Artificial intelligence and drug development.. Are we on the cusp of a revolution?
The use of artificial intelligence in the field of drug development is causing a wide debate among various stakeholders, from scientists, researchers, investors, and the general public, and this debate oscillates between cautious optimism and deep skepticism.
Some researchers and companies are adopting an ambitious view that sees artificial intelligence as a driving force to transform the course of the pharmaceutical industry, pointing to the significant increase in research and investments in this field in recent years.
The success of models like Google's AlphaFold, which won the 2021 Nobel Prize in Chemistry for its ability to predict protein structures and design new proteins, is strong evidence of the potential of artificial intelligence to accelerate the pace of drug development.
However, some veteran experts in the pharmaceutical industry have warned against overestimating the potential of artificial intelligence, describing the use of artificial intelligence in drug discovery as nonsense, and demanding a realistic examination of its potential to accelerate drug discovery.
This skepticism is based on the fact that drugs developed using artificial intelligence have not yet proven their ability to reduce the high failure rate of new drugs in clinical trials, which is 90%, and they compare the success of artificial intelligence in other areas, such as image analysis, and its hitherto unclear effect in drug development, raising questions about its effectiveness in this complex field.
So the question is: Can Artificial Intelligence really revolutionize drug development and significantly improve success rates
First, applications of artificial intelligence in drug development:
Researchers apply artificial intelligence and machine learning at all stages of the drug development process, from identifying drug targets in the body, screening potential drug compounds, designing drug molecules, predicting toxicity, to selecting patients who may respond best to drugs in clinical trials.
We have already seen some tangible successes in using artificial intelligence to accelerate drug development, during the period from 2010 to 2022, 20 startups specializing in artificial intelligence were able to discover 158 potential drug compounds, 15 of which reached the clinical trial stage.
More importantly, some of these compounds were able to complete preclinical tests and move on to human trials in just 30 months, compared to the typical duration of 3 to 6 years. This achievement clearly demonstrates the potential of artificial intelligence in accelerating the process of drug development and reducing the time needed to roll out new treatments to patients.
However, while AI models show a remarkable ability to quickly identify promising drug compounds in controlled laboratory environments, or in animal models that simulate some aspects of human diseases, the success of these compounds in the critical stages of clinical trials, conducted on humans, remains highly questionable, as clinical trials are the stage where the majority of new drugs fail, showing unexpected side effects or ineffectiveness required.
Although the alphafold model is an important achievement in the field of predicting protein structures, which is an essential step in understanding how drugs work, but its accuracy in designing effective drugs is still uncertain, because very small changes in the molecular structure of the drug, which may seem at first glance not important, can significantly change how the drug interacts with the body, and then completely affect its therapeutic effectiveness and ability to treat the target disease.
Secondly, what are the challenges of applying artificial intelligence in drug development?
Similar to artificial intelligence, previous innovations in the field of drug development, such as: Computer-Aided Drug Design, The Human Genome Project, and high-throughput screening technologies, have seen noticeable improvements in specific steps of the development process over the past forty years, however, the failure rate of drugs has not decreased significantly, suggesting that the problem may be deeper than just optimizing individual steps in the process.
Most AI researchers have a high ability to tackle specific tasks within the drug development process, especially when provided with high-quality data and clear and specific research questions, but these researchers often lack a comprehensive understanding of the full scope of the complex drug development process, which leads them to reduce challenges to problems in pattern recognition and optimize individual steps in the process.
On the other hand, many researchers with long experience in the field of drug development lack the necessary training and knowledge in the fields of artificial intelligence and machine learning, and this knowledge gap forms effective communication barriers that hinder fruitful cooperation between the two sides, prevent them from bypassing the mechanisms of current development processes and identifying the root causes of drug failure.
It can be said that current methods of drug development, including methods using artificial intelligence, have fallen into a trap known as the survival Bias, which consists of focusing excessively on the least important aspects of the process, ignoring the main problems that contribute most to failure.
This situation is similar to the repair of damage to the wings of aircraft returning from the battlefields of the Second World War, neglecting fatal weaknesses in the engines or the cockpit of aircraft that never returned.
Similarly, researchers often focus on how to improve individual drug characteristics, such as efficacy or toxicity, rather than addressing the root causes of failure,
This limitation confirms the persistence of a 90% failure rate of drugs in clinical trials, despite 40 years of improvements in the process, and suggests that the problem lies not only in the efficiency of individual stages, but in the overall approach to drug development.
Addressing the root causes of drug development failure:
The failure of drugs in clinical trials is not limited only to the design of studies; the selection of incorrect drug candidates for testing is a decisive factor contributing to high failure rates, and new AI-guided strategies offer promising solutions to meet both challenges: the design of studies and the selection of candidate drugs.
Third: Addressing the root causes of drug development failure:
Dosage: some drugs fail because it is difficult to determine the ideal dose; low doses may be ineffective, while high doses may be toxic or cause dangerous side effects.
Safety: other drugs fail due to their high toxicity or lack of safety for Human Use, and these drugs may cause organ damage or lead to dangerous health complications.
Effectiveness: many drugs fail because they are considered ineffective in treating the target disease, often due to the inability of the drug to reach the biological target or not react with it in the required form, or because it is not possible to increase the dose given to the patient to a higher level without causing harm or unlikely side effects.
Fourth, What does the future hold for artificial intelligence in drug development
Dr. Duxin Sun and Dr. Christian Macedonia propose a system based on machine learning to contribute to the selection of candidate drugs with high accuracy, by predicting their three crucial characteristics, namely the appropriate dosage, safety, and efficacy, based on five features of drugs that are often ignored in traditional methods.
This system allows researchers to use artificial intelligence models to analyze these features and predict the likelihood of drug success in clinical trials, the five features are:
The extent to which the drug is associated with known and unknown targets: the system analyzes the extent to which the drug is associated with known biological targets causing the disease, in addition to its ability to associate with other unknown targets that may affect its effectiveness or safety.
The level of these targets in the body: the system determines the level of concentration of biological targets in various tissues and organs of the body, which helps in understanding how the drug interacts with these targets.
How much the drug is concentrated in healthy and diseased tissues: the system predicts how much the drug is concentrated in healthy and diseased tissues, which helps in assessing how much the drug reaches the place of injury and avoids impact in healthy tissues.
Structural properties of the drug: the system analyzes the chemical and structural properties of the drug, helping to understand how it interacts with biological molecules in the body and predict its pharmacological properties.
Pharmacokinetic properties of the drug: the system analyzes how the drug is absorbed, distributed, metabolized, and excreted by the body, helping to determine the optimal dosage and avoid side effects.
These five features of drugs generated using artificial intelligence can be tested in the so-called (Phase 0+ trials) phase 0+ trials, where extremely low doses of the drug are used on patients with both severe and mild diseases. This approach can help researchers challenge the ideal drugs to continue developing them, while reducing the high costs of the current (test and control) approach used in clinical trials, which relies on trying a wide range of drugs and evaluating their results.