We're sorry but this page doesn't work properly without JavaScript enabled. Please enable it to continue.
Feedback

Open science/open research - How Do Knowledge Graphs Contribute to Understanding COVID-19 Related Treatments?

00:00

Formal Metadata

Title
Open science/open research - How Do Knowledge Graphs Contribute to Understanding COVID-19 Related Treatments?
Title of Series
Number of Parts
6
Author
License
CC Attribution 3.0 Germany:
You are free to use, adapt and copy, distribute and transmit the work or content in adapted or unchanged form for any legal purpose as long as the work is attributed to the author in the manner specified by the author or licensor.
Identifiers
Publisher
Release Date
Language

Content Metadata

Subject Area
Genre
Computer animation
Computer animation
Transcript: English(auto-generated)
So now let's see how we can see Knowledge Graph as a formal structures that integrate not only a scientific publications, but other type of data sources. So let's start here. So if we think in terms of the research questions that have been around COVID, we
will see that, okay, there are several friends and there are several results or statements that says, okay, the people who suffer certain comorbidities are more likely to actually
have fatalities or people that have certain age or certain diseases as cancer, but at the end, we try to understand what happened. What happened with the treatments? What happened with what have been published around the treatments and why?
Because at the end, we want to improve the patient treatment, the outcomes, prevent infections, and also equip healthcare system to respond more effectively to these type of events. So as all of us have discussed today, we have a great number of publications, but only
publications are presented in the way that they are human readable. So the amount of publications that are available is growing and growing and is almost impossible to address or to analyze all of them, okay? But also there are many scientific databases as for example, Drop Bank that presents information
about drops and information about the interactions between drops that this information is extremely relevant in order to understand the potential or in order to understand the side effects
that a particular treatment might have. So but again, in order to go and extract this information, a lot of manual work is required. What is our proposition and what we are doing is that we are transforming data into knowledge. And how we do this, because we have implemented this framework that receives data from different
data sources, and well, this is what we call scholarly data, and this data is collected then by using different test mining techniques, we extract a concepts and these concepts
are entities and relations. And this, of course, there are several tools that can be used, for example, in MetaMap or Semdrep or BioWork or existing mining tools as for example, Falcon, but the point
is that these entities and these relationships that are extracted are described based on ontologies. And these ontologies provide frameworks, formal frameworks, to describe not only the meaning about these entities and predicates that have been extracted, but also the relationship
between these concepts. Then we have more tools for extracting knowledge about these entities, and this is taken from the saurios, for example, UMLS or existing ontologies, and this is what is helping
us to integrate. So it's not only that we have publications or we have what have been extracted from scientific databases as in DropBank, but it's that we are seeing everything together. And when we are seeing everything together is to create the knowledge graph, and on top
of the knowledge graph is that different analytics can be performed. So, for example, the analytics that were described in the different panels today. So what is a knowledge graph? Formally, a knowledge graph is a data structure that represents a factual knowledge and represents factual knowledge in using a graph model and provides a formal specification, in this
case, of the biomedical knowledge, also taxonomies between entities, common understanding of the meaning of entities, and enables management, knowledge management and discovery. So we have entities that are the nodes, we have relationship between them, and what
is interesting is that in the knowledge graph, they are living together in data and metadata. So this is one of all these ontologies, UMLS ontologies in the biomedical domain, but there are several, and this is what we are transforming.
We are transforming human readable representation of scientific publications into a graph-based machine and human readable representation. So instead of having the human readable, now we have a graph, and this is done not
only for the publication, but also for what is in store in the scientific databases. And we are extracting not only keywords, we are extracting concepts and the associations between these concepts. So for example, in this publication, we know that these drugs are drugs that are related
to COVID, but that they have certain effects, and these effects are particularly the QT prolongation, and this is extracted from the publication and expressed based on the ontology.
The same here from the publication, we are extracting that these drugs are mentioned in this publication, but the relationship between these drugs as well. At the end, everything is together, why? Because we are representing these entities in terms of the same vocabulary, in terms
of the same ontology, and we are able to put what is in the ontology, what is extracted from the publications, what is extracted from the databases, and everything is together. Okay? And why this is there? Well, because we want to find patterns. So for example, we have these three drugs that have been used for, or have been
used for treating COVID, and we want to see what are the potential adverse events that may occur when a person with diabetes is treated. And here we have an explanation, right?
We have chloroquine, hydroxychloroquine, acetaminophen, and if we have a person that is taking metformorphine, well, that may have a problem. Why? Because there is an interaction between actually the drugs that we are having, so hydroxychloroquine with acetaminophen, these three drugs in combination with metformorphine
will bring adverse events. But we need to explain these patterns that are extracted from the knowledge that we have in the knowledge graph. And we are using the publications that are also presented there.
So based on that, we have knowledge in the knowledge graph that supports why these patterns are created. So in this case, we have the pattern, and because the publications that are also mentioning the interactions between the drugs, the side effects of these drugs,
and so on, can be used for having explainable patterns. So in numbers, what we have in our knowledge graph, we have built until now a knowledge graph that has this number of biomedical entities.
We have more than 50,000 publications. We are representing drugs. We are representing conditions, toxicity between drugs. And also, the knowledge graph is a link, a link to other knowledge graphs, so to DBPDR, to Bio2RDF, to drug pack.
At the end, what we can have on top of that, which type of analysis you can imagine in addition to discovering patterns. So you see all these drugs are drugs that are usually analyzed for COVID.
And these are publications. This is just running a simple query on top of our knowledge graph. So we can compare what happens on top of that. We can cluster the drugs based on the publications that actually are mentioned in the works or are mentioned in these semantic entities.
Or we can have what are the drugs, for example, that are mentioned in the publications that are associated with this set of entities. And in this case, clearly, what we see is that, okay, vitamin D or sulfasatonine
or aspirin or pretinol, they are drugs that are actually very well mentioned in all these and are describing these papers. Also, the disease is because we have the semantic type of all these papers that are
annotated in our knowledge graph. And so we are able... Sorry to interrupt, Maricela. We have 10 minutes, if you could. We are summarizing this so we can also cluster the drugs based on the interactions and infer new relationship within them.
So we have participated in this hackathon in the European Commission in April. And now we are going into a distraction of new interactions, okay? No interactions only between drugs, drugs, but also between other types of interactions.
And linking to the open research knowledge graph, because that is bringing more us also semantics, and implementing a hybrid approach that allows for not only validating based on the publications, but also in other types of studies. For example, in vitro, in vivo, and in C. I'm well-deceived, my group, and many things.