The use of artificial intelligence for drug discovery and development is a promising idea.
Both excitement
Scientists, investors, and the general public are sceptical about artificial intelligence .
Artificial intelligence is
Some companies and researchers claim that AI is taking over drug development. In the last few years, research and investment have exploded in AI-driven platforms that design drugs and optimise trials. AI-driven platforms such as
AlphaFoldwas the winner of the
The 2024 Nobel Prizewill be awarded for its ability predict the structure of proteins, and design new ones. This will showcase AI’s capacity to accelerate drug discovery.
AI in drug discovery
Some industry veterans warn that this is “nonsense “. They say that “AI’s ability to accelerate drug discovery is a potential that needs to be explored.”
“Reality check” as AI-generated drugs are yet to demonstrate their ability to address the
90% failure ratefor new drugs in clinical trial. AI has not been as successful as it was in the past.
Image analysis and its effect on drug discovery is still unclear.
Nortonrsx/iStock via Getty Images Plus (#19659006) We have been monitoring the use of
AI in drug developmentin our work as a
Pharmaceutical scientist both in academia and industry, and as a
Former program manager at the Defense Advanced Research Projects Agency (or DARPA). We argue that AI is not a game changer in drug development, nor is it utter nonsense. AI is not some black box that can transform any idea into gold. We see AI as a tool, which, when used intelligently and competently could help address the causes of drug failures and streamline the process.
The majority of AI work in drug development aims to reduce the
Time and money are both factors that determine the time and cost to bring a drug to market. Currently, it takes 10 to 15 years to bring a drug to market and between $1 billion and $2 billion. But can AI really revolutionise drug-development and improve success rates.
Researchers are using AI and machine learning in
Every stage in the drug development process. This includes identifying target organs, screening candidates, designing drug molecule, predicting toxicity, and selecting patients that might respond to the drugs best in clinical trials.
From 2010 to 2022.
20 AI-focused startups found 158 drug candidates. 15 of these advanced to clinical trials. Some of these drug candidate were able to finish preclinical testing and enter human trials within 30 months.
3 to 6 years This achievement demonstrates AI’s ability to accelerate drug development.
Unlike fields such as language processing and image analysis, where large datasets are available for training AI models, the AI used in drug development is limited by small datasets of low qualityIt is difficult to create drug-related datasets for cells, animals or human beings. AlphaFoldhas been a breakthrough for predicting protein structure, but its precision in drug design is still unknown. Minor changes in a drug’s structural makeup can have a significant impact on its effectiveness in treating disease.
Survival bias
Past innovations in drug development, such as computer-aided designthe Human Genome Project ( ) and high-throughput testing have improved individual steps in the process over the past 40 years. However, drug failure ratesstill haven’t decreased.
When given high-quality data or specific questions to answer, most AI researchers are able to tackle specific tasks within the drug development process. They are not familiar with the full scope of drug development and they often reduce challenges to pattern recognition problems or refinement of specific steps in the process. Many scientists with experience in drug development are not trained in AI or machine learning. These communication barriers can prevent scientists from identifying the root cause of drug failures and moving beyond the mechanics.
Current drug development approaches, including those that use AI, may be falling into a survivorship biased – focusing too much on less critical aspects while ignoring major problemswhich contribute to failure. This is similar to repairing the damage to the wings on aircraft returning from World War II, while ignoring the fatal vulnerabilities of engines or cockpits in the planes that did not make it back. Researchers tend to focus too much on the individual properties of a drug rather than its root causes.
Martin Grandjean and McGeddon from the US Air Force/Wikimedia.com CC BY SA
Currently, drug development is done in an assembly line-like manner relying on a checklist approach with extensive testing for each step. AI may reduce the cost and time of the preclinical lab stages of this assembly-line, but it is unlikely to increase the success rates of the more expensive clinical stages which involve testing on people. This limitation is highlighted by the 90% failure rate for drugs in clinical trials despite 40 years’ process improvements.
Addressing the root causes
Drugs failing in clinical trials in clinical trials cannot be attributed solely to the design of these studies. The wrong drug candidateschosen to test in clinical trial are also a major contributor. New AI-guided approaches could help to address both of these issues.
At present, three interdependentfactors
are responsible for most drug failures – dosage, safety, and efficacy. Some drugs fail due to being too toxic or unsafe. Some drugs fail because of their ineffectiveness, or because the dose cannot be increased without causing harm.
Our colleagues and I propose a machine-learning system that can help select drug candidates. It predicts dosage, safety and efficacy based on five previously ignored features of drugs. Researchers could use AI models in order to determine the specificity and potency of the drug in binding to known and unknown target, the concentration of the drug in healthy and diseased tissue, and its structural properties. These AI-generated drug features could be tested using phase 0+ trials (also known as 19459161) in patients with mild and severe disease. This could help researchers find the best drugs and reduce the costs associated with the current “test and see” approach to clinical trial.
AI alone may not revolutionise the drug development process, but it can help to address the root cause of why drugs fail, and streamline the lengthy approval process.
Duxin Sun, Associate Dean for research, Charles Walgreen Jr. Professor of Pharmacy and Pharmaceutical Sciences at the University of Michigan (19459161) and Christian Macedonia (19459161), Adjunct professor in Pharmaceutical Sciences at the University of Michigan (19659029). This article was republished by The Conversation using a Creative Commons License. Read the original article.
