QSAR classification and regression models for β-secretase inhibitors using relative distance matrices

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Alzheimer's disease (AD) prevalence is increasing globally.
• Developing effective β-secretase (BACE1) inhibitors is critical for AD treatment.
• Robust Quantitative Structure-Activity Relationship (QSAR) models are needed to predict inhibitor activity.

Purpose of the Study:

• To propose a novel QSAR modeling strategy using relative distance matrices.
• To evaluate the efficacy of this approach for predicting β-secretase inhibitor activity.
• To develop accurate classification and regression models for BACE1 inhibitors.

Main Methods:

• Utilized relative distance matrices as input features for QSAR algorithms.
• Employed machine learning algorithms including Support Vector Machine (SVM), Tree Complex, and Gaussian Process.
• Applied rigorous validation techniques such as cross-validation, bootstrapping, and y-randomization.

Main Results:

• Achieved classification accuracy and Area Under the ROC Curve (AUC) values close to 100%.
• Obtained regression R-squared (r²) values close to 1.0 and Root Mean Square Error (RMSE) values around 0.1.
• Demonstrated high performance in both training and external validation sets.

Conclusions:

• The proposed QSAR method using relative distance matrices is highly effective for predicting β-secretase inhibitor activity.
• This approach shows significant promise for accelerating the discovery of novel Alzheimer's disease therapeutics.
• The validated models provide a reliable framework for virtual screening and lead optimization.