Computational solution to assessing chemical toxicity
The pharmaceutical and cosmetics industries consume large amounts of energy and produce enormous quantities of waste. If not properly disposed of, some chemicals used during production, such as solvents, can end up in the soil, or in water. “Another issue is that when chemicals from pharmaceutical and cosmetic products are excreted from our body, they end up in the sewage system,” explains Eco-CosmePharm project coordinator Rafael Gozalbes, CEO of ProtoQSAR, Spain. “Not all sewage treatment plants are designed to remove these chemicals. This can lead to pharmaceutical and cosmetic end products being released into the aquatic environment.” Identifying which chemicals induce harmful effects on aquatic life is critical if we are to effectively address this situation. A key challenge however is that toxicity testing can be expensive and time-consuming.
Screening for toxicity
The Eco-CosmePharm project, which was undertaken with the support of the Marie Skłodowska-Curie Actions programme, sought to apply technology to provide a convenient and cost-efficient alternative to lab testing. “Our aim was to computationally identify the impact of potentially hazardous pharmaceuticals, cosmetics and solvents,” says Gozalbes. “We did this by using data analysis and chemoinformatics to predict levels of human and environmental toxicity.” The project applied a powerful computational technique for property prediction, called Quantitative Structure-Activity Relationship (QSAR). This technique uses machine learning algorithms to understand the structural features or patterns responsible for the properties of molecules (in this case, aquatic toxicity). “The multitasking QSAR models we built were then used to screen marketed pharmaceuticals and cosmetics, to identify potential chemicals that might be harmful to aquatic life,” adds Marie Skłodowska-Curie fellow Pravin Ambure. “Some 35 chemicals were selected, and their level of toxicity validated in the lab to confirm the accuracy of the modelling.” The models pioneered through the Eco-CosmePharm project are ‘multitasking’ in the sense that they can take on board a number of parameters and variables, such as aquatic conditions. This means that scientists can apply the models to fit specific conditions of interest. “The knowledge gained from Eco-CosmePharm has already enabled us to classify some existing marketed pharmaceuticals and cosmetics according to their toxicity with regards to aquatic environments,” continues Gozalbes. “Through chemoinformatics, we can then design novel analogues of selected toxic chemicals, or identify alternative chemicals that might show similar desirable properties with less or no ecotoxicity.”
Marketable software products
The success of the Eco-CosmePharm project has led to the development of two pre-market AI-based software tools, called ‘ProtoML-Basic’ and ‘ProtoML-Mixture’. These tools can be used to carry out several QSAR and machine learning tasks, either for individual chemicals or mixtures. “These tools will be provided by ProtoQSAR as part of a panel of software products,” explains Ambure. “We hope they will help policymakers and industry to become more aware of the potential risk of some chemicals, and to choose the safest ones where possible.” The next steps include the development of a user-friendly and interactive online web platform, which could be used to predict aquatic toxicity for any given set of query chemicals. This will generate detailed toxicity reports for users. Gozalbes envisages this online tool being used to screen compounds ahead of regulatory approval, as well as for selecting compounds for further experimental testing.
Keywords
Eco-CosmePharm, ProtoQSAR, pharmaceutical, cosmetics, sewage, toxicity, solvents, aquatic