Distinguishing compounds with anticancer activity by ANN using inductive QSAR descriptors.

Area of Science:

• Computational Chemistry
• Pharmacology
• Artificial Intelligence in Drug Discovery

Background:

• Accurate prediction of anticancer drug activity is crucial for efficient drug development.
• Traditional methods for drug screening can be time-consuming and costly.
• The need for advanced computational models to aid in identifying potential anticancer agents.

Purpose of the Study:

• To develop and validate an artificial neural network (ANN) model for predicting anticancer drug activity.
• To assess the efficacy of quantitative structure-activity relationship (QSAR) descriptors in conjunction with ANN for drug classification.
• To provide a robust computational tool for the early-stage identification of potential anticancer compounds.

Main Methods:

• Utilized a feed-forward neural network with a back-propagation training algorithm.
• Employed 30 'inductive' quantitative structure-activity relationship (QSAR) descriptors.
• Trained and tested the model on a dataset of 380 drugs (122 anticancer, 258 non-anticancer).

Main Results:

• Achieved 84.28% accuracy in separating anticancer from non-anticancer compounds using 30 QSAR descriptors.
• The ANN model demonstrated high performance with Q(pred) = 74.28%, sensitivity = 0.9285, specificity = 0.7857, and MCC = 0.6998.
• The model successfully assigned anticancer character to literature-based trial drugs, confirming its predictive power.

Conclusions:

• Artificial neural networks, combined with QSAR descriptors, offer a powerful approach for predicting anticancer drug activity.
• The developed model provides a validated and accurate method for drug classification, aiding in drug discovery pipelines.
• This computational strategy can significantly accelerate the identification of novel anticancer therapeutics.