We’re delighted to announce that we have appointed a couple of senior industry figures to be deepmirror’s new scientific advisors. Today, we sit down with Mark Murcko, a senior lecturer at MIT and legendary drug-hunter, to find out why he’s joined the deepmirror journey:

Tell us about your background and expertise in drug discovery‍

I’m a PhD chemist, with close to 40 years in the pharmaceutical and biotech industry, and my work has mostly focused on discovering new ways to discover drugs. I’ve been fortunate in my life to have worked with some great teams, and I’ve been involved in bringing 11 new drugs onto the market.

The metric for success in this industry ultimately is drugs reaching the market. And the unfortunate truth is that we’re really bad at it. Some estimates put the number of people working in drug discovery at up to a million people globally, and yet we only produce about 50 new medicines a year. So we need new technologies, new ways of thinking, and that’s what brought me to deepmirror.

‍Why did you decide to join deepmirror? 

Drug discovery is incredibly complex, which means there’s a tremendous opportunity for computational tools to improve outcomes.   I’ve used computation a great deal in my career and I’m a big fan of trying to incorporate computational approaches wherever it makes sense.

It’s that complexity that makes an AI-native company like deepmirror so powerful; it’s been built from the ground up to focus on the use of machine learning predictive models to optimise the back end of drug discovery. 

The other thing that really excites me about deepmirror is the fact that they’re building the infrastructure. Everybody in the world is going to need ways of organising their data, constructing models, deploying those models, and validating them, and that is not trivial. Lots of people are focusing only on the creation of AI models, but it’s really important to create the infrastructure that enables anybody in the world to get the maximum benefit out of these tools. That gives deepmirror a unique asset.

‍How is deepmirror different from the tools you used in your career?

Very few people were even thinking about machine learning models ten years ago, so the idea of generating proprietary data to improve results was not really a thing. It was far more common to see an individual computational chemist at a big pharma company, who would develop a specific model using some kind of machine learning, and deploy it to their specific project. 

But that kind of tool would only work if that specific person was running it - I would call them ‘the wizard’. These computation tools had limited utility because of that reliance. There’s no scale there, so it’s fundamentally not a very useful technology for drug discovery. What we have now with deepmirror are scalable tools that don’t need a wizard; where any chemist or project scientist can work with the model and it will perform reliably. 

Beyond the utility of the tools, there’s also a far greater acceptance now that you need to add new data to build the best models. You can't just scrape the literature and then build a model and expect it to be perfect; the data is so sparse and often error prone. An integral part of what deepmirror is doing is generating new data from scratch, as well as building collaborations which give them access to data that can be used to improve models.

‍What are the biggest challenges in modern drug discovery?

Certainly optimizing the ADME and Tox properties of molecules is extremely difficult. Powerful data exists within each pharmaceutical and biotech company that could be combined to enable the creation of improved predictive models.  However, there’s a long-standing unwillingness from pharmaceutical companies to share their data, and this attitude is unfortunate because we know that the benefits will far outweigh whatever theoretical risk there might be.  Every  big pharmaceutical company is sitting on hundreds of thousands of compounds worth of data, which if made available to everyone, would allow all of us to discover drugs far more effectively.  So, sharing makes great long-term sense for everyone.