Application of Artificial Intelligence in drug development

Artificial Intelligence in drug development: Artificial intelligence (AI) is playing an increasingly important role in drug development. It is applied in drug molecule design, metabolism prediction, and toxicology assessment to enhance research efficiency and reduce failure rates. Machine learning models help define structure–activity relationships, predict metabolic sites, and identify toxic metabolites. AI platforms such as GLORYx, DeepDILI, and DeepCarc have shown promising results in metabolism and toxicity prediction, while tools like SafetAI support safety evaluations and the FDA approval process. Although challenges remain in model interpretability and reproducibility, AI-driven approaches are revolutionizing drug discovery, optimization, and safety assessment in pharmaceutical research.

Artificial Intelligence and Drug Discovery

Artificial intelligence is a potentially powerful tool in the chemical toxicologist’s toolkit. AI-based platforms have been developed to aid in the design of drug-like molecules, predict the propensity of compounds to form toxic metabolites, and expedite the FDA drug approval process.

A major task in drug discovery is defining the structure-activity relationship (SAR) of a class of molecules. SAR information is crucial for assessing the therapeutic potential and potential toxicity of a class of compounds. Generally, the greater the selectivity of a given compound for its intended target, the lower the likelihood of off-target side effects. ML models have been developed that can simulate the number of drug interactions within protein targets using molecular dynamics simulations.

Another application of AI is the de novo generation of molecules through automated drug discovery, which offers advantages such as reproducibility, scalability, 24/7 operation, and the elimination of user bias. However, fully automating the design of new chemical entities without human intervention remains challenging. Full automation of drug discovery requires building trust in the machine learning process, overcoming technical challenges, and considering the synthetic feasibility of the final molecules. In current machine learning applications, human input helps define the machine learning search target, analyze the cascade design, assess relevance, and debug human interventions during the learning process. In these approaches, machine learning is an essential component of the collaborative work between computational and experimental drug design.

Artificial Intelligence and Drug Metabolism

Computational prediction of small molecule metabolism falls into two categories: site of metabolism (SOM) identification and metabolic structure prediction. GLORYx is a machine learning method used to predict and rank the chemical structures of metabolites produced by oxidation, hydrolysis, and conjugation metabolism. First, SOM prediction is performed using FAME3 software. This is achieved by using a professionally annotated dataset and a set of unique molecular ring atom type descriptors as molecular fingerprints. GLORYx uses the SOM predictions to score and rank the predicted metabolites. Challenges in predicting conjugated metabolites include high false positive identification rates and difficulty distinguishing reaction types occurring under the same SOM. Comparison of the SOMprediction model with a high-quality, manually curated test set revealed that conjugation of SOM predictions can improve the prioritization of metabolites produced by metabolic processes.

Artificial Intelligence and Toxicology Prediction

Machine learning can be used to connect metabolic and toxicology models to understand how metabolite bioactivation leads to hypersensitivity reactions and hepatotoxicity. A study reported the development of a machine learning model to predict the formation of quinones in drug metabolism, successfully improving the ability to rationally engineer drugs to prevent quinone formation. Furthermore, this machine learning process was compared with generic structural alerts for identifying potential chemical sites susceptible to bioactivation. Compared to broad structural alert approaches, machine learning provided more accurate toxicity predictions. This modeling framework has been applied to epoxidation, nitroaromatic reduction, and thiophene sulfoxidation reactions and could be further extended to other metabolic pathways. Furthermore, this approach can be combined with a machine learning platform for inferring intermediate metabolites, potentially helping to identify drug bioactivations with unexpected side effects that could lead to drug discontinuation in clinical settings. Jonathan Goodman of the University of Cambridge discussed the application of AI in predictive toxicology in his publication

AI-Based New Drug Approval Process

SafetAI is designed to assist in providing safety profiles for clinical trial (IND) applications, thereby helping drug candidates advance into Phase I clinical trials (https://www.fda.gov/about-fda/nctr-research focus-areas/safetai-initiative). SafetAI utilizes a deep learning architecture designed to improve toxicity assessments to facilitate drug safety studies and is currently being expanded to multiple organ systems. This tool can provide critical safety information during the IND review process and eliminate the need for intensive animal studies under the FDA’s New Drug Innovative Science and Technology Approach program.

Examples of AI tools used to assess drug safety include the deep learning-powered platforms Drug-Induced Liver Injury (DeepDILI) and DeepCarc. These tools were designed and evaluated as preclinical screening methods for drug-induced liver injury (DILI) and cardiotoxicity, respectively, in potential drug compounds. These tools use machine learning algorithms trained on drugs approved before 1997 to predict DILI and cardiotoxicity in drugs approved since then. The DeepDILI model has also been used to predict potential DILI issues in drug repurposing candidates. Such tools are publicly available at https://github.com/TingLi2016/DeepDILI and https://github.com/TingLi2016/DeepCarc.

The Development of Artificial Intelligence in Drug Development

To reduce the financial costs and failure rates associated with drug development, pharmaceutical companies are turning to artificial intelligence (AI). Many pharmaceutical companies have invested and continue to invest in AI and are collaborating with AI companies to develop essential healthcare tools. A rough estimate of leading international AI pharmaceutical platforms includes Schrödinger, Exscientia, AbCellera Biologics, Recursion Pharmaceuticals, Atomwise, Benevolent, Insilico, Insitro, and Cyclica, among others.

While key advances have been made in these areas, continued innovation is still needed. Currently, no single AI model can accurately predict all conditions, but when used appropriately, specific models can glean extremely useful information. Furthermore, AI tool designers should understand whether their platforms are truly “realistic” and ensure that detailed data management protocols are in place to accurately reflect toxicity profiles for specific endpoints. Careful model evaluation is essential throughout the machine learning development process to ensure accurate predictive power and assess model adaptability. Reproducibility is another key consideration in machine learning design, as factors such as random seeds and software package versions have been found to be sources of variation. Measures to improve reproducibility include using Docker containers to maintain consistent workflows and ensuring that all data and code management platforms are publicly accessible. Overcoming these challenges is expected to advance the development of AI platforms.

Conclusion

Artificial intelligence has transformed the landscape of modern drug development, offering powerful tools for accelerating discovery, improving safety, and reducing costs. By integrating machine learning with computational chemistry, toxicology, and pharmacology, AI enables researchers to predict molecular behavior, design drug candidates more efficiently, and anticipate potential side effects before clinical trials. While technical challenges such as data quality, model transparency, and reproducibility remain, continuous innovation and collaboration between AI developers and pharmaceutical scientists are driving significant progress. As AI models become more robust and interpretable, they will play an essential role in shaping the next generation of precision medicine and predictive toxicology, ultimately leading to safer and more effective therapeutic solutions.