AI-assisted Toxicity Prediction Services

Because the use and production of chemical compounds may be directly harmful to humans, animals, plants, or the environment. Chemical toxicity and adverse effects have become key regulatory aspects for a multitude of industries. The main factors that determine the toxicity are the route of exposure, the amount of chemical substance, and the properties related to absorption, distribution, metabolism, excretion/elimination (ADME), etc. Currently, a growing number of authorities are calling for the use of in silico computational models to replace traditional in vivo toxicity tests carried out on laboratory animals. Considering the ethical aspects, economic and time efficiency, artificial intelligence (AI)-based models have been used to predict toxicity with greater reliability and robustness than traditional tests. Creative Biolabs has established an advanced AI-assisted platform to support toxicity prediction for customers worldwide.

Fig.1 Perspective of artificial intelligence-based methodologies applied to chemical toxicology. (Isaksson, et al., 2020)

AI-assisted Potential Developmental Toxicity Prediction

Developmental toxicity is defined as the adverse effects that are produced by exposure prior to conception, or during pregnancy and childhood. Through the prenatal developmental toxicity study, the chemicals that may pose a risk to the developing fetus can be identified. The REACH2 regulations in European Union and the United States Environmental Protection Agency (USEPA3) include developmental toxicity as one of the most important toxicological endpoints. Traditional methods to detect the developmental toxicity potential (LEL) of chemicals are complex, time and resource intensive. Also, the experiments need to be done on animals, which raises ethical issues. Therefore, it is necessary to establish AI-assisted models for predicting the developmental toxicity potential of chemicals.

AI-assisted Irritation and Sensitization Prediction

Many chemicals and allergens can cause skin sensitization and irritation such as cosmetics, drugs, fragrances, pesticides, metals, and preservatives. Therefore, the assessment of skin irritation and sensitization is of enormous importance to a host of different industries, which are related to customer and worker safety. For most risk assessment and regulatory objectives, animal studies have been the preferred testing approach. Currently, various non-animal testing methods for skin sensitization are developed to replace them. Because of the high speed and low cost, the use of in silico methods is particularly appealing and advantageous. Based on the molecular structures, computational methods promise the ability to predict the skin sensitization potential of substances.

AI-assisted LD50 Prediction

Safety is always the most important issue during drug development. Generally, a variety of toxicities and adverse drug effects should be evaluated in preclinical and clinical trial phases. The commonly used methods are in vitro and in vivo tests, and there are also some efforts to develop low-cost models. In silico methods have shown great advantages because they are fast, green, accurate, cheap, and most importantly they could be done before compounds are synthesized. At present, there are many computational models for drug safety assessment, such as the quantitative structure-toxicity relationship (QSTR) models, which can predict the median lethal dosage for rodent oral acute toxicity (LD50). It is a standard piece of data used to classify substances based on the risk they pose to human health following acute exposure.

Fig.2 Scheme of building QSAR or structural alerts models for prediction of chemical toxicity. (Yang, et al., 2018)

AI-assisted Mutagenicity and Carcinogenicity Prediction

The induction of permanent transmissible alterations in the amount or structure of the genetic material of cells or organisms is referred to as mutagenicity. Carcinogenicity refers to a carcinogen's ability or proclivity to cause such radionuclide or radiation. Mutagenicity and carcinogenicity are the most important and complex toxicological endpoints that require thorough evaluation during the industrial chemical registration process. Experimental tests on animals are expensive and time-consuming, hence there is a push to develop in silico approaches for assessing carcinogenicity. For example, the European chemical regulation REACH encourages the use of Quantitative Structure-Activity Relationship (QSAR) models and read-across as alternate approaches for assessing the physical, chemical, and biological properties of compounds.

AI-assisted Biodegradation Performance Prediction

Under typical environmental conditions, biodegradation is the decomposition of materials into environmentally acceptable products such as water, carbon dioxide, and biomass by naturally occurring microbes. Biodegradation is the principal environmental dissipation process and a key factor to describe the long-time effects of chemicals being decomposed in the environment. In silico approaches for determining the biodegradable profiles of chemicals are encouraged and are an active current research area because of a lack of complete experimental data, expensive study costs, and time-consuming studies. It is important to use various computational tools to predict possible degradation pathways.

References

1. Isaksson, L.J.; et al. Machine learning-based models for prediction of toxicity outcomes in radiotherapy. Frontiers in oncology. 2020, 10: 790.
2. Yang, H.; et al. In silico prediction of chemical toxicity for drug design using machine learning methods and structural alerts. Frontiers in chemistry. 2018, 6: 30.