Genomic Associations Interface
Thursday, March 4th, 2021
At the intersection of AI and machine learning, Mastermind is recognized as a premier research tool within the genetic variant interpretation space. By identifying every genomic association within the existing literature, as well as drawing informative connections between them, Mastermind’s intuitive approach offers unmatched coverage, convenience, and efficiency to clinical investigators.
Although Pro users appreciate Mastermind’s quality of information, the quantity may sometimes leave you struggling to identify a good starting point. In response to this, we have released the Genomic Associations interface, exclusively for Mastermind Professional Edition.
This new capability elevates variant interpretation with a flexible user experience that not only allows your search to begin from any point, but also evolves with your criteria to uncover hidden information that could lead to a research discovery or solving a clinical case. As a result, Mastermind is more than just a database – it’s a virtual colleague.
In this webinar, Genomenon co-founders Dr. Mark Kiel and Steve Schwartz, are joined by Elizabeth Varga, Director of Customer Success, who introduced this exciting new functionality, as well as explored how Genomic Associations can best serve you and your variant interpretation goals.
We discussed how to:
- Focus and sort your concepts of interest through intuitive search and filtering options,
- Easily compare data on associated genes, variants, diseases, phenotypes, therapies, and CNVs, and
- Apply your new skills to a variety of clinical use cases.
WEBINAR TRANSCRIPT | Q&A
CANDACE: Hello, and thank you for joining us today for our Mastermind masterclass featuring our new genomic associations interface. I’m Candace Chapman, and I’m part of the team here at Genomenon. I’ll be moderating today’s webinar.
Before we get to introductions, let’s do some quick housekeeping — you can submit questions in the question box in your GoToWebinar control panel at any time, and we’ll have a Q&A session with the presenters after the presentation. Today’s masterclass is being recorded and will be posted on the webinars page of our website by the end of the day today. We’ll also email you with a link to the recording of this presentation, and we should have the Q&A posted on our website by early next week, but if you have any questions, reach out to us at any time on our website or at firstname.lastname@example.org.
Alright, let’s get to it! Once again, we’ve gathered our two amazing founders to discuss Mastermind: Dr. Mark Kiel, our chief science officer, and Steve Schwartz, our chief technology officer. I want to extend a big welcome to our director of customer success, Liz Varga, who will lead the Mastermind demo.
Steve, I’m going to let you kick things off and get back to our powerpoint. Take it away!
STEVE: Awesome, thank you, Candace! So this is a masterclass on our genomic associations interface powered by genomics language processing. Mastermind indexes the genomic associations that are supported in the primary evidence of the medical literature. It does that by identifying every gene, variant, CNV, disease, phenotype, therapy and categorical keyword in the medical literature in every way that an author can describe them. It then analyzes this information and draws the associations between the different genomic concepts, and then makes that available in the user interface in order to find the information that you’re looking for based on the medical literature. So we’re now making this information easier to find through our new genomics associations interface, which is on the next slide.
This is a preview of what you’ll be seeing, of what Liz will be showing you how to use, this new genomics associations interface. One of the things that you’ll notice in yellow is that we’ve separated the inputs and the outputs of your searches of the genomic information, again, supported by the medical literature. What I mean by that is you can search by any combination of genes, variants, diseases, CNVs, therapies, phenotypes and keywords in the upper left in the search bar, and then on the associations page, you can then view all of a given association type that are related to your search criteria. So, for example, you can search by disease and see related diseases, which what you’re seeing in this screenshot here. For each related disease in the lower left here, you can see the number of other associations that correlate with that disease on that row along with the disease that you’ve searched for in the search bar. If you select different tabs, which is what you see highlighted in red, you can view other association types that correlate or correspond with your search criteria.
So instead of seeing a list of diseases, you can see a list of genes, variants, phenotypes, therapies or CNVs in this case, and then on the right side, again in yellow, you can add any of those associations to your search criteria to further refine the results. You’ll start to see how useful this becomes in the use cases that Liz will demonstrate, in how you can not just search for things that you already suspect may exist, but start to explore the data and iteratively build up your search criteria. When you’re ready to see the evidence that supports those associations, you can click what’s highlighted in the orange box here, under “View Evidence.” You can view the articles that support the associations that you’re seeing and go straight to the information in the literature to see what they say about those correlations or associations.
Having said that, I will now turn it over to my co-founder and our chief science officer, Mark Kiel.
MARK: Thanks, Steve, and welcome everybody! Thanks for joining today’s masterclass. It’s always fun to be on these masterclasses. They’re fairly informal. We like to have a lot of interaction, and so I don’t have much of a formal presentation to give, no slides or anything, but I do want to emphasize that Steve and Liz and I will be available for any questions that you have as we’re introducing this new powerful capability that is now putting in your hands some of the internal capabilities that we’ve had that we would apply toward doing some of the discovery work that we do with custom project development. We’re now making that available to our professional users.
The reason this is important and the context that I want to set for you is that, up until this point, Mastermind has predominantly been most useful and most applicable to situations where you know what you’re looking for. Most of you probably use Mastermind when you have a variant in your workflow queue for which you need to find the evidence supporting your variant interpretation process. Obviously, that’s a very powerful capability, but there are certain situations — particularly when you’re using a larger panel, a new panel that you’re less familiar with — where you don’t necessarily know all of the context or the nuanced functional information for any gene that comprises that panel. You know as well as we do that the trend toward larger and larger panels culminating ultimately in whole-exome and even whole-genome sequencing is making this situation where you’re not exactly sure what information to look for, or what the nature of this gene is, or even what disease you’re trying to diagnose. In those situations, the Mastermind association page is now applicable for more discovery work.
Liz will be going through some representative searches to apply to different situational use cases, where you may wish to learn new things about the diseases, the phenotypes, the therapies or the other genes or genetic variants that are associated with a given search. She’ll walk you through some of those combinatorial searches, and she’ll showcase how to apply those searches, and what Steve demonstrated a little bit earlier with respect to the association page made possible through the GLP capability that Genomenon has. She’ll show you how to navigate those search results to maximal effect in those different circumstances. So I will be back for the Q&A. We usually go round robin, but I invite you as Liz is walking through the demo to submit your questions. I’d be happy to answer anything, and at great length, to to give you a sense for how this association information came to be, how to put it to use, and what the future of this kind of capability looks like in Mastermind.
So with all that said, the context that I wanted to set, let me introduce our director of customer success, Liz Varga, who’s going to walk you through a demo that she has drawn up here. I’ll see you all later on the Q&A session. Liz, if you want to take it away.
LIZ: Yeah, thanks so much, Mark! So I have the opportunity to show you some of the demonstrations that we use to showcase some of the associations features. In my time working with customers, one of the common questions that we would get was the desire to see a certain gene list or variant list that might be associated with a disease or phenotype. Given that common use case, I wanted to start there and show you an example. Just like you’re used to with your Mastermind searches you’ve done in the past, you’ll see our general user interface here. In this case, I’m going to start with a phenotype search. I will enter craniosynostosis as my phenotype and I will select that and search. This time, because I didn’t start my search with a gene, variant, or CNV at the start, that’s automatically recognized by Mastermind as something we want to explore an association with.
Immediately, when I select my search, you’ll see our associations interface page. It starts out with our diseases tab so that you can see other associated diseases, but you have the opportunity now to toggle between these different variables. So if I want to explore genes, I will select the genes tab, and at this point, I will get a numeric list of the genes that are associated in the literature with craniosynostosis. You can see that there are a number of gene mentions, and they are organized by numeric order, starting with those that have the most associated articles, in this case, TWIST1. You’ll immediately recognize that these are common genes that you would see on curated gene panels as associated with craniosynostosis. That is a common use case, and if I would like to now explore any individual gene and its association with the disease, I can simply click the “Add to search” button and then select “View evidence,” and that will take me to view the associated literature between TWIST1 and our phenotype, craniosynostosis. The interface will then load our associated articles (it will still organize them by relevance) and you can also apply different filtering using our filter categories as needed.
That is a very basic use case. The next one that I wanted to show you was a little bit more complex, which is the really nice addition with this association because you can layer your searches, as was previously mentioned. One thing that I was asked a lot when I was in clinical practice as a genetic counselor — I worked in pediatric hematology/oncology — often I would have a physician that would come to me after going to a conference where they would learn something new about a therapy, particularly for underlying genetic disease, and maybe some genetic associations with the impact of that therapy in the context of the disease. The example we’re starting with here will be based on the therapy, in this case, Vincristine, a chemotherapeutic agent.
I’ll select my therapy, and then, in this case, I want to explore any association between Charcot-Marie-Tooth disease. I will add that as a disease, select, and start my search there. Now in this case, I will see the other diseases mentioned in the literature as related to Vincristine and Charcot-Marie-Tooth. If I was being asked by my clinician to think about diseases that also might be associated with Vincristine toxicity in the context of Charcot-Marie-Tooth, what I would see is literature at the top related to the PMP22 gene, which in fact is the most common gene associated with Vincristine toxicity and Charcot-Marie-Tooth. If I want to then explore this a bit more, I can add that to the search. If I want to now explore specific variants that I might be interested in related to that search, I’ll go to the variants tab. Now these will be organized by the number of articles that connect these three parameters together. I’m immediately taken to this top variant, and if I want to see the two articles that relate to that variant, I can just select on the number two, immediately go through those articles, and just as I can always do, I can now see the sentence fragments and explore the information in more detail.
The last use case that I wanted to demonstrate to you is actually one in which you would start with a phenotype you already know, a certain genetic variant that’s present in a patient, and you might want to explore therapies that could be relevant in that context. The example I will use is a cardiac condition called Arrhythmogenic Right Ventricular Dysplasia. In this case, my patient has a specific variant in the DSP gene, so I will add DSP. I will enter my variant, in this case, by protein position, select, and then when I perform that search, because this could be a standard Mastermind search that you could do even before the associations page was in existence, this time you are brought to your typical user interface for Mastermind.
If I want to explore the association from here, I’ll click the “Explore Associations” button in the corner. Now, I will again be brought to the associations page, and I can explore further genes, variants, diseases, phenotypes, therapies and CNVs. Since my question of interest relates to therapies, I will select that tab. So now I’m brought to a list of therapies that are mentioned within the same articles that contain my three parameters of interest. This allows me in a more exploratory fashion such as research, where I’m not really sure of what I’m looking for at the outset, now I can peruse these different therapies and decide what to explore from there. So again, at any point, if there’s a therapy that I want to add to my search, I will just click the “Add to search,” hit “View evidence,” and then you can see that the therapy is added here at the top. Now, I will be able to load my articles of interest, again, sorted by relevance, and then when I’m exploring all of the associations in the sentence fragments, I will select “All matches,” and if I want to see an even a larger view so I have more real estate, I can apply Focus View, and now I can look at all of the different associations where they are mentioned within the article and quickly toggle through my different examples.
So these are some of the main capabilities of the associations page. I walked through those three examples, and the key points that I just wanted to summarize for all of you, your take-homes for today and what you can do with associations: You can start from any parameter that we cover in Mastermind when you’re exploring an association. You can add and remove search criteria. The best way to add if you don’t know going into your search what exactly you’re looking for will be to just click the plus button, add it, and then at any point, you can view evidence by either clicking this button or the one in the upper right corner. You can toggle between the associations tab and the evidence. You can also sort things on the associations page in different ways. One thing that users have asked for routinely is to see a list of phenotypes organized alphabetically, and so that was something that we took into account when we created this page. Within any of the categories, you can simply select this top button and your information, while it starts out organized by those with the most article associations, it can be reorganized alphabetically. The last thing I just wanted to mention was that these list headings up here display for you right at the outset how many gene associations have been identified with how many different articles, so those numbers can immediately give you some helpful insights.
Those were the main use cases I had. We wanted this presentation to allow for a lot of interactivity, so that you could ask very specific questions of us today. At this point I will move over to any questions that you all have.
CANDACE: Thank you so much, Liz! That’s a great demonstration, and even though there’s just three examples, I think people can see how how useful this is for so many different use cases. One question that came through: Can you use free text search in this interface?
LIZ: Yes! That is a fantastic question, and I’m glad that you asked. That is certainly one that I have tried to demonstrate for users. Let me give you an example for primary immunodeficiencies, where I will look for associations. So the great thing about our associations page is that you still have all the capabilities that you had previously; nothing is missing. As I mentioned, you can add filter categories, so if you wanted to see certain article types, that can be applied, but in this case, let’s say I’m exploring the genes associated with immunodeficiency. I have this listing here of the genes of interest, but then I want to refine it with a text field. One example might be that you have an interest in a certain population, so in this case, the Amish, people are predisposed in this population to develop immunodeficiencies. I can add that as a text search here, just reapply my search, and in that case, now the gene list is going to be reorganized, taking into account my text field. Sure enough you will see here that this gene is the most common gene that is mutated, in the case of immunodeficiencies in the Amish population.
CANDACE: Awesome, thank you. Another thought: How could the associations feature be used, or could it be used, to find gene fusion events?
LIZ: In that case, we still have the capability to do Boolean search. So if you wanted to start out with ETV6, you can see automatically here a fusion between ETV6 and NTRK3 pops up, which is a very common fusion. You could just select on this fusion as an example to start your search, but in the more standard situation where I want to perform a fusion search, what I will do is I’ll enter ETV6 and then NTRK3. One new capability on our interface is that, as you can see, it’s saying ‘Select to replace, or Shift+select to add gene to current search.’ What I will do for a fusion is hold down my shift bar, I’ll select, and now I have both of my genes within the search. I’ll apply it, and then to look for fusions, one thing that is helpful in that context is to go to “Filter Categories” to “Genetic Mechanism.” There’s a fusion events tab here, and so what you can do if you’re interested in fusions is apply any of these terms related to fusion and submit with the chosen filter.
Before the association options, this is how I would normally perform a search where I’m interested in exploring the ETV6-NTRK3 literature, but now all I have to do to further investigate is select the “Explore Associations” tab. Let’s say I want to look at the diseases that have been reported in association with this fusion. Now I have a list, and one that I’m interested in is sarcoma, for example. I can add that to my search very easily, and then just view the evidence there. You’ll see that my filters are still applied, so it’s still recognizing that I’m interested in the fusion data, as is related now also to sarcoma. It’s just layering that search. I can keep adding different filters, exploring more and more associations just by toggling back and forth until I’ve really been able to investigate it at a level where it’s been narrowed down enough that I feel confident in my search.
One thing I didn’t demonstrate was the CNV association search. I really do love the CNV associations as well. In this case I’m looking at fusions, I’m looking at sarcoma, but I can also see co-occurring CNVs that might be in the same literature. We know that often times in somatic cancer or cancer mutation landscapes, there will be multiple different types of genetic factors or alterations that are coming into play within that cancer. With the associations page, now you can immediately see what the top associations are within the literature as relating to this context or this scenario, and again, choose now to investigate that. So if I also want to see the intersection of literature with my fusion sarcoma and this MDM2 amplification, I can simply select on the 40 here, and that will now take me to the 40 articles that combine that constellation. Then, again, everything is ranked by relevancy and I can explore it within my sentence fragments, just as I always have.
CANDACE: Wonderful, this is so awesome, Liz. I’d like to invite Steve and Mark to join us again, and perhaps they have some thoughts about the questions that were asked or the demo that they can that they can share.
STEVE: Sure, yeah.
MARK: Thank you, Liz, for that answer to the question about fusions. One interesting thing to showcase: if you delete the CNV that you’ve searched on and delete the ETV6 gene, when you go back to the association page, suppose you don’t know what fusion partner the NTRK3 gene is associated with in the context of sarcoma. If you go to the associations page and you go to the genes page, that will show the additional genes that are associated with mentions of NTRK3 and sarcoma with the fusion keywords. If I’m not mistaken, one of the first things that should come up is ETV6, which is the known fusion that Liz has searched on.
As I suggested in my preamble, this associations page is a powerful way for you to perform some discovery work. So in the context of, say, designing a new panel, this would be a way to apply Mastermind and all the richness of the associations coming from the genomic language processing for you to design that panel based on these fusion genes that more routinely occur. Obviously, these genes are sorted in descending order, so ETV6 is quite likely to be the most prevalent and/or the most important for you to design your fusion panel around. The same thing would be true for other features, like the variant list or the gene list, if you’re not looking specifically for fusions, but rather looking for which genes are recurrently mutated. If you perform the search on the disease and go to the variant tab, that will showcase the genes and their various mutations that are most recurrently associated with that disease. In trying to get the audience to think about circumstances in which the associations page would be most applicable, I’ll say that those are the kind of contexts in which there’s the most power behind this new capability.
STEVE: Something else that I’ll point out that is often useful in these types of use cases or scenarios is the relevance and the value of some of the other counts that are in the middle columns here. One of the things that’s interesting when you’re looking at this data, we have sarcoma and we have NTRK3, and we’ve got the fusion event categories and keywords enabled, so everything that you see here is filtered by those search criteria. When we’re looking at the gene candidates for the fusion event in the left column here, of course we see ETV6 at the very top, but one strategy that’s also interesting is looking the prevalence of other associations with each of these gene candidates. As you look down the list, you start to see some interesting patterns, for example, there are more papers that cite NTRK3 together with NTRK1 in the context of sarcoma and our category keywords. There are more papers that co-cite NTRK1 than co-citing with the RET gene or TP53, but when you look at TP53, within those 187 articles that co-cite sarcoma, NTRK3, our categories and TP53, there are other genes discussed within those articles. Without even knowing that it’s TP53, that would tell us that the other gene has more evidence and more discussion in how it relates or how it co-occurs with publications about other genes, which makes sense for anyone who knows TP53.
You can also see, for example, how the number of CNVs for each association decreases with the number of articles that talk about those associations, until you get to RET and TP53, and then the number of co-occurring CNVs jumps. Without really discussing what that means for this particular data, that’s one of the other things that I’m often looking at in the associations page: with an eye toward what information I’m looking for, once I identify a fusion partner, am I looking for a therapy? Am I looking for how those partners may relate to CNVs or other phenotypes or diseases? That’s where some of these patterns in the count data in the middle can start to become very valuable, which is why it’s also useful to be able to sort your results by each of those count columns. It depends on your use case and what you’re looking for as to what patterns you’re seeing actually become interesting for your use case.
CANDACE: Awesome. Alright, we have a new question, a very good clarifying question: You show the number of articles in the column on the left. What do the other columns show, and how do I use them?
STEVE: Absolutely, yeah, that’s a great question. It kind of plays along with what I was describing. I’ll precisely explain what those are. The first column, the articles column, that is the number of papers or articles which describe your search criteria altogether, along with that item in the list. For example, that 378 on the second row there means there are 378 articles that cite sarcoma, NTRK3, and all of my selected category keywords, all of those together with ETV6. So that is what that count is. The remaining counts are, within those 378 articles, how many other genes they discuss, how many variants they discuss, and how many diseases, phenotypes, therapies and CNVs those 378 articles discuss.
CANDACE: Alright, that sounds good. I’m going to turn this one to Mark and give you a break: Can we apply the same steps as infusion for searching digenic inheritance?
MARK: Yes, absolutely. If I’m appropriately understanding what you mean by digenic inheritance — I’m presuming that you mean there’s two mutations that are causative or associated with disease causation in two different genes — the way to get to that type of answer would be to invoke the Boolean search capability that Liz showcased for fusions, gene one and gene two. Then you’d be able to apply category keywords, such as “heterozygosity,” say, or looking for clinical cases in which you might have seen those two genes described. The association page, with those two genes and those appropriate categories, would then allow you to see the phenotypes that were associated with those articles, as well as the genetic variants, which you might see from among your patient variant list. Alternatively, you can directly search for the variants that you’re seeing in your patients to identify if there’s ever been a paper that specifically called out those two variants in the context of the disease that you’re suspecting is caused by those mutations.
The principles apply in the very same way, and that question actually allows me to reinforce the idea that, when you have a very broad search, your answers are going to be more broad, less specific or less well-defined. If you, for instance, just have one gene in your search, you’re going to have the whole universe of content resulting in your search, whereas if, in subsequent searches, you decorate your search with additional terms, you will enhance the specificity of your search result and thereby enhance the value and importance of your search results, but there’s always a give-and-take with sensitivity and specificity. When you’re in discovery mode, you can use the power and the comprehensiveness of our database and all that has been indexed to really start to drill in on the most prevalent and the most important features to learn, but as soon as you’ve learned what you need to learn by taking that specific approach, then you can start to back away and tailor your search more toward the clinical case that you’re organizing your questions around. So again, the associations page is largely meant to be an exploratory capability, so there should be an expectation that you would iterate a bit as you learn and as you add more terms to your search and learn more and more about which areas you want to be more specific on in your search terms.
CANDACE: Awesome, thanks! I just want to take a moment and thank the people that have shown their excitement in the comments. We think this is a pretty awesome new way to look at things, and what I’m hoping is that we will hear stories about how you can use this in your clinical practice. Do you all have thoughts, or have you heard from customers about how they plan to use this in their practice?
MARK: I’ll start. Liz and Steve, I’ll invite you to jump in, but I’ll say personally, in the work that I do, supervising projects for pharmaceutical clients and biotech, this kind of capability is usually a preamble to the deep curation that we do across entire diseases or for whole gene pathways. This type of capability, as I mentioned in my introduction, was up until this point just an internal capability. We’re releasing some of the power and capability now to the hands of Mastermind professional users to apply to their clinical workflows in those circumstances that I that I described at the beginning: panel design, dealing with the novelty of a larger gene panel for which you may not have a pre-existing database, or for which you may not have as much professional familiarity with the disease or the gene, the idiosyncrasies of the transcripts or legacy nomenclature. With the associations capability, this is a situation where you can learn more quickly about those different new circumstances so that you can build out your database faster or streamline your workflow much faster, or in the event that you’re looking at, say, exome sequencing results, get to the information that you need as quickly as possible, even in the absence of the need to build out a database.
CANDACE: Awesome. Liz, do you have anything to add?
LIZ: Well, I just wanted to speak more to the clinical perspective. Certainly, there’s a lot of value in the laboratory, but from a research and clinician perspective, there is also value in the fact that you’ll have unexpected or rare associations that you may have a very difficult time navigating in the literature through traditional means. One thing I found this extremely helpful for was in exploring those rare phenotypic associations with a specific gene and variant, where I just wasn’t sure if it was present, or thinking about a context where something comes up that’s unexpected in a personal or family history, now you can go and investigate that to answer a question. The power of associations to me is really the ability to explore any combination of variables. I think it’s great for testing different hypotheses, for example.
STEVE: I would say, just generally too, I think Mastermind has always been really good at answering the question from a clinical perspective, “has this been seen before?” Generally, what the associations interface allows is moving beyond that. The question is now, “if this has been seen before, what exactly has been said about it? What kind of evidence is there?” It’s moving beyond the original question. Going from “has this been seen before” to “what therapies have been described that may be able to treat this, or that this may be resistant to?” Starting to answer questions like that, whichI think would probably sum up the general use case.
CANDACE: Thanks. I also want to point out that this capability is available in our advanced API, so you can do this kind of work in your home. There’s a lot of power there, and I’m just really proud of the team for what they’ve put together here. I hope to hear stories about how you’re using it. I would love to invite you to contact us. We are always here, hello@Genomenon.com. We’d love to hear more feedback, we’d love for you to share this with others in your organization, colleagues, we want your work to be easier. I’d love to thank Mark, Steve, and Liz, all of you for joining us for the masterclass! This has been great. Keep an eye out for an email from us with a recording of today’s event. That’s it! Have a great day. Keep the questions coming. Bye-bye.