Developing novel computational prediction models for assessing chemical-induced neurotoxicity using naïve Bayes...

Area of Science:

• Computational toxicology
• Drug discovery
• cheminformatics

Background:

• Evaluating chemical neurotoxicity in early drug design requires reliable in vivo alternatives.
• Computational methods offer efficient prediction of potential drug-induced neurotoxicity.

Purpose of the Study:

• To develop and validate computational prediction models for drug-induced neurotoxicity.
• To identify a reliable virtual screening tool for early-stage drug design.

Main Methods:

• Utilized the naïve Bayes classifier to develop 110 classification models.
• Selected eight key molecular descriptors and 10 types of fingerprints (ECFP_10) with varying diameters.
• Evaluated model performance using training and external test sets.

Main Results:

• The NB-03 model, combining eight molecular descriptors and ECFP_10 fingerprints, achieved 90.5% training accuracy and 82.1% external test concordance.
• Naïve Bayes classifier outperformed recursive partitioning for neurotoxicity prediction.
• Identified structure alerts for characterizing neurotoxicity.

Conclusions:

• The NB-03 model serves as a reliable virtual screening tool for predicting neurotoxicity in early drug development.
• The identified structure alerts can guide chemists in optimizing chemical structures to reduce neurotoxicity risk.