Advancing Drug Discovery with Structured and Intelligent Data

Michael Wilson and Craig Knox, co-founders of DrugBank, sat down with me to share how their comprehensive drug knowledge base can be used together with AI and machine learning for drug discovery and repurposing, clinical research, uncovering of patient insights, precision medicine, and much more.

Here is a sneak peek of our conversation:

Q: We hear a lot about drug discovery, which is a kind of early identification in silica coming from a computer algorithm. We hear a lot about quantum computing being able to do these kinds of amalgamations and identifications very quickly. I was wondering how you described this. It’s almost like the ubiquitous building blocks, which are these chemical entities. Are the algorithms able to put lego piece 1 with lego piece 2 in the new confirmations that nobody’s ever thought about? Then, the AI enables you to determine this new way that this new chemical entity that you created with lego block 1 and lego block 2 that we’ve never thought about before. Is this the way Drugbank is working?

A: What you’re describing would be kind of an approach, one way you might try to predict new drugs. We actually enable our customers to try many different approaches. There are all these different stages of drug discovery that Craig was kind of touching on there. When you’re looking at a new approach and you’re looking at an example, like what you’re saying, where you want to take this modular approach; where you have something like almost lego blocks for chemicals, you want to make a new configuration and then the next step you want to do is predict if that is going to work or is that going to cause side effects? How is it going to be metabolized? All of these different AI approaches are basically different ways to predict something about that drug.

Some of our customers will be trying to predict many different things at once. In other cases, they’re focused on a more narrow question, maybe they’re just focused on trying to predict side effects, for example. But in any of those cases, you need the knowledge in order to train your machine learning models with all the machine learning approaches. You can’t really just program it in or tell it what to do. You have to show it, give it examples, data, and knowledge so that it learns from that data what to do and what predictions to make.

One of the really important things that we do is provide that knowledge for those AI algorithms to learn from. Because there are so many different possible ways to approach that problem and people come up with new ways every single day, we actually enable that piece of innovation to increase. Now you have a new idea, you want to try a new experiment, a new in-silica experiment, instead of the first step being “Collect all this data and figure out how I can do this.” and then “Okay, now finally after doing all that work I can try that experiment.” With DrugBank, you can just do the experiment today. Try and experiment, iterate, tweak again, try bringing this other data and bring our data about metabolism, bring our data about side effects, whatever it is.

There are hundreds and hundreds of different types of data that we capture, that you have at your fingertips to do those experiments with. We don’t make predictions about what drugs are going to work but we enable all of our customers to take and experiment with different approaches to make those predictions.

For more of our discussion, you can watch the whole Fireside Chat with Michael Wilson and Craig Knox, or listen to the podcast version, below.

To check out previous Fireside Chats and to make sure that you don’t miss any future updates, subscribe to our newsletter or follow us on YouTube, LinkedIn, Twitter, Facebook or our podcast. If you enjoyed this episode, kindly leave a review on iTunes.