Prediction of aromatase inhibitory activity using the efficient linear method (ELM)

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Drug discovery

Background:

• Aromatase inhibitors (AIs) are crucial for breast cancer treatment.
• Existing in silico prediction methods like artificial neural networks (ANN) and support vector machines (SVM) have limitations in simplicity and interpretability.
• There is a need for improved quantitative structure-activity relationship (QSAR) models.

Purpose of the Study:

• To develop a simple and interpretable QSAR model for predicting aromatase inhibitors.
• To introduce an efficient linear method (ELM) with a built-in feature importance estimator.
• To compare the performance of ELM with other machine learning techniques.

Main Methods:

• An efficient linear method (ELM) was employed to build the QSAR model.
• Optimal ELM parameters were determined using a genetic algorithm.
• The model's performance was validated using 10-fold cross-validation and compared against ANN, SVM, and decision tree methods.

Main Results:

• The ELM model demonstrated robust predictive performance with Matthews Correlation Coefficient (MCC) values of 0.64 for steroidal AIs and 0.56 for non-steroidal AIs.
• Comparative analysis confirmed the effectiveness of the ELM approach.
• Key molecular descriptors influencing AI activity were identified, providing mechanistic insights.

Conclusions:

• The ELM method offers a simple yet powerful approach for predicting aromatase inhibitors.
• Molecular shape, polarizability, terminal primary C(sp3) functional groups, and electronegativity are important factors for AI activity.
• This study provides valuable insights for the design of novel breast cancer therapeutics.