Recursion Pharmaceuticals, Inc. (NASDAQ:RXRX) develops therapies using AI-assisted clinical research. This creates a large pool of successful drug candidates in lab and animal testing, statistically increasing the chance of success of its programs in clinical trials. The company has solid partnerships with big companies, some early and interesting data, and a number of data catalysts in the months ahead. However, their cash reserves are fast depleting.
Currently, there are five programs in Phase 2 with upcoming data readouts. These are – “REC-994 in cerebral cavernous malformation (CCM’) in Q3 2024, REC-2282 in neurofibromatosis type 2 (NF2) in Q4 2024, REC-4881 in familial adenomatous polyposis (FAP) in H1 2025, and REC-4881 in AXIN1 or APC mutant solid tumors in H1 2025.” Some of these programs represent billion dollar opportunities according to the company (see 10-K, page 9, cited below).
Thus far, RXRX is a standard biotech, but the company calls itself a “TechBio” (this term has some general usage, see here and here). Let me throw a few facts at you before I explain that term. These facts are –
NVIDIA Corporation (NVDA) recently invested $50mn in RXRX, funds that will be used to expand BioHive (they have models 1 and 2), the company’s on-premise supercomputer, and turn it into the fastest supercomputer owned by any biotech company in the world.
The company built a full-stack artificial intelligence technological solution called Recursion OS to automate the drug discovery process.
Recursion OS consists of a wet lab and a dry lab component connected by LOWE (Large Language Model-Orchestrated Workflow Engine), a natural language interface that streamlines complex drug discovery tasks. Wet lab, as you may be aware, is where biological experiments are done, and a dry lab is where computational and theoretical work is done on data generated from the wet lab.
The company deployed Large Language Models (LLMs) specifically trained to map scientific literature with internal data to find areas of unmet need.
They also deployed Phenom-1, which analyzes billions of proprietary biological images using AI. The company says this is the largest phenomics foundation model in the world. This is a type of AI model designed to analyze and interpret complex phenotypic data. Phenomics models can identify novel drug targets by observing how genotypes affect phenotypes.
(This material is adapted from here, page 6, with explanations where relevant.)
Thus, a TechBio is a high-technology company exclusively focused on biotech and drug development. Specifically, it’s a company that uses artificial intelligence to speed up and facilitate the drug discovery and development process. TechBio has two types of intellectual property work products: One, the standard biotech material, molecules that target diseases; and two, the AI system that they have used to arrive at this biotech pipeline.
How does AI assist in drug discovery and development?
Artificial intelligence is being increasingly integrated into the core processes of biotech, from research and development to diagnostics and manufacturing. One important component of that integration is leveraging AI in drug discovery and development. There are at least three elements of use here – identifying drug candidates, modeling their efficacy based on indication parameters, and optimizing their chemical structure based on target proteins and outcome models.
The drug discovery process is complex. Millions of molecules may be run through the high-throughput screening (HTS) process in the initial screening phase. This could take a few hundred thousand molecules in the hit identification stage, and perhaps produce a few hundred lead molecules. These are run through a lead optimization phase, where maybe a few dozen molecules get through to lab testing, and finally, maybe one or two get to the IND stage for a particular indication. The process is time-consuming and expensive.
AI has a huge role to play here. Its vast natural language processing ability, image processing prowess, and general neural network-enabled decision-making ability enable it to speed up the entire workflow. The result is that you get more numbers of successful molecules at every stage of drug discovery, and at great cost savings.
The BCG study
But do those higher numbers translate to more successful clinical trials? According to a Boston Consulting Group study, the percentage of success in the human trial phase remains more or less the same for these AI-assisted molecules. That is to say, while AI molecules may have an 80%-90% success rate in Phase 1, the rate comes down to around 40% in Phase 2, which is similar to what we also get in non-AI molecules.
I have doubts about this conclusion because the data is lopsided, given that in these early days of AI-assisted research, a much higher percentage of molecules are in Phase 1 trials than in later-stage trials. If you go to this link, you can see BCG’s list of AI-assisted assets in clinical trials. As you can see, most Phase 2 trials are ongoing, while a much higher number of Phase 1 trials have been completed. We do not yet know how these Phase 2 trials will turn out. Hence, the results may be skewed.
If those results are skewed, that tells me RXRX will have a higher chance of success in their later stage trials, like they claim (see here, page 8, para 4). This makes for a very interesting investment thesis for the stock. I cannot underscore enough, though, that a biotech is only as good as its pipeline. Even when it comes to a TechBio, their AI or related technology has to be proven on the battlefield of clinical trials, just as their molecules have to be.
Chances of success in RXRX Phase 2
RXRX has only Phase 1 data for some programs, and as we know, Phase 1 is usually pk/pd and safety data, while Phase 2 is efficacy data in a small patient population. Therefore, there’s no direct relation between Phase 1 and Phase 2 data, and it’s difficult to predict one from the other, except with indicative pharmacodynamic (PD’) data. Yet, there may be signs, especially when a drug-target interaction is logical or otherwise proven, and what we are left to see is the magnitude of that interaction. So, let us quickly look at the available data to see if we can find something interesting.
REC-994 is the lead molecule, which is targeting cerebral cavernous malformation (CCM’) in a Phase 2 trial, and will have data in Q3. CCM is a rare disease occurring in 15 out of 100,000 people on average and is characterized by the formation of anomalies in the small blood vessels of the brain and the spinal cord, which, in extreme cases, can cause fatal hemorrhagic strokes. There are currently no pharmacologic treatment options that address the root cause of the disease, which is loss-of-function mutations in certain genes. There are three early-stage pipeline drugs – OV-888, a ROCK2 inhibitor from Ovid Therapeutics Inc. (OVID), NRL-1049, another ROCK inhibitor from Neurelis, and Atorvastatin; all these are in Phase 1 trials.
REC-994, also called tempol, a molecule discovered by Russian scientists in the 1960s, is thought to reduce oxidative stress by mimicking the activity of superoxides. REC-994’s use for CCM is a good example of how Recursion OS works, delivering drug candidates selected using automated software that “outperformed those chosen by human analysis in subsequent orthogonal screens.” Here, there was a multistep screening process to identify potential therapeutic drugs for treating CCM. This was done by analyzing structural and functional phenotypes in both cell and animal models from a list of 2100 repurposed drugs.
The primary screen used machine learning coupled with automated immunofluorescence, a powerful tool to stain and identify specific proteins within endothelial (blood vessel surface) cells. A secondary screen narrowed down potential molecules based on their ability to restore or improve endothelial barrier function. The tertiary screen moved to animal models, identifying those drugs that reduce microvascular leak. The fourth step screened for a reduction in CCM lesion burden by the identified molecules. There were further steps which investigated REC-994’s effect in mouse models of human CCM disease using AI.
In a Phase 1 trial, endpoints were PK and safety. Safety and a robust PK profile were adequately demonstrated, with both tablet and solution formulations showing that plasma concentrations of REC-994 increased with increasing doses in both the SAD and MAD portions of the trial (SAD-single ascending dose, MAD-multiple ascending dose). There were no serious adverse events or SAEs, and the AEs that did occur were either mild or, in a few cases, moderate, but transient.
However, nothing about efficacy can be gleaned from the study, as expected. Having said that, the company points to preclinical and historical data of similar molecules in diseases with similar pathophysiology to indicate REC-994’s potential efficacy.
Early and historical efficacy signals
The basic hypothesis here is that an increase in ROS, or Reactive Oxygen Species, causes various diseases, including CCM. REC-994 is a free radical scavenger, meaning it can neutralize highly reactive free radicals that could otherwise cause oxidative stress and damage lipids, proteins, and DNA. Oxidative stress in CCM patients causes endothelial hyperpermeability (basically, the blood vessels leak).
There are historical instances of drugs with similar molecular structures that are known to reduce ROS. Some of these drugs are approved, for example, Edaravone, which is approved in the US to treat amyotrophic lateral sclerosis, “another disease with underlying pathophysiology mediated by ROS.” These drugs work in the same manner that REC-994 does, by reducing ROS, and they are approved in indications similar to CCM in the underlying pathophysiology.
This is the key argument for REC-994, and this drug was identified using the company’s AI-assisted biotech workflow model.
Indicative valuation
While it’s widely acknowledged that efforts to numerically evaluate early-stage biotech are exercises in fallacy, there’s no gainsaying that investors want to see hard numbers. Take REC-994 in CCM, for example. CCM, the company says, is a billion-dollar opportunity. That may well be true; but if REC-994 does well in CCM, and maybe gets approved, should we value this company using that billion dollar market opportunity? That can’t be right because REC-994 in CCM is just a proof of concept – proof of an idea that is quite revolutionary. This idea is not just related to the biotech Recursion Pharmaceuticals, but also to the TechBio Recursion Pharmaceuticals. An approval will validate their AI-assisted methodology, which can potentially have a much higher value than just the sum of parts of their pipeline molecules.
The current market cap of RXRX is just a little above $2bn. This is a robust figure and shows the market’s faith in the company. This is also a mere 2x the figure for the market opportunity of their lead indication. Let’s not forget, however, that besides the value of the technology itself, they also have 4 other Phase 2 programs yielding data in the next 12 months or so. That gives us a rough indication of the potential value of this company. Trading volume is robust, at around 4 million.
I covered this company very briefly 18 months ago, and there has been little movement in the market cap since then. This is a steady stock, and the catalytic events are stacked up over the next few months.
Cash position and related data
The company had cash and ST of around $296mn as of the last report. In June, the company did a $200mn stock offering, causing a fall in the stock price. They also had a $13mn revenue last quarter. Research and development expenses were $67.6 million for the first quarter of 2024, while general and administrative expenses were $31.4 million. Thus, they had about three quarters of cash remaining as of March 2024. At that rate, they should have no more than a quarter’s cash runway today, ignoring the June offering.
Partnerships
Recursion has multiple valuable partnerships with pharma giants; they also have an interesting partnership with Nvidia.
Recursion has formed strategic partnerships with Bayer Aktiengesellschaft (OTCPK:BAYRY) and Genentech, a subsidiary of Roche Holding AG (OTCQX:RHHBY) in fibrosis and neurosciences, respectively. The Bayer collaboration includes an upfront payment of $30 million and a $50 million equity investment, with potential milestone payments of up to $1.2 billion across approximately 12 programs. Additionally, there are mid-single-digit royalties on sales. Bayer is going to be the first beta user of LOWE.
The Genentech partnership entails a $150 million upfront payment, research milestones up to or exceeding $500 million, and potential per-program milestones around $300 million for up to 40 programs, along with mid-single-digit royalties on sales.
The NVDA partnership, where NVDA invested $50mn in a private placement, is a rare collaboration between a biotech and a tech giant. Recently, NVDA’s CEO Jensen Huang presented at a joint conference between the two companies, signifying the value they put to the partnership. NVDA has BioNeMo, its own drug discovery platform, which will host Recursion-built computational and data tools.
Other working partnerships include ones with Tempus AI, Inc. (TEM) giving Recursion access to over 20 petabytes of patient oncology data, a deal with Enamine to develop enriched screening libraries, and a recent partnership with Helix to access hundreds of thousands of patient records.
Competitive space
The AI-involved biotech development space has seen increasing interest. Key companies besides RXRX are Schrodinger, Inc. (SDGR), Absci Corporation (ABSI), and Exscientia plc (EXAI), while pharma giants like Bayer, Roche, and Bristol Myers Squibb Company (BMY) have taken an interest, as well as tech giants Alphabet Inc. (GOOG) (GOOGL) and Nvidia.
Each company works in different but related fields. Thus while Schrodinger’s expertise is in computational chemistry, Exscientia is known for being the first company to bring an AI-driven molecule into the clinic. Absci’s proprietary AI system is called the Denovium Engine, which is used to analyze biological data, which helps design novel protein therapeutics. Each company has a different focus on the disease area or target. RXRX is the largest company by market cap while closely following Schrodinger is the oldest.
Investment thesis
If you’re a believer in the future of AI, especially as it pertains to biotech, you will want to invest in a basket of AI-driven biotech companies; just like Bayer, for example, has invested in both Recursion and Exscientia. The list might include the four companies I mentioned in the last section. In this article, I have given some reasons for including Recursion in that list. Some of these reasons are not exclusive to Recursion, but together, they form its distinctive profile.
These reasons are a) RXRX is the largest of the TechBios, b) it has major midstage data updates this year, which may positively impact the stock, c) it has managed to get into partnerships with big pharma, even big tech, and d) early preclinical and safety data are clear and possibly even impressive.
Risks
This investment is not without its risks. While that’s nothing unusual, I must list them here so you are aware of these risks. First, we have a high burn rate that is eating away what is otherwise a decent amount of cash. That they have just about one quarter of cash runway today (before the offering, which adds another two quarters) is not reassuring.
There’s also the fact that RXRX is a pioneer in a yet-to-be-proven field, although with immense potential. They have no clinical efficacy data yet. Until they do, this investment is considered speculative.
Editor’s Note: This article discusses one or more securities that do not trade on a major U.S. exchange. Please be aware of the risks associated with these stocks.
Read the full article here