Home PublicationsData Innovators 5 Q’s for Yiannis Kiachopoulos, CEO of Causaly

5 Q’s for Yiannis Kiachopoulos, CEO of Causaly

by Eline Chivot

The Center for Data Innovation spoke with Yiannis Kiachopoulos, co-founder and chief executive officer of Causaly, a research platform based in London that uses AI to mine and analyze scientific biomedical information. Kiachopoulos discussed how Causaly’s AI and natural language processing supports scientists by accelerating the research process, thereby enabling rapid innovation and progress in healthcare.

Eline Chivot: What is your background, and what led you to found Causaly?

Yiannis Kiachopoulos: My co-founder Artur and I are both computer scientists and avid book readers. Although reading by itself is enjoyable, when we wanted to find specific information, we found ourselves reading for many hours just to find small but important nuggets of knowledge. In these situations we wanted to find a way to extract that knowledge faster, capture its essence, and document the main ideas of an article. A tool that helps users to connect the dots between multiple disciplines and find evidence faster.

We started to explore the possibility of teaching a computer to read at superhuman speed, and extract cause-effect evidence that had been encapsulated in text by the author. To test our hypothesis, we developed a prototype which we ran on a few thousand Wikipedia articles and developed our first plan to design a machine-reading platform.

After we validated the technological feasibility, we researched possible applications and decided to focus on the biomedical domain because of its importance in human health and the big challenge of coping with millions of biomedical publications. Furthermore, in the domain of natural language processing (NLP), there had already been decades-long efforts to make biomedical domain knowledge more accessible and structured through the development of expert ontologies, wide scale integration, and standardization. We set out to machine-read all the knowledge that exists in biomedicine to create a cause and effect network, a representation of how the biomedical world works based on the written word.

To validate the evidence in our platform, we started very early on to work with experts in their fields such as the pharmaceutical company Novartis in Switzerland, working on different use cases. This played a pivotal role in getting us where we are today.

Chivot: How does using AI change the process of scientific discovery, and what are the benefits they provide to scientists and researchers?

Kiachopoulos: While AI can be used to solve a variety of problems, at Causaly we aim to solve how humans can find evidence quickly from millions of documents. There are about 30 million biomedical publications, and if you are trying to find out, for example, which protein is related to which side effects, the evidence is difficult and very time consuming to find.

Our AI reads the text and flags which substance or proteins are related to which side effects, and highlights the documents with the evidence. We are solving a search problem to find relevant evidence that can further be used for creating hypotheses, and lead to better decisions and outcomes. For now, our focus is on the development of machine-reading technology and innovative ways to visualize and present evidence for our users. This same evidence can be used to make valuable predictions such as the network analysis we did for repurposing drugs for COVID-19.

Chivot: Causaly uses NLP to capture relationships between disorders, chemicals, drugs, and genes, to discover new molecules, and to validate hypotheses. Could you give an example of how this works? What is the advantage compared to traditional testing methodologies, and how do you ensure accuracy of the results?

Kiachopoulos: Say you wanted to find all possible side effects for a drug. Finding this information would require you to read hundreds, if not thousands, of documents. With Causaly it only takes a few seconds, and you find all side effects that have ever been documented. In addition, our system can also infer possible side-effects by looking at connected relationships in our knowledge graph. If a drug affects a particular protein which is also known to be associated with a side effect, scientists can quickly find this knowledge and use it to develop novel hypotheses.

This is important in preclinical drug safety or when a drug is introduced in the market, to ensure patient safety and comply with regulatory requirements. With Causaly, you can answer even more difficult questions, such as “how does obesity affect hormones that cause breast cancer?”

Finding this evidence is very time intensive, and involves looking at the entire document corpus of obesity and breast cancer which spans hundreds of thousands of documents. Singling out hormones which could potentially connect both diseases is like searching for a needle in the haystack. Causaly, in just a few seconds, finds the evidence and answers a question that would otherwise take researchers weeks and even months of work.

There are different areas where this applies—drug safety and drug discovery as mentioned, but also for commercial functions: What are the drugs or therapeutic procedures that can treat a disease? Users can get a bird’s eye view on the treatment landscape, and understand entire disease domains quickly and in a visual format.

I had already mentioned that in the biomedical domain there have been many efforts over the last decade to structure information in taxonomies and ontologies. There are several well-known ontologies, such as those maintained by the National Library of Medicine in the United States, which for example classifies various types of cancers. These expert-developed ontologies are used to organize the knowledge in our system, so that our users can find what they need.

There are two main ways in which we ensure accuracy. We test our algorithm against abstracts that have already been read and annotated by scientists. Using this “gold standard,” we can assess the extent to which the AI finds what humans find. This process helps us to identify how to improve our system over time. The second way is that in our platform, our users can always verify whether a statement is correct themselves. All evidence is linked back into the primary publications, thereby traceable and verifiable.

Chivot: Causaly was founded in 2017. What are your accomplishments so far? What have been some of the biggest obstacles to developing your capabilities?

Kiachopoulos: Let’s start with accomplishments. As co-founders, Artur and I are proud to have brought together a stellar synergistic team of passionate people. We are also fortunate to be working with some of the world’s largest and most prestigious pharmaceutical companies and research institutions such as our partnership with University College London (UCL) to help accelerate their research into COVID-19.

We had some great moments when we received user feedback on how much Causaly helps them in their daily jobs, not just in pharma or epidemiology, but also in clinical care and clinical research.

In terms of development obstacles we have overcome, it is always challenging not just to develop technology, but to put oneself in the place of your customer, and look at real problems. It is important to try to understand their workflow and provide a solution that is helpful, rather than just building a tool. The relationship with our customers is like a partnership where we listen carefully to what their challenges are and quickly develop solutions together. Our scientists have almost daily discussions with our users, and are available to help whenever they are needed.

Chivot: Your technology could be used in multiple industry sectors. Which other areas could it move into, and how could it be applied in practice? 

Kiachopoulos: The information found in biomedical publications is important for all areas that touch upon human health. Primarily this applies to pharmaceutical and life science companies or clinical care, but it also plays a major role in other sectors such as food, cosmetics, or agrichemical. For example looking at human safety, you can use Causaly for questions in drug safety, or whether a cream might cause skin irritation, or how different food affects the microbiome, or how pesticides affect human health as well as the environment.

Health and life science research affects many different areas in our lives and we are humbled to be in a position to help with the scientific pursuit of finding evidence to make good decisions.

We are currently supporting any non-commercial research into COVID-19, with complimentary access to Causaly. If we can help accelerate your work, you can request COVID-19 open access! 

You may also like

Show Buttons
Hide Buttons