- R&D 100 Awards
Since it first began in 1950 with Alan Turing, the movement to imitate human intelligence in machines has expanded greatly. Today artificial intelligence (AI) has the potential to help select precise treatments for individual patients and develop new and better practices in drug discovery.
Matthew Clark, PhD, Director of Scientific Services at Elsevier, said that by tapping into AI, machine learning and big data, researchers and doctors could someday enhance treatments for cancer and rare diseases. Clark also said that these tools could allow researchers to better predict how people will react to certain drugs, aiding in drug discovery and development.
Clark will present these ideas on Nov. 15 at his talk “AI, Machine Learning and Big Data for Life Sciences: The Good, the Bad and the Ugly,” at the 2018 R&D 100 Conference held at the Waldorf Astoria in Orlando. To provide a sneak peak of what to expect, Clark recently granted an exclusive interview with R&D Magazine where he discussed his upcoming talk and how AI can revolutionize life sciences.
R&D Magazine: Can you explain a little about your background?
Clark: I come from a diverse background. I have a PhD in chemistry, then I worked for a drug discovery software company for some time and then in the information industry. Then I worked in pharma for 15 years, doing big data, analytics, informatics. I was the head of an analytics and computation group.
R&D Magazine: What is one key takeaway you are hoping participants at the R&D 100 Conference get from your talk?
Clark: Everyone is talking about AI, especially now, and big data as an art as well as a science. It is not as mathematically precise as it is portrayed in the media. It requires some artistry to apply it well.
R&D Magazine: How is AI being used in drug discovery?
Clark: For anything that you’d like to predict, people have made predictive models even before computers. Machine learning is a more powerful way to combine data to predict almost anything, like what drugs could be used to treat a rare disease or what kind of adverse effects the drug might have early in its progress. It will allow them to more easily predict that drug A is likely to be toxic, but drug B is not. Then we should focus our research on drug B and make those types of decisions earlier in the process.
R&D Magazine: Why do you think this is an important topic?
Clark: It is involved in all types of research that will allow people to predict things about, not only science, but also people are using it for business. It is a really good emerging technology and it is about understanding data. I think it is a very exciting technology and people are mixing all different kinds of data together to learn from. Images, text, numeric data, and it is like throwing them all into a big pot, stirring them up and seeing how they interact with each other and what you could learn. It is something that wasn’t possible for older prediction systems.
R&D Magazine: How have doctors reacted to the idea of using AI/machine learning in their practice?
Clark: Memorial Sloan Kettering and MD Anderson have big projects—these are major cancer centers—and they have had some success using AI to learn what the best treatment is.
I have to say that this particular use of AI has not been met with really good success. It is a bit general, the place where it will work more is when the question is a bit more defined, more precise “like can you help me select a drug that will treat a rare disease? Or given the chemical structure, what adverse events will this likely cause in humans?” The smaller scale predictions will be more accurate than the large-scale predictions. That’s part of the art of data science of arranging the data in the right way so that you can answer the precise question that you want.
R&D Magazine: How do you see the use of AI in life sciences expanding or changing in the future?
Clark: I think it will grow with lots of data and experience. One of the things AI needs is data and it is well known in the medical field that we lack things like outcome data. If I have a patient with a disease, which of the five treatments that people use is the best one, which has the best outcome? There is actually not enough data in some cases to actually train systems on that. So what we need is more data, specifically so you can answer questions like that. Medicare and hospital records have what the treatments were for those patients but it does not necessarily gather what the outcomes were.
R&D Magazine: Is there a learning curve in the industry using this type of technology?
Clark: Right now, we are sort of in a phase where everyone is throwing everything into deep learning.
They are trying to use deep learning for every possible thing and some of it works and some of it doesn’t. I see that as a learning process; it is not a bad thing. As the field progresses and people try everything, they’ll find some things with work better than others.