Artificial intelligence and other advanced analytical tools are increasingly popular with pharma firms and their research partners. Such technology can be useful for sorting through tall mountains of data to determine which candidates might offer hope to patients in need of a novel treatment for their particular condition.
However, an AI-based approach might miss the mark by leaving out a human touch. David Harel, CEO and co-founder of Cytoreason, spoke with Outsourcing-Pharma about what tasks AI is best suited for, where it might fall short, and how to refine your approach toward drug discovery.
OSP: Looking back on news OSP has shared about CytoReason, you’ve had an interesting few months. Could you please tell us about some of the notable collaborations, projects, and other announcements you’ve shared over the past year or so?
DH: We’re continuing to execute our core strategy - developing computational models of human disease, based on proprietary and public data. The big news for this year is that we introduced more models with more features, which allowed us to significantly grow our customer base.
Among the partnerships that made the news this year is our collaboration agreement with Merck KGaA. We’ll be focusing on an immuno-oncology drug, with the goal of improving Merck’s understanding of the drug’s mechanism of action and identifying patient populations and tumor types where it is most effective.
In Japan, we’re running a PoC with one of the country’s largest pharma companies to help it gain a better understanding of how its drug behaves on a molecular level, relative to similar drugs on the market.
We’re also engaged in a handful of collaborations with large pharma companies like Sanofi and mid-size companies like Ferring, in addition to providing multiple disease models to Pfizer.
OSP: Please share your perspective on how the use and understanding of AI in the life sciences, in general, has evolved in recent months/years, and, specifically, in drug development.
DH: It’s clear that big data is transforming the pharma industry. Every two minutes, a new study comes out in the life sciences.
For years now, access to data is no longer the bottleneck, but rather our ability to draw value that has become the bottleneck. And the more data the pharma companies generate or acquire, the worse the problem becomes. They’ve grown their teams, but they’re still struggling to analyze all the data they have.
So the main challenge is interpreting the data, i.e. how do you make heaps of data, public or proprietary, mean anything useful?
You can hire one more bioinformatician, ten more biologists, fifty more annotators. But it will never be enough. There’s really no way humans can catch up, and make sense of it all.
The only way to deal with ever-growing data is to create technologies that bridge the data-insight gap. Technologies that can collect, organize and integrate all that data. It’s a paradigm shift that’s already happening. There are platforms that speed up the work of bioinformaticians, which are only getting faster and better.
OSP: Could you share your thoughts on what’s great about harnessing AI in pharmaceutical development? What are some of the benefits of using the technology, compared to more conventional tech?
DH: AI can solve the most complex challenges in the pharma industry today. Using AI, pharma and biotech companies can do a myriad of actions, in both pre-clinical and clinical phases, much faster and more accurately. They can prioritize new targets, find biomarkers, profile combinations, identify new treatment opportunities, shorten trial phases, reduce development costs and increase the likelihood of drug approval.
OSP: Then, you mentioned in our earlier conversation that while AI has its pluses, “it is not necessarily the best technology to apply in the primary phase of drug discovery.” Could you please elaborate?
DH: AI is often used in target discovery, as companies seek out novel targets and mechanisms that are beyond the trodden path. But a growing number of scientists are beginning to realize that maybe it’s not the best approach here. That’s because finding the proteins, genes, or pathways that drugs can target is not a straightforward process. It’s an open-ended, multi-disciplinary endeavor that humans cannot succinctly define.
OSP: To take it further, you also mention that AI is better for use on tasks that humans can define, rather than those we cannot. How does this apply to drug discovery, and why are humans better for this stage?
DH: That’s right. As long as humans cannot explain the specifics of the process - the input required, the decisions to be made, the output expected - AI will not be able to do the job. It can support humans in certain aspects, but it cannot replace humans. It can define and match patterns across hundreds of diseases, but it cannot define for humans what makes a good target.
OSP: At what point should researchers pass the baton to AI when working on drug discovery?
DH: That’s a great question. I can speak only to biology, not to medicinal chemistry. I believe that when researchers have determined the criteria for a good target, it’s time to pass the baton. In other words, when they have clarity regarding their expectations, AND could potentially carry out the process independently, in the absence of machines. AI will do the work in a fraction of the time and cost.
OSP: Please share some of the ways in which CytoReason is poised to help with drug discovery.
DH: CytoReason’s computational disease models are designed to continuously aggregate all available data that can shed light on the disease, and make it understandable to researchers of all levels in the organization.
Guided by our scientists, the CystoReason system takes those disease models and makes them an integral part of our customers’ R&D processes.
Pharma and biotech companies can then use these disease models to prioritize new targets, find biomarkers, profile combinations, and get a full mechanistic understanding of diseases and treatments.
It’s a scientific portfolio management tool serving both scientists and program directors. It allows them to prioritize and compare all their drug programs across patient populations and competing treatments, and to do this without losing the ability to drill down to the individual dataset and individual patient.
