Most people would be shocked to learn how little is known about most chemicals. Only 3 percent of industrial chemicals, primarily drugs and pesticides, undergo comprehensive testing. The majority of the 80,000 to 140,000 chemicals found in consumer products either haven’t been tested at all or have only been superficially examined for local harm at high doses. I am a physician and have previously led the European Center for the Validation of Alternative Methods of the European Commission from 2002 to 2008. I’m committed to developing quicker, cheaper, and more accurate methods for testing chemical safety. I now head a program at Johns Hopkins University aimed at overhauling safety sciences. As part of these efforts, we’ve created a computer-based method of testing chemicals that could potentially save over $1 billion annually and spare more than 2 million animals. These alternative methods are especially important as government regulations on the chemical industry become less strict, posing risks to human and environmental health.
Our computer tests are feasible partly due to REACH (Registration, Evaluation, Authorizations, and Restriction of Chemicals) legislation in Europe, which was the first to systematically log existing industrial chemicals. Between 2008 and 2018, chemicals produced or sold in Europe at over 1 ton annually had to be registered, with required safety tests increasing with production volume. Our team published a critical analysis in 2009 concluding that to meet these legislative demands, new methods of chemical analysis were essential.
Although Europe does not track chemicals produced or sold at less than 1 ton per year, the size-comparable U.S. chemical industry brings about 1,000 chemicals annually to market in that range. Despite this, Europe does a better job of requesting safety data. This underlines the need to assess numerous new substances produced in small quantities, which aren’t regulated in Europe. Cost-effective and rapid computer methods are well-suited for this task. Our team utilized public safety data from REACH to create the largest toxicological database to date in 2016, logging 10,000 chemicals and linking them to 800,000 related studies. This foundation allowed us to test the reproducibility of animal tests, traditionally deemed the gold standard for safety testing. Some chemicals were tested repeatedly in the same animal test, such as over 90 tests on rabbit eyes for two substances, and 69 substances tested more than 45 times.
This excessive use of animals enabled us to assess whether animal tests provide consistent results. Our analysis showed that these tests, which consume over 2 million animals worldwide each year, are not very reliable. Animal tests revealed a chemical to be toxic only about 70 percent of the time. These tests adhered to OECD guidelines under Good Laboratory Practice, representing the best available standards, yet still demonstrated overrated quality, indicating a need for new strategies in toxicity evaluation.
In alignment with the vision for “Toxicology for the 21st Century,” a movement led by U.S. agencies to modernize safety testing, important work was conducted by my Ph.D. student Tom Luechtefeld at the Johns Hopkins Center for Alternatives to Animal Testing. Partnering with Underwriters Laboratories, we expanded our database and utilized machine learning for toxicity prediction. As reported in the journal Toxicological Sciences, we’ve developed a novel algorithm and database for chemical analysis, termed read-across structure activity relationship (RASAR).
To develop this, we compiled an extensive database of 10 million chemical structures from public databases, creating a map of the chemical universe through 50 trillion chemical pairs. A supercomputer plotted chemicals close to each other if they shared many structures and far apart if they didn’t. Usually, molecules near a toxic one are also dangerous, and this pattern holds even if many toxic substances are clustered, thereby isolating harmless substances.
Analyzing substances involves placing them in this map, requiring half a billion mathematical calculations per chemical to determine its position. This map focuses on 74 characteristics for predicting a substance’s properties. By considering the properties of neighboring chemicals, we can predict whether an untested chemical poses a hazard. For example, our program predicts eye irritation based not only on data from similar chemicals tested on rabbit eyes but also using skin irritation data, as what irritates the skin often affects the eyes.
The method is designed for new, untested substances. However, by applying it to chemicals with known data and comparing predictions with actual results, we evaluate its accuracy. We did this for 48,000 chemicals, well-characterized for at least one aspect of toxicity, discovering toxic substances in 89 percent of cases, outperforming animal tests that get it right only 70 percent of the time. Our RASAR method will be validated by an interagency committee of 16 U.S. agencies, including the EPA and FDA, using chemicals with unknown outcomes. This step is crucial for its acceptance by various countries and industries.
The RASAR approach, grounded in chemical data from the 2010 and 2013 REACH deadlines, has enormous potential. If estimates are correct, and chemical producers had adopted RASAR instead of registering chemicals post-2013, it could have saved 2.8 million animals and $490 million in testing costs, while also improving data reliability. This hypothetical scenario underscores its potential value in other regulatory programs and safety assessments.