Written by Anthony Zhao
Machine Learning (ML) is a powerful tool for researchers to develop new drugs.
To begin, one needs a general understanding of how drugs work:
There are many mechanisms of drug action, but current examples of efficient medicines use two key steps:
A molecule attaches to a receptor in the body, bound together like glue.
Once bound to the molecule, the receptor instigates a reaction in the cell or tissue.
Additionally, one needs to understand the concept of ML:
ML is a subtype of artificial intelligence programs. According to UC Berkeley's School of Information, ML programs have three distinct commonalities:
They can independently predict, calculate, or identify key concepts.
They each have an “error function”, which allows them to evaluate their work/outputs.
They can improve their predictions, calculations, or classifications using that error function.
Now, let’s put all that into the context of drug development! Before ML, drug developers tested plants known to have helpful effects on tissues. Then, they would isolate molecules, test them, and identify a drug–almost like a guess-and-check (Office of the Commissioner, 2018)! As you can see, this would require a lot of testing, expensive resources, and time to find a drug.
One way that ML can help is by optimizing the search for drugs from known molecules. ML programs go through thousands of known molecules that have been isolated before and simulate them interacting with a specific receptor or tissue in the body. Molecules that interact somewhat optimally would be marked as “hits”, those that do not will be marked as “non-hits”, and the program will not look for molecules like those (Catacutan et al., 2024).
Because ML can find patterns and learn from its past results, it can look for similarities in the “hit” molecules and search for similar molecules in the database; therefore, a scientist can isolate/synthesize and generate a future drug more quickly!
ML can help drug developers in the opposite direction! Instead of helping researchers determine whether preexisting molecules could be potential drugs, researchers could generate new hypothetical molecules and see if they would cause therapeutic effects in cells or tissues in machine-learning-aided simulations.
Publicly accessible programs, such as Nanome, use thousands of preexisting observations of protein shapes to predict how a receptor or protein may behave when a part of a molecule, called a “functional group,” interacts with it. Drug developers are also aided by the fact that Nanome can be used in virtual reality, making visualization of interactions much easier! Using this, drug developers may assemble molecules that behave ideally and attempt to synthesize them for laboratory testing (McCloskey, 2019).
It should be noted that ML programs serve only as a tool to help expedite the search for drugs. Still, sometimes, their predictions are not always accurate, or they may be accurate in very specific conditions. In other words, further laboratory work, testing, and approval are needed before manufacturing a drug to be sold.
Among the many ways that AI and ML are integrated into everyday life, their integration in drug discovery may be one of the most promising. A process that once took countless days of case studies and months of testing may be reduced to mere hours of computing, potentially expediting the production of life-saving drugs for many worldwide.
References:
Catacutan, D. B., Alexander, J., Arnold, A., & Stokes, J. M. (2024). Machine learning in preclinical drug discovery. Nature Chemical Biology, 20(8), 960–973. https://doi.org/10.1038/s41589-024-01679-1
McCloskey, S. (2019, February 5). Collaborative structure based drug design in virtual reality. Medium. https://blog.matryx.ai/collaborative-structure-based-drug-design-in-virtual-reality-e35892e74bbc
Office of the Commissioner. (2018, January 4). Step 1: Discovery and development. U.S. Food And Drug Administration. https://www.fda.gov/patients/drug-development-process/step-1-discovery-and-development
UC Berkeley School of Information. (2022, April 19). What is machine learning (ML)? - I school online. UCB-UMT. https://ischoolonline.berkeley.edu/blog/what-is-machine-learning/
留言