QSPR modeling to predict surface tension of psychoanaleptic drugs using the hybrid DA-SVR algorithm

Area of Science:

• Computational Chemistry
• Medicinal Chemistry
• Drug Design

Background:

• Surface tension is a critical physicochemical property influencing drug behavior.
• Psychoanaleptic drugs (psychostimulants and antidepressants) require precise property prediction for effective development.
• Existing methods for predicting surface tension in drug molecules are limited.

Purpose of the Study:

• To develop a robust Quantitative Structure-Property Relationship (QSPR) model for predicting the surface tension of psychoanaleptic drugs.
• To identify key molecular descriptors that govern surface tension in this drug class.
• To provide a reliable tool for the rational design of new psychoanaleptic medications.

Main Methods:

• Utilized a dataset of 112 psychoanaleptic drug molecules.
• Applied genetic algorithm (GA) with feature selection methods (OLS, PLS, SVM) to identify ten AlvaDesc descriptors.
• Constructed predictive models using the Dragonfly Algorithm with Support Vector Regressor (DA-SVR).

Main Results:

• The GA-SVM-based DA-SVR model demonstrated superior predictive performance (R² = 0.98142, Q²LOO = 0.98142).
• Model validation using an external test set confirmed its reliability and applicability.
• Ten key AlvaDesc descriptors were identified as crucial for surface tension prediction.

Conclusions:

• The developed QSPR model accurately predicts surface tension for psychoanaleptic drugs.
• The model can guide the design of novel psychoanaleptic drugs with tailored surface tension characteristics.
• This approach enhances the efficiency of drug discovery and development.