Machine learning meets pK _a.

Area of Science:

• Computational Chemistry
• Drug Discovery
• Machine Learning Applications

Background:

• Accurate prediction of pK_a is crucial for drug design and chemical research.
• Existing computational tools vary in accuracy and accessibility.

Purpose of the Study:

• To develop and validate a novel, open-source Python tool for predicting small molecule pK_a values.
• To assess the performance of machine learning models for pK_a prediction.

Main Methods:

• A random forest machine learning model was trained on a curated dataset of monoprotic compounds.
• The model's performance was evaluated using five-fold cross-validation and two external validation sets.
• Performance metrics included mean absolute error (MAE), root mean squared error (RMSE), and correlation coefficient (r²).

Main Results:

• The random forest model achieved an MAE of 0.682, RMSE of 1.032, and r² of 0.82 in cross-validation.
• The tool demonstrated comparable performance to commercial software (Marvin) on external datasets.
• The developed model outperformed a recently published open-source pK_a prediction tool.

Conclusions:

• The Python-based pK_a prediction tool provides an accurate and accessible solution for researchers.
• Machine learning, particularly random forest, is effective for predicting small molecule pK_a.
• The open-source nature of the tool and data promotes wider adoption and further development.