Stream: genomics
Topic: Functional Annotation / Variant Consequence / Functional Eff
Patrick Werner (May 12 2020 at 16:06):
discussion on the component "functional annotation" vs Functional Annotation / Variant Consequence / Functional Effect
Patrick Werner (May 12 2020 at 16:07):
Original Comment by @Rachel Kutner :
For Mode of Inheritance, there's a LOINC code suggested in the link I originally included, but the LOINC's answer list isn't fully inclusive of the terms we're wanting to create. Would the best approach be to submit a request to update the LOINC code to include the full list of terms we're interested in? The list is here: https://ftp.ncbi.nlm.nih.gov/pub/GTR/standard_terms/Mode_of_inheritance.txt
Or is there some reason a new LOINC code might be warranted if we want to expand the scope?
For Functional Annotation, we have a few questions. Based on the description of the term in the link, it points to a SO code which indicates Structural_Variant, which seems incorrect. http://sequenceontology.org/browser/current_svn/term/SO:0001537
It possibly intended to reference the Functional Effect list here, which corresponds to our concept of "Functional Effect": http://sequenceontology.org/browser/current_svn/term/SO:0001536
If it did intend to reference the Structural_Variant SO, that is what we're considering "Variant Consequence" in our own data structure: https://m.ensembl.org/info/genome/variation/prediction/predicted_data.html
Which term is intended to be referenced by "Functional Annotation"?
Rachel Kutner (May 12 2020 at 21:52):
Hi all, I'm looking to see what we want to do as next steps on this.
We can summarize the feedback we get while developing our project (incorporating "Variant Consequence/Structural_Variant" and "Functional Annotation/Functional Effect" into our data model based) and share it with the group.
Are there other suggestions for how we could help move these decisions/clarifying these concepts forward?
Kevin Power (May 12 2020 at 23:05):
Summarized feedback is a great way to start the process.
Ultimately we need to develop specific proposals and log trackers. So we come up with small(ish) statements of work, like "We need to rename this component" or "We need to create a new ValueSet which constrains possible answers from SequenceOntology" or "We need to split up the Variant profile like ...." or "We need to update the LOINC code's description / answer list" or "We need to write up guidance/examples that articulate X,Y,Z" -- I will admit I was multi-tasking during the call today so perhaps I missed it, but I am not sure we have many (any?) specific proposals yet?
We can collaborate on developing those specific proposals here.
Rachel Kutner (May 28 2020 at 14:43):
My proposal would be to rename the current "Functional Annotation" component to "Variant Consequence" and create a new component for "Functional Effect" which corresponds to the Functional Effect list in SO.
Rachel Kutner (May 28 2020 at 14:50):
I submitted Jira trackers for all 3 change suggestions - expanding the answer list for Mode of Inheritance, updating Functional Annotation to Variant Consequence, and adding a component for Functional Effect.
Jamie Jones (Jun 01 2020 at 14:28):
Trackers look great Rachel, hoping to introduce them on call today and potentially vote on them next week
Kevin Power (Jun 01 2020 at 16:02):
Just to link it here, this is the document @Jamie Jones put together where we can try to collect more details about these topics:
https://docs.google.com/document/d/16pECfXFwsqrQDTcMqSCSQPax08v1pPDF8Pec1UWgoS0/edit
Liz Amos (Jun 04 2020 at 19:43):
As I was reading into the definitions and list of SO terms, I wonder what is the difference between the concept of molecular/variant consequence and what we already have with DNA Change Type (48019-4) and Amino Acid Change Type (48006-1)?
Arthur Hermann (Jun 04 2020 at 20:02):
Liz - would you please be kind enough to give an example? Sorry, this is out of my area of knowledge, so it would help to see an example...
Liz Amos (Jun 04 2020 at 21:41):
@Arthur Hermann I don't have a concrete answer or example, this is also out of my area of knowledge. I am asking what is the different from our concept DNA Change Type (https://loinc.org/48019-4/) vs what we are trying to define for Functional Annotation (http://build.fhir.org/ig/HL7/genomics-reporting/variant-definitions.html#Observation.component:functional-annotation)
Jamie Jones (Jun 12 2020 at 18:43):
Hi all, hoping to get our Lab ask here as focused as possible after folks have had a chance to review Clinvar's glossary and dictionary. It seems the real meat here is understanding labs' workflows for attaching SO terms to variants. To that end questions I'd pose are:
- Do they have one or more lists of SO terms that they pick from? Are other ontologies preferred?
- Do they follow clinvar's "Molecular Consequence" approach and separate out calculable protein product effects and location-based terms from other functional effects?
- Do they value annotating some functional terms as "predicted" vs "observed"?
What are others' thoughts here?
Jamie Jones (Jun 12 2020 at 19:47):
I'm not an expert, but it looks like every term under SO's "structural variant" (http://sequenceontology.org/browser/current_svn/term/SO:0001537) may be able to be calculated explicitly from the type and location of the variant. We should confirm this and then we could push for using the larger list as it aligns with Clinvar's "Molecular Consequence."
However, Clinvar's guidance on the "Functional Consequence" concept is very spotty, and we need to be careful there. @Liz Amos do you have any contacts related to that effort? Kevin and I uncovered some discrepancies... Additionally, their data dictionary lists several terms that are also under "structural variant" in SO, so the move to binding this concept to the sister SO term "functional_effect_variant" (http://sequenceontology.org/browser/current_svn/term/SO:0001536) should be deliberate, if made.
We may need to separate out consequences not based on where they are in SO but on how they were asserted. I'm looking to the IM work to help out here, and think it should be included in our "lab ask."
Jamie Jones (Jul 13 2020 at 17:42):
@Liz Amos found the v0 GA4GH variant annotation model spreadsheet with comments on molecular consequence and other related topics!
https://docs.google.com/spreadsheets/d/1zQU-Yv7gB7IHKIOVsTh-74BwdtgB9KQpKcWkSHZOa-Q/edit#gid=2097512597
They narrowed down the SO list to a handful of terms and left it extensible, rather than pointing to the full list of ~244 terms, though they haven't finalized that decision yet.
Of note, they have an experimental "Functional Impact Statement" listed as TODO, and a couple other similar concepts split out in the "Final Statement Model" tab. I encourage others to take a look.
Kevin Power (Jul 13 2020 at 18:30):
Good reference. So, it seems we are left with 4 options in this space (ordered I think from no work (1) to the most work(4) ):
-
Do nothing (doesn't seem right)
-
Make the proposed change from today:
add functional effect - binding: SO:functional_effect_variant
add variant consequence - binding: SO:structural_variant (ensuring we have AA change type covered)
remove AA change type -
Create new components for Molecular Consequence / Functional Consequence as closely to how ClinVar defines them today, likely still remove AA change type.
-
Pull in the pieces that make the most sense from the ga4gh_va_model in process work, acknowledging it is still a work in process, but hopefully we end up ahead of the game.
At this point, I sort of lean toward (3). I want to say (4), but not sure this is something HL7 should build upon - at least not yet? Having said that, it does seem like "Molecular Consequence Statement" is close to / exactly like ClinVar's Molecular Consequence - so that is good.
Kevin Power (Jul 14 2020 at 13:04):
@Rachel Kutner / @Patrick Werner have looked at this the most I think - curious what they have to say?
Arthur Hermann (Jul 14 2020 at 16:01):
@Tim Owen are you able to take the above information to any lab folks at Invitae to get their input? If the question isn't clear enough, perhaps the team could help clarify
Kevin Power (Jul 16 2020 at 02:05):
@Tim Owen (or anyone else in the lab world). Without getting bogged down in what we have done in our IG today (in case you simply can't figure it out :slight_smile:), I would ask the questions like this: Do you report data like what is found in the following three branches of SequenceOntology?
http://www.sequenceontology.org/browser/current_release/term/SO:0002072 (Sequence Comparison)
http://www.sequenceontology.org/browser/current_release/term/SO:0001537 (Structural Variant)
http://www.sequenceontology.org/browser/current_release/term/SO:0001536 (Functional Effect Variant)
If you do report things like those, what do you call the fields you use, and how do you define those fields?
PS - Don't worry too much about how SO defines those branches, but instead look at how terms they have.
Kevin Power (Jul 16 2020 at 02:09):
And we won't be surprised if the answer is "Yes, we have data like that, but we send it very differently" - that is OK. Any feedback we can get would be appreciated.
Tim Owen (Jul 16 2020 at 17:53):
@Kevin Power @Arthur Hermann Thanks! I will forward these questions to a couple of our lab directors and report back ASAP.
Rachel Kutner (Jul 17 2020 at 19:14):
@Kevin Power @Liz Amos I asked our Molecular Pathology advising committee about Functional Effect. The responses I got were that Functional Effect and Molecular Consequence are very different concepts and both need a landing zone in the spec.
They described the pipelines generating Functional Effect as "highly complex and context dependent", but they did say that the SO's Functional_Effect branch is the correct starting point to work off of and those values are clearly applicable to the concepts we're trying to represent - and the split in the SO makes sense.
I asked if the tools would output values like in the SO or a calculated score (like in the Functional Impact in GA4GH) and the response was "it depends on the discretion of the lab and the workflow in question, and the solution needs to have the ability to represent either, possibly through multiple fields available to represent Functional Effect in different ways, like they're trying to do in GA4GH with the Functional Impact & Pathogenic Mechanism fields". We pretty
Examples of different contexts seemed to be germline vs somatic testing, disease being evaluated, methods of evaluation. The values in the Functional_Effect_Variant in the SO include values that would be applicable in both Somatic and Germline contexts, and the field would need to be structured such that it was clear which context the value is being sent in.
They also stated clearly that these fields should both be Variant-level concepts.
Liz and I are working on getting an updated proposal to discuss and vote on at a future meeting - we'll update the JIRA trackers as well.
Kevin Power (Jul 17 2020 at 22:30):
Good stuff, thanks for the continued great info @Rachel Kutner ! Your points about "context dependent" makes me wonder a bit - do others agree with that?
As we know, we don't have a great way to really indicate relationships between the different components of an Observation. We already have quite the hodge-podge of components today, and it makes many of us uncomfortable. This is just another one on the pile I suppose, but we should consider that.
Of course, we could do a new profile of Observation for Functional Effect? We would have to ask how we make it distinct from our Implication profiles? Or is it distinct? We already have the "context" in our Implications. However, this effect doesn't quite feel the same as what we put into Implication today.
Having said all this, it is end of day on a Friday, so I will stop there and let others jump in. :slight_smile:
Rachel Kutner (Jul 17 2020 at 22:44):
I think adding it in the same location that "Functional Annotation" is documented here makes sense. Is this an outdated resource? I can never seem to find the right parts of the FHIR Specification when I google it. http://hl7.org/fhir/uv/genomics-reporting/variant.html
Note that in this case, if a Functional Effect is sent in, the Variant it is on can also have "Germline" or "Somatic" noted in a related component, which provides the needed context.
I also think he meant that what was being evaluated changes the values you would expect to return, how you evaluate it, and what tools you use. This was in reference to asking about the pipeline/annotation methods that result in the assignment of a particular value of Functional Effect/Significance. But the SO branch for Functional Effect has the terms that would be expected to be returned in various contexts, and that it is a highly curated expert set of values and that the organization in it makes sense to him as a pathologist.
He also supported this being extensible so it is able to adapt to evolution of the terms in the future.
Kevin Power (Jul 17 2020 at 23:25):
You found a good place. That is the STU1 release location.
image.png
We do have a component for germline or somatic, so I suppose if that is the only context we need, we can say we are covered (still need some good documentation). Seems like there might be cases that isn't enough and we need even a disease?
Tim Owen (Jul 21 2020 at 15:39):
@Kevin Power @Arthur Hermann Response from one of Invitae's lab directors:
We use SO terms that are generated through a program called SnpEff - you can see a table of their terms here: http://snpeff.sourceforge.net/SnpEff_manual.html We use these to inform our variant naming and evidence placement, but we do not report them out directly. In fact, many of them we group together or ignore entirely if they're too broad or otherwise irrelevant for the type of testing that we do. For the SO terms that we do use, they're back-end annotations in [our homegrown report writing applications] and are not included in our report data.
Kevin Power (Jul 21 2020 at 16:13):
Interesting, thanks for the response @Tim Owen . Is there anything remotely similar in your reports?
Arthur Hermann (Jul 21 2020 at 16:23):
@Kevin Power @Tim Owen - Kevin - just a note to remind you that Invitae does not do Somatic reporting.. only germline, just to make sure everyone is clear...
Kevin Power (Jul 21 2020 at 16:47):
@Arthur Hermann - Fair point. One of questions along the way was how specific was this to somatic versus being applicable to germline.
Bret H (Jul 27 2020 at 13:18):
Glad to see you @Tim Owen be sure to take a peek at the diagnostic and therapeutic implications too. @Kevin Power it has been suggested a number of times, to use a single component: 'variant-level annotation' and leave the value set to the sender. This would simplify the number of components but it makes it hard to identify the component. That being said, there is no reason someone cannot slice on component with a specific component.code Maybe we should be giving guidance on the component.codes to use for different situations as a guide. A FHIR server can handle components on its own, so the guidance to slice on component with a list of .codes would produce data that is queriable on a FHIR server. General LOINC Codes can be useful to provide meaning.
Bret H (Jul 27 2020 at 13:19):
that is my way of saying: slicing component means that there is already a possibility of a large 'grab-bag' of components. we've not committed a foul
Kevin Power (Jul 27 2020 at 14:02):
RE: one component - It seems there really are two concepts here, but not sure we have them quite nailed down yet. I hope we have a more concrete proposal soon that can be shared with the group.
Kevin Power (Aug 07 2020 at 21:50):
@Rachel Kutner / @Liz Amos - Any updates on this topic?
Liz Amos (Aug 11 2020 at 13:45):
@Kevin Power , I think @Rachel Kutner is still waiting on a response from Tempus.
Rachel Kutner (Aug 11 2020 at 13:50):
@Liz Amos @Kevin Power Correct. I've been sent to a few different people now, I'll check in again.
I've also been pretty caught up in a project the last week, so haven't had much time to follow up since I last heard from them.
Kevin Power (Aug 11 2020 at 13:58):
OK, thanks for the update.
Last updated: Apr 12 2022 at 19:14 UTC
