A computational model platform of artificial intelligence for the toxicological and environmental evaluation of molecules has been developed since 2017 by the company Altox in São Paulo as part of the Innovative Research Project in Small Companies (from Portuguese PIPE), winner of one of the FAPESP awards. In 2019, Altox started a partnership with companies from the cosmetic, pharmaceutical, agrochemical, food, and related sectors as well as toxicology laboratories for a system of data cooperation and service provision.
The platform, called iS-Tox® (In Silico Toxicology Platform), aims to rationalize the use of animals in tests and provide safety data on new molecules, ingredients, and/or impurities. It is composed of varying cores that simulate different effects on the human body and in other species, containing artificial intelligence and other models validated according to OECD criteria, which predicts toxicological effects.
According to Altox director and the researcher responsible for the project, Carlos E. Matos dos Santos, the company’s performance has contributed not only to the rationalization of animal use or as a solution in computational toxicology services, but also as a facilitator for regulatory frameworks that have been implemented in recent years as part of policies to rationalize the use of animals in testing and the modernization of outdated regulations. To the extent that computational models and in vitro tests to evaluate new chemical ingredients are increasingly composing safety assessments for regulatory purposes, and because we already have standards and guidelines that recommend the use of computers instead of animals in specific contexts (examples of impurities in medicines, agrochemicals, and food), the productive sector and regulators will have a more open way to establish these advances without losing competitiveness, as we have done since 2011 in Brazil, he adds.
Platform interfaces were based on the applicable legislation for each molecular context, including new molecules, cosmetic ingredients, foods, medicines, biocides and agrochemicals, and impurities and contaminants. According to the researchers Rodolpho C. Braga and Iury Tércio Simões de Sousa, in addition to its scientific robustness, the tool is user friendly; the models have broad structural coverage, facilitate the interpretation of the mechanisms of action and the structural contribution in the predictions, and ultimately generate a standardized, transparent, comprehensible, and complete report. In the development coordinator’s view, Luiz Augusto Mota Filho, the platform integrates the most current technologies, with a modern, cognitive, and secure structure.
Some tools, such as the adverse outcome pathway-based model for skin sensitization (AOP-Sens), combine the most advanced methods in toxicology such as the use of artificial intelligence models and structural alerts and an adverse outcome pathway (AOP) of sensitization, with mechanistic relevance and predictability of individual models.
The Genotox-iS tool performs the toxicity evaluation of impurities using statistical models of artificial intelligence and alert verification according to the criteria of ICH M7-Assessment and Control of DNA Reactive (Mutagenic) Impurities in Pharmaceuticals to Limit Potential Carcinogenic Risk, one of the topics provided in the Common Technical Document (CTD) and Manual CTD-3.2.S, recently published by the National Agency of Sanitary Surveillance for the registration of active pharmaceutical ingredients (APIs). The tool differential is the generation of a robust report, with predictability and applicability domain measures, in addition to transparency and mechanistic relevance.
This concept involves building and exploiting chemical spaces with data from toxicological tests and artificial intelligence models, identifying patterns that enable the prediction of the toxicity of new molecules by measuring the predictability and applicability domains.
Figure: Chemical and toxicological space for dermal sensitization, with experimental data of combined biological tests (of humans and animals), direct peptide reactivity assay (DPRA), human skin, KeratinoSens, local lymph node assay (LLNA), and h-CLAT, with compounds grouped by structural similarity.
