Drug toxicity testing is one of the key steps in the process of new drug discovery, aiming to screen compounds that are safe and effective for humans. Based on statistics, more than 30% of candidate drugs are expected to cause study interruptions due to toxicity issues.
Traditional drug toxicity prediction usually relies on in vitro and in vivo studies, which is generally time-consuming and costly. Artificial intelligence, especially machine learning and deep learning algorithms, provides more accurate and effective alternative methods for drug toxicity prediction. Here, we cite a review of literature outlining recent advances in six key toxicity attributes and Tox21 detection endpoints for AI-based drug toxicity prediction, each of which corresponds to a specific physiological mechanism and health risk.
hERG encodes the Kv11.1 protein, an alpha subunit of the potassium channel. hERG channels play an important role in cardiac action potentials, and their inhibition may lead to arrhythmias. AI models were used to predict the inhibitory properties of hERG to avoid drugs that might cause cardiac disease.
LD50 is a standardized index to compare and measure drug toxicity in toxicology. The AI model classifies the chemicals of drugs into toxic and non-toxic categories by binary classification to predict their lethal risk to the organism.
Drug-induced liver injury (DILI) is one of the common causes of drug recalls. AI models can identify compounds that are likely to cause liver injury through deep-learning algorithms, which is essential to minimize the risk of failure later in drug development.
The Ames test is the most widely used method to assess the mutagenicity of compounds. The AI model can achieve high accuracy prediction across large datasets, effectively overcoming the limitations of the traditional Ames test.
Carcinogenicity is a crucial toxicological endpoint for chemicals, which may cause cancer by damaging DNA or disrupting the ability of cells to metabolize processes. AI models predict whether a compound is carcinogenic by machine learning and deep learning methods.
Skin sensitization prediction is a critical endpoint in assessing the safety of novel compounds or chemical combinations. A variety of AI models have been used to predict whether compounds will cause cutaneous allergic reactions.
The Tox21 project aims to quickly and efficiently determine whether compounds have the potential to alter human physiological systems, involving a broad range of toxicity endpoints, which exemplifies the potential of AI models to predict multiple toxic effects.
Fig. 1. AI in drug toxicity prediction. (Tran T T V.; et al. 2024)
These advances demonstrate the advantages of AI in identifying potential toxicities early in drug development, including reducing the need for animal testing and improving the efficiency and safety of drug discovery.
AI extracts toxicity relationships from drug structure and physicochemical properties through learning and reasoning. Our company integrates industry-advanced AI technology, and our services aim to simplify the process of drug toxicology testing and improve accuracy and efficiency of toxicity prediction. We understand that each product is unique to our customers, so we provide customized solutions based on their specific needs. If you are interested in our services or have a question, please feel free to contact us for more details.
Original Article:
Tran T T V.; et al. (2023). Artificial intelligence in drug toxicity prediction: recent advances, challenges, and future perspectives. Journal of chemical information and modeling. 2023, 63(9): 2628-2643.
Services Related in the Article:
AI-guided Toxicology and Safety Testing
We harness the power of artificial intelligence to transform the landscape of drug discovery. Our mission is to accelerate the development of life-saving medicines by leveraging cutting-edge AI technologies.
