AI in Drug Development: A Comprehensive Review of the Latest Advancements
Drug discovery has historically been a laborious, expensive, and inefficient process. Developing a new drug takes on average US$2.6 billion and between 12 to 15 years with clinical trials having less than a 10% success rate. However, new advancements in artificial intelligence (AI) have the potential to change all that. Large Language Models (LLMs) and generative AI are becoming mainstream AI tools that are already revolutionizing pharmaceutical research.
A review published by Kang Zhang et al. discusses how AI can transform drug discovery, development, and real world practice. The authors review state-of-the-art AI applications spanning the entire drug discovery and development pipeline - from target identification to post-market surveillance.
Revolutionizing Drug Discovery Pipeline
AI technologies can meaningfully improve the efficiency and effectiveness of many steps in the preclinical drug development pipeline.
Identifying & Validating Targets
Labor-intensive target identification often fails to pan out. Using AI algorithms, researchers can interrogate large datasets embedded in complex biological networks. By building knowledge graphs of multi-omics data and biological relationships, researchers can pinpoint disease-relevant molecular sub-patterns and causal associations.
Virtual Screening and De Novo Design
Virtual screening rapidly assesses hits and guides further prioritization in ultra-large libraries of potential lead compounds. AI-driven receptor-ligand docking models (including equivariant neural networks) can predict ligand-induced fit spatial transformations and generate accurate binding poses.
De Novo Drug Design can automatically be accomplished with deep generative models. Modern generative models, like diffusion models and chemical language models, are rule-based engines that design novel molecules that adhere to desired molecular features from scratch. This methodology has led to small-molecule inhibitor and functional protein design.
ADMET Prediction and Synthesis Planning
AI can also optimize later stages.
- ADMET prediction
Deep learning models are replacing traditional descriptors in predicting Absorption, Distribution, Metabolism, Excretion and Toxicity (ADMET).
- Synthesis planning
Computer-aided synthesis planning (CASP) tools apply deep learning to retrosynthetically assess target molecules and break them down into smaller building blocks. This process coupled with automated robotic synthesis speeds up the overall "Design-Make-Test-Analyze" (DMTA) cycle.
Fig. 1. Overview of AI applications in the drug development pipeline. (Zhang K.; et al. 2025)
AI for Clinical Trials and Real World Practice
The review emphasizes that AI's utility extends well beyond the laboratory into clinical settings.
- Identification of diagnostic and predictive biomarkers
AI models are used to predict and develop biomarkers using digital pathology slides and large multi-omics data sets allowing clinicians to accurately stratify patients for personalized treatment plans.
- Drug repurposing
Repositioning previously approved drugs can be accomplished using AI to parse through large biomedical data sets and real-world evidence (RWD), in hopes of revealing new therapeutic value of these drugs cutting down both time and money spent on development.
- Optimizing clinical trials
AI is being used to optimize trials from design, recruitment, all the way to Digital Twins - virtual avatars of each participant enrolled in a trial.
- Continuous monitoring of drug quality, safety, and efficacy
Machine learning algorithms can be deployed after approval to continuously monitor adverse events.
How Protheragen MedAI Can Help
AI has a crucial role to play in helping drug developers overcome the rigorous and intricate path that drug discovery has become. Here at Protheragen MedAI, we pride ourselves on bridging the gap between these concepts and real world applications that can provide pharmaceutical companies and research institutes with measurable results.
- Target identification
Using network analysis of multi-omics data and knowledge graph enrichment, we can identify and validate targets.
- AI-powered drug design
Our generative AI models coupled with deep learning allow rapid hit-to-lead optimization and de novo molecule design.
- AI-based ADMET prediction
Get ahead of costly attrition by predicting pharmacokinetic profiles early in the drug discovery process.
- Drug repurposing
Find new indications for your assets using real world data and drug perturbation signatures.
Contact us today if you'd like to learn how you can implement these strategies into your drug discovery pipeline or have any questions.
Original Article:
- Zhang K.; et al. (2025). Artificial intelligence in drug development. Nature medicine. 2025, 31(1): 45-59.
Services Related in the Article:
