Developing a new therapeutic compound is a lengthy and costly process that often spans 10-15 years and requires investments of several billion dollars. Despite this commitment, nearly 90% of drug candidates fail during clinical testing, often due to weak target validation, poor pharmacological properties, or unforeseen safety concerns.
Image credit: Gorodenkoff/Shutterstock.com
Artificial intelligence (AI) is emerging as a powerful tool to address these inefficiencies, and its use in drug discovery has grown substantially. By analyzing large and complex datasets from chemical libraries, omics data, and electronic health records, AI can reveal complex patterns that traditional methods struggle to detect, thereby accelerating preclinical research and clinical development.
Particular attention has focused on its role in hit identification and lead optimization and its potential to enhance clinical trial design and streamline integration across the drug development pipeline. Applications now include large-scale virtual screening, de novo molecular design, trial protocol optimization, and early exploration of emerging approaches such as adaptive trial strategies.[1]
AI-Powered Hit Identification and Lead Optimization
Traditional high-throughput screening usually evaluates millions of compounds. AI can offer a more efficient route by using machine learning models, such as neural networks, random forests, and support vector machines, to predict which molecules are most likely to bind to their target and display acceptable safety profiles effectively.
The application in de novo design is an interesting development. Rather than drawing from existing chemical libraries, AI can facilitate the design of molecules from scratch, optimizing them simultaneously for multiple properties such as affinity, solubility, and synthetic accessibility. This expands the range of chemical structures available and reduces dependence on trial-and-error methods.[2]
Protein structure prediction also enhances hit discovery. Tools like AlphaFold, when paired with generative design pipelines, provide accurate models of protein conformations, enabling more precise docking simulations. These advances allow the identification of promising binding interactions more reliably than conventional approaches.
Building Trust Through Explainable AI
One barrier to using AI in drug discovery is the lack of transparency regarding why an algorithm produces a particular prediction. Explainable AI (XAI) can address this issue by making model outputs interpretable.
Techniques such as SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) highlight which parts of a molecule contribute most strongly to predictions of activity or toxicity. This interpretability helps medicinal chemists prioritize compounds with greater confidence and supports regulatory requirements for transparency.
By integrating XAI into validation workflows, it is possible to get a more explicit rationale for selecting compounds for experimental testing. Consequently, explainability tools are gaining traction across chemical, biological, and translational research, highlighting their importance in building trust in AI systems.[3]
De-Risking Clinical Trials with Predictive AI
Clinical testing is one of the most expensive and failure-prone stages of drug development. AI can reduce these risks by supporting trial simulations, improving recruitment, and informing adaptive study designs.
In silico trials, an emerging approach, seek to model therapeutic effects in virtual patient cohorts, incorporating pharmacokinetic and pharmacodynamic variability. While not yet widely adopted, these methods are under investigation and may help to highlight potential toxicity or dosing issues before human trials begin.
AI also improves patient recruitment and stratification. Algorithms applied to electronic health records can reduce recruitment delays by quickly identifying eligible participants. Stratification tools can distinguish likely responders from non-responders, enabling smaller, more focused trials.
Improving trial success rates is another area where AI shows promise. A recent analysis found that drugs discovered through AI methods progressed through Phase I with 80-90% success rates, well above traditional averages of about 50%.[4]
While Phase II outcomes were comparable to industry norms (around 40%), the combined data suggest that end-to-end success rates could nearly double, significantly impacting R&D productivity.
A notable real-world example is Rentosertib (ISM001-055), a TNIK inhibitor discovered by Insilico Medicine’s AI platform, which entered Phase IIa trials in 2025. This represents one of the first AI-designed drugs to reach mid-stage clinical development and provides early evidence that AI-driven pipelines can translate into clinically viable compounds. [5]
How AI is accelerating drug discovery - Nature's Building Blocks | BBC StoryWorks
Video credit: BBC StoryWorks/Youtube.com
Integrating AI into the Drug Discovery Pipeline
Initially confined to the early discovery stage, AI’s role is becoming embedded across the complete research and development pipeline, from target identification to clinical deployment. For example, predictive models can simulate multiple trial designs in advance, allowing investigators to choose the most promising configuration regarding dosing, duration, and endpoints.
Real-time AI analysis supports adaptive trials, allowing researchers to adjust enrollment or dosing mid-study without undermining scientific rigor. AI is moving from a supplementary tool to a core component of pharmaceutical pipelines.
Challenges and Limitations of AI in Drug Discovery
Despite evident progress, there are some limitations to the widespread use of AI in this field. Data quality and accessibility are limiting factors. Many available datasets are incomplete or inconsistent, with the risk of compromising predictive accuracy.
Bias also presents challenges, as algorithms trained on unrepresentative data risk producing inequitable outcomes, such as inadvertently excluding certain demographic groups from trial eligibility.
Regulators require reproducibility and explainability, although the “black box” nature of many deep learning models undermines trust and broader adoption. Integration of XAI methods will be essential to satisfy compliance requirements.
In addition, although early evidence suggests improved success rates for AI-discovered compounds, these findings are based on relatively small datasets. Therefore, more extensive evidence will be needed to confirm whether these trends persist on a larger scale.
Conclusion
AI is emerging as a practical tool to accelerate hit discovery, provide transparent validation mechanisms, and contribute to clinical trial design in ways that could improve efficiency. Early indicators suggest AI-designed molecules may progress through development at higher success rates than traditionally discovered drugs.
However, challenges need to be addressed if AI is to be embedded responsibly across drug discovery pipelines. High-quality data, interpretable models, regulatory readiness, and ethical safeguards are all prerequisites for effective integration.
AI can be responsibly embedded across drug discovery pipelines when these conditions are met, supporting more efficient, transparent, and reliable therapeutic development.
References
- Rehman, A. U., Li, M., Wu, B., Ali, Y., Rasheed, S., Shaheen, S., Liu, X., Luo, R., & Zhang, J. (2025). Role of artificial intelligence in revolutionizing drug discovery. Fundamental Research, 5, 1273-1287.https://doi.org/10.1016/j.fmre.2024.04.021. Available: https://www.sciencedirect.com/science/article/pii/S266732582400205X
- Fang, Y., Pan, X. & Shen, H. B. (2023). De novo drug design by iterative multiobjective deep reinforcement learning with graph-based molecular quality assessment.Bioinformatics,39.10.1093/bioinformatics/btad157. https://academic.oup.com/bioinformatics/article/39/4/btad157/7085596?login=false
- Ocana, A., Pandiella, A., Privat, C., Bravo, I., Luengo-Oroz, M., Amir, E. & Gyorffy, B. (2025). Integrating artificial intelligence in drug discovery and early drug development: a transformative approach. Biomark Res, 13, 45.10.1186/s40364-025-00758-2. https://biomarkerres.biomedcentral.com/articles/10.1186/s40364-025-00758-2
- Kp Jayatunga, M., Ayers, M., Bruens, L., Jayanth, D. & Meier, C. (2024). How successful are AI-discovered drugs in clinical trials? A first analysis and emerging lessons. Drug Discovery Today, 29, 104009.https://doi.org/10.1016/j.drudis.2024.104009. Available: https://www.sciencedirect.com/science/article/pii/S135964462400134X
- Su, J., Zhu, X., Gao, Y., et al. (2025). Discovery of ISM001-055 (Rentosertib), a potent TNIK inhibitor advanced through AI-driven drug design. Nature Biotechnology.https://doi.org/10.1038/s41587-025-02012-x. https://www.nature.com/articles/s41591-025-03743-2