Review annotations to co-localizes_with GO:0032991 protein-containing complex & children

[pgaudet]

The Protein Complex Working group has decided that annotations to GO:0032991 protein-containing complex & children should NOT be allowed with the colocalizes_with qualifier, see #1500.

There are quite a few EXP annotations (non EXP manual are below).

The review spreadsheets are sorted by group, and then by year. https://docs.google.com/spreadsheets/d/1TPSWpRTqElK0f05IzRjYSXiBVC0Bh98e5wfFH2vt190/edit#gid=0

Group	# annotations
UniProt + UniProtKB	482
SGD	267
FlyBase	81
BHF-UCL	69
ParkinsonsUK-UCL	30
ARUK-UCL	28
WB + WormBase	24
MGI	21
dictyBase	16
RGD	10
CAFA	7
HGNC	7
AspGD	4
PomBase	4
AgBase	2
CACAO	2
CGD	2
GO_Central	1

Here are the manual, non-EXP annotations:

https://docs.google.com/spreadsheets/d/1hanprhIk5PJ8VQeUeGVv8G-S8gJKSJPgRnVatorQYIs/edit#gid=0

Group	# annotations
UniProt	485
BHF-UCL	25
ARUK-UCL	12
GO_Central	12
ParkinsonsUK-UCL	9
AgBase	5
HGNC	5
dictyBase	5
PomBase	1
SGD	1
WB	1

[ValWood]

I only spotted this ticket by change, nobody is tagged in it? PomBase ones were already done via https://github.com/geneontology/go-annotation/issues/1500 https://github.com/pombase/curation/issues/1940 I updated the spreadsheet.

[pgaudet]

I think @bmeldal will send out an announcement to fix this ? We said we'd start from the most recent ones but she can probably provide a more precise plan. I'm assigning a few groups (unfortunately we're limited to 10 assignees).

pascale

[bmeldal]

I don't mind who sends the announcement. Tbh, shouldn't it come from an editor? Or I copy/paste the start of this ticket?

[slaulederkind]

All manual RGD annotations have been fixed or removed.

[pfey03]

All dictyBase EXP annotations have been changed or removed and updated in Google table

[pfey03]

only now saw the NON EXP. Took care of those few also.

[hattrill]

All ours have been updated.

[rachhuntley]

So what are our options for updating these. Take the example from #1500:

SGD annotate CDC5 polo like kinase to "cohesin complex" with "colocalizes_with" It isn't a bona fida subunit but it binds and phosphorylates cohesin during DNA damage

In this case, should a new term be requested "cohesin complex binding"?

[bmeldal]

SGD annotate CDC5 polo like kinase to "cohesin complex" with "colocalizes_with" It isn't a bona fida subunit but it binds and phosphorylates cohesin during DNA damage

Shouldn't that be something like 1. protein-containing complex binding | IPI | WITH "cohesin complex" If there's binding info it's more than co-localisation. 2. kinase activity (or child) | IDA | "some extension to indicate substrate" (sorry, I don't know them by heart) "cohesin complex"

[pgaudet]

You can use the protein complex ID to annotate, using 'with', or yes, request the complex binding term (we had decided we would create these until we have a better structure to capture the information).

Pascale

[ValWood]

SGD annotate CDC5 polo like kinase to "cohesin complex" with "colocalizes_with"

that is the one which originated this ticket https://github.com/geneontology/go-annotation/issues/1490 (because it was used in annotation transfer). But I don't see this annotation at SGD any longer.....

I think that in S. cerevisiae polo phosphorylates kleisin subunit in meiosis so in this case it might be possible to just add the substrate to the actual subunit.....

[RLovering]

Help required: PMID:24665357 shows co-localization of NOS1AP (Fig.2) with two ion channels. Fig 2F provides a graph of the degree of co-localization for a variety of channels. Plus NOS1AP silencing in cardiac myocytes reduced significantly the amplitude of electrically evoked calcium transients and the authors suggest functionally relevant interactions with the ion channels that regulate the action potential duration through action on the L-type calcium channel, and potassium channels, probably through S-nitrosylation. However, no specific binding is shown. I think it is important to capture this data, however we have deleted these annotations in line with the proposal.

Please advise how this information can be captured

Thanks

Ruth

[pgaudet]

@RLovering

How about 'regulation of ion channel activity'? If no binding is shown, then no binding should be annotated ;)

[ValWood]

They aren't really showing this direct connection though.... maybe they are altering something about the membrane composition, or something else, if there is nothing to demonstrate a physical connection...

Is there a particular membrane region they are both associated with? maybe the localization to the same region, in the same cell type is sufficient.

There are already lots of annotations to positive regulation of potassium ion transmembrane transport positive regulation of delayed rectifier potassium channel activity positive regulation of voltage-gated potassium channel activity involved in ventricular cardiac muscle cell action potential repolarization regulation of calcium ion transmembrane transport via high voltage-gated calcium channel regulation of high voltage-gated calcium channel activity

with without an actual interaction with a specific complex at this resolution it is really providing more support for the argument than the experiments provide....

[RLovering]

yes but as you say these effects can be mediated by very distant regulatory processes. The close proximity of these proteins suggests that there is a more intimate interaction required

[ValWood]

but the figure just shows "vague" statistical colocalization in the same cell types over other transporters "NOS1AP co-localizations in neonatal cardiac myocytes" and not to a particular structure, or region?

Is it GO-worthy?

[bmeldal]

See also "x binding term" NTR from Karen for reannotation: https://github.com/geneontology/go-ontology/issues/15692

[RLovering]

Hi Val

I am somewhat surprised to find that published experimental data with statistical support is being described as 'vague'. This is how the majority of immunofluorescent images are interpreted. The protein is not shown to co-localise with the other channels that it doesn't regulate and does co-localise with the channels it does regulate.

I am not convinced that removing all these annotations is the right way to proceed.

Ruth

[ValWood]

Maybe it's because I'm not used to looking at images of human cells and I don't know what I'm looking at. To me this seemed to be 'everywhere' including the nucleus. Perhaps that isn't what is being shown? What is the subcellular localization where these two proteins are colocalizing in this image?

To me, the information does not add so much value, because it also 'colocalizes' with the other complexes but to a lesser degree....

If we keep them, we have the problem that this becomes associated often as a member of the ion channel complex, and we don't really want that?

[ValWood]

this is what the text says

"showed a high degree of co-localization of NOS1AP with the Kir3.1 potassium channel (Pearson coefficient 0.7 ± 0.03), and the L-type calcium channel (0.68 ± 0.018). Much lower was the degree of co-localization with RyR2, a sarcoplasmic reticulum protein (0.4 ± 0.011) and alpha-sarcomeric actin, a myofilament protein (0.35 ± 0.032) and minimal in the case of connexin 43 (Cx43), an intercalated-discs protein (0.15 ± 0.01). These results suggest possible functional interactions with ion channels that influence the cardiac action potential and less relevant interactions with the sarcoplasmic reticulum, sarcomeres and intercalated discs."

it says there is higher colocalization with ion channels than a sarcoplasmic reticulum protein?

Isn't this just a subcellular location?

[bmeldal]

Summary from call on 9th May 2018:

@vanaukenk @deustp01 @sandraorchard @pgaudet @hdrabkin @krchristie @ValWood @tberardini

Summary:

• carry on removing the qualifier and replace as follows:

• either remove the annotation completely if no interaction evidence present

• or annotate to [x complex binding] - request new term if needed

• or annotate as "[GP] | protein-containing complex binding | with [Complex Portal AC]

• or annotate as "[GP] | protein-containing complex binding | AE has_direct_input [GO complex term]

[RLovering]

Thanks for this very helpful summary.

I don't think the AE relation 'has_substrate' is right, based on existing guidelines, I think it would be has_direct_input.

Also how about adding the option or annotate as: "[GP] | GO CC annotation | AE adjacent_to [GO complex term] or [Complex Portal AC] This would enable me to make the annotation I would like to make. although I am not sure how GOC pipelines would translate this extension. But I don't think it would end up causing a new annotation to be created. I'll give it a try and see what happens ;)

[bmeldal]

I don't think the AE relation 'has_substrate' is right, based on existing guidelines, I think it would be has_direct_input.

That makes more sense! (I put the ? for a reason)

[GP] | GO CC annotation | AE adjacent_to [GO complex term] or [Complex Portal AC]

Can you give an example, please?

[RLovering]

I think the model in Figure 1 in https://www.ncbi.nlm.nih.gov/pubmed/25221472 is an example of where co-localized would previously have been used. NOS1AP binds NOS1 (nNOS) which binds PSD95 which binds the NMDAR (GO:0017146 NMDA selective glutamate receptor complex).

Note this review discusses developing drugs to prevent the NOS1AP interaction NOS1 and thus prevent NMDAR dysregulation. Admittedly this review discusses the controversy about the role of NOS1AP but it is established that NOS1AP binds NOS1, and that NOS1 regulates the activity of NMDAR through localised synthesis of NO (I think via S-nitrosylation).

So perhaps we could create: NOS1AP GO:0031234 extrinsic component of cytoplasmic side of plasma membrane adjacent_to GO:0017146 NMDA selective glutamate receptor complex Instead of NOS1AP co-localizes_with GO:0017146 NMDA selective glutamate receptor complex

[ValWood]

But which component term would you use?

We discussed this briefly last night... what "sub-cellular localization" are they demonstrating with this experiment?

The general feeling seemed to be that some colocalization experiments which required a statistical approach to show a specific colocalization, but where the actual subcellular location was not clear from the image was not clear were probably not "GO annotatable"

[ValWood]

I now see your comment above. So can you see that this is

"GO:0031234 extrinsic component of cytoplasmic side of plasma membrane" from this figure?

[ValWood]

This is the image https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3961100/figure/fig02/

[bmeldal]

Hmmm, staining for NOS1AP shows an almost uniform cellular distribution and where it overlaps with certain receptors might just be coincidental given it's uniform distribution. In A and E it seems to accumulate a little stronger on the nuclear membrane - where it shows NO overlap with any of the receptors in question.

Ignore the box plot - it seems totally irrelevant, one can shoehorn significance into any data by simply correlating A with B and ignoring all outliers ;-) (I've seen it done first hand in a previous lab!)

[ValWood]

That's what I thought....I would not annotate this to a GO CC term.... possibly "nuclear membrane" but I'm not familiar with human experiments so I wasn't sure ... I would not use this data for any association with a specific complex.

[BarbaraCzub]

Hello,

This conversation reminded me about one of my annotations, which I believe, captured the biology best with the colocalises_with qualifier.

The figure below (Figure 4 from PMID:16923168) shows perfect co-localisation of tubulin complexes (GO:0045298) [anti-tubulin-beta3] with TTBK1 (UniProtKB:Q5TCY1; UniProtKB:Q6PCN3; tau-tubulin kinase 1) [7589].

Generally the paper focuses on characterisation of TTBK1, and it demonstrates the phosphorylation of the microtubule-associated protein tau by TTBK1. However, it shows no direct interaction with tubulin; only the co-localisation.

I would assume that it does not bind to tubulin directly (but it binds to tau, which in turn binds to tubulin). And it certainly is not a part of the tubulin complex. Yet, the figure clearly demonstrates an association between TTBK1 and tubulin, and the colocalises_with qualifier captures this relationship well, and it reflects the biology well.

So, I was actually wondering what was the rationale for deciding that annotations to GO:0032991 protein-containing complex & children should NOT be allowed with the colocalizes_with qualifier? And what is the recommended alternative? (Sorry, @bmeldal , I am not sure where to find the minutes from the Protein Complex Working group's calls?)

Thanks, Barbara

cc @RLovering @rachhuntley

[bmeldal]

Minutes are in this ticket above Ruth's comment from yesterday.

I have to disagree that your figure shows much annotatable data. The authors say:

"Frozen sections of human and mouse cortex were probed with antibodies to a neuronal marker, tubulin‐β3 (Figs 4a, d, e, f, i and j in green), and 7589 (Figs 4b, d, e, g, I and j in red) for confocal laser‐scanning microscopic imaging. The 7589‐immunoreactive cells were mainly co‐localized with tubulin‐β3‐positive neurons in human (Figs 4a–d, arrowheads) and mouse cortex (Figs 4e–h, arrowheads), and the immunoreactivity was diminished by pre‐incubation of 7589 with the antigen peptide (data not shown). These data indicated that TTBK1 is present in neurons."

NB: Ab7589 detects TTBK1 as shown in F3.

Tubulin-beta3 was used as marker for neuronal cell showing that TTBK1 was also present in such cells (As summarised by the authors, in bold above). Yes, there's overlap between the 2 proteins but that's because tubulin-beta3 is globally distributed in the cytoplasm while TTBK1 seems restricted to certain areas. So the overlap is coincidental. You can stain a cell with any 2 proteins and if one is globally distributed you will find some overlap but that doesn't say there's any functional relationship between the 2.

The other reason why I would like to see colocalizes_with being retired is that most user tools will strip the qualifier and then the analysis reads

"TTBK1 | part_of Tubulin complex | IDA"

which you state above is wrong. In fact, the figure doesn't provide details for the tubulin complex, even that's an assertion.

Screenshot from AmiGO:

[ValWood]

I agree... ~you could possibly annotate to "microtubule cytoskeleton"......but not tubulin complex.~

Let me go back to the qualifier. The "colocalizes_with" qualifier was NEVER intended for "coloclizes_with gene product" type experiments. It was really created for "colocalizes_with" subcellular structure type experiments. This is not clear in the documentation which does refer to complex (but during discussions the examples were for things like "spindle" and "ribosome" and other larger macromolecular structures. Over time the usage got subverted (mainly because the docs are not very clear).

We really want to get rid of the use os "colocalizes_with" for complexes. This would not be a big annotation loss. You should, if the annotation was warranted be able to replace by a complex binding, or a direct annotation to a cellular structure.

I eradicated the qualifier completely at PomBase with no loss of information..... This might not be possible but I think it is a worthy aim at least for the future. If it is only required for a handful of dubious annotations it really isn't worth the hassle it has caused for curators, end users, and software developers, or the amount of time we had spent discussing its use. We need to be pragmatic.

[BarbaraCzub]

Hi @bmeldal, I agree that it is very misleading, when user tools strip the qualifier. Thank you for bringing this to my attention. I've updated the annotation with a more appropriate GO term 'GO:0005875 microtubule associated complex', which does not require the qualifier.

Please note that Figure 4 has not been used in isolation to provide evidence for this annotation (because this is only the co-localisation), but it complements findings from Figures 5 and 6 and from Table 1 by providing cellular context.

I agree that any two proteins can often show signal overlap, even if there is no functional relationship between them. However, this is why we curate whole papers rather than individual figures.

[bmeldal]

Thanks, Barbara. I have to admit I only looked at the figure mentioned in the ticket and not the whole paper. If there are better evidences than of course they should be captured, the discussion here was if and how the colocalisation could be captured - and opinions are divided on that ;-) (they still were during my call-in to the GOC mtg yesterday).

[pgaudet]

Action point at the NYC 2018 meeting

[hdrabkin]

MGI done

[pgaudet]

From the NYC 2018 GOC meeting:

• ACTION ITEM: Review the definition of colocalizes_with in the RO; add new relations, if needed, to more accurately describe the biology.
• ACTION ITEM: Review current use of colocalizes_with; remove where the qualifier does not reflect ambiguity about localization. Discuss edge cases.

[bmeldal]

Timeline required. 6 or 12 months?

[pgaudet]

up to you @bmeldal

[RLovering]

Hi

I am not convinced that tools strip the co-localizes with and leave the GO term, I think they strip out the whole annotation row.

Also I have looked at the annotations the UCL team have made using co-localizes with, we have these associated with 27 different complexes: alpha9-beta1 integrin-vascular cell adhesion molecule-1 complex AP-2 adaptor complex calcium channel complex catenin complex Derlin-1 retrotranslocation complex ESCRT III complex flotillin complex HOPS complex Hrd1p ubiquitin ligase complex insulin receptor complex integrin alpha8-beta1 complex integrin complex inward rectifier potassium channel complex ionotropic glutamate receptor complex L-type voltage-gated calcium channel complex MHC class II protein complex mitochondrial outer membrane translocase complex mitochondrial respiratory chain complex I NMDA selective glutamate receptor complex oligosaccharyltransferase complex retromer complex ryanodine receptor complex Sec61 translocon complex SNARE complex vacuolar proton-transporting V-type ATPase complex VCP-NPL4-UFD1 AAA ATPase complex voltage-gated potassium channel complex

The problem is how big is a cellular component? a ribosome could be a cellular component or a protein complex. A lot of our annotations are to specific receptor or channel complexes. Could we have a term for the receptor complex and associated signaling and regulatory molecules? (and ditto for channels?).

Many proteins are not uniformly distributed along a plasma membrane but cluster around the complexes that they are involved in regulating. Therefore, this information can help confirm that a protein does have a role in regulating a process directly (or certainly within shouting distance) rather than being very downstream. The ingenuity tool is popular because it draws graphs that localise proteins within the cell, by removing these annotations, many of which will go because the interaction data to confirm complex binding annotations won't exist, we are removing manual data that is valuable and may be useful in the future for system biologists. So please could you explain what exactly is gained by removing this data.

Thanks

Ruth

[ValWood]

All of the slimming and enrichment tools I have used so far ignore this qualifier!-Including GOs!

How do you know that being in the same vicinity as a protein implys a role in regulating ? In fission yeast proteins cluster in the plasma membrane into different zones (based on membrane composition). So obviously, many things will cluster with protein complexes in growth zones and the like. Also for the papers we looked at, the colocalization seemed to be possibly coincidental.

In the list above many of the complexes are ER transport complexes, or the mitochondrial translocase, so if you use co-localizes with these you are most likely annotating cargoes?

Doesn't colocalizes with "mitochondrial respiratory chain complex I" just mean localizes to the "mitochondrial inner membrane"

If it is necessary to capture gene products being in the same membrane region, a better approach would be to annotate to specific membrane regions or sub compartments as these are becoming quite well-defined.

We shouldn't use colocalizes_with because this wasn't the intended use.

[ValWood]

The "mitochondrial respiratory chain complex I" are assembly factors for mitochondrial respiratory chain complex III assembly

and they have the assembly annotation and the appropriate mitochondrial annotation. They just don't need a co-localizes with annotation. It would be inconsistent, unless we did the same for every single assembly factor everywhere?

[ValWood]

So please could you explain what exactly is gained by removing this data.

annotation consistency, less confusion for users, more accurate analyses

[bmeldal]

If it is necessary to capture gene products being in the same membrane region, a better approach would be to annotate to specific membrane regions or sub compartments as these are becoming quite well-defined.

The annotation would then be "GP part_of CC" right?

Filter by GPs that are annotated with same CC and if CC is granular enough you get the potential overlaps. In IntAct (or more accurately, in IMEx) we only annotate colocalisations between 2 GPs if their physical interaction has also been shown by a proper PPI methods in the same paper.

[ValWood]

Probably the way forward here is to remove the annotation which are clearly misleading/not useful, like those associated with transport complexes etc. etc. After the clean-up individuals can summarize what they have left, and discuss how best to capture it.

Note that this qualifier was NEVER EVER intended to be used for "colocalization experiments" of this type. This is a result of bad documentation. It was intended to capture when a gene product clearly localized with some cellular structure (i.e membrane) but you did not know really know if it was part of it beause of the resolution (i.e it could be cell cortex).

I don't like the intended use, but it helped historically. The more recently acquired use for statistical overlap between gene products and complexes will (does) cause problems for consistency and QC.

If there is useful information we will need to find a way capture that distinctly from co-localizes with.

[selewis]

We should also get someone in Paul's group at USC to fix their enrichment tool to cull these qualified annotations out.

On Tue, Jun 12, 2018 at 10:10 AM, Val Wood notifications@github.com wrote:

Probably the way forward here is to remove the annotation which are clearly misleading/not useful, like those associated with transport complexes etc. etc. After the clean-up individuals can summarize what they have left, and discuss how best to capture it.

Note that this qualifier was NEVER EVER intended to be used for "colocalization experiments" of this type. This is a result of bad documentation. It was intended to capture when a gene product clearly localized with some cellular structure (i.e membrane) but you did not know really know if it was part of it beause of the resolution (i.e it could be cell cortex).

I don't like the intended use, but it helped historically. The more recently acquired use for statistical overlap between gene products and complexes will (does) cause problems for consistency and QC.

If there is useful information we will need to find a way capture that distinctly from co-localizes with.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/geneontology/go-annotation/issues/1940#issuecomment-396665100, or mute the thread https://github.com/notifications/unsubscribe-auth/ABcuEFvMA1aSJWzkqMMJUreQS16nKO0Qks5t7_YSgaJpZM4TgXWD .

[ValWood]

This might not be ideal either.

If used correctly  (i.e contributes to for emergent functions of complexes, like DNA polymerase,  proton-transporting ATP synthase activity, rotational mechanism etc), and colocalizes_with  for gene products which appear to localize to a sub-cellular structure, they can be useful/informative  in enrichment and slimming.

I still think the best way forward is  to tidy/re-evaluate  to get everything correctly annotated.  Ideally we would eventually phase out "colocalizes with" It would be better if tools only needed to  be concerned with relationships and not also with qualifiers. The qualifier should only qualify the annotation, not change it's meaning.

v

On 12/06/2018 19:31, Suzanna Lewis wrote:

We should also get someone in Paul's group at USC to fix their enrichment tool to cull these qualified annotations out.

On Tue, Jun 12, 2018 at 10:10 AM, Val Wood notifications@github.com wrote:

Probably the way forward here is to remove the annotation which are clearly misleading/not useful, like those associated with transport complexes etc. etc. After the clean-up individuals can summarize what they have left, and discuss how best to capture it.

Note that this qualifier was NEVER EVER intended to be used for "colocalization experiments" of this type. This is a result of bad documentation. It was intended to capture when a gene product clearly localized with some cellular structure (i.e membrane) but you did not know really know if it was part of it beause of the resolution (i.e it could be cell cortex).

I don't like the intended use, but it helped historically. The more recently acquired use for statistical overlap between gene products and complexes will (does) cause problems for consistency and QC.

If there is useful information we will need to find a way capture that distinctly from co-localizes with.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub

https://github.com/geneontology/go-annotation/issues/1940#issuecomment-396665100, or mute the thread

https://github.com/notifications/unsubscribe-auth/ABcuEFvMA1aSJWzkqMMJUreQS16nKO0Qks5t7_YSgaJpZM4TgXWD .

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/geneontology/go-annotation/issues/1940#issuecomment-396689745, or mute the thread https://github.com/notifications/unsubscribe-auth/AHBLKEqG8jEAiNPLOzm879IyXxODsqaNks5t8Aj4gaJpZM4TgXWD.

-- University of Cambridge PomBase http://www.pombase.org/ Cambridge Systems Biology Centre http://www.sysbiol.cam.ac.uk/Investigators/val-wood

[pgarmiri]

Hi, The UniProt annotations from EBI have been reviewed and the changes have been reported in the spreadsheets. I haven't done anything for the ISS annotations as they will be mirroring the changes of the experimental parent. The rest UniProt annotations are from SIB ( some have the label in the end). @sylvainpoux could you please take care of them? For PMID:16943429 I have requested a new complex term that is more suitable ( GO:0140261 BCOR complex ). Thanks, Penelope

[sylvainpoux]

I'm Sorry, but we honestly have no time and resources to handle this currently.

When curation guidelines that affect so many annotations are being changed, a mechanism should be found to automatically update entries. Reviewing hundreds of annotation is simply not sustainable.

Sylvain

[ValWood]

Maybe there are some ways we could reduce this list?

But at a quick glance many are quick to fix and should just be deleted.

For example

Transitional endoplasmic reticulum ATPase subunits annotated to FACT complex or ATP synthase subunit alpha, mitochondria annotated to COP9 signalosome.

Or if they are clearly part of the complex, like

26S proteasome non-ATPase regulatory subunit 9 | colocalizes_with | proteasome regulatory particle, base subcomplex

just delete the qualifier.

Somebody could probably just go through and delete the obvious anomalies in an hour or so, and this would maybe not leave many to revise? It seems quicker for a human to do this (because it requires a Y/N curator decision) than to think of a way to automate, because its a one off-fix, then future errors should not accumulate...

@pgaudet will we have a GO rule to prevent this?