Cutting through the hype: Brandon Allgood talks AI and machine learning
Posted 28th September 2018 by Jane Williams
When we last spoke, Brandon Allgood told us about The Future of AI. This week, we caught up with Brandon and spoke about drug discovery and machine learning algorithms.
Can you tell us a little bit about your current position ?
I am the Chief Technology Officer at Numerate and one of the co-founders of the company. I am responsible for the development of Numerate’s cloud-based machine learning platforms. I run the software team and the data science team, and I’m also the technical lead on all of our internal drug programs, as well as external collaborations.
How did you get started in AI and Machine Learning?
My PhD is actually in theoretical cosmology and computational astrophysics. I thought I was going to be an academic, but then got somewhat disillusioned with academia in graduate school, so I decided that I wanted to look outside of academia for a career path. Being next to Silicon Valley, there were a lot of opportunities, many of which, I was not interested in.
Out of graduate school, I received offers from a number of bigger companies and a small start-up applying machine learning to drug discovery. I thought about my passions, and my passions are not just machine learning, math and computers. I wanted to have the freedom to pursue my interests – what I wanted to do, do cutting-edge science, and still have somewhat of an academic life. What I found was that the start-up life allowed for that. So I ended up joining the start-up over the other offers, and that was the start of my path.
The first thing I saw at the start-up was that I studied the wrong science. I studied the science of last century, physics. The science of this century is biology and chemistry, largely due to the advances made in physics.
You say you were disillusioned about academia. Could you tell us a little bit about that?
I wanted to be an academic since I was 10 years old. I was always led to believe that academia was a meritocracy, but if you just look at sheer numbers, there aren’t enough positions that open up each year in relation to the amount of PHD graduates for that to be the case.
In a situation like that, this is no longer a meritocracy. You have to not only be good and smart – you have to be good, smart, and lucky. I figured I’d have a better chance at being successful in the private sector.
Could you briefly run through the topic of your presentation at the upcoming congress?
My goal is actually to try and cut through the hype, to specifically talk about what machine learning algorithms we use, how we use them and give some concrete examples. I want to inform the scientific community and show them what it is exactly that machine learning can do, or at least the machine learning that we have developed. Good scientists can smell the hype a mile away, and so what I’m seeing is that good scientists aren’t looking at this because all they see, is the hype, so it’s time to cut the through the hype.
There are a lot of companies out there now, especially younger companies that are coming out and applying machine learning to the drug discovery space. I think we’re all attacking different problems. In many cases, the problems we’re attacking are orthogonal even though we’re perceived as competitors. For all of us, I think what we can offer is a faster, cheaper, drug discovery process.
There are however nonlinear increases afforded by machine learning. Based on our work we have shown that these are in improving translation, both from biology to drug program and discovery to the clinic.
On the early translation side, long gone are the days where you inhibit an enzyme, you lower blood pressure, and you have a multibillion drug. That type of low-hanging fruit is gone. These days, we’re faced with more complex diseases, diseases of neurodegeneration, oncology, and inflammation. Modern biological assays being developed against these diseases often involve cells showing phenotypic signs of the disease. These assays are not amenable to a typical screen and ween process, but they are uniquely unlocked through the use of machine learning. I will discuss examples of this.
On the other side, machine learning is starting to have an impact on the translation rate into the clinic. In my view, every decision you make today in a drug discovery program needs to be informed by every piece of data you’ve ever generated. This is currently often very difficult because it requires chemists and biologists to rely entirely on memory or to constantly be querying the database over and over again with the hopes of getting the right query to pull the right piece of data to inform them on their current program. Every piece of data you’ve ever generated will need to be used to train a set of machine learning models that should inform every decision you make to improve success. Here I am talking about ADME and toxicity modelling and I will give a few examples of how we are helping companies to improve later stage design decisions.
Are there any speakers that you are looking forward to meeting or are there any presentations in particular that you are looking forward to?
I am looking forward to hearing John Baldoni’s and Willam Van Hoorn’s talks. They always have insightful things to say. The talk by Georgia McGaughey also looks very interesting. Data visualization can often be the key to enabling scientific breakthroughs.
Brandon Allgood is Chief Technology Officer at Numerate. He will be speaking at the 2nd Global Pharma R&D Informatics & AI Congress.
To find out more about AI and machine learning, join us at this eagerly anticipated event. View the full agenda here.
Leave a Reply