Optimal Modeling of Anti-Breast Cancer Candidate Drugs Based on Graph Model Feature Selection

Area of Science:

• Oncology
• Computational Chemistry
• Bioinformatics

Background:

• Breast cancer remains a leading cause of mortality in women worldwide.
• Estrogen receptor alpha (ERα) targeted therapies significantly improve breast cancer patient outcomes.
• Urgent need exists for efficient methods to discover and develop novel ERα-inhibiting anticancer drugs.

Purpose of the Study:

• To abstract the drug screening process into an optimization problem.
• To develop a robust computational model for predicting ERα biological activity and ADMET properties of anticancer compounds.
• To enhance the accuracy and efficiency of anticancer drug discovery.

Main Methods:

• Utilized a graph model to extract low-dimensional features from compound attributes.
• Employed kernel functions for feature mapping to high-dimensional space.
• Constructed quantitative analysis and classification models using support vector machines (SVM).
• Applied Sequential Least Square Programming (SLSQP) for model optimization.

Main Results:

• The graph model demonstrated reduced error rates (MAE, MSE, RMSE) compared to Principal Component Analysis (PCA) on anticancer datasets.
• Achieved significant improvements in classification prediction, with enhanced recall (19.5%) and precision (12.41%) over PCA.
• Identified optimal biological activity values, including IC50_nM of 34.6 and pIC50 of 7.46.

Conclusions:

• The proposed graph model offers a superior approach for feature extraction in anticancer drug screening.
• The developed SVM-based models provide accurate predictions for ERα biological activity and ADMET properties.
• This computational strategy effectively aids in the identification of potent anticancer drug candidates.