Drug toxicity predicted by differences between preclinical models and humans

In the UK, there was a case in which TGN1412, an immunotherapy under development, caused a cytokine storm in humans within hours of administration, leading to multiple organ failure. As another example, the stroke drug candidate Aptiganel was highly effective in animals, but was discontinued in humans due to side effects such as hallucinations and sedation. Drugs considered safe in preclinical testing can be lethal in human clinical trials. Machine learning-based technology has been developed to learn these differences and proactively identify potentially dangerous drugs before clinical trials.

A research team including Professor Kim Sang-guk of the POSTECH Department of Life Sciences and Graduate School of Artificial Intelligence, Dr. Park Min-hyuk and Mr. Song Woo-min of the Department of Life Sciences, and Ahn Hyun-soo of the Graduate School of Artificial Intelligence has developed a technology to predict the side effects of drugs in humans using machine learning. The study was recently published online in an international medical journal. e-biomedicine.

During the development of new drugs, drugs that pass preclinical testing often exhibit unexpected toxicity to humans. This problem arises from differences in the biological responses of humans and animals. For example, chocolate is generally safe for humans, but toxic for dogs. Similarly, a drug that is safe for mice may not be safe for humans. Until now, this “difference between species” has been the main cause of failure in new drug development.

The research team focused on “genotype-phenotype differences (GPD),” the biological differences between cells, mice, and humans. They analyzed how genes targeted by drugs function differently in humans and preclinical models, focusing on three key factors. Second, patterns of gene expression in different tissues. Third, the connectivity of genes within biological networks.

Validation using data from 434 hazardous drugs and 790 approved drugs revealed that GPD characteristics are significantly associated with toxic drug failure in humans. Predictive power was significantly improved over relying on drug chemistry data, with area under the curve (AUPRC1) increasing from 0.35 to 0.63 and area under the curve (AUROC2) increasing from 0.50 to 0.75. The developed AI model demonstrated superior predictive performance compared to existing state-of-the-art models.

The team also demonstrated the practicality of “time-series verification,” which alerts users to drugs facing market withdrawal due to toxicity. When a predictive model was trained only on drug data up to 1991, it accurately predicted which drugs were expected to be withdrawn from the market after 1991, achieving 95% accuracy.

The importance of this study is to bridge the “translation gap” between preclinical and clinical studies by quantifying biological differences in cells, preclinical animal models, and humans. Pharmaceutical companies can reduce development costs and time by screening high-risk candidates before clinical trials, while also improving patient safety. As more relevant data and annotations accumulate, the effectiveness of the model is expected to increase.

Professor Sanguk Kim said, “This is the first attempt to incorporate differences in genotype-phenotype relationships into drug toxicity prediction. Our framework enables early identification of high-risk drugs in clinical development.” He added, “This approach holds promise for reducing development costs, improving patient safety, and increasing the success rate of drug approval.” Co-first authors Dr. Min-hyuk Park and Woo-min Son said, “Human-centered toxicity prediction models will be a very practical tool in new drug development.” Pharmaceutical companies will be able to proactively screen for high-risk drugs at the preclinical stage, which they hope will improve development efficiency. ”

This research was supported by funding from the National Research Foundation (NRF), the Korean government (MSIT), the Medical Device Innovation Center, and the Synthetic Biology Human Resources Development Program.