the file.

b) The jbrowse appears to be crashing in your code where it says TypeError. This will effectively halt the javascript interpreter and make it look like it is infinite loading. Also, the warning message seems to indicate that the "label" is undefined. Perhaps one of the tracks in trackList.json is missing the "label" parameter? Recall that the "label" is actually the "primary identifier" that jbrowse uses, and "key" is more like the trackname that is displayed.

c) The italics is again an interesting scenario. I think you have the right idea about the trade-offs between htmlfeatures and canvasfeatures. One possibility is that you could have italics in the mouseover but not in the label by using the new customized mouseover labels that are in the master branch. An example of this is here https://github.com/GMOD/jbrowse/commit/c13d4142cc54e67796f6815a59fdaed2146ec0e0

d) if you need the mixed fonts to be rendered directly to the canvas i would open up a new github issue, I don't think it's currently possible

e) if you can narrow down the issue with featureScale and histScale, file a github issue for that too!

I can't say exactly why the API for CanvasFeatures and HTMLFeatures is so different since i haven't been involved for too long but I think your assessment is pretty accurate.

Hope that helps,

Thanks,

-Colin