Next-Generation Artificial Intelligence for ADME Prediction in Drug Discovery: From Small Molecules to Biologics

Area of Science:

• Pharmacology and Computational Chemistry
• Drug Discovery and Development
• Artificial Intelligence in Medicine

Background:

• Pharmacokinetic (PK) behavior, encompassing absorption, distribution, metabolism, and excretion (ADME), is critical for drug discovery, dose optimization, and safety.
• Predicting human PK early in development remains a significant hurdle, leading to high clinical attrition rates and inefficiencies in pharmaceutical pipelines.
• While AI and ML have advanced ADME predictions for small molecules, challenges persist for emerging modalities like peptides and biologics.

Purpose of the Study:

• To review the evolution of computational frameworks for predicting ADME and PK properties.
• To highlight methodological advancements in AI/ML for predicting drug behavior across diverse therapeutic modalities.
• To discuss emerging trends, limitations, and future perspectives of AI-driven ADMET predictions in drug design.

Main Methods:

• Summarizing advancements from traditional descriptor-based QSAR and classical ML to deep learning, GNNs, and chemical language models.
• Examining multimodal frameworks integrating experimental data, structural information, and biological context.
• Analyzing AI approaches incorporating sequence-, structure-, and mechanism-aware representations for biologics.
• Discussing foundation models for unified cross-modality ADMET modeling.

Main Results:

• AI/ML methods have significantly improved ADME predictions, especially for small molecules.
• Multimodal and sequence/structure-aware AI approaches enhance predictability for peptides, oligonucleotides, and antibody therapeutics.
• Foundation models offer unified representations and improved generalization for cross-modality ADMET modeling.
• These advancements show promise for more accurate and interpretable predictions.

Conclusions:

• AI-driven computational frameworks have evolved significantly, enabling more accurate PK/ADMET predictions from small molecules to complex biologics.
• Emerging multimodal and foundation-model approaches are crucial for addressing challenges in predicting the behavior of novel therapeutic modalities.
• The practical implementation of AI-driven ADMET predictions holds substantial potential to accelerate rational drug design and development.