Immunoprecise Antibodies - AI Meets Biology
A Retail Investor's Chance To Invest Like Peter Thiel
Disclaimer:
I’m not a scientific expert, nor even a financial expert. As such, this is not a comprehensive article. It sacrifices a lot of exciting complexity and nuance for simplicity and clarity. Readers should consider this article a primer, a first step in their own due diligence. That’s how I’ve used it.
Introduction
Immunoprecise Antibodies (IPA) has the potential to be to drug discovery what Amazon is to online shopping, YouTube is to videos, or Microsoft is to enterprise software; IPA could supply a leading platform used by drug developers to stay competitive.
Despite having comparable, if not superior, technology relative to much more richly valued peers, it has been left for dead by the market and trades with very little liquidity, meaning institutional investors can’t buy meaningfully and should they step in, the price increase would be expontential.
IPA stock is an opportunity to be an early-stage investor in a disruptive company, an opportunity rarely found in public markets.
Big Picture
Healthcare is a massive market, especially as populations age and become sicker; the leading causes of death, cancer and heart disease, continue to increase in prevalence and be problems for researchers. Unlike past medical innovations like antibiotics and vaccines, which eradicated polio and death by strep throat or example, more recent inventions, like pacemakers, insulin pumps, and chemotherapy, treat the symptoms of the disease rather than solve the disease itself. Though many people have been helped, a true paradigm shift, like that of the invention of antibiotics, has not occurred for around 100 years.
Additionally, it takes at least 10 years to bring a drug to market and costs a fortune (for the developer and patient). Most drug candidates will fail, resulting in millions and sometimes billions of dollars spent in addition to years in wasted efforts. Billions of wasted dollars and decades of wasted time is an inefficient way to improve healthcare. There is room for improvement.
In short, society needs improved healthcare solutions, and the drug discovery market is ripe for disruption. Immunoprecise Antibodies is positioned precisely at this juncture.
Science Background
Antibodies combined with AI are seen by many as a potential way to shift the paradigm. Antibodies are the body’s defenses that fight off colds, the flu, covid, bacterial infections, cancers, and more. Sometimes they don’t bind to (recognize) harmful agents, which can result in cancer development for example. Other times, they overreact, resulting in things like allergies, arthritis, and eczema. When working properly, an antibody (FBI agent) will bind to an antigen (foreign agent), and flag it for removal (execution/deportation). In 2021, 4 of the top 10 therapeutics were antibody treatments: antibodies work.
What makes medical researchers excited about the potential of antibodies is the sheer diversity of them. The human body can make an antibody to fight off almost anything, with researchers suggesting the body could create up to 1 quintillion, or a billion billion (1,000,000,000,000,000,000), unique antibodies.
And that’s just in humans. Other animals have their own antibody genomes that could be used to combat disease, making the number of potential unique antibodies essentially infinite. Indeed, there are more antibody possibilities than atoms in the universe.
When you have nearly infinite disease fighting possibilities, an effective treatment for cancers and other diseases is likely to be found; antibodies could be the next antibiotics, if we can speed up and narrow the search.
Commercial Opportunity and Competition
With such a vast number of possibilities, being able to sort through and make sense of the data becomes of primary importance; it doesn’t do you any good to have a list of a quintillion trillion potential cancer treatments. Recall that drug development currently occurs mostly in labs and takes at least a decade and billions of dollars; the sun would burn out before we could test all the possibilities.
A handful of companies have recognized this and are working towards AI computational solutions for modelling experiments. The idea is that if a software platform can be created for drug developers to reliably and quickly identify promising antibodies likely to pass clinical trials without ever having to enter a physical lab, time and money would be saved (and probably lives). The fees charged for such a platform would substantial.
A variety of companies, like Exscientia (EXAI), Atomwise, Schrodinger (SDGR), Abcellera (ABCL), InSilico Medicine, and AbSci (ABSI), are all trying to create this platform. They are data-scientists, physicists, and chemists, using algorithmic, structural, and molecular approaches to parse antibody data. Though the antibody discovery platforms made through these approaches are far better than the status quo, their main limitation is that antibodies are biological, and data-science is struggling to model that. The real world, biology/nature, is famously messy and complicated, and these theoretical approaches struggle with that.
For example, InSilico Medicine has a platform they call pharma.ai, which does exactly what an antibody drug development platform should: it screens for, identifies, helps design, and develops probabilities of success in the clinic for antibodies. However, they recently began looking at incorporating quantum computing into their platform to boost computing power. It sounds sexy, but to an informed observer, it signifies that they haven’t devised an effective enough system to organize the infinite number of unique antibodies. They have more data than they can parse.
Let’s use an analogy to better understand: a disease is a Facebook account you’re trying to hack, and the antibody is the password. InSilico Medicine has found ways to exclude a certain number of possible passwords and has got some clues about what could be in the password, but the number of possible passwords remains so high that they don’t have a chance of guessing it, so they’ve had to recruit friends to help them guess. Simply, instead of working smarter and finding a better way to reduce the number of possible passwords, they’re working harder, and trying to brute force the password through sheer number of guesses. Most companies in the AI-powered antibody discovery platform space are facing this problem.
Frederic Chabot, IPA’s head of corporate development argues that any successful platform will need to take a biological perspective to antibody development because nature doesn’t necessarily reduce to man-made rules of data-science, chemistry, and physics. Immunoprecise Antibodies, and to some extent Recursion Pharmaceuticals (RXRX) (“Digital Biology” is their catchphrase), are the only two that have taken a biological approach from the start, though more will follow as the limitations of more theoretical approaches become clear.
Why Immunoprecise Antibodies?
IPA was founded in 1989 as a Contract Research Organization (CRO), meaning they did diagnostics on antibodies in the form of contracts from pharma companies as those companies began to outsource their R&D. As a result, IPA has a huge repository of antibody data and techniques (intellectual property) that others do not. Considering an AI is only as good as the data its trained on, having 25 years of proprietary antibody data is extremely valuable. According to 3rd-party research firm Root’s Analysis, IPA’s wet-lab capabilities are ranked number one in the world. This is corroborated by the fact 19/20 of the largest pharmaceutical companies in the world are partnered with IPA. As such, their CRO is strongly profitable, and growing revenue.
In the mid 2010s, a group of investors retired the founder of IPA and took the company public. Rather than doing diagnostic research on drugs for others, they pivoted towards drug discovery and entered the space themselves with a branch they called Talem. For most of the time the company has been public, it has attracted investors because the profitable CRO partially funds the drug discovery pipeline. Unlike most biotechs, rather than betting that one lottery ticket drug candidate does or doesn’t work out, with IPA you have a lottery ticket machine that can survive failure. Why have one lottery ticket when you can have many? The value proposition is clear.
What has got me most interested though, is the 2022 acquisition of Biostrand, a biotech artificial intelligence company with incredible intellectual property that could be used to create the most effective antibody discovery platform in the market.
What Biostrand’s co-founder and Head of Innovation, Dirk von Hyfte, stumbled upon, were patterns at the base of all genetic information in the biosphere, which he began to document and index for future reference. What he describes as his “eureka moment” was after finding 660M of these patterns, he couldn’t find anymore. This is significant simply because it’s finite — far less than the inifinite possibilities. Hyfte went on to investigate the protein structures formed from these 660M sequences, and found there are around 20M, again, far less than infinity. He termed this finite set of patterns nature uses to code antibdodies “HYFT Universal Fingerprints.”
A simplified analogy is that Biostrand claims to have the Rosetta Stone for antibodies. Like the Rosetta Stone gave Egyptologists the tools to decode hieroglyphics, Biostrand’s “HYFT Universal Fingerprints,” give IPA scientists the tools to decode antibodies. IPA has the letters (660M patterns at the genetic sequence level) used to form words (20M patterns at the structural level) which nature/biology uses to write an antibody’s meaning (code an antibody’s function). The knowledge is not complete, they still must decode many of the “hieroglyphics,” but with enough quality data and AI assistance from large-language models like ChatGPT, IPA could become fluent in nature’s antibody-language while others continue to make quantum-powered educated guesses at what their quintillions of data points may mean.
IPA recently patented this technology. This means they aren’t just making this up, the patent office agrees it is something. Secondly, it means that for competitors to catch up, they would have to discover HYFT Patterns on their own (unlikely) and index them (takes years). While that’s going on, Biostrand would have improved their knowledge, making their AI smarter than any competitor could be. IPA is in a strong competitive position.
The most important part of this article is IPA’s key claim: through their patented index of HYFT patterns (their Antibody Rosetta Stone), they are becoming literate in nature’s antibody-language. IPA is building a platform that could predict the function of an antibody by connecting the sequence-structure relationship (“letters/words/sentences”) to a specific function (“meaning”).
Significance, Implications, and an Example
I’ll briefly explain three near-term implications and provide an example after. Thanks for being patient.
Firstly, this technology is significant because they can build more predictive and accurate computer models in their drug discovery platform, allowing for better-quality drug candidates found quicker. What currently takes 10 years, billions of dollars, and often results in failure, Biostrand’s HYFT-trained, AI-powered drug discovery/screening platform can do more accurately and in a couple weeks. IPA sent out a few case studies to some partners highlighting their platform’s AI-enhanced capabilities, and all recipients replied that they wanted to be a part of trials. In the race to build the best platform, IPA would have a major advantage.
Secondly, it means they could reverse engineer the discovery process to create bio-similar drugs. In pharma, the genetic sequence of a drug is patented, not the protein structure. So, instead of starting by searching for sequences or structure to perform a medicinal function, they could start with the function, then use HYFT Universal Fingerprints to build/find a similar structure that uses an unpatented sequence. Though the function would be similar, because the sequence behind the structure is different, the new drug would not be a patent infringement, thereby creating a biosimilar drug. Imagine being able to essentially make generics out of on-patent drugs, highly disruptive.
Thirdly, just officially announced, they can use this reverse engineering to model the humanization of wildtype animal antibodies. Until now, the transgenic animal model are the most utilized for discovery work. The traderoff from these animals are a limited antibody diversity whereas using wild-types could greatly improve that diversity and yet still be able to have them in their human format.
Here’s an example to put it all together:
IPA took a rabbit’s antibody that had bound to SARS, and instead of having to alter it to be human and test it in the lab on transgenic animals, they reverse engineered the process by identifying the HYFT patterns in the structure (words) and sequence (letters) and used their AI to generate a list of human candidates that resemble those patterns. It took just a couple days, no lab required. The transgenic animal testing market was entirely circumvented.
The structurally biosimilar human antibodies did indeed have a similar function like that of the rabbit’s, even though the genetic sequence underlying it was different, based on their model. It hasn’t been proved in the wet-lab, but from my research, if it could be proved, the speed of the results, this reverse engineering of the discovery process to create biosimilars, and ability to use non-human antibodies as sources is not claimed by any other company. Again, highly disruptive.
There are further implications, but I’m sticking with these three near-term uses of Biostrand’s AI on antibody discovery for now, but the plan is likely one, prove de novo antibodies can be designed insilico, two, improve those abilities, and three, work towards personalized medicine.
That said, this company is so innovative and with so many possible uses for their tehnology, it’s hard to predict what application may be next. If we think like a venture capitalist, this is what we’re looking for.
Drawbacks and Catalysts
If the technology does what they say it does, IPA will be a leading drug discovery platform worth billions (check out the comps in the next section). The current issue is the "if." The patent is certainly a step in the right direction, partner interest in the software is optimistic, and the case studies they have disclosed (like humanizing the rabbit antibody) are certainly promising, but they are not clinical success. We lack certifiable proof of concept, though management is very confident.
Frederic Chabot pointed to an ongoing program with BriaCell, a cancer therapy company, to build new antibodies ("de novo") entirely discovered in their software ("in silico"). These de novo antibodies discovered in silico are currently being tested in the wet-lab at IPA. If the antibodies are able to bind to their target, IPA would be one of the only, if not the only, company to do so.
Institutional ownership of the stock is nil. The shareholders are mostly people like me, the mob, though there exist some long-term private investors with significant holdings. With almost no liquidity and an expired prospectus, institutions can’t and won’t buy the stock. IPA is stuck in purgatory until those issues are fixed.
A retail investor could see this as an opportunity because a tight float means any significant purchase of the stock could multiply its value. If the technology can be confirmed to be as disruptive as it appears, institutions could pile in, exponentially increasing the share price.
The company is trying to fix low liquidity and mob ownership by partnering with Jefferies, an well-known investment bank that can enable an ATM (At The Market) trading facility, allowing institutions to buy large blocks of stock and IPA to issue stock at pre-set prices. This way the stock can trade without wild volatility, they can avoid bought deal banking fees, and won't be shorted by hedge funds like they would if they had to announce a stock issuance.
Proof of concept and fixing their stock’s trading and ownership are the two main drawbacks currently, but any positive news regarding them would be strong catalysts to unlock value, with potentially exponential upside.
There is a lot of other stuff going on that is less pressing but still important which I’ll leave up to the reader to do their own due diligence on, but here are a few:
According to the most recent earnings call, they are also in talks to partner with “a highly respected multibillion-dollar global leader in the field of oncology” to finalize an AI-driven research collaboration.
Their Covid antibody cocktail they spent a ton of money on hasn’t managed to get off the ground, for reasons that are not perfectly clear. The data generated has been reportedly helpful in training Biostrand’s AI however.
Talem, IPA’s drug development branch, has not yet had clinical success, but continues to have a reportedly strong pipeline.
Biostrand has been asked to draft a proposal to use their AI to make a data management platform for a large pharma company. 30 candidates were contacted, 3 were asked for a formal request for proposal, Biostrand being one of those three.
In summary, there is substantial interest in IPA’s technology and services from partners, but there are some drags the company needs to reverse to see value unlocked.
Financials and Valuation
Take off your traditional value investing cap, forget free cash flow yield and focus on disruption potential: you’re in the world of venture capital now.
Here’s how sales, profits, and margins are looking:
Sales have been growing strongly, driven by the industry leading CRO, and with very recent additions from Biostrand. Though sales have plateaued, the more important issue for the future is whether their technology is superior/competitive to competition, and it appears to be.
Margins, from a traditional security analysis point of view, are catastrophic. We’re investing in a long duration asset though, so we care primarily about R&D. Most of the recent increase in spending was R&D expenses related to attempting to bring their covid antibody cocktail, “polytope,” to market. That seems to have stalled with no financial benefit, so they are no longer spending on it. Disappointing, but next year’s cash burn should be back to a more normal ~$7-9M from last year’s nearly $20M.
Fortunately, they have enough cash for next year, meaning they don’t have to do a raise this year. The share price is very low, raising at these levels could be more dilutive than is pleasant.
As we can see the price-to-sales ratio appears lofty at around 5 despite burning a lot of cash next year. From a traditional point of view, this seems unfavourable. But how does it compare to publicly traded peers in the AI-powered antibody discovery platform market?
Very favourably it turns out. IPA’s Head of Corporate Development, Chabot, considered RXRX (Recursion Pharmaceuticals) to be their most naturable comp, and RXRX trades at a whopping 7x higher price-to-sales ratio than IPA. The next closest in the group is Abcellera (ABCL), which no longer even advertises itself as an AI platform. Something is up.
We need more details:
Take a moment to closely read this table. IPA burns far less cash than most peers, the exception being ABCL, whose only revenues are a covid antibody therapy slowly being phased out. AbSci has 1/3rd the sales, but twice the market-cap. The median P/S for the sector is 25x, while IPA trades at 5x. It’s clear the market likes the idea of AI-powered drug discovery platforms, just not IPA, even though its technology could be superior, or at least competitive, with all of them. Relatively speaking, IPA is a value stock.
Perhaps this is because the AI component of IPA’s platform was only added in 2022 and is just truly being integrated, most investors are probably unaware of IPA as a player in the AI-powered antibody discovery platform market. It could be that a large part of the market’s undervaluation of IPA is likely to do with the unproven nature of the platform and an illiquid stock prohibiting institutions from investing. The market may simply disbelieve management. It’s hard to know.
Notably, most of the concerns around IPA are short-term. If we think with a 5–10-year time horizon, IPA appears to present a very favourable risk/reward. Unless pharma companies cease wanting to innovate or IPA experiences a catastrophe, it’s hard to see long-term downside in the stock at these levels, especially when we consider its trading at a substantial discount to peers, in a risk-off market. Should risk come back, and/or IPA close the valuation gap, investors would have a nice return. If IPA can do even half of what it claims, then investors should have a better-than-nice return.
Conclusion:
I like the company because IPA has the world's best CRO that is growing profitably, a proprietary drug pipeline at Talem which could win the lottery at any moment, and potentially the world's only AI-enabled software to make de novo antibodies completely in silico, which would change the entire landscape of drug discovery. Moreover, these branches mutually benefit each other. There is a lot more that can go right than can go catastrophically wrong.
I like the company’s stock because there is limited downside and expontential upside at the current price. Sellers appear to have tired, and with such low liquidity, when/if institutions buy the stock, the price will rise exponentially. At $3/share, the stock is in a catapult, ready to launch. It's just awaiting the order.
A goose-egg is always possible, but I consider this much less likely with IPA than other biotech companies for all the reasons listed above. Tom Hayes likes to say, “Amateurs deal in absolutes, and experts deal in probabilities,” and I like the probabilities here.
To take a large position I would need to see the platform validated and the stock tradeable by institutions, but at these levels I’m happy to dip my toe in. If their drug discovery platform is even close to what they say it could be, this stock is worth billions. The company thinks so and the market prices its competitors that way.
At the current market cap of $75M, even a toe-dip could make a dramatic difference in a portfolio, and if it doesn’t work, then you lose, at most, a toe.
I also think the science is super amazing and interesting, and love being a part of it.
Hi - I came across this write up today and found it very useful. BTW what do you make of IPA new AI model , any substance ?
Thanks for the candid take . btw great write up on Moberg, that seems to be going places . Wish the management was a bit more aggressive with marketing and upbeat about prospects , maybe just a Scandinavian thing .