How Keystonemab is using Watson to revolutionize drug discovery

By | 5 minute read | April 15, 2021

Medical research

Tushar Satav was about to supplement his career in pharmaceutical research with an MBA in data analytics when he was diagnosed with a massive brain tumor. Using his relevant experience, he began to research every available treatment option for the relatively rare type of brain tumor, a vestibular schwannoma. Unfortunately, because of its rarity, this type of tumor lacked treatment options.

Drug companies aren’t incentivized to develop new drugs when the potential patient pool is very low. It can take many years to discover, develop, and research a drug. Then comes a protracted clinical development and a phase where the drug has to be tested for pharmaceutical adverse effects in humans and is reviewed by governmental agencies. Due to these constraints, drug development requires massive outlays of capital, and even a promising path of research can miss the target of commercialization.

Fortunately, Satav’s tumor was benign, and it was successfully removed. But the experience got him thinking about how the traditional drug discovery process might be improved. He came up with the idea of using technology to identify existing or failed drugs for new indications. Satav’s need for expertise in technology and business development led to his meeting co-founders Tjerk Geersing and Dr. Roland Meisel. Geersing is a technology development veteran, while Meisel is a pharmaceutical business development expert with more than 20 years’ experience in business development and creative deal-making. The trio co-founded Keystonemab, a startup that’s using AI to find hidden links among information extracted from millions of scientific papers that can be used to develop new treatments. Usually, it can take decades for drug discovery scientists to review the relevant scientific literature in a given field. But Keystonemab’s software can quickly unearth real-time actionable insights with IBM Watson’s powerful Natural Language Processing (NLP) capabilities.

Cheaper, safer drugs — faster

In addition to discovering new drugs, researchers can use technology to find two or more existing drugs with synergistic effects, allowing pharmaceutical companies to experiment with new drug combinations. This approach is called “drug repurposing” or “drug repositioning.” It brings substantial improvements in treatment efficacy, cost savings as well as a shorter time to market, which can be lifesaving in scenarios like the current COVID-19 pandemic. It also allows drug companies to more safely and cost-effectively target rare conditions. Combinations of drugs can also be less toxic and more effective than a large dosage of a single drug.

According to Satav, the development timeline for repositioned drugs can be 30 – 60% lower than those for de novo compounds, and overall development costs can be reduced by as much as 60%, since existing drugs have already undergone expensive safety profile trials.

Satav and team are confident that there are scores of viable treatments for uncommon diseases just waiting to be discovered. But the only way to find them is to make connections between different drugs that are not typically considered to be complementary. Their business model is based on vast troves of pharmaceutical research data, newly available to the public, that include information about drugs, diseases, drug target proteins, biomarkers and signaling pathways. Researchers can use AI to discover meaningful semantic connections between two or more drugs that possess complementary characteristics, like finding multiple matching needles in a very large haystack.

Trusted, secure and approachable AI

When Satav began to look for a technology platform, he discussed his use case with fellow entrepreneurs and specialists in his industry. “They actually guided me toward Watson,” he says. “Watson is a proven technology. There are a lot of startups who are in the pharmaceutical domain who are using Watson, so I don’t need to sell clients on why we’re using this particular technology, because it’s all really proven.”

Watson’s security was another factor. Pharma clients demand a high level of security for their data due to the highly regulated nature of the industry. “We work with pharmaceutical clients, and data security is very important for us,” Satav says.

Thirdly, Satav chose Watson because he’s not an AI expert by training. “It’s very user friendly,” he says. “I am not an artificial intelligence expert, I am more of a wet-lab scientist. I’m used to working in labs doing drug discovery research, with an extension into big data analytics, but I was not the artificial intelligence or software guy. So for me this user friendliness was important.”

Keystonemab is using Watson Knowledge Studio and Watson Natural Language Understanding (NLU) to build and train custom models that identify millions of entities and relationships, determine the strength of these connections, and ignore the weakest links. It took Keystonemab about a year to build this solution, and the company plans to spend six more months to bring it to market.

Disrupting Big Pharma and beyond

Keystonemab is selling primarily to chief scientific, chief development officers and other C-level executives at small and large drug discovery companies, but the goal is to provide the platform to commercial drug manufacturers, lab scientists, physicians, and other users further down the supply chain. For now, the team is focused on presenting partners with findings that they wouldn’t have been able to unearth with their current internal staff of researchers.

But first, Satav and his co-founders must convince decision makers that AI isn’t just hype. “The pharmaceutical industry is not very aware of how AI works, and it’s because they are conservative about using new technology that they are not experts in,” Satav says. “So you have to explain it to them, or it has to be simple enough to make them understand that it’s not a black box. It is just a technology that can improve their day-to-day working life a lot.”

Satav and Meisel’s pharma backgrounds go a long way toward persuading these potential customers.

“If you don’t speak their language, it’s very difficult to convince them,” he says. “Historically, my industry is not very accustomed to change, so you cannot disrupt it so easily, but if you are from the same community, and if you can explain the technology to them in their own terminology, then they are more likely to adopt.”

Keystonemab’s long term goals are to see its insights actualized in the form of new treatments, and eventually to expand use cases beyond pharma into nutrition, healthcare and other industry sectors. The same ability to find connections between drugs could be theoretically employed to derive insights from scientific literature relating to the manufacture of any kind of complex chemical products, as well as the way nutraceuticals can support patients, specifically in nutrition deficits caused by, for example, gastric diseases or diabetes. Similar approaches are also envisioned in the bioprocessing industry, as the dependence of cells from nutrition media, when cultured, is comparable with the dependence of patients suffering from lack of specific trace elements or other nutrition deficits.

But for now, Keystonemab is focused exclusively on drug discovery, and helping to de-risk the highly risky process that drug research companies take when exploring new treatments. This effort serves as a fantastic use case for AI search, demonstrating how finding ways to crawl documents quickly and more thoroughly can quite literally result in lives saved.