Drug Discovery & AI Predictions ft. Stephen MacKinnon

In this episode of Innovation Capital

How will the drug discovery process and experience change in the coming years? In this episode of Innovation Capital, we explore this fascinating topic in detail alongside scientist and innovator, Stephen MacKinnon.

Listen now

Episode transcript

Ray Chohan: Steven, welcome to Innovation Capital! We’re really excited to have you on this episode today. In fact, we’ve been looking forward to chatting with you for quite a while because we’ve been following Cyclica as an organization for a while, and are close admirers of this space.

We’d love to kick it off with a bit about your backstory and how you ended up in the wonderful world of biotechnology, Stephen. Then if you could delve into Cyclica’s mission and vision, we’ll go from there.

So, Stephen, over to you!

Stephen MacKinnon: Thanks for having me on board! I’m always glad to talk about what I’ve done, and what the organization has done. My name is Stephen MacKinnon. I’m the VP of research and development (R&D) at Cyclica.

My background is in computational biochemistry and I hold a Bachelor’s degree, as well as a PhD in this field. However, as opposed to studying biochemistry with beakers and experiments in the lab, my way of studying has been by looking at large databases of information and trying to glean new insights by uncovering trends in the data. My PhD was in structural Bioinformatics, which is a matter of looking at the 3d structures of proteins. And again, looking for patterns in how they interact with one another, entirely computationally.

I joined Cyclica at the very early stages of the organization, in the pre-seed stage where we were all in the basement, so to speak. I was the first staff scientist at the time. We set out with an idea to make predictions about how drugs might be interacting with proteins. But, not in the sense of saying, “Here’s a new protein target, what are all the drugs that might be interacting with this protein target?” Instead, our approach was to develop an all-encompassing solution that could solve not only the problem we were designing the drug for, but also discover all the protein targets this drug might be interacting with.

This follows the idea that there are roughly 20,000 different protein encoding genes in the human genome. No matter how well we develop a drug to interact with only one protein target to induce a therapeutic effect, we know that drug is interacting with many different things. It’s being metabolized, it’s being absorbed and moved from one place in the body to another. So, we know that there are many different proteins that a drug is interacting with, and we want it to be able to use computational approaches to try and map all of those relationships.

Now, I’ve been with the organization for about seven or eight years now, and seeing it progress from a pre-seed organization to where it is now, I’ve had the opportunity to build a team of scientists. Initially, I was the main developer, doing a lot of the programming and research directly. Then, I started to build a team. I found other like-minded scientists to join us, and we have been collectively working on this problem together.

Ray: That makes sense. Stephen, is it fair to say that in November of last year when DeepMind AlphaFold was able to predict a protein’s 3D shape from its amino acid sequence and solve that 50-year-old grand challenge in biology, does that moment segue into Cyclica’s vision of building out drugs which are in essence like software built to spec?

Stephen: I think the progress DeepMind made was certainly exceptional. I particularly like the way in which they think and combine many different forms of information to solve a specific problem.

The protein structure prediction, or at least the de novo prediction of protein structure, is certainly a form of Holy Grail in the academic community for long periods of time. Having an understanding of how proteins fundamentally work, how they associate from these trains into 3D structures that are little machines inside the cell self-assembling from a blueprint into a machine capable of doing the most complex tasks we can imagine, even beyond our macroscopic views. On that front, it’s certainly a great achievement.

There are a lot of proteins that are going to be very well served from a technology able to predict structures, in terms of the human proteome. For instance, from about 20,000 gene encoding proteins, around four to 5,000 have been solved in some capacity experimentally, of which around another 10,000 proteins have some degree of existing cousin structure that can be used to model and when a technology like AlphaFold comes in to look into those dark regions of the proteome, in particular, the ones that have not been that aggressively study by researchers, and has an idea what those 3D shapes might look like.

It’s an important and meaningful addition in that context, as well as for identifying protein structures in species that haven’t been studied very much yet. For example, bacterial species that might be pathogenic, and are very important to drug discovery and to human health, or the gut microbiome, which also plays a key role in in human health. There is definitely a lot of potential for structural prediction technologies to impact those areas.

Ray: That make sense, and it was something we observed closely here at PatSnap last year. Now, we’re entering this glorious paradigm on how AI, data analytics and classic bench biology are converging together. In your professional opinion, where’s the backstory of this new paradigm?

Everyone has a slightly different nuanced lens on it. For our audience to really understand the beginning of this paradigm, can you tell us where it all started? Also, by 2026 where do you think we’ll be in terms of patient lives?

Stephen: Machine learning (ML) has been part of drug discovery processes for a very long time. There’s a particular paper that I mentioned before, and in it the authors were talking about the impacts of ML on drug discovery in the 1980’s.

There’s a long history of trying to use information as we as we can. In a lot of cases, the information we have is limited, and the scope of the problems we’re trying to solve is so large. So, there’s always been this motivation to use all of the information that’s available as much as possible in order to inform where experiments should go.

For instance, chemical space — the number is so large, I don’t think that there’s an actual word for numbers this large, but 10 to the power of 60 is one of the numbers that tends to be cited for the number of molecules that look like drugs that could possibly exist, or that could possibly be created. So, when the magnitude or when the permutations of all the possible experiments that you could do in a program are so massive, it’s very beneficial to have these computational predictions to help narrow that scope or to help focus where you’re going to get the best bang for your buck in terms of research.

A few things that have ignited the more recent advances are robotic-type approaches that were really common and rapidly development between the early 2000’s and the 2010 timeframe. So, when people were collecting data and performing experiments, instead of doing it on small numbers like 10s or hundreds of data points, instead they were collecting tens of thousands to hundreds of thousands of different data points. So, it’s not just about how many different measurements are being taken — it’s also about the data within those data points as well. There’s much more dimensionality to those measurements.

For instance, next gen sequencing, the technology that really started picking up in the 2008 – 2010 timeframe — one experiment could measure thousands of different data points. There are more experiments being performed, and the information within those experiments has much higher dimensionality. With that comes the need for computational predictions to make sense of information content that’s too high for individuals to consume in its entirety and try to make informed decisions on that basis.

Ray: Stephen, would you say the rapid cost decline is also an opportunity? For example, if you look at the cost of sequencing the genome in the early 2000s, this was close to a billion dollars. As of today, it’s less than $1,000. Does that play a big role moving forward as well, when you have this deflationary effect on how AI and the other moving parts around robotics and computational biology? In essence, is this becoming more scalable and accessible?

Stephen: Yes. Sequencing is one of the great examples of a biotech enhancement that’s become much cheaper, and as a result can be done at higher scales for less cost. However, there are multiple contributors that play a role in that.

For example, there’s an entire academic community performing those types of measurements and sharing that data openly with the world. This has gone a long way toward providing a basis to develop new types of technology as well. So, it’s not just the cost, but also the community effect. Combined, these elements have contributed to large public databases in the life sciences and have paved the way for companies develop new technologies and try new things. These companies can test what works and what doesn’t work, based on the information available.

Secondly, the accessibility not just to data, but also to algorithms has been a game-changer. There are a lot of great public toolkits dedicated to machine learning that make it more accessible to start developing new technologies.

Ray: Looking at this exciting space where you’ve got the convergence of DNA sequencing, synthesis, machine learning, computer vision, automation, and you’ve just touched upon robotics, what we’re also seeing in the marketplace is this new generation of biotech company. These function like a platform business model, no different from maybe an Uber, or an Airbnb, where they’ve got elements like data network effects, where other companies can build on their IP and platform. Is this something you have observed at Cyclica or are proactively participating in?

Stephen: That’s certainly a question that came up in the earlier stages of our development. At our inception we were coming up with new predictions that we expected could bring value to drug discovery. Our focus was around that central philosophy that I mentioned earlier that we really wanted to know or be able to account for, not just what a drug is intending to do with its primary target, but also everything on the podium, or as much of that as possible, up front as part of the design process for new for new compounds.

So, at one point, we develop a prediction. The prediction is good, it’s got a good predictive signal that we’re happy with. Eventually this begs the question, “How does this get commercialized?” In other words, how do you go from here (a predication and some software code) to a full-fledged company with a business model?

Now, I can’t really speak on behalf of other industries, but if you look at AI for drug discovery as an example — the drug discovery product lifecycle is very complicated. There’s not a single path to go from, “I want therapy for this particular condition to here’s an FDA approved drug for the condition.” There’s not a single path, there are multiple pathways along the way. Additionally, there are hundreds of different steps involved in that research and productization.

So, there are a lot of different areas where computational techniques can have an effect, and a lot of different places in which a technology could insert itself. Naturally, there are many companies globally that are being formed in this space. Surprisingly, not that many are directly competitive in terms of offering two tools designed for the same task. There is actually a lot of diversity in terms of the different tasks each of the technologies are developed to address and build.

At one point in developing a company, there is a stage where we have to consider, “Do we want to focus on solving one problem, doing it very well, and then broadening that problem?” or “Do we want to focus on a specific inflection point that drives value within drug discovery and start building some drug discovery programs centered around that?”

That’s the tradeoff between a more horizontal type business structure versus a more vertical type business structure. And there are these two schools of thoughts operating with different companies emerging in this space as to whether you take a technology and use it to design better drugs, or do you build a technology and partner and create a platform of partnerships for a lot of different people to help and assist a number of different programs in their own research? That’s an identity crisis every company is going to face as they develop to some extent.

Ray: Where are we at now, Stephen, with the split between a problem solution building (which is pointed on a specific number of cases) and the opposite which is more of a platform where you’ve got this core IP capability around data, and you’re letting other folks build of top of you, and you go from there?

Stephen: That’s a good question. There have been successes in this industry in both areas. The more the focus tends to be on the point solution, the more that particular technology can be refined. The more it’s refined, the better it can serve. However, the counter to that is at some point there ends up being diminishing returns on the development and tinkering with one specific thing.

Perhaps there is actually an intersection that is better in which start with an effective point solution and try to move gradually upstream and downstream to improve on that and then support partner projects for a longer period of time.

Ray: This is so interesting, Stephen. As you’re talking, I’m visualizing our space, which is the classic software space where you’ve got platform companies, you’ve got thousands if not 100,000+ point solution players. They’re all driving value in different ways, and that ecosystem has led to so much productivity within the enterprise and in a B2B context, or B2C context.

In terms of journey on the roller coaster, where are we within this new paradigm within biotech? Specifically, how far in are we if you were to compare it to say the dot come, early internet where in 1994 to 1995, those were the really early stages, 1998 started the crazy dot come book, and then the rest if history? In your opinion, where are we in terms of building this out and making it a game-changing impact on healthcare?

Stephen: I don’t want to say the 2000 timeframe or even anything before that because it’s almost predicting there’s a bubble emerging. But I think to some extent comp bio might be in the nascent pre dot come bubble with big growth.

That said, I don’t quite think that’s the case because I think there’s more foundation in comp bio than there might have been in that really early-stage internet in the 1995 to 2000 timeframe. And I think that is because there has been so much good work done in comp bio for a long period of time, just under different namesakes. It wasn’t under the AI namesake at the time. Various terms have come and gone such as computation chemistry, and there was a lot of focus on simulations.

Furthermore, there’s this long history of simulations and predictions, predictions through the form of simulations or machine learning within that particular space. So, I actually think we’re closer to the 2006 to 2007 era where the crazy hype cycle may have come and gone. But, we’re kind of looking at this bit of democratizing drug discovery, democratizing computational biology, and making that more accessible broadly, and contributions from many different organizations globally. So, I’m thinking more of the rise of the YouTube and the Facebook, and the internet 2.0 type stage as opposed to the pre dot come bubble.

Ray: Wow. You’re making me smile with that. So, 2006 to 2007? I thought you were going to say 2001 to 2002! Like we’re early, but we’re getting there.

Can you talk to us about the big force multipliers, the core technologies which set us up beautifully for that 2006 to 2008 kind of era where we’re really going to accelerate and start making a positive impact?

Stephen: Oddly, I don’t really feel that that question is answerable through the technology. I really think it’s the culture of the scientists and their readiness to adapt and to contribute. Also, there has been this desire for smaller companies and biotech companies to start up globally, outside of these traditional hub zones.

I think it’s more of a culture of readiness and the mentality of scientists towards predictions. It’s like the content creation shift that began in 2007 where people went from being an internet consumer to an internet content creator. Scientists globally, instead of just generating basic science, are taking their ideas and applying them and creating therapeutics.

It’s all the students and the professors in labs globally, making this amazing research and these great discoveries about the effects of certain proteins on cells. They’re the ones seeking out the next stage and asking the important questions so as, “Why can’t I be the one who develops a new therapeutic? Why can’t I be the one who spins something up?”

I think that appetite is what makes us very excited, right? If we can find a way to help democratize drug discovery then I think we’d be in that great position where these internet companies were at the time when people were ready for it.

Ray: It seems like we’re opening up a brand new metaverse here, Stephen. Where if we democratize it in this way there could be for example, a bright PhD student on a campus at an institution in South Africa who is building on a platform of a biotech based in North America and really driving value in that process. Is that the potential world we’re going to be executing in the next one to two years, or is it already happening?

Stephen: I think that’s already happening, and there’s the excitement and the appetite. The types of problems researchers are looking into are rare diseases, and you mentioned South African so also tropical diseases. Perhaps it’s forcing the industry to broaden the scope of things it’s looking at as well by doing this.

Ray: What are your favorite examples of teams or individual organizations who are really making an impact right now?

Stephen: Well, I’d like to say us (laughs).

Ray: Of course, I know you guys are!

Stephen: Certainly, there are many other companies within this space we hold to a high regard. But that’s what we aspire to do, and what we want to be able to help with. We’ve set up academic partnership programs for that very reason. We believe that training and that experience is invaluable to the partners and the students, and the postdoctoral fellows we’re working with. It provides them with exposure to an industry, creates interest and an appetite to keep moving.

We released an archive paper as a collaboration with two other academic institutions. One that helped us on the technology side and developed a new technology to identify protein targets through Bo Wang at the Vector Institute. his student Haotian, and postdoc Mehran, who helped us with a COVID-19 Drug Discovery Program where they would help identify targets. Then, we take our technology we developed, Polyform DB, which is a drug repurposing database that works by predicting drug target interactions of around 10,000 clinically tested compounds.

So, these are compounds that have some phase one data, phase two data or phase three, or have been FDA approved relative to about 8,000 human proteins. It’s kind of an all by all precomputed database of drug target interactions and we cross reference their targets, the predictions that we made, and with the help of another academic group, Costin Antonescu from Ryerson University and postdoc Michael from his group, they developed an accurate virus infectivity essay for using real human cells and using live viruses to be able to really see thousands of compounds relative to some robotics. We can see with this low throughput scientific screen that’s accurate. So we were able to purchase around 26 compounds for testing based off the production, and we had a few hits. It led to the discovery of a new candidate for COVID-19, but instead of acting directing on the virus it acts on human cells. The hope with that approach is it might be more resistant to mutations if subsequent experiments show that this is indeed a viable candidate for clinical trials.

I wanted to point that out because I thought it was a good example of the appetite from academic institutes specifically regarding COVID-19. When the pandemic first hit back in March to April 2020, everybody wanted to be able to help in this space. So, we partnered with a couple of academic institutes, and everybody played a vital role. We found not just repurposing compounds, but new targets that could possibly be useful as well as we move down the line and perhaps develop some new compounds that might be good against future coronaviruses, or other viruses.

Ray: The more you talk about that example where you had these wonderful partnerships with certain academic institutions, it’s so akin to the way certain software’s developed. We are seeing this huge convergence where the modern-day biotech company has an operating rhythm no different from a best-in-class cloud-based software business.

Stephen, has a lot changed on the detail of execution? So when you are partnering with X academic institution in say, New Zealand, and the process to get things moving and do all of the boring stuff in terms of due diligence, certain legal requirements, and whatever process you typically have to go through, have you guys made big changes around working with folks and fast-tracking how things are done and arranged?

Stephen: There certainly was a bigger barrier at the beginning. If we weren’t that persistent about working with academics, and figuring out what those processes look like, then perhaps things may have looked a little bit differently.

Fortunately, we have a great strategic partership team that was really committed to getting our technology out there and working with academics to help co-develop technologies and field test them to help us get validation and so forth. The the first few were certainly a challenge, but after a while we learn what the academic institutions want. And because these institutions are generally looking for more applied research and partnership, there’s an increased appetite.

That appetite also comes from universities and government programs as well, who are doing their best to help promote industry academic collaborations. I know in Canada, there’s a lot of very good resources that we’ve had the advantage of working with where they help fund some collaborative research programs for us. This provides us with an opportunity to dabble in something we otherwise may not have.

For the academics, it provides them with what they need in terms of developing their research and getting publications. Then for the students, they’re provided with this unique opportunity that sets them up for a greater chance of success when they graduate. So, I think it’s coming from a lot of different places, and it’s not just more receptivity from the academic institutes and their IP management divisions, who are also increasingly flexible. Furthermore, the government programs aid in this process substantially. Lastly, there is still really strong need for basic science, but that need is being balanced with the need for applied science as well. So, it’s the whole ecosystem evolving and improving that’s making things more accessible.

Ray: Focusing on the academic and government side, has there been massive cultural shifts in the past one to two years where the red tape is reduced? I know in the past there were challenges in getting things going and people executing. But, can you now get going on a partnership and try to execute and deliver results?

Stephen: It’s never red tape cleared. I don’t know if there are systematic improvements, or our own team that’s been getting better at working with it, but it’s certainly not much of a barrier anymore. I think the will is stronger than whatever red tape exists, so even when they happens, we get through it.

Ray: Is it fair to say internally at Cyclica having that intelligence workflow is key to your day to day now, because you’re looking outside more than ever?

Stephen: When we were much smaller, seven to 10 people, it was certainly hard to navigate and keep track of all of that. But you know, as we’ve grown, especially now we’re past 50 people, it’s a little bit easier to have different people responsible for monitoring different things. And within that group of 50, there’s many individuals who are very interested in certain elements such as those government programs, academic institutes and partnerships. So as an organization, certainly, it helps maintain awareness because as individuals like it can be hard to keep.

Ray: Just having some fun now and looking at utopia, we’ve got this brand new paradigm within the biotech space and Life Sciences from a commercial aspect as a whole. Now, as you mentioned earlier, we’re kind of approaching that 2006 to 2007 internet era within this space.

What do the monetization models look like? Because my head is spinning thinking about the way you can scale a business now. Are there certain revolutionary new ways you can scale revenue and democratize monetization to enable a wider set of stakeholders to have skin in the game and benefit from doing great work and affecting healthcare?

Literally, going wild with this on Stephen — not just as an industry, but the space as a whole. Are there certain monetization models which are being talked about as wacky new ideas or actually happening right now?

Stephen: In the 2007 to 2008 timeframe when YouTube content creators started coming out and Facebook users were booming, and thinking about different monetization schemas. The models that succeeded in that space were the ones in which both the platforms and the individual contributors were benefiting. For example, the rise of the YouTube stars on YouTube illustrate this perfectly.

I think the emerging solution that’s going to come out of this space will follow similar lines. Academics who are coming up with the basic science and identifying new interesting targets and developing disease models and cellular disease models are the ones who are going to be able to benefit from that research. So yes, I definitely think there’s a lot of hope and excitement for how biotechs are going to be seated and developed in the upcoming five to 10 years.

Ray: With the rapid rise of the event blockchain technology to central decentralized ledger technology really accelerating over last, and on a more republic level the huge spike un Bitcoin and Ethereum, it’s kind of having its 1998 moment right now. We’re actually reading and speaking to prospective partners here about how if the development of a drug is democratized, you could potentially incentivize, say, an early stage academic researcher who may be has a predictive algorithm which is used by X biotech company. But they’re now more open to sharing that on maybe a marketplace where if you’re a bright mind at an institution in say, South Africa, you just upload your idea and It’s tokenized. It’s stored on the blockchain.

So if X biotech in San Diego says, “Wow, we love this predictive model it actually enables this capability within our company. We’re going to timestamp this and reward you.” Are you seeing areas like blockchain, and non-fungible tokens (NFT) bleeding into biotech?

Stephen: Well, that’s definitely an interesting idea of having a sort of app store of comp bio. Maybe we should look into that and start something on the side.

Ray: Well, if we’re linking this back to that internet story, it’s inevitable, right? Are you seeing anything on a broad level outside of your organization, just generally, a 500,000 foot overview is this being discussed?

Stephen: With blockchain in particular I would say a little bit less so. But, I have certainly heard about uses of blockchain for electronic notebooks. I can’t really speak to how that is going in particular, but a way of kind of assuring that information isn’t changed within experiments and things that are recorded at least for the sake of being audited or verified afterwards. So I have heard of applications of blockchain in those areas.

Ray: Well, Stephen, it’s been fascinating connecting with you. We could probably talk for hours because I think what you and the team are doing there at Cyclica is amazing. You’re in a marketspace we’re deepy passionate about here at PatSnap. We have well over 1,000 customers who are directly or indirectly operating in this space. We really apprecaite you participating in Innovation Capital today.

Now, just to wrap things up with a fun quick fire round, completely off topic — alien lifeform believe or non-believer?

Stephen: Believer. For sure.

Ray: Why?

Stephen: I think alien life is probably not the way we imagine it. But, to say there’s nothing that might look like a bacteria of some sort, on some planet somewhere is a bit out there. Every time I hear something new about the world of astronomy it’s that the universe is 10 times bigger than we thought it was before. So, if the size of the universe keeps being larger than we expected then the likelihood of aliens has to be more likely than we expected it to be as well.

Ray: Where do you think the comp bio space will be in 2030 in terms of time to market, drug cost and exciting developments?

Stephen: I think on the technology side, I think there will be a lot more experiments run just for the sake of training models. I’m thinking cloud computing clusters like AWS and Google Cloud, they tend to have these instances that are used during the downtime if nobody’s otherwise purchasing the VMs, during that particular time renting the VMs, the downtime is just used for other types of computations. I think lab equipment, instead of focusing on specific research programs, especially robotic stuff, could be just generating data in dark spaces that have proteins that under study just for the sake of generating data.

In terms of the long-term vision toward patient outcomes, I think what this entire field will do — especially if there is a big emphasis on boosting up these early stage biotechs that are coming out of university, is it will benefit the more underserved conditions. I also think there will be a lot more support for tropical diseases, rare diseases, and perhaps things that are a little less canonical. I think there’s going to be a big impact for those particular patients, perhaps in areas that weren’t otherwise drug targets, before that. There is a lot of potential to expand the scope of what people consider druggable or what people consider as therapeutic opportunities with this approach.

Ray: The classic headline of the average cost of developing a drug is $2.5 billion with an average time to develop of eight to 10 years. Where do you think we are in 2030 in terms of time and cost?

Stephen: I do think there will be an impact. But, I also think by doing a lot more trial and error, there will be a lot more error. My assumption is it will increase the diversity of things. But, to put a specific number on how expensive it might be, it should definitely be less, but it’s hard to put a specific number on how much less. I’d like to say cut in half.

Ray: We’ll wait and see. Stephen, I really enjoyed the exchange today and look forward to seeing you again soon. Thank you so much.

Stephen: Thank you.