[Prediction of drug bioavailability by genetic algorithm and artificial neural network]

Area of Science:

• Computational chemistry
• Pharmacokinetics
• Machine learning in drug discovery

Background:

• Accurate prediction of drug bioavailability is crucial for drug development.
• Traditional methods for bioavailability assessment can be time-consuming and costly.
• Artificial neural networks (ANNs) offer a promising approach for predictive modeling.

Purpose of the Study:

• To develop and optimize an artificial neural network (ANN) system for predicting drug bioavailability.
• To enhance the predictive accuracy of ANNs through optimization using a genetic algorithm (GA).

Main Methods:

• An artificial neural network model was constructed for drug bioavailability prediction.
• A genetic algorithm (GA) was employed to optimize the weights of the ANN.
• Key molecular descriptors including volume, refractivity, lgP(c), hydration, polarizability, E(HOMO), and E(LUMO) were used as inputs.

Main Results:

• The GA-optimized ANN model achieved a high prediction precision of 95.9% for average drug bioavailability.
• The GA effectively optimized the ANN, leading to improved predictive performance.

Conclusions:

• The developed GA-ANN model demonstrates significant potential for accurately forecasting drug bioavailability.
• This computational approach can aid in the early stages of drug development by predicting bioavailability efficiently.