AI is being applied to medicine in fields such as disease transmission prediction and drug discovery...
Artificial intelligence (AI) has already had a broad application in manufacturing previously, where it enables a machine to perform a single task, such as product inspection or assembly, very well. Recently, another type of AI, which allows a machine to perform tasks that involve more sophisticated analysis, is being applied to medicine in fields such as disease transmission prediction and drug discovery.
AI and disease transmission prediction
The modeling and prediction of disease transmission are critical because they offer insights and potential strategies for disease prevention and patient quarantine and treatment.
In recent years, academic researchers have started to apply AI to disease prediction.
For example, a group at the Massachusetts Institute of Technology used AI and mobile data to simulate and predict dengue fever spread in Singapore. Also, a group in the Philippines used machine learning to identify weather and land-use factors that affect the transmission of the same disease.
Another example is HIV. Identifying HIV patients to prevent the spread of the disease is a pain point. In the United Kingdom, one-quarter of the HIV carriers do now know that they are infected. A group of investigators employed algorithms to locate this group of people more effectively and prevent 5% of new infections without altering the behavior of the HIV carriers. Lastly, with machine learning, a team in the U.S. used animal behavior and ecology data to predict which type of bats were most likely to spread Ebola.
In 2020, the Covid-19 pandemic, caused by the SARS-CoV-2 virus, has propelled the use of AI in disease prediction to the global stage.
In as early as late 2019, BlueDot, a Canadian company that has successfully predicted the spread of the Zika virus in Florida in 2016, was the first to identify a cluster of “unusual pneumonia” cases in Wuhan, China; the cluster of cases eventually snowballed into the Covid-19 pandemic. Later, using diverse datasets such as global airline ticketing data, BlueDot was able to identify cities that were most likely to receive Covid-19 cases. On the other hand, Tencent AI lab used AI to predict the likelihood of a patient infected with Covid-19 to develop critical symptoms.
Right now, AI is used to project the acceleration, peaking, and decline of Covid-19 cases in the U.S. Such information will help governments allocate resources to specific regions or cities or redistribute patients to balance the burden on multiple healthcare systems. On the other hand, a 2017 study used AI simulation to identify different strategies to treat the hepatitis C virus within budgets of varying sizes. When the budget was limited, it was best to focus on early treatment; when the budget was bigger, more money on screening and testing would be effective.
In the future, AI is likely to have a presence in the prediction of future outbreaks.
AI and drug discovery
In addition to helping design vaccines, AI is also assisting in the design of drugs against Covid. Actually, AI has the potential to accelerate drug discovery in general.
The drug discovery process starts with the identification of a target gene that is involved in a disease. The process aims to identify a compound that can interact with the gene and interrupt the disease’s formation.
Drug screening, or the process of identifying the appropriate compound, used to involve the testing of many libraries of compounds to identify a “hit”; such strategy often has a low rate of success. Drug development is enormously expensive because many drug candidates fail during the clinical trial, making the previous investment in them a sunk cost. By one estimate, a drug takes over $900 million to develop, obtain regulatory approval, and launch in the market.
As a result, rational drug design is increasingly considered a more cost-effective approach to maximize the match between a drug target and the drug candidate and increase the chance of a drug to succeed in clinical trials.
AI can help researchers mine the knowledge in biomedical databases and literature, such as protein structures, protein-ligand interactions, and genetic sequences, to optimize drug design. Meanwhile, in the lab, AI and robotic automation can help scientists test more compounds quickly and accurately.
More than ten years ago, a pair of robots, aptly named Adam and Eve, were already able to predict the function of a yeast gene and identify a common ingredient in toothpaste as a potential treatment for drug-resistant malaria parasites.
Now, hundreds of small biotech companies are using AI for their drug discovery process. In 2017, more than 200 startups were already using AI in drug discovery. For example, researchers at Berg, an American company, use its AI platform to analyze an immense amount of data from patients to identify previously unknown cancer mechanisms. With such insight, they were able to discover new drug targets and drug candidates.
Sometimes there is insufficient information on an identified target protein. Previously, investigators could only wait for more research on the protein. Oxford-based Exscientia’s algorithm enables researchers to compare what is known about the target protein against a massive database of protein interactions so that researchers can identify possible compounds that might work on the target despite its limited information. Also, since different proteins often work together in a mechanism, it is sometimes insufficient to target one protein. By analyzing a massive amount of genetic data from across the globe, Toronto-based Cyclica is screening for compounds that work on multiple target proteins at a time.
Furthermore, AI can help to repurpose existing drugs to treat a different disease.
Repurposing drugs will not only accelerate clinical testing but also reduce the cost. For example, BenevolentBio, a London-based startup, used AI to analyze data from primary research publications, patents, clinical trials, and patient records to identify existing drugs that may be used to treat neurological diseases such as amyotrophic lateral sclerosis or ALS. They narrowed about 100 existing compounds to five and found that four out of the five compounds showed promise.
Many big pharma companies, such as Pfizer, GSK, Genentech, have started collaborations with AI companies, hoping to increase the odds of success of their chosen candidates and reduce the overall cost of drug development.
AI has tremendous potential in medicine in terms of disease transmission prediction and drug development. However, AI is utterly dependent on Big Data for training and analytics. Global, open, and comprehensive data sharing will be essential, and it will be up to international, national, and local governments to drive data sharing.
Also, the sharing of know-how in data training is critical. Sharing pre-trained and validated AI models can expedite the application as well as reduce the cost of AI, especially in resource-poor regions.
This article was first published on EE Times Europe