Launch HN: PostEra (YC W20) Medicinal Chemistry-as-a-Service and Covid Moonshot

[morraa]

Hey everyone! We’re Alpha, Matt and Aaron, co-founders of PostEra (https://postera.ai/). The title above is quite a mouthful (we used all 80 characters) so we'll begin by breaking down what it means.

What is medicinal chemistry? It’s part of discovering new drugs. A drug hunter decides what disease to focus on and then selects ‘targets’: usually proteins whose activity is key to the disease. Then they look for a molecule that can ‘hit’ that target and stimulate a response which will hopefully have beneficial effects. Developing such a molecule that is potent and safe is medicinal chemistry.

Despite it being a crucial part of drug development, this field has relied on trial-and-error approaches—a very expensive way to muddle toward a drug. Where computational tools have been used, they have emphasized the 'best' designs without any awareness of what it would take to physically make the drug in a lab and test it. Our approach is to apply computational methods that know how to make these designs.

We’ve been working on developing machine learning tools to advance the field for the last 3 years. Alpha formed a lab at Cambridge in 2017 to apply machine learning to drug discovery. Matt joined the group and soon some exciting results began to emerge, particularly in the area of how to make molecules. We published the first model to outperform trained human chemists in predicting the outcomes of chemical reactions. Alpha then got Aaron, his former mathematics classmate and debate partner at Oxford to leave his job for the world of drug discovery.

We decided to focus on the one challenge that exists at almost every step: molecules need to be made. No matter how clever it looks on paper, a molecule is worthless unless it can be tested in a lab. The task of actually making molecules, known as chemical synthesis, is often a challenging problem, involving the combinatorial explosion of games like Go with moves that can’t be defined in a simple rulebook.

You start with a set of simple molecules which can be combined through chemical reactions (a ‘move’) to form more and more complex molecules, known as the ‘route’, until you arrive at your desired drug candidate. But how to combine these molecules? Trial and error is not an option, given the enormous cost of doing chemistry, and just enumerating all options to a client is unhelpful given that your average molecule can have hundreds of theoretically-possible routes. Searching this tree of routes and scoring the viability of such routes is where ML becomes very powerful.

We developed a machine-translation approach which takes in reactants and outputs the product of a reaction; an approach very similar to how Google Translate operates. This allows us to score the viability of each move. We combine this with fast tree search algorithms, used in models like AlphaGo to efficiently search the large combinatorial space of possible reactions.

To get this technology in front of users, we're building a cloud-based platform. Clients input the molecule they want to be made, our system designs a route for how to make it, and then the client can order this molecule through our platform. We don’t own a lab, but we partner with chemical manufacturers around the world who execute the routes we design. Combining automated chemical synthesis with compound ordering creates a better experience for the drug hunter who wants to focus on their science and just wants a vial with their compound without the cumbersome process of figuring out how to make it and where to get it from.

All that is what we were working on until the pandemic hit... and now we can answer the second part of the title: COVID Moonshot.

We had just finished YC W20 when a tweet from a team of scientists quickly changed our travel (and company) trajectory. A team of scientists at Diamond Light Source in the UK had shown that a selection of chemical fragments were effective at binding to a key part of the COVID virus. We realised there were hundreds of chemists sitting at home, with their projects on hold, who could help take these fragments and turn them into genuine drug candidates—an open-science approach to crowdsourcing a new drug. We created a platform where designs could be submitted and hoped for maybe 50 to 100 submissions. In the first few weeks, we’ve received over 4000 submissions from 200 scientists around the world.

This was the start of a COVID Moonshot initiative that we are now helping lead. It is an international consortium of scientists drawn from academia, biotechs, and pharma, all working pro bono or at cost with no IP claims on any resulting drug candidates. The aim is to find an antiviral candidate for COVID-19 by the end of the year—a ‘moonshot’ of a time frame compared with the standard drug discovery paradigm.

That standard paradigm is unfortunately broken when it comes to pandemic-related diseases. Biology and chemistry are hard enough, but things become even intractable when there are little or no commercial incentives to develop new therapies. Sadly, this explains why promising antibiotic companies like Achaogen go bankrupt and why, even after SARS-CoV brought the Far East to a halt in 2003, we still didn’t invest in coronavirus therapies during the last 17 years. For therapies that only become critical once every few decades, we need a new approach to developing drugs.

We think that drug hunters can learn something from the CS community and its embrace of open source. Similarly to open-source software development, someone has to manage the roadmap and triage suggestions. For Moonshot, the candidate drug submissions are great but we obviously can’t make and test all of them, so how do you pick the most promising ones? Here is where our technology comes in: it can identify which candidates can be synthesized easily. Since in a pandemic you need to move quickly, prioritizing compounds that can be synthesized easily is a natural triaging mechanism. Where a human chemist would take 3-4 weeks, we were able to design synthetic routes for all submissions within 48 hours. The top route designs were then passed on to our chemical manufacturing partner for synthesis. We’ve now experimentally tested over 500 compounds and found several promising candidates which we are now testing further. All data is publicly available on the site: https://postera.ai/covid

Inspired by open-source software, we’re seeing advantages of open-science collaboration in areas where market incentives are lacking. We started with the opportunity to connect drug hunters with the latest ML, but have expanded this into a platform that helps connect scientists with each other. This is particularly needed when it comes to drug discovery logistics—the fragment screens are conducted in Oxford and The Weizmann Institute in Israel, computational methods are done by PostEra in California and Memorial Sloan Kettering Cancer Center in New York, and chemical synthesis is carried out across several countries. Many of the features we are rolling out, such as automated alerts on suggested drug designs, open forum discussions, and live data uploads, feel very akin to a ‘GitHub for drug development’.

Identifying biological mechanisms of diseases and forecasting clinical outcomes are huge problems, but we believe that the chemistry stage of drug discovery can become a reliable industry rather than an artisanal craft. Machine learning tech is a key part and we're still working on it, but our clients have been constantly reminding us that just the logistical aspects of drug discovery are a great source of pain. Science software is also notoriously hard to use so we've learned that combining good UI with good ML should be our ambition. Our current mantra is: ordering a molecule through PostEra should be as easy as ordering a pizza!

We need more researchers, coders and chemists to help us on this journey and we’d love to hear from you if our vision sounds like something you could get on board with! Here are the open positions within the company we are now actively hiring for: https://www.workatastartup.com/companies/13332

Over to you, HN! We're eager to hear your feedback, questions, ideas and experiences in this area.

[kayhi]

Great idea! If I have a list of drug candidates and your software can apply a confidence level on the best route to synthesize a product that is better than chemists you're company is very valuable.

Inital thoughts - defining easy of synthetic pathway seems hard. If I'm an organic chemist and the recommended molecule involves a cost, technique or equipment that's inaccessible then I'd be stuck. However, if your network of manufactures can make it then labs may consider outsourcing it. I commonly see 10 to 1000x pricing differences in custom synthesis so there's also a risk of one vendor quoting too high and development prematurely stopping. For context and those that maybe outside this field, you can get a quote for over a million dollars for scaling up a product such as 100mg to 10g (not even a new route). You may need large amounts of the molecule for assay, animal testing, humans trials and going to market. Technique and equipment may not be an issue if your network is large enough, It will be interesting to see if you charge the drug companies as like a SaaS play and/or percent of sale from manufacturers like the YC company Science Exchange. Another idea is offering the software for free and make a percentage on the patent (IP) related the pathway.

[morraa]

Yes absolutely, the variance of pricing across different custom synthesis providers can be surprising. This we think though is actually an opportunity to serve our end users (drug hunters) better. Generally the cost of synthesis is a function of raw material costs and the labor time (translates to number of synthetic steps and risk of each step). All these parameters are datapoints we have access to or can model through ML. As such we've found that the ability to search large chemical databases is a huge saving when many clients may often just default to use a single CRO provider who may not be the best provider for their synthesis request. We are aiming to build a good network of reliable CRO providers to address the shortcoming and work with them to improve algorithms to optimize for reaction conditions etc.

Regarding pricing model this is something we are still working on. And yes you've hit the nail on the head -- We could charge the drug hunter (SaaS or IP) or we could charge the CROs that we send the custom synthesis to. Love to hear your thoughts here.

[JunkDNA]

A few random thoughts I'm curious about (full disclosure: I worked in anti-infectives R&D as a bioinformatician early in my career for a major pharma)

1) One big challenge in synthesis is ensuring compound purity. Even when I was working in pharma, it was often the case that some of the compounds in the screening library could be contaminated with intermediates. This is murder for any kind of anti-infectives research because you end up with false-positives for toxic intermediates. Since your assay is often, "does the compound kill the bug?" the answer for most chemicals is, "yes!". How do you ensure the purity of what you deliver to your customers? If I'm a medicinal chemist wanting to try this, I want to know that I'm not getting a vial of brick dust back.

2) Just because you have a mechanism to synthesize, doesn't mean the yield is going to be great. Does your algorithm factor in yield when selecting the route?

3) When I started reading your post I thought, "Hats off to these folks, this is a super hard problem that no shortage of extremely smart people have spent years trying to solve." Then I got to the moonshot section! The number of small molecule antiviral drugs with efficacy is vanishingly small. I understand why you would try to tackle this, but it truly is a moonshot.

4) I can't help but wonder what Derek Lowe (https://blogs.sciencemag.org/pipeline/) thinks of all this, have you guys tried to reach out to him?

[morraa]

1. We ensure that the compound is pure using analytical methods like NMR and LC/MS. In our Moonshot project, the assay cascade comprises biochemical assays against the main protease (2 different assay methodologies, run in Oxford and Weizmann Institute) and live virus assays, thus we should be able to infer whether the activity is caused by impurities killing the virus. In addition, we also perform high-throughput x-ray crystallography to determine the structures of all the protein-ligand complexes, which serves an an orthogonal assay.

2. Yes, our algorithm does factor in the yield when it decides which reaction to use.

3. You're absolutely right. It is very ambitious but we've realized that even if we don't get our compounds into human trials (currently aiming for in-vivo testing in next few weeks) that we will still have generated a lot of useful data that is there in the open for when the next pandemic comes around. This has been a real weakness from prior pandemic where research wasn't continued and certainly wasn't stored in clean accessible ways. As I'm sure you know SARS has super high genetic similarity to current COV-2 so having prior data accessible and cleaned would have given researchers a real head start.

4. Yes Derek is aware of COVID Moonshot and is also of the opinion that is it both ambitious but sadly necessary. We continue to follow his posts as healthy skepticism particularly in the area of AI for drug discovery is always helpful.

[kayhi]

4) He has touched on this topic to varying degrees: https://blogs.sciencemag.org/pipeline/archives/2018/01/30/au...

[nlh]

This is an EXCELLENT write-up. Thank you for both doing what you’re doing and for explaining it so well. I’m a software guy and have always found biosciences to be a mysterious black box, and you’ve cracked it open it in a wonderfully enlightening way.

Best of luck to your team and the project!

[morraa]

Thanks so much for the positive feedback! Forgive the very long read but yes biosciences can be a little opaque at times.

[coding123]

Definitely an excellent write up if not one of the best Launch HN write ups I have seen.

[101008]

Wow, this seems very interesting. As someone who did his CS Master thesis with vaccines, medicinal chemistry is something that I always enjoy, so I chcked your open positions. Sadly, I saw that even for Remote positions you are asking for a VISA. Can I ask why? Thank you!

[morraa]

Ah thanks for pointing this out; it is slightly misleading. Visa is only required if applicants want to work with us in our in-person office in the Bay Area, CA. If they are a non-US citizen we sadly can't sponsor a visa. Though certainly for remote work we wouldn't require it.

[lend000]

I love the idea: in particular, the arbitrary molecule generation. I've been thinking about is using the mechanical structures already provided by DNA to generate antibodies by reading the DNA tape in the same way that human cells assemble proteins. But it would be great to find a more general solution based on chemical computation that doesn't rely on and isn't limited by organic chemistry.

[morraa]

Yes, certainly an interesting idea. For now our models are trained on large corpora of organic chemistry reactions so our tech has some way to go before extrapolating outside this space.

[voicedYoda]

Kudos on the launch!

If I read this correctly, your team isn't discovering new drugs, but addressing the logistical need of building new compounds to make drugs, using ML to circumnavigate the most efficacious route to generate said compounds. Once complete, you actually outsource to manufacturers to build the compound molecules using your ML generated map.

Is that close?

[morraa]

That's a great description. For now we are focused on helping others discover novel drugs (rather than developing our own) and we think using ML to design 'recipes', while solving logistical challenges of shipping molecules around the globe, is a really key part of that process. And yes the manufacturers do use our ML generated recipes to make the compounds.

Outside of this we also engage clients on more in-depth partnerships where we help design, make and test new drug candidates but again our real value-add/USP here is the 'make' stage.

[figo22]

This is incredibly cool. As someone looking to try out some ideas around applied ML, can I ask what your stack looks like?

[mwcvitkovic]

Employee here.

React/Django/Postgres on AWS for APIs and websites. Terraform to manage the infrastructure. cortex.dev for serving some ML models, AWS Lambda for serving others. ML models are all PyTorch at the moment, with RDKit doing the chemistry heavy lifting. Data obtained through various means, including some tools from nextmovesoftware.com

There's a bunch more tech involved in supporting the scientists in the COVID moonshot, but that's basically everything ML-related.

Any particular part of the stack you were curious about?

[selimthegrim]

Alpha, you had a nice arxiv preprint on spin glasses and ML. The Wales lab strikes again!

[alphaalee]

Thanks! That was a fun project to work on! We’re also doing some work now to map out how the energy landscape changes as a function of network depth, and it appears that the landscapes of DNNs look dissimilar to typical structural glasses.

[drc007]

Sounds interesting and I'll be following you with interest. That said in my experience deciding how to make a compound is not really the challenge, the real challenge is deciding what to make.

[morraa]

Certainly we can appreciate that better designs are important though we chose to focus on synthesis as there is an oversupply of algorithms in literature/industry that can suggest designs ideas (we published a few of our own) while the synthesis space was less explored and we think provides the biggest leverage point to speed up cycle times.

Ultimately when partnering with drug hunters, outside of our cloud-based platform, we offer the integration of molecular design with chemical synthesis as we believe computational approaches are at their most useful when these two aspects are coupled.

[1996]

When/how will this be exploited by the "designer drugs"/"research chemicals" community to start a drop-shipping psychoactives business?

[morraa]

As a company we certainly work hard to ensure our technology is used for approved drug development. We have internal alerts as a company as to the risk of compounds being entered into our system and certainly suppliers and chemical manufacturers have high regulatory oversight. But yes there is always such an unfortunate risk but something we do spend time thinking about.

[adsodemelk]

Would PostEra be interested in using their platform to cure a neglected disease? https://deeplearningindaba.com/grand-challenges/leishmaniasi...

[morraa]

This looks very similar to our Moonshot open-science initiative and we love the idea of using crowdsourcing when such diseases lack commercial funding. If you know the folks who are organising this we'd be very happy to discuss with them.

[panabee]

this is an awesome idea and a very lucid explanation. good luck!

apologies if this is wildly inaccurate, but is it fair to characterize your startup as helping people create custom drugs the same way ARM helped people create custom chips? from the outside, it seems like there could be intriguing parallels.

[mc-robinson]

Thanks, really interesting parallel -- though I can't say I'm intimately familiar with how ARM's model works. I'll definitely do some reading.

But yes, we are very interested in helping people get made what they want to get made. It often falls into two situations:

(1) If the customer knows exactly the custom design they want to get made, we can help them find the best way to purchase or synthesize it. In many cases, customers may have trouble coordinating with CROs themselves, finding the best building blocks and route to the molecule, and dealing with logistics. We try to help ease that pain.

(2) The customer has a specific target they want to hit and they need just the right small molecule to "fit" in it. We also help with this, mainly through partnerships. And our thinking is that good design of small molecule inhibitors (such as one targeting the COVID main protease) involves expert knowledge of what can be quickly made and tested to help guide further design.

Lastly, we also work on suggesting molecules that may be slightly different from what the customer thinks they want, but may show similar activity -- and will be much easier to make.

[poroppo]

Nice!