Machine Learning Could Make Animal Tests Obsolete
Animal advocates everywhere know that the fight against animal experimentation has long hinged on a tradeoff between ethics and efficacy. For a long time, advocates have waited for science to “catch up” to social consciousness; there has always been a tension between society’s reactions of horror at animal experiments, and scientists wringing their hands and explaining that these tests must happen because there is no other option.
This study was conducted to determine the possibility of replacing animal testing with technology and statistical models. The method in question is called RASAR, standing for “read-across structure activity relationships.” A library of known chemicals and their properties was created and made to be machine-readable. RASAR then takes the subject chemical and compares it to the library, giving us an idea of the subject’s chemical’s properties, including toxicity to humans, damage to water or air quality, volatility and flammability, carcinogenicity, and danger to eyes, skin, or respiratory organs.
According to the authors, this process of RASAR testing was able to correctly predict these properties 80% of the time, and reproduce the results accurately. The scientists note that “’Simple’ RASARs obtain cross-validated sensitivities above 80% with specificities of 50%–70%” which they say “is on par with the reproducibility of the respective animal tests.”
Animal testing is often redundant to the point of excess. While results must be replicated to be considered good science, there is a point at which it becomes pointless to test any further. While excess in some scientific experiments will just result in wasted grant money or extra time in the lab, excessive animal testing causes actual harm to the animals used as subjects as well.
Moreover, animal testing can be quite expensive, and the results are not necessarily useful. Six tests in particular – the “toxicological six-pack” – accounted for 55% of all animals used for testing in 2011. With regard to these tests, the computational model was able to outperform animal testing significantly, reproducing the results 89% of the time compared to 70%.
Animal testing for toxicology in Europe alone costs around $3 billion Euro annually, but is not always the best choice for determining the effects of a chemical. In the U.S., the government alone spends $15 billion each year on wasteful experiments. While animals mirror the biological complexity of humans better than a computer can, testing on them is an arduous and lengthy – not to mention cruel – process.
If we are able to replicate or improve on animal testing in a fraction of the time while consuming fewer resources and harming fewer animals, why wouldn’t we? Every test that’s done on a computer means one less animal caged and experimented on, and that’s something that every animal advocate should celebrate.