For a long time, the practice of testing new products, particularly pharmaceutical drugs, on certain species of animals to determine safety of use in humans has been opposed by animal lovers. Help may finally be at hand, with artificial intelligence and machine learning systems offering a way out, according to a BBC report.
Testing of new products on animals is one of the dark realities of scientific research. The testing has continued though in some cases it is practically useless, a classic instance being arthritis drug Vioxx. It passed the animal testing stage and went on sale, but had to be withdrawn when studies showed that long-term use led to an increased risk of heart attack and stroke in humans. The painkiller aspirin, on the other hand, would have failed in animal tests, as it is toxic to rat embryos.
One thing artificial intelligence and machine learning systems are proving effective at is to trawl through all existing and available animal testing results worldwide to prevent the need for unnecessary new tests, the BBC report said.
It can be difficult and time-consuming for researchers to sift through decades of data to find and analyse exactly what they are after, Joseph Manuppello, senior research analyst at the Physicians Committee of Responsible Medicine, a US nonprofit, told the BBC.
“I’m very excited about the application of AI models like ChatGPT to extract and synthesize all of this available data, and getting the most out of it,” Manupello said.
Thomas Hartung, toxicology professor at Johns Hopkins University in the US and director of the Center for Alternatives to Animal Testing, said, “AI is as good as a human, or better, at extracting information from scientific papers.”
A primary reason for animal testing is to test new chemicals, Prof Hartung said. With more than 1,000 new compounds entering the market every year, a lot of testing is involved. But machine learnings systems are beginning to be able to determine a new chemical’s toxicity, he said.
“Having tools available where we can press a button and we get a preliminary assessment, which is giving us some flags of ‘here’s a problem’ … will be enormously helpful,” he said.
A more exciting machine-learning project is AnimalGAN, which is being built to try and replace the need for future animal testing. Developed by the U.S. Food and Drug Administration, the software aims to be able to accurately determine how rats would react to a given chemical. The system was trained using data from 6,442 real rats across 1,317 treatment scenarios.
A similar international project called Virtual Second Species is creating a virtual dog, which is being trained using data from historic dog test results.
As more such systems are developed and brought into use, the need for animal testing will, animal lovers hope, be reduced, if not be eliminated entirely within our lifetime.
