Explainable artificial intelligence for molecular design in pharmaceutical research.

Area of Science:

• Artificial Intelligence
• Molecular Design
• Computational Chemistry

Background:

• Machine learning (ML) models, particularly deep learning, are advancing molecular design.
• The "black-box" nature of these ML models hinders understanding and acceptance of their predictions.
• Explainable AI (XAI) is essential for bridging this gap, especially in experimental science.

Purpose of the Study:

• To examine challenges and opportunities for XAI in molecular design.
• To evaluate the benefits of incorporating domain-specific knowledge into XAI.
• To discuss limitations in evaluating chemical language models for molecular design.

Main Methods:

• Review of current XAI methods in the context of molecular design.
• Analysis of domain-specific knowledge integration for XAI.
• Discussion on the evaluation of chemical language models.

Main Results:

• XAI methods need to provide human-centered, transparent, and interpretable explanations.
• Domain knowledge can refine ML models, aid experimental design, and support hypothesis testing.
• Current evaluation methods for chemical language models in molecular design are limited.

Conclusions:

• XAI is vital for the practical application of ML in molecular design.
• Tailoring XAI with domain expertise is key to unlocking its full potential.
• Further development is needed for robust evaluation of AI tools in drug discovery.