Generation of Molecular Counterfactuals for Explainable Machine Learning Based on Core-Substituent Recombination.

Area of Science:

• Medicinal Chemistry
• Artificial Intelligence
• Computational Chemistry

Background:

• Black box machine learning models hinder interdisciplinary research due to lack of transparency.
• Explainable AI (XAI) is gaining importance to rationalize predictions and identify model limitations.
• Counterfactuals (CFs), or contrastive explanations, are minimally modified instances with opposing predictions, offering intuitive insights.

Purpose of the Study:

• To introduce a novel methodology for the systematic generation of CFs in medicinal chemistry.
• To address the limited investigation of CFs in this field despite their chemical intuitiveness.
• To provide a transparent and computationally simple approach for generating CFs.

Main Methods:

• Developed a new methodology for systematic CF generation.
• Centered the approach on well-defined structural analogues of test compounds.
• Ensured the method is transparent and computationally straightforward.

Main Results:

• Successfully generated a wealth of CFs for test sets.
• Demonstrated the effectiveness of the new methodology.
• The approach is computationally efficient and easy to implement.

Conclusions:

• The introduced methodology offers a transparent and efficient way to generate CFs in medicinal chemistry.
• This approach can enhance the interpretability of machine learning models in drug discovery.
• The method is freely available, promoting wider adoption and further research in explainable AI for chemistry.