First in Human Episode #33 featuring Aaron Morris

For episode 33, we connect with Aaron Morris Co-Founder & CEO at PostEra. Find out what the difficult process of drug discovery may look like in the future, and how PostEra is looking to drastically speed it up through the application of AI. First In Human is a biotech-focused podcast that interviews industry leaders and investors to learn about their journey to in-human clinical trials. Presented by Vial, a tech-enabled CRO, hosted by Simon Burns, CEO & Co-Founder & guest host Co-Founder, Andrew Brackin. Episodes launch weekly on Tuesdays.

Simon Burns: [00:00:00] Aaron, thank you for joining us on First In Human.

Aaron Morris: Hey, Simon, great to be here. Thanks for the invite.

Simon Burns: Tell us a little bit about your background. How did you get started in this crazy field and how did it all come together about joining and starting, PostEra. 

Aaron Morris: Sure. I have a very unorthodox background. I come from a mathematics and then eventually machine learning. I finished my postdoc studies at the University of Oxford in the UK where I met actually one of the future co-founders of PostEra, Dr. Alpha Lee. We became good friends.

He went off into the academic field for six or seven years. He began researching AI and drug discovery. I actually went into the finance industry. I was applying machine learning as a trader on the stock market at an investment bank in London. We decided after several years of just talking and catching up that there was a huge amount of excitement around the academic work that had come from his lab. I felt that I would be able to help him commercialize what had begun as academic research. So, PostEra really formed at the end of 2019, building upon a lot of the years of work that was done at Cambridge University in the UK.

Simon Burns: Let’s go one click deeper. We’d love to hear about your machine learning platform, Proton. What’s the novel approach you’re taking? How is it different than some of the other approaches being taken to plot machine learning to medicinal chemistry?

Aaron Morris: The first thing you picked up on is that the technology itself, at least to date, has been very much squarely centered on medicinal chemistry as opposed to doing AI for biology where you’re trying to pick new targets or AI for clinical trial design. We’ve tried to brand ourselves as the world leader in AI for chemistry.

The real innovations, if you will, are across a couple of fronts. Firstly, there was a huge amount of frustration in the field around the early AI models that could generate ideas and generate molecules, but had very little, if any, practical consideration as to the necessary synthetic investment needed in order to make the molecule.

This is important if you can accelerate drug discovery because synthesis is the biggest bottleneck. It is the longest period in the traditional medicinal chemistry cycle just making the darn molecule in some lab around the world. Pursue was really the first company to innovate not just on machine learning for designing molecules, trying to get optimal properties, trying to make sure that they were safe as well as efficacious, but also taking a very serious, practical view of how to actually make molecules. The upshot being that advertising ourselves as the first company to cover what is referred to as the design make test cycle of medicinal chemistry. All three stages really driven by ML in a single unified approach. Certainly we felt that this would address some of the shortcomings that the earlier approaches to AI for chemistry adopted.

Simon Burns: Seems quite obvious, but that AI actually proposes molecules that are synthesizable, but… [Laughs]

Aaron Morris: Oh, it’s incredibly painful to get an algorithm to do that. [Laughs] Secondly, even if you can get an algorithm to dream something up that is theoretically synthesizable, what you need is the next step, which is, not only how do I make it, but who is best placed to make it. A huge amount of Proton over the last three and a bit years has been integrating the technology with all the world’s major CROs. 

We have over six and a half billion molecules updated in our inventory every single month. We know the lead times of these CROs. We know the FedEx shipping times of each CRO, getting a very granular understanding of synthesis, logistics and building block inventory and catalog availability is all the unsexy stuff that you need if you’re going to have a high functioning, synthesis driven approach in med chem.

Simon Burns: Let’s talk about what hasn’t been so obvious. What were some of the surprising challenges and obstacles in building a startup? I hear building a startup isn’t easy. [Laughs] What were the hard parts?

Aaron Morris: No, apparently it takes a bit of effort. [laughs] There are few aspects that come to mind. The first I would assume is generic and applies to every startup found, and then maybe one that’s more pertinent and specific to the industry that I’m in. 

Culture is sadly this buzzword that gets thrown around on PowerPoints but is very real and very tangible. If you get it wrong, it’s very hard to undo. What you find, as a founder, is the culture of your company becomes an overflow of your own personality. Your own strengths and weaknesses become the strengths and weaknesses of the company. It’s almost like looking in the mirror every day. 

A large part of what is difficult about a founder is admitting the flaws in your company are a flaw in yourself and that you might need to change the way that you view the world and do things. I know that sounds very abstract and aloof, but I think that very real practical feedback mechanism that you get every day from clients, investors, scientific experiments, and from your employees, becomes a very quick reflection on, you as a person and how you’re leading the company. That’s difficult and challenging. It’s humbling. It’s something you try and get better at. 

The second I certainly found difficult in the early days of PostEra was converging on a business model. It’s quite non-obvious what you do if you have [00:05:00] an advanced technology platform of some form. You can package and sell it as software, try and offer a kind of tech-enabled services model, or just go vertically integrate, and be a player in the industry or you can try a mix of the above. I spent a lot of time trying to learn from other people in the industry, other companies, and I’m super happy with where we ended up, but that was difficult in the early days. What is the business model here?

Simon Burns: You guys have done some interesting work in the covid space, and it looks very different. The preprint model, open science, just the general collaborative nature of Covid research looks quite different. Tell us a little bit more about that experience and how does it apply or doesn’t apply to the rest of the field?

Aaron Morris: That’s a great question, Simon. We launched a project called Covid Moonshot. The idea behind Covid Moonshot was to take a preliminary fragment screen that had been done at Diamond Light Source in the UK. They actually published all the data, on Twitter, and we quickly realized that actually if you’re going to take these early, basic fragments, they’re not even drugs, they’re fragments of a potential drug.

And you want to actually develop an antiviral cure for patients. Synthesis is going to be really important here cause you need to be able to merge these fragments together in a way that is fast and reliable and iterative. What we did was, we invited anybody and everyone to submit ideas for how to merge these fragments and then use PostEra’s ML to design the synthetic routes to score the molecules and then get them made.

I don’t know, we thought twenty, maybe, thirty people might visit our little website. We were three months old as a company. It blew up quite exponentially. We had over 400 scientists contribute about 20,000 different design ideas, and we chose to keep everything in the open and we wanted to do that firstly for speed. We didn’t have time to find lawyers and put IP patents and set up a whole clinical strategy.

Honestly, at the time it was just, we need to keep moving fast. At worst, we have valuable data that is in the public domain that we hope can help other people. There are actual drugs in clinical trials now that reference the Covid Moonshot data as their source of inspiration for some of their compounds.

 Thirdly, the majority of the developing world is unlikely to get access to vaccines as early as the more rich, wealthy western nations. These nations need a line of defense against covid and antivirals is the best they’re going to get. You don’t have the cold supply chain constraints. You have a readily available pill, and if we do it open source, we don’t have investors demanding a huge ROI, so we can do it cheap. We can provide an accessible antiviral for nations that are unlikely to get access to vaccines soon. 

That was the motivation. And yes, that project went unbelievably well, and quick from just this fragment screen to a development candidate in about eighteen months. It’s now going through IND-enabling studies, and we’ll start phase one trials as what we believe the world’s first ever crowdsourced drug, which is incredibly exciting.

In terms of your second question, I take a bit more of a sober view that this approach could be used for any and every disease. There are certainly areas you imagine, underserved research areas where often the economic incentive isn’t strong enough to pull RND into that space. Think antibiotics, for example. Some of what we learned from the open source model could well be applied in that area. 

I would say though, we had this unique moment in time where scientists around the world just couldn’t go to work or operate their labs, and they were looking for something to do, and an open science project was a good use of their time. I’m not sure if you decided tomorrow to crowdsource a drug for Parkinson’s that you might get that same global focus, the same level of just charitable donations; companies devoting assays and synthesis resources to us for free at the speed and the pace that we did. I hope we can benefit the broader community. There is still huge value in open science, particularly as it pertains to putting data in the public domain. If only we had that data from SARS COV one, we could have moved faster on SARS COV two. But I want to be sober and I don’t think it’s a silver bullet for every disease.

Simon Burns: You have a series of projects, really quite remarkable speed, that you’ve been moving things forward, and seemingly moving things towards the clinic at a rapid pace. Tell us more about some of the projects ongoing. You partnered with Pfizer, tell us more about the experience working with the big pharma company.

Aaron Morris: We have, almost, three prongs of drug discovery we’re doing. There’s the open science covid moonshot work, which is fantastic. Obviously, there’s no economic benefit for PostEra, but it’s a great social good, and we’re very enthusiastic and committed to that. The US government just gave us and a couple of our collaborators a $68 million grant to continue this work for other viruses, that open science virology part of what we do as a company. 

Then, as you’ve mentioned, there were the partnership-based approaches. Pfizer really selected PostEra as the kind of leader in AI for chemistry and their primary partner about three years ago. We’ve now done one academic partnership and now two[00:10:00] commercial partnerships with them, the latest of which is called an AI lab. It sounds quite grand, but the point is that the sandbox approach will be set up within Pfizer to really test if AI can move the needle. 

Pfizer are effectively asking PostEra to work on certain programs and certain targets. We use our proton platform to drive those programs forward. It’s difficult to disclose a huge amount, certainly the most recent projects that we’ve done have gone very successful and up to the board level of Pfizer. It’s been a fantastic case of being able to show the speed and improvement that an AI approach can bring.

We’ve learned a huge amount from Pfizer in the process. I’m glad that we’ve done these partnerships. Not so we can come up with a nice case study or, a PR release, but, there’s huge value in working with scientists who’ve got twenty-five to thirty years experience in the field and to take those learnings and put them, hopefully, into a more machine readable format. The Pfizer partnerships are exciting. There’ll hopefully be other partnerships announced very shortly. 

The third angle is our internal pipeline. That’s a little bit newer. We kicked that off toward the end of last year; start of this year. We’ll be disclosing the indications on the things we work on in the coming months. But, again, here is an opportunity for PostEra to drive programs end-to-end and to use proton to move cures toward patients in disease areas that we care about as a company.

Simon Burns: I’d love to zoom out. You spent a lot of time thinking about the future of drug discovery. Where’s the space going? All of these innovations are being done in parallel. Once it all gets wrapped in, what do you think drug discovery looks like in five or ten years?

Aaron Morris: In five-to-ten years, what we talk about at PostEra, is not just what does science look like? But what does a scientist look like? When we’re hiring these chemists, biologists, and drug discovery folks, I advertise that AI is not going to replace your job, entirely, in the next five years. But scientists who can use AI will replace the jobs of the scientists who can’t. I imagine PostEra being at the bleeding edge of creating the modern twenty-first century scientist where there is a seamless partnership between the scientists and the engineers working together, collaborating to build modern technology that drives drug discovery. 

There is so much within the traditional drug discovery paradigm the AI is going to take over. There are lots of aspects such as logistics, compound ordering, trying to identify the correct CRO to work with, changing certain aspects of a scaffold in order to fix an ADME issue.

All of these things are approaches where we really believe in the next five years PostEra can entirely drive by AI. There are aspects in which humans will continue to be needed. I ultimately want my chemist and drug discovery team to become more strategic rather than more in the weeds and tactical. That is what I hope Proton can help with. 

 I don’t think AI is going to get to a point in the next five years where it can say, okay, this is the target product profile. This is the killer experiment. Here’s an orthogonal assay that we should run for verification. Here’s a portfolio view of the PostEra pipeline.

Here’s where we need to invest more money. Here’s where we should slow down. That type of high level strategic decision making is something I want my chemists and biologists to be free enough to be able to work on. While the AI can handle the grunt work of medicinal chemistry.

That’s where the field is moving. You’re going to see AI applied there across the entire field beyond what PostEra is doing. AI for biology, picking new targets, rather than necessarily having biologists walking through hundreds and hundreds of papers trying to piece things together. Trying to get AI to come up with coherent suggestions. And again, in the clinical trial, thinking about optimal patient recruitment and things. I don’t expect humans to be absent in this field in the next five to ten years but their role will look very different.

Simon Burns: A lot of computer scientists are now trying to get into tech bio to better understand machine learning and AI’s impact on biology and how they can do something about it. Do you have any advice for anyone looking to get into the field from your experience?

Aaron Morris: Yeah, I’d advise a couple things. I wouldn’t advise textbooks. I don’t think sitting and reading textbooks is the most constructive. I would say if you’re in the ML space, you sat here today listening to me and you on Twitter, or Meta, and you’re thinking about using your talent in the field of healthcare and in medicine. There are plenty of open source datasets. I encourage you to start using your ML and your CS skills on the open source chemistry and biology datasets and just play around and tinker and see where you can go get used to the data, understand the challenges of that type of data. There’s a bunch of fantastic blogs. I even post things now and again, if you really want to read those. 

My second line of advice would be speak to practitioners, speak to the people working in the industry. [00:15:00] Try and see what are the fundamental differences in the tech biospace or in biotech in general, as opposed to fintech or social media or wherever you’ve been using your other skills.

The third thing I recommend is to try internships. Take three months to jump in a company. Ideally, a small one, so you get into the weeds as fast as possible and just see if you enjoy the work, enjoy the people. I came from another industry. I am incredibly glad to be in biopharma, now. I find the mission and the people incredibly motivating. There’s a high barrier, a lot of jargon, and a very different domain to what people must be used to, but there are resources to help you get there.

Simon Burns: Sounds good. No textbooks. Maybe a little bit of learning from Twitter, a little bit of learning from your blog posts. I like that. Good feedback. [laughs] Well with that, Aaron, thank you for joining us. Really appreciate the time.

Aaron Morris: My pleasure. Thanks again, Simon.

Contact Us

By submitting, you are agreeing to our terms and privacy policy
This field is for validation purposes and should be left unchanged.