MolP-PC: a multi-view fusion and multi-task learning framework for drug ADMET property prediction

Area of Science:

• Computational chemistry
• Pharmacology
• Artificial intelligence in drug discovery

Background:

• Accurate prediction of ADMET properties is vital for reducing drug development failure.
• Current deep learning models struggle with data sparsity and information loss from single-molecule representations and isolated tasks.

Purpose of the Study:

• To develop a multi-view fusion and multi-task deep learning framework (MolP-PC) for precise ADMET property prediction.
• To overcome limitations of existing deep learning approaches in drug property prediction.

Main Methods:

• Proposed MolP-PC framework integrating 1D molecular fingerprints, 2D molecular graphs, and 3D geometric representations.
• Utilized an attention-gated fusion mechanism and multi-task adaptive learning strategy.
• Incorporated ablation studies and interpretability analyses for validation.

Main Results:

• MolP-PC achieved optimal performance in 27 of 54 ADMET prediction tasks.
• Multi-task learning (MTL) significantly enhanced predictive performance, surpassing single-task models in 41 of 54 tasks.
• Demonstrated effective generalization in predicting pharmacokinetic parameters for anticancer compound Oroxylin A.

Conclusions:

• MolP-PC offers an efficient and interpretable approach for ADMET property prediction.
• Multi-view fusion captures multi-dimensional molecular information, enhancing model generalization.
• The framework provides a novel approach for molecular optimization and risk assessment in drug development, with potential for further improvement in predicting volume of distribution.