ChemStable: a web server for rule-embedded naïve Bayesian learning approach to predict compound stability.

Area of Science:

• Computational Chemistry
• Chemical Informatics
• Drug Discovery

Background:

• Compound chemical stability is critical for accurate bioassay results, as instability can lead to false positives or negatives.
• Existing methods for predicting stability are limited, necessitating new computational approaches.

Purpose of the Study:

• To develop and validate an in silico prediction service for organic compound stability.
• To improve the accuracy of stability predictions and minimize false negatives in bioassays.

Main Methods:

• Utilized the COMDECOM dataset of experimental stability data for 9,746 compounds.
• Applied rule-embedded naïve Bayesian learning based on atom center fragment (ACF) features.
• Developed a structural pattern recognition algorithm to assign stability probabilities to ACFs.

Main Results:

• Achieved an Area Under the Curve (AUC) of 84% for stability prediction.
• Obtained a tenfold cross-validation accuracy of 76.5%.
• Integrated a rule-based module to further reduce false negatives.

Conclusions:

• The developed in silico method provides a novel and effective approach for predicting organic compound stability.
• This service has the potential to enhance the reliability of bioassays by identifying unstable compounds early.
• This represents the first in silico prediction service for organic compound stabilities.