In Silico ADMET Profiling: Evolution from Traditional Models to Deep Learning Techniques

Area of Science:

• Computational chemistry and cheminformatics
• Pharmacology and toxicology
• Artificial intelligence in drug discovery

Background:

• Traditional experimental methods for ADMET profiling are expensive, slow, and limited in scalability, contributing to high clinical trial failure rates.
• Machine learning (ML) offers a powerful approach to model complex structure-activity relationships for ADMET properties, driven by growing chemical and biological data.
• Integrating explainable AI, transfer learning, and active learning can address data scarcity and improve regulatory acceptance of ML models.

Purpose of the Study:

• To review and evaluate state-of-the-art machine learning methods for predicting ADMET properties.
• To highlight the importance of molecular representations, data quality, and model evaluation for building reliable predictive models.
• To emphasize recent advancements like multi-task learning and data-driven feature engineering for improved prediction accuracy.

Main Methods:

• Examination of various machine learning algorithms including support vector machines, gradient boosting, random forests, and graph neural networks.
• Assessment of software tools used for forecasting ADMET properties.
• Discussion of techniques for creating trustworthy predictive models, focusing on molecular representations, dataset quality, and evaluation metrics.

Main Results:

• Machine learning models demonstrate significant potential in predicting complex ADMET properties.
• Advancements in multi-task learning and feature engineering enhance prediction accuracy across multiple ADMET endpoints.
• Web-based tools and hyperlinks are provided to facilitate in-silico ADMET research.

Conclusions:

• Machine learning-based ADMET prediction is becoming increasingly reliable, streamlining drug development.
• These methods contribute to the creation of safer and more effective therapeutics.
• The study provides resources to aid in-silico ADMET research for drug candidates.