“Some people call this artificial intelligence, but the reality is this technology will enhance us. So instead of artificial intelligence, I think we’ll augment our intelligence.” Ginni Rometty.

Insilico Medicine,  a generative artificial intelligence-driven clinical-stage biotech company announced on June 27, 2023, that it has completed its first dose in patients in Phase II clinical trials. In addition to INS018_055, multiple drugs have used AI at some point in their discovery and some will also be fully discovered and developed by AI.

Last Week’s Recap

To start, we have a brief recap from last week's article on drug development and drug repurposing. Drug development is the process where a new drug goes through 5 major stages to become an FDA-approved drug available in the market. Drug repurposing is the process where an existing drug finds a new purpose and is tested before being approved by the FDA. Drug repurposing was initiated as an accident as aspirin was developed to be a painkiller but was found to be effective in preventing heart disease. There are various cases of both successful and unsuccessful drugs being repurposed. Corticosteroids are an example of a successful drug being repurposed. During the COVID pandemic, COVID was identified as a respiratory disease and scientists investigated if the anti-inflammatory action of inhaled corticosteroids might reduce the illness from hyperinflammation in COVID-19. After going through the repurposing process, it was confirmed by the Cochrane Collaboration that people with COVID-19 and mild symptoms were able to use inhaler devices to reduce the chances of admission to the hospital or death for up to 14 days.

The main difference between drug development and drug repurposing is highlighted in the picture above. New drugs that enter the market go through extensive testing to determine what illnesses they can target and extended clinical phases to ensure it is safe for human use. Repurposed drugs have already either been through or were in the process of drug development, so the previous data gives a headstart making it both faster and cheaper. In some rare cases, repurposed drugs can skip to Phase 1 of clinical trials, shortening the time and cost of the process even more. In both cases, the FDA will still review the drug to ensure its safety and approve it once deemed safe for the general public. Quantitative metrics reviewing costs and time required for each process can be seen from the diagrams. The average R&D cost to develop an asset from discovery to launch in 2023 was about $2.284 billion and according to Pillai and Wu’s et al. , they estimated the cost of repurposing a drug to be approximately $300 million. The average period for a repurposed drug is between 3-8 years, while the de novo drug development period ranges between 10-15 years. Advancements in science and the evolution of AI helped in lowering the cost and reducing time in both drug development and repurposing processes.

Introduction to AI

AI consists of various techniques that are branches of machine learning algorithms. Machine learning is a branch of AI and computer science (CS) that focuses on using data and algorithms to enable AI to imitate the way humans learn, improving over time with more data and iterations. There are three major steps in its program: decision process, error function, and model optimization process. As seen in the picture below, there are 5 main branches of machine learning: unsupervised learning, neural networks, ensemble learning, supervised learning, and reinforcement learning.  First, unsupervised learning is when machine learning algorithms analyze and cluster unlabeled data sets. It is called unsupervised because it does not require human intervention to operate. It can be applied in the earlier stages of drug development to identify, group, and develop information on new drugs to aid scientists. Secondly, a neural network is a machine learning program that makes decisions mimicking the human brain using processes to weigh options and arrive at conclusions. Neural networks rely on training data to learn and improve over time similar to an actual human. Neural networks are part of most major companies in biotech such as DeepChem, AlphaFold, PathAI, and many more. The uses of neural networks are limitless and will continue to evolve.

Third, ensemble learning combines multiple learners (neural networks) to produce better predictions. Ensemble learning techniques such as bagging, boosting, and stacking show the differences between sequential, parallel, homogenous, and heterogeneous types of ensemble methods. Ensemble learning is used to detect, monitor, and predict diseases due to its ability to process larger amounts of data compared to the other algorithms. The multiple learners act like a small sample size to create accurate predictions. Fourth, supervised learning is the process that uses labeled data sets to train algorithms that classify data or predict outcomes accurately. The main difference between supervised and unsupervised learning is the type of data they use (one is labeled the other is not). Supervised learning would be used while researching repurposed drugs as the algorithm can predict new uses for existing drugs and optimize their safety and efficacy. Finally, reinforcement learning is a process that focuses on decision-making by autonomous agents. Autonomous agents are any system that can make decisions and act in response to its environment without instruction from a human. Reinforcement learning can be used in various situations in biotech from genomic sequencing to looking for patient recruitment.

Adapted from Anbarjafari - An extended tree map shows the five major branches of machine learning with each branch listing examples on practical applications.

AI’s Impact on Biotechnology

AI’s impact in biotechnology is limitless from a simple task of a kiosk to advanced deep learning with drug sequencing, AI has integrated itself as a necessity. Ever since AI’s introduction, it has helped the entire world progress and the numbers speak for themselves. CRISPR or Clustered Regularly Interspaced Short Palindromic Repeats, is one of the technologies in the healthcare system. It allows scientists to make precise gene editing changes to living organisms and has become a cornerstone in both biological research and biotechnology. CRISPR and AI are paving the future in research and genetic manipulation with increased rates of accuracy, precision, and efficiency.

As seen in the example mentioned above, AI can reduce time and costs while increasing accuracy, precision, and efficiency, and will continue to as it evolves with time. It has revolutionized personalized medicine, optimized medication dosages, enhanced population health management, established guidelines, provided virtual health assistants, supported mental health care, improved patient education, and influenced patient-physician trust. AI can reduce the time required and costs of new drug development by 25-50% quoted in multiple articles. AI is also taking over the drug discovery stage as some programs can analyze whole-genome sequencing data faster than traditional methods, which contributes to accelerated identification. For example, Stanford cardiologist Euan Ashley and his team received a Guinness World Record for sequencing a full human genome in just over five hours. This process helps quicken the process of finding genetic markers linked to diseases, which then allows treatment plans to become more tailored to the individual. AI’s algorithms can be used to share, find and analyze data.

According to Asruddhi Yardi’s article, 38% of medical providers use computers as part of their diagnosis and more are joining as AI is considered the future in the industry. AI algorithms can achieve diagnostic accuracy rates of over 90% in medical imaging and up to 50% with early detection of diseases such as cancer with sensitivity rates compared to traditional diagnostic methods, surpassing the performance of human radiologists in certain tasks. An example of this is a CBC study conducted by Ji-Jung Jung, MD, where AI demonstrated a comparable sensitivity of 59.6% compared to the 56.5% of the reviewing radiologists. AI once again bests human diagnosis by also being able to detect CBC at least six months before diagnosis. During the study, “The researchers noted the AI software had an area under the receiver operating characteristic curve (AUROC) of 83.6 percent and a 91 percent specificity rate for CBC.” (Hall 2024). AI has also assisted in the creation of synthetic organisms by accelerating the design and testing process which helps cut costs for new bioengineered solutions. These statistics demonstrate the transformative potential of AI in biotechnology, driving significant advancements across various domains and leading to more efficient, cost-effective, and personalized solutions.

Current News with AI in Biotech

Steve Lohr’s article, How A.I. Is Revolutionizing Drug Development, opens with, “The laboratory at Terray Therapeutics is a symphony of miniaturized automation. Robots whir, shuttling tiny tubes of fluids to their stations. Scientists in blue coats, sterile gloves, and protective glasses monitor the machines.” At a lab at Terray Therapeutics, millions of interactions are recorded daily, generating about 50 terabytes of raw data that has to be filtered into groups, and with the help of AI this process occurs simultaneously while the data is being input by scientists. With that data, machine learning can predict and suggest different sequences. When talking about drug development, some companies use AI for the first stage, drug discovery and development. “For example, AI platforms like AtomNet use structure-based drug design to predict how different drug molecules will interact with the target, thereby enhancing the precision of drug development.” (Hayes-Mota 2024). Generative AI has developed enough that it can digitally design a drug molecule and test its interaction with a target protein. Major companies and industry leaders are trying to implement AI into their processes to help quicken and improve their methods. Sanofi has cooperated on AI with biotechnology companies such as Owkin, Escientia, Insilico Medicine, Amunix Pharmaceuticals, Atomwise, and Aqemia to quicken the drug discovery and development process with accurate models and predictions with the help of AI. Insilico Medicine is a biotechnology company that is powered by generative AI in drug development, chemistry, and biology and currently has the first AI-discovered and developed drug in Phase 2 of clinical trials. Insilico Medicine first mentioned using generative AI for the design of novel molecules in 2016, which laid the foundation for Pharma.AI, a commercially available platform for biotech. Dutch startup Cradle leverages generative AI to assist biologists in designing enhanced proteins and accelerating research and development. This technology nearly automates the initial phase of drug development. Over the past year, the company has collaborated with nine leading industry partners.

Nvidia and Recursion Pharmaceuticals have partnered to develop AI-driven platforms for drug discovery. Nvidia is investing in Recursion and licensing its models through the BioNeMo generative AI cloud service. If this partnership is successful AI can be used to accelerate drug development processes, which would reduce costs and time significantly. Xaira Therapeutics founded in 2023 is a startup company that formally launched with $1 billion in venture capital. Xaira seeks to build on Baker’s RFdiffusion model, which is capable of designing protein drugs that bind to influenza molecules in just weeks. As a young company Xaira has found a stroke of luck, “Xaira has several big names attached to it. Several months ago, the start-up’s cofounders tapped Marc Tessier-Lavigne, a former Genentech scientist and onetime president of Stanford University, to lead Xaira as CEO.” (Walrath 2024). Xaira plans to design protein drugs from scratch and reverse-engineer the process to design an antibody that can bind to it well and will hopefully be able to apply the same process to smaller molecules down the line. Biotechnology AI companies continue to improve their algorithms, hopefully, one-day making clinical trials autonomous and shortening the drug development process along the way.

Written and Constructed by Joshua Minami, Christopher Korban, Christian Chung

• Author links open overlay panelMahintaj Dara a, a, b, c, d, AbstractCRISPR, Barrangou, R., Bhattacharjee, G., Kafle, S., Komor, A. C., Mitra, S., Munsie, M., Piecyk, R. S., Qiu, T., Xu, L., Adli, M., Ahmar, S., Alipanahi, B., Balatti, G., … Esteva, A. (2024, May 22). Convergence of CRISPR and Artificial Intelligence: A paradigm shift in biotechnology. Human Gene. https://www.sciencedirect.com/science/article/abs/pii/S277304412400041X#:~:text=The%20fusion%20of%20CRISPR%20and,et%20al.%2C%202022)

• Bean, R. (2024, May 9). How moderna is Embracing Data & AI to transform drug discovery. Forbes. https://www.forbes.com/sites/randybean/2024/03/25/how-moderna-is-embracing-data--ai-to-transform-drug-discovery/

• Measuring the return from Pharmaceutical Innovation 2024. Deloitte United States. (n.d.-b). https://www2.deloitte.com/us/en/pages/life-sciences-and-health-care/articles/measuring-return-from-pharmaceutical-innovation.html

• Hayes-Mota, G. (2024, March 4). Council post: AI is rapidly transforming drug discovery. Forbes. https://www.forbes.com/sites/forbesbusinesscouncil/2024/02/29/ai-is-rapidly-transforming-drug-discovery/#:~:text=By%20enhancing%20target%20accuracy%2C%20AI,saving%20millions%20in%20research%20costs

• InSilicoMeds. (n.d.). Insilico medicine completes patient enrollment in its phase IIA study in China investigating ins018_055 for idiopathic pulmonary fibrosis (IPF). EurekAlert! https://www.eurekalert.org/news-releases/1048870

• Senthil, A. (2024, June 28). See Ajith Senthil’s activity on linkedin. See Ajith Senthil’s activity on LinkedIn. https://www.linkedin.com/feed/update/urn:li:activity:7212439249474736128/

• Two new patents for using AI to accelerate biotechnology. Teselagen. (2023, November 6). https://teselagen.com/in-the-news/new-patents-for-using-ai-to-accelerate-biotech/

• Walrath, R. (2024, April 25). Backed by $1 billion, Xaira Therapeutics is readying AI-generated drugs. Chemical & Engineering News. https://cen.acs.org/business/start-ups/Backed-1-billion-Xaira-Therapeutics/102/i13

• Watson, J. L., Juergens, D., Bennett, N. R., Trippe, B. L., Yim, J., Eisenach, H. E., Ahern, W., Borst, A. J., Ragotte, R. J., Milles, L. F., Wicky, B. I. M., Hanikel, N., Pellock, S. J., Courbet, A., Sheffler, W., Wang, J., Venkatesh, P., Sappington, I., Torres, S. V., … Baker, D. (2022, January 1). Broadly applicable and accurate protein design by integrating structure prediction networks and Diffusion Generative Models. bioRxiv. https://www.biorxiv.org/content/10.1101/2022.12.09.519842v2

• Wellcomeopenresearch.org. (n.d.). https://wellcomeopenresearch.org/articles/7-150

• What is a neural network?. IBM. (2021a, October 6). https://www.ibm.com/topics/neural-networks

• What is ensemble learning?. IBM. (2024, February 9). https://www.ibm.com/topics/ensemble-learning

• What is supervised learning?. IBM. (2021b, September 23). https://www.ibm.com/topics/supervised-learning

• What is unsupervised learning?. IBM. (2021, September 23). https://www.ibm.com/topics/unsupervised-learning

• Yardi, S. (2024, April 2). Medical Biotechnology Statistics 2024 by growth, gene editing. Market.us Media. https://media.market.us/medical-biotechnology-statistics/