Some Manhattan Plots don't show up

[jielab]

Dear Peter:

Sorry to bug you on this. I am very puzzled that some of my GWAS do NOT have Manhattan plot displayed. Please see the screenshot below. Even after I clicked the "Manhatten" link, nothing changed.

I don't see any error message on the console either. Please see the screenshot below.

Can you please give me a little hint, so that I know where to look for problems?

Thanks!

JH

[pjvandehaar]

Browser devtools console

…

On Wed, Jun 19, 2024 at 12:00 PM Jie Huang ***@***.***> wrote: Dear Peter: Sorry to bug you on this. I am very puzzled that some of my GWAS do NOT have Manhattan plot displayed. Please see the screenshot below. Even after I clicked the "Manhatten" link, nothing changed. image.png (view on web) <https://github.com/statgen/pheweb/assets/26947455/9c513e18-76ef-4ebb-b1de-caf32f30b4df> I don't see any error message on the console either. Please see the screenshot below. image.png (view on web) <https://github.com/statgen/pheweb/assets/26947455/9fe21c56-4118-4924-ab6a-dea0227bb1fd> Can you please give me a little hint, so that I know where to look for problems? Thanks! JH — Reply to this email directly, view it on GitHub <#230>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAGSPCIBKVPQATPFWEY3G63ZIGTKLAVCNFSM6AAAAABJSK2AKGVHI2DSMVQWIX3LMV43ASLTON2WKOZSGM3DENZRGM4TQMY> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[jielab]

Thanks, Peter!

Please see the screenshot below. The console shows "Manhattan XHR failed".
How could I find the cause and a solution for this?

Thanks!
JH

[pjvandehaar]

Look at browser devtools network tab. Maybe something went wrong loading. Maybe just rerun pheweb processing?

…

On Thu, Jun 20, 2024 at 10:58 PM Jie Huang ***@***.***> wrote: Thanks, Peter! Please see the screenshot below. The console shows "*Manhattan XHR failed* ". How could I find the cause and a solution for this? Thanks! JH image.png (view on web) <https://github.com/statgen/pheweb/assets/26947455/890af095-3c42-4b0a-9c96-4c38b83ef68a> — Reply to this email directly, view it on GitHub <#230 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAGSPCLFJCL7HG5JMYKBZPTZIOJFTAVCNFSM6AAAAABJSK2AKGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCOBRHEYDINBXGA> . You are receiving this because you commented.Message ID: ***@***.***>

[jielab]

Dear Peter:

I did rerun pheweb processing a couple of times, but still the Manhattan plot does not show up.

To facilitate this communication to identify the cause, and not wasting too much of your precious time, I now post my GWAS data at this link https://drive.google.com/file/d/1wmZiGEzb5Guw6qBH6QKIjB4R_JpKqUJI/view?usp=sharing.

And below is my config.py file

hg_build_number = 19
assoc_max_maf = 1
show_correlations = False
show_manhattan_filter_button = True
cache = "/mnt/d/data/dbsnp/"
field_aliases = {'CHR':'chrom', 'POS':'pos', 'NEA':'ref',  'EA':'alt', 'BETA':'beta', 'SE':'sebeta', 'P':'pval'}

At your convenient time, can you please download this GWAS file and test at your side? I hope this exercise could identify an issue that is useful for many other pheweb users as well.

Many thanks!

JH

[jielab]

I have been so puzzled by this behavior. I would greatly appreciate if someone could clarify this mystery. I hope that it is not just me who got this manhattan plot display problem.

Thanks!
JH

[kmetsalu]

We have a same problem. Most phenotypes work but some don't.

[kmetsalu]

We discovered that some manhattan .json files contained NaN values. Testing if this may be a reason.

[jielab]

Hi,

Thank you very much!

Yes, indeed, I found NaN value in the .jason file for the trait that can't display a Manhattan plot!

This .json file only has one line. My original GWAS file does NOT have any NaN values. Do you know the cause and fix for this NaN issue?

Best regards,
Jie

[kmetsalu]

Sry. No answer yet - people will start investigation in upcoming days. It happens somewhere in data processing pipeline and now the question is, whether it's something related to input data or something unexpected in pheweb processing step.

[jielab]

That's great!

Looking forward to a further investigation and an answer :-)

Best regards,
Jie

[hoogla]

Dear @kmetsalu and @jielab , did you ever come up with a solution to this or at least some more insights what might be the cause? I am running into the same issues with some of my traits. Thanks!

[hoogla]

I found out that all the traits that fail to show a manhattan plot have some signals with p-values of 0 in them.

In /pheweb/load/manhattan.py , these 0 p-values get replaced with math.inf in the _bin_variant function:
qval = math.inf if variant['pval'] == 0 else self._rounded(-math.log10(variant['pval'])

And in the _rounded function, these inf values then end up as NaN in:
x = qval // self._qval_bin_size * self._qval_bin_size + self._qval_bin_size / 2

This pull request #188 provided the solution that works for me. In the above-mentioned _bin_variant function, replace
qval = math.inf if variant['pval'] == 0
with
qval = self._rounded(320) if variant['pval'] == 0
Since in the rest of the pheweb interface a p-value of 0 is also shown as <e-320, this fix seems actually consistent.

The weird thing is that I have other traits with p-values of 0 and also the example that @pjvandehaar mentions in the pull request (https://pheweb.org/UKB-TOPMed/pheno/275.1) seems to show up just fine. I assume that the explanation is in the selection logic of the _get_qvals_and_qval_extents function, which frankly I don't fully understand. But it seems to me that for some traits the NaNs get selected as qval_extent (with the result that the manhattan plots for these traits will not show, due to the NaN in the json), while for other the NaN doe not get selected as qval_extent and all is fine. Maybe this is impacted by how many other significant signals there are in the region surrounding the p-value = 0 signal, no idea.

I know that development on pheweb has stopped, but I am still writing this here for all the other old users, who might stumble upon this issue.

[pjvandehaar]

I believe I just fixed this with 3d5f912 . If not please re-open this issue.