2023-07-17 17:04:40 ET
Summary
- Recursion, a biopharma company, uses a data-driven approach to drug discovery, launching four clinical trials and planning a fifth in oncology.
- Recursion's software platform, the Recursion OS, is used to construct and navigate vast datasets, enabling the identification of patterns in the data that would be impossible for humans to detect.
- Despite significant revenue growth, Recursion's ability to generate revenue is still very hard to predict, and therefore the dilution risk looms large for shareholders.
The efficiency in biopharma has been flat for the past 20 years , despite the efforts of countless scientists. In contrast, the technology industry has seen rapid improvements, leading to significant changes across various sectors.
In 2013, a company by the name of Recursion ( RXRX ) was established, recognizing a gap between the two prevailing trends. The inefficiency in the biopharmaceutical sector is largely due to the intricate nature of biology, which is currently beyond our full understanding.
To tackle this complexity, Recursion employs a unique approach to drug discovery that is akin to using a map. Rather than leaning on existing literature to form preliminary target hypotheses, Recursion allows data to guide every hypothesis. This data, generated on a large scale, is designed to be comparable over time, facilitating learning and iteration both within and across programs.
Pipeline
In 2022, Recursion has made noteworthy progress using their distinctive approach to drug discovery. They've launched four clinical trials in the first three quarters of the year and are planning a fifth in the field of oncology. Additionally, they have several other oncology programs progressing towards Investigational New Drug-enabling studies, with over a dozen more projects in the pipeline.
Recursion has also announced collaborations with Bayer in the area of fibrosis and Roche Genentech in neuroscience and oncology, and has made significant strides in these partnerships. They've constructed one of the largest relatable, proprietary datasets in biology, which they believe gives them a competitive advantage . They argue that while machine learning can become commoditized, combining it with a definitive dataset can create a powerful tool and a competitive moat.
Recursion has five programs at the clinical stage, focusing on oncology and rare diseases. The first program, REC-994, is in the second phase of clinical trials for a condition called cerebral cavernous malformation. This condition affects the brain and spinal cord and is more common than cystic fibrosis. Using their platform, Recursion discovered an unexpected target and developed a molecule that has shown encouraging results in lab and animal tests.
Their second program is a combined Phase II/III trial for neurofibromatosis type 2, a condition that causes tumors to grow on nerve tissue. They discovered a specific interaction of a group of inhibitors (HDAC inhibitors) with the NF2 system and have licensed a molecule that was previously in clinical trials to use in this trial.
The third trial, REC-4881, is for familial adenomatous polyposis, a condition that causes polyps to grow in the large intestine. They discovered an unexpected interaction when they deactivated the APC gene and identified a significant effect from a group of inhibitors (MEK inhibitors). This program was initially part of a collaboration with Takeda, but was brought back into Recursion after Takeda shifted away from rare diseases.
The fourth program, REC-3964, could potentially prevent recurrent disease and be used as a secondary preventive therapy in high-risk patients with C. difficile infections. C. difficile infections are a leading cause of diarrhea induced by antibiotics and a major cause of illness and death. C. difficile-induced diarrhea arises from the disruption of normal bacteria in the colon, with toxins A and B secreted by the bacterium causing significant illness.
REC-3964 was identified using Recursion's phenomics approach, which identified changes in cells associated with the disease-causing changes resulting from exposure to C. difficile toxins. It could be used to prevent recurrent disease and potentially used as a secondary preventive therapy in high-risk patients, including elderly patients with weakened immune systems in long-term care facilities. Unlike antibiotic treatments that can eliminate the gut bacteria and further enhance C. difficile infection, this toxin-targeted mechanism would not be expected to negatively impact the gut microbiome.
They have a fifth clinical program for cancers caused by mutations in either the AXIN1 or APC genes, where they are using REC-4881. They are planning a trial and are doing a lot of modeling because they want to go after a basket trial looking at different tumor types that could be driven by mutations in either AXIN1 or APC outside of colorectal cancer.
In addition to their clinical programs, Recursion has two advanced new chemical entities (NCEs) in the preclinical space in oncology.
All-in-all, they have a prolific pipeline, even if most of these therapies are years away from commercial stage.
The role of software
Recursion is on a mission to transform the traditional drug discovery process, which typically resembles a narrowing funnel, into a more efficient 'T' shape. The aim is to identify failures early on when they are less costly and to accelerate the overall process. This is achieved by merging various layers of data and developing software that visualizes their interconnections.
A key asset of the company is its software platform, the Recursion OS. This platform is used to construct and navigate their vast datasets. The process begins in a large automated lab where robots perform tasks equivalent to a PhD's workload every 15 minutes. This enables Recursion to carry out up to 2.2 million experiments per week, generating over 21 petabytes of unique biological data. This data volume is comparable to that of some of the largest pharma companies, but it is specifically designed for machine learning and training neural networks.
This setup allows them to identify patterns in the data that would be impossible for humans to detect. To date, they have made over three trillion predictions about relationships across biology and chemistry. When they uncover intriguing insights, they validate them using scaled transcriptomics, a high-dimensional assay based on transcription. They can do this on a large scale, with 15,000 samples a week near whole exome levels. They have worked on more than 48 human cell types, including primary cells, and have scaled the production of iPSC-derived neurons to a high level.
Recursion has discovered that only a small percentage of genes have a small molecule that replicates the knockout of that gene. This suggests that a significant amount of functional biology is not being targeted by existing small molecules. To address this, they are using machine learning to expand their small-molecule library and potentially using other modalities in the future to target these genes. This innovative approach likely attracted Nvidia ( NVDA ) to invest in the company .
Looking to the future, Recursion is exploring the potential of using machine learning to predict what molecule they should order next. They are also considering investing in automated microsynthesis, which could significantly reduce the time it takes to run a SAR (Structure-Activity Relationship) cycle. This would involve synthesizing molecules in-house that their AI systems suggest they should explore next, which could dramatically shorten the period of time invested in translating each of the molecules from their platform.
Valuation and Risks
Evaluating Recursion's revenue is a complex task. The company has received cash payments for research that is still in progress, which complicates the accurate definition of the revenue stream.
Recursion seems to be pioneering a research-as-a-service model for business development. Despite experiencing significant revenue growth over the past three years, the company is still in a cash-burn phase.
The cash-flow metrics add to the confusion. The operating cash-flow loss was less than the earnings loss, not because of substantial stock-based compensation, which is often the case with emerging stocks, but due to a large amount of cash received that corresponds to unearned revenue tied to future research and development milestones. This adds a layer of complexity and uncertainty to the company's evaluation.
The balance sheet also presents a puzzling picture due to the presence of unearned revenue. For example, the current ratio appears to have worsened compared to last year because of this. On a brighter note, the current cash position seems sound, with cash on hand covering more than five times the company's current liabilities.
However, estimating the true value of the cash-burn is challenging due to the difficulty in estimating the recurring nature of the revenue. The first quarter results revealed a $73 million cash-flow loss, a $150 million degradation relative to the prior year. Without revenues from research milestones, I estimate the company could burn over $200 million per year. In that case, its cash holdings would only suffice for about two and a half years.
Considering the company is now valued at $2.4 billion, without a solid benchmark for recurring revenue and likely to be highly dilutive in the coming years, the current valuation may prove excessive unless it demonstrates the ability to derive revenues from its research-as-a-service model.
Nvidia's PIPE deal with the company adds $50 million in cash but also introduces dilution. While it is a vote of confidence in the company's technological approach, I don't see it as sufficient to warrant an investment at the current levels. I'll continue to monitor this one and may revise my opinion later, as I did with Arcturus (ARCT), which is pursuing a similar approach to revenue generation, and in the last few quarters, it has shown a great ability to de-risk by generating revenues from its research programs.
For further details see:
Recursion: Shaking Up The Biopharma Landscape With A Bold, Data-Driven Blueprint