From Black Boxes to Actionable Insights: A Perspective on Explainable Artificial Intelligence for Scientific...

Area of Science:

• Chemistry
• Artificial Intelligence
• Data Science

Background:

• Black-box machine learning models in chemistry lack transparency.
• Explainable Artificial Intelligence (XAI) is emerging to address this gap.
• XAI provides justification and actionable insights for chemical predictions.

Purpose of the Study:

• To survey and categorize XAI techniques applicable to chemistry.
• To illustrate the practical utility of XAI through case studies.
• To discuss challenges and future directions for XAI in chemistry.

Main Methods:

• Literature review of XAI techniques adapted for chemical applications.
• Categorization of XAI methods based on technical details.
• Presentation of case studies demonstrating XAI's practical use.

Main Results:

• A range of XAI techniques suitable for chemical applications have been identified and categorized.
• Case studies showcase XAI's utility in identifying carcinogenic molecules and guiding molecular optimization.
• Challenges include developing reliable explanations, ensuring robustness, and customization.

Conclusions:

• XAI is crucial for advancing machine learning in chemistry.
• Addressing current challenges and embracing new techniques like large language models will solidify AI's role.
• XAI empowers chemists with interpretable and actionable insights from complex models.