Rare diseases affect around 6-7% of the population in the developed world (defined as fewer than 1 in 2,000 in Europe and fewer than 200,000 individuals in the US).
By their very definition of serving a relatively small population, brand new drugs for this audience (or orphan drugs) can be prohibitively expensive—yet legislation in the U.S. (FDA Orphan Drug Act, 1983), Japan, Australia, and Europe incentivizes treatment development.
So what’s a pharmaceutical company to do? Is there a more cost-effective way to reach cures faster?
On the surface, drug repurposing promises much–known safety profiles of existing drugs, a reduced development timeline, and, as a result, a significantly reduced cost to market (we’re talking about bringing expenditure down from billions of dollars to millions of dollars here).
There’s still much research, which is time-consuming and resource-heavy. This is why drug companies are currently focusing on automated literature analysis.
Let’s look at the example of Arteriovenous Malformation (AVM), which has been in the news recently in the UK. It’s a condition that affects hundreds of thousands of people worldwide, causing abnormalities in blood vessels. These abnormalities can result in dangerous complications and disfigurements. Now, researchers have identified drugs that could target the underlying cause of the condition.
Take a look at this diagram, which simplifies the repurposing pipeline from this piece of research.
Once these genes were identified, the next step in this particular repurposing study was to screen for drugs that targeted the relevant proteins. In this case, several candidate drugs were already used in cancer therapy.
Here, the disease in question has been taken as the starting point, and faulty genes have been identified on the RAS/MAPK pathway, which controls cell growth.
In this case, we see that treating AVM-BRAF mutant zebrafish with the BRAF inhibitor Vemurafinib restored blood flow in the AVM.
Drug repurposing relies on making connections, but as mentioned earlier, this is not easy when you’re faced with millions of documents, all with unstructured text.
Wouldn’t it be helpful if a computer could recognize key scientific information in unstructured text, such as scientific papers? Of course, the answer is yes, but one of the main hurdles with this approach is getting the computer to do this quickly while being able to process scientific synonyms and ambiguity.
Building on this is semantic search. A tool that allows a researcher to find relevant information about their target. In this case, we’re looking for drugs that inhibit BRAF. As you can see, the search tool also picks up synonyms, ensuring you don’t miss out on potentially valuable data. Contrast this with a conventional search engine, where if you search for “drug,” you’ll get results that mention the word “drug.” However, with a semantically enriched search engine, the computer knows that this means anything that is defined as a drug.
Building on this is semantic search. A tool that allows a researcher to find relevant information about their target. In this case, we’re looking for drugs that inhibit BRAF. As you can see, the search tool also picks up synonyms, ensuring you don’t miss out on potentially valuable data. Contrast this with a conventional search engine, where if you search for “drug,” you’ll get results that mention the word “drug.” However, with a semantically enriched search engine, the computer knows that this means anything that is defined as a drug.
The results go beyond highlighting individual entities, allowing you to extract information about relationships between entities, such as gene-phenotype or drug-target. Extrapolate this over 28 million Medline abstracts, and you have an incredibly powerful tool.
These relationships can then be built into networks, providing a computer-readable framework for searching the data and making new connections.
Labeling the entities in the text with unique identifiers allows you to take this a step further and map to other data systems, connecting related diseases, adverse events, pathways, and drug labels.
Of course, this method can be turned on its head to discover new examples; you could compare diseases based on their phenotype profiles once you know that two diseases are strongly related if there’s a drug that information. If you treat one of these conditions, you can hypothesize that you have a potential repurposing candidate on your hands for the other condition.
If you’d like to know more about semantic analytics in drug repurposing, our whitepaper explores this theme in much more detail.
If you’d like to discuss how the SciBite Platform can transform your data, we’d love to hear from you at [email protected].
Richard is a seasoned marketing professional with over two decades of experience in the information services and life sciences sectors. Currently, he is the Senior Manager, Portfolio Marketing at Elsevier’s SciBite, where he drives strategic campaigns and harnesses data-driven strategies to amplify the platform’s online visibility and impact.