Automation and Machine Learning: a Look into Recursion Pharmaceuticals
““...tangible goal of discovering and repurposing drugs for a hundred rare diseases in just ten years...””
Artificial intelligence (AI) has the potential to surpass the limitations of human knowledge.1 Since IBM’s AI called “Deep Blue” famously defeated the world chess champion Garry Kasparov in 1997, the field of “machine learning” has advanced tremendously.2 With processing chips getting smaller and faster each year, and data storage becoming more affordable, corporations such as Google are integrating AIs into almost every facet of their company and job tasks previously performed by humans such as data mining and statistical analysis, are being replaced by machine automation.3 Recursion Pharmaceuticals, a start-up pharmaceutical company, based in Salt Lake City, is an emerging biotechnology and drug discovery powerhouse that is integrating advanced AI technology into experimental rare disease research.4 With an ambitious but tangible goal of discovering and repurposing drugs for a hundred rare diseases in just ten years, Recursion Pharmaceuticals has already raised 57.8 million dollars from both private and public funding since its founding in 2013.4
““In fact, only one in 5,000 novel therapeutics developed will make it to the commercialization stage.””
The current research paradigm for rare disease therapeutic development can be summarized into five main processes: 1) identification of gene(s) related to disease pathology, 2) elucidation of pathological molecular mechanisms, 3) identification of drug targets within the molecular pathway, 4) development of drug compounds, and 5) clinical trials to test drug efficacy.5 Drug discovery and development in the context of this paradigm, however, typically costs over a billion dollars and more than a decade of research and testing for a drug to move from the bench to the market.6 In fact, only one in 5,000 novel therapeutics developed will make it to the commercialization stage.6 Many of these drugs, which took decades of research and millions of dollars of funding to develop, are currently not utilized. Recursion hopes to make use of both drugs currently on the market and drugs in stock to screen for their utilization as a potential rare disease therapeutic rather than developing entirely novel drug compounds.6 This process is called the repurposing of drugs, and will significantly cut down the cost and time of rare disease therapeutic research. Additionally, if the efficacy and safety of the drug have already passed the clinical trials phase of FDA approval, then the drug can be utilized to treat the rare disease of interest immediately.
““Human biology is incredibly complex and involves many interactions between cells, molecules, and intracellular signaling cascades that scientists are still trying to uncover.””
The current research paradigm for rare disease research only identifies therapeutics that target the mechanisms that are known to scientists. The gene(s) related to disease pathology may produce protein products that have an unknown pathological impact to mechanisms that are currently undescribed. Human biology is incredibly complex and involves many interactions between cells, molecules, and intracellular signaling cascades that scientists are still trying to uncover. With an incomplete understanding of these interactions, rare disease research is limited to what is known rather than embracing the complexity of what is not known in human physiology. One can argue that our preconceived understanding of human biology may limit rare disease research because scientists look for what they suspect is happening rather than having an agnostic approach.4 However, with latest advances in machine learning and robotic automation that allows a computer algorithm to analyze many images and data, one can run experiments and perform analysis for hundreds of diseases simultaneously, free of inherent biases of what is known. In other words, integrating an artificial intelligence program that performs data analysis typically conducted by trained biologists may provide a novel means to research rare diseases.4
““In other words, integrating an artificial intelligence program that performs data analysis typically conducted by trained biologists may provide a novel means to research rare diseases.””
Recursion Pharmaceuticals is bypassing steps three and four in the five paradigms of rare disease research through the integration of high-resolution microscopic imaging and high-throughput screening with artificial intelligence.4 High throughput screening is the process of drug discovery where many chemical compounds are assayed in parallel against disease targets through automated machinery. It allows pharmaceutical companies to screen thousands of drugs per week and drastically speed up the drug discovery process.7 The genetic engineers at Recursion Pharmaceuticals first create a human rare disease cell model by inserting mutations into the gene associated with pathology (step one).8 As of March 2016, the company has developed over 200 rare disease cell models.8 These cells can be plated and used to test the efficacy of a diversity of drugs using automated high-throughput screening technologies. Instead of elucidating the mechanisms of the drug targets, Recursion Pharmaceuticals determines the disease cell model morphology and phenotype using microscopic imaging technology and an open-source analytical software called CellProfiler, which was developed by researchers at Harvard and MIT to provide biologists with means to measure and analyze cells.4 This process is known as “cellular fingerprinting” and involves taking thousands of high-resolution pictures from different angles to determine the aberrant morphology of the cells that is unique to the rare disease of interest.4 By taking follow-up microscopic images of the cells after treatment with a potential therapeutic, one can determine the efficacy of the drug to restore normal cellular morphology.8 This allows one to avoid elucidating the full molecular pathway of disease pathology, and instead, utilize image recognition technology to look at changes to cellular features as a means to determine the effectiveness of drugs. This is where artificial intelligence comes into play. The analysis of thousands of cellular features in conjunction with the applications of a wide variety of drug types requires a self-learning software that can process coinciding parallel information simultaneously. Recursion Pharmaceuticals currently take around 200,000 images weekly and screen more than 2,000 known chemical compounds for each rare disease model.8 In less than a year since their founding in 2013, they identified 12 potential therapeutics for seven different rare diseases and are currently moving towards pre-clinical development.8
Christopher Gibson, the founder of Recursion Pharmaceuticals, witnessed the potential of mixing artificial intelligence and rare disease research first hand while he was studying cerebral cavernous malformation (CCM), a rare disorder where blood vessels in the brain weaken and break, causing blood leakages that lead to strokes.8 In an interview with the National Institute of Health, Gibson recalls how he and his team created a CCM cell model using genetic engineering, where blood vessel cells exhibit abnormal structure.8 He then had a pair of professional cell biologists compete against an AI system to screen for drugs that would reverse the CCM cellular defects. Both the biologists and the AI tested the efficacy of the drugs by analyzing microscopic images of the CCM cells before and after the application of a potential therapeutic and ranked the drug’s behavior to normalize the aberrant cellular behavior of CCM. While both the AI and the biologist duo identified 39 drugs to have therapeutic potential against CCM, a follow-up study determined seven of the compounds identified by the computer system to reverse the effects of CCM whereas only one compound from the biologist set had a notable efficacy as a potential therapeutic. “That moment convinced me that the computer was seeing what humans can’t,” said Dr. Gibson in an interview with the NIH.8 By integrating the latest image recognition technology and high-resolution microscopic imaging in combination with state of the art automation and self-learning computer algorithms, Recursion is successfully cutting down the time and cost of drug discovery.8
Recursion Pharmaceuticals represents the future of pharmaceutical research and is a new beacon of hope for millions of rare disease patients devoid of treatment options. By integrating the rapidly advancing field of artificial intelligence into their drug discovery process, the diversity of rare disease treatments is set to increase exponentially within the next decade. As machine learning and computer processing get faster and more sophisticated, researchers are given the proper means to tackle the deficiency of rare disease therapeutics currently in the market. The price of these therapeutics can be provided at a much lower cost to rare disease patients because of the drug repurposing process utilized by Recursion, making treatments more affordable. The intersection of computation and biology offers a novel approach to rare disease research and provides a tangible means to look for therapeutics for multiple rare diseases simultaneously.
Works Cited:
1. Naik G, Bhide SS. Will the future of knowledge work automation transform personalized medicine? Appl Transl Genomics. 2014;3(3):50-53. doi:10.1016/j.atg.2014.05.003.
2. Kasparov G. Strategic intensity: a conversation with world chess champion Garry Kasparov. Harv Bus Rev. 2005;83(4):49-53,132.
3. Tsafnat G, Glasziou P, Choong MK, Dunn A, Galgani F, Coiera E. Systematic review automation technologies. Syst Rev. 2014;3:74. doi:10.1186/2046-4053-3-74.
4. Recursion Pharmaceuticals. http://www.recursionpharma.com/. Published 2017.
5. Pandey S, Pandey P, Tiwari G, Tiwari R. Bioanalysis in drug discovery and development. Pharm Methods. 2010;1(1):14-24. doi:10.4103/2229-4708.72223.
6. O’Brien A. Old drugs, new uses. Long Short. 2017. http://thelongandshort.org/life-death/old-drugs-new-uses.
7. Liu B, Li S, Hu J. Technological advances in high-throughput screening. Am J Pharmacogenomics. 2004;4(4):263-276.
8. Recursion Pharmaceuticals Helps Launch New Platform for Rare Diseases Drug Discovery. NIH. 2016. https://sbir.nih.gov/statistics/success-stories/recursion.
Cite This Article:
Chon J., Zheng K., Chan G., Ho J. Automation and Machine Learning: a Look into Recursion Pharmaceuticals. Illustrated by P. Taarea. Rare Disease Review. September 2017. DOI:10.13140/RG.2.2.19921.43368.
