Artificial intelligence is reshaping how pharmaceutical scientists evaluate absorption, distribution, metabolism, excretion and toxicity – collectively called ADMET. Poor ADMET properties cause a large portion of late-stage drug failures, and traditional assays are costly and slow. Machine‑learning (ML) models now offer rapid, cost‑effective alternatives. A 2025 review of ML applications in ADMET prediction found that ML-based models can outperform some quantitative structure–activity relationship models and integrate seamlessly into drug-discovery pipelines【182517202011111†L170-L178】. These models provide reproducible predictions and reduce experimental burden【182517202011111†L170-L178
/image
】.
Why ADMET matters
ADMET summarises the journey a compound takes through the body. Absorption describes how molecules enter the bloodstream; distribution tracks their movement between tissues; metabolism captures chemical transformations, often in the liver; excretion covers elimination via the kidneys or bile; and toxicity concerns off‑target effects. Early identification of ADMET liabilities helps medicinal chemists prioritise safer candidates.
Machine learning meets pharmacokinetics
In supervised learning, algorithms such as support vector machines, random forests and neural networks are trained on labelled datasets of molecules and known ADMET outcomes. These models have demonstrated significant promise in predicting solubility, permeability, metabolism and toxicity endpoints【182517202011111†L170-L178】. Unsupervised and self‑supervised methods learn latent representations of chemical structures without labels and can be fine‑tuned for specific tasks. Graph neural networks and transformer‑based architectures operate directly on molecular graphs. Multitask learning predicts multiple endpoints simultaneously, improving performance by sharing representations.
Data and descriptors
Large public databases—ADMETlab, ChEMBL, TOX21 and DILIrank—contain millions of molecules annotated with pharmacokinetic and toxicity outcomes. Models leverage classical descriptors like molecular weight and polar surface area, fingerprints encoding functional groups, and learned embeddings. Deep-learning models trained on these descriptors achieve high accuracy predicting intestinal absorption, hepatotoxicity and plasma protein binding.
Hybrid approaches combine mechanistic models with ML. Physiologically based pharmacokinetic simulations describe how drugs move through the body; ML complements these simulations and improves clearance predictions across species【182517202011111†L170-L178】.
Benefits, challenges and next steps
AI-driven ADMET prediction delivers clear benefits: reduced cost and time, decreased reliance on animal testing, and the ability to explore chemical space beyond what is possible in the lab. Yet challenges remain. Data bias can cause models to perform poorly on novel chemotypes; deep networks are often opaque; and regulators require transparency and validation【182517202011111†L182-L187】. Continued integration of ML with experimental pharmacology and efforts to improve interpretability are essential.
Several tools make ADMET prediction accessible, including ADMETlab, pkCSM, ProTox‑II and ADMET‑AI. Open-source packages such as DeepChem and RDKit help researchers build custom models. Ultimately, predictive models should complement—not replace—wet‑lab assays. High-throughput experiments and automated laboratories can validate predictions and refine models, creating a closed-loop discovery pipeline.
Conclusion
Predictive ADMET modeling shows how artificial intelligence is transforming drug discovery. Machine‑learning algorithms offer rapid, reproducible predictions of absorption, distribution, metabolism, excretion and toxicity, outperforming traditional approaches and reducing experimental burden【182517202011111†L170-L178】. However, success depends on high‑quality data, interpretability and regulatory acceptance. With careful integration of mechanistic knowledge, experimental validation and ethical oversightst content starthical oversight, predictive modeling can accelerate the development of safer and more effective therapeutics【182517202011111†L182-L187】.
Sources
- Leveraging machine learning models in evaluating ADMET properties for drug discovery and development (2025)【182517202011111†L170-L178】【182517202011111†L182-L187】.
Related Reading
- AlphaFold and Drug Discovery: How AI Is Finding Molecules That Extend Lifespan
- AI for Science: How Artificial Intelligence is Revolutionizing Drug Discovery, Biology & Research
- Future of Drug Discovery: AI and Peptide Innovation
- The Rise of Agentic Scientists: How AI Agents Are Transforming Discovery