A new artificial intelligence model that can predict whether a drug will be toxic to humans by analyzing the biological differences between humans and animals has been developed by a research team in South Korea. This new technology has the potential to significantly reduce the number of drugs that fail in clinical trials due to unforeseen side effects, which would save time and money in the drug development process and, most importantly, make new medicines safer for patients. The AI model was developed by a team led by Professor Sanguk Kim at the Pohang University of Science and Technology (POSTECH).
The new AI system addresses a long-standing problem in pharmaceutical research: the fact that drugs that appear safe and effective in animal studies often turn out to be toxic to humans. These failures are largely due to what scientists call “interspecies differences,” or the biological variations between different species. The new AI model is designed to identify these differences and use them to predict which drugs are likely to cause problems in humans. This could help to weed out dangerous drugs before they are ever tested on people, which would be a major step forward for drug safety and for the efficiency of the pharmaceutical industry.
The Challenge of Interspecies Differences
For decades, drug developers have relied on animal testing to get a sense of how a new drug will affect a living organism before it is given to human volunteers in a clinical trial. This has been a crucial step in the drug development process, but it is far from perfect. One of the main reasons for this is that animals are not perfect stand-ins for humans. There are many subtle and not-so-subtle biological differences between species, and these can have a big impact on how a drug works.
A classic example of this is chocolate, which is a harmless treat for most people but can be poisonous to dogs. In the world of medicine, a drug that is safe for a mouse or a monkey might have devastating effects on a person. The opposite can also be true: a drug that is toxic to an animal might have been a lifesaver for humans, but it was discarded in the early stages of development because of the results of animal testing. These interspecies differences are a major reason why so many drugs that look promising in the lab end up failing in clinical trials.
A New AI-Powered Solution
To address this challenge, Professor Kim and his team at POSTECH developed an AI model that is specifically designed to account for the biological differences between humans and the animals that are commonly used in preclinical research. The model is based on a concept that the researchers call the “Genotype-Phenotype Difference” (GPD). This refers to the idea that the same gene can produce different effects, or phenotypes, in different species. The AI model is trained on a massive amount of data about genes and their effects in both humans and animals. It uses this information to learn the subtle ways in which a drug that targets a particular gene might have a different effect in a person than it does in a mouse.
The Three Pillars of GPD Analysis
The POSTECH team’s AI model analyzes GPD along three main axes. The first is gene essentiality, which is a measure of how important a gene is for survival. A gene that is essential in a mouse might not be essential in a human, and vice versa. The second axis is tissue-specific gene expression. This refers to the fact that genes are turned on and off in different tissues and at different times. The AI model looks at the differences in these patterns between humans and animals. The third axis is gene network connectivity. This is a measure of how a gene is connected to other genes in a biological network. The AI model analyzes the differences in these connections between species.
Impressive Predictive Power
The researchers tested their AI model on a large dataset of drugs that included 434 drugs that are known to be risky and 790 drugs that have been approved for use in humans. The model was able to distinguish between the safe and unsafe drugs with a high degree of accuracy. The researchers found that the model’s predictive power was significantly better than that of older models that only looked at the chemical structure of a drug. The new model was also found to be more accurate than other state-of-the-art AI models that are used to predict drug toxicity.
Chronological Validation
To further test the power of their AI model, the researchers performed a “chronological validation.” They trained the model using only drug information that was available up to the year 1991. Then, they asked the model to predict which drugs would be withdrawn from the market after 1991 due to toxicity. The model was able to do this with 95% accuracy. This is a strong indication that the model could be used in the real world to identify potentially dangerous drugs before they are given to patients.
The Future of Drug Development
The development of this new AI model is a significant step forward for the field of drug discovery and development. By combining artificial intelligence with bioinformatics, the researchers have created a tool that could help to make the drug development process more efficient and more reliable. This could lead to the development of safer and more effective drugs for a wide range of diseases. It could also help to reduce the number of animals that are used in preclinical research. As AI and machine learning continue to be integrated into the pharmaceutical industry, we can expect to see more innovations like this one that will help to improve human health.
The Research Team and Publication
The research was led by Professor Sanguk Kim of the Department of Life Sciences and Graduate School of Artificial Intelligence at the Pohang University of Science and Technology (POSTECH). The research team also included Dr. Minhyuk Park, Mr. Woomin Song, and Mr. Hyunsoo Ahn. The results of their study were published in the international medical journal eBioMedicine.
