Filling the Gap in and Evaluation for Saturated Fluorine-Containing Derivatives With Machine Learning

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Lipophilicity (logP) and acidity/basicity (pKa) are critical physicochemical properties influencing drug behavior.
• Accurate prediction of logP and pKa is essential for successful early-stage drug discovery.
• Standard prediction methods struggle with saturated fluorine-containing compounds, limiting their utility.

Purpose of the Study:

• To develop accurate predictive models for lipophilicity and acidity/basicity of saturated fluorine-containing compounds.
• To overcome limitations of existing prediction methods for this specific chemical class.
• To provide open-source tools for targeted molecular design.

Main Methods:

• Compiled a unique dataset of fluorinated and non-fluorinated derivatives with experimental logP and pKa values.
• Evaluated, trained, or fine-tuned over 40 machine learning models (linear, tree-based, neural networks).
• Utilized substructure mask explanation (SME) to validate model findings and the role of fluorine.

Main Results:

• Developed highly accurate machine learning models for predicting logP and pKa in saturated fluorine-containing compounds.
• Confirmed the significant impact of fluorinated substituents on lipophilicity and acidity/basicity.
• Demonstrated the consistency and reliability of the developed predictive models.

Conclusions:

• The developed models offer an optimal approach for predicting lipophilicity and acidity/basicity in challenging fluorinated compounds.
• Open-sourced resources (GitHub, pip, conda, KNIME) enable public access for targeted molecular design.
• This work facilitates the development of novel fluorinated drug candidates.