Classification of 5-HT1A receptor agonists and antagonists using GA-SVM method
Abstract
Aim: To construct a reliable computational model for the classification of agonists and antagonists of 5-HT1A receptor.
Methods: Support vector machine (SVM), a well-known machine learning method, was employed to build a prediction model, and genetic algorithm (GA) was used to select the most relevant descriptors and to optimize two important parameters, C and r of the SVM model. The overall dataset used in this study comprised 284 ligands of the 5-HT1A receptor with diverse structures reported in the literatures.
Results: A SVM model was successfully developed that could be used to predict the probability of a ligand being an agonist or antagonist of the 5-HT1A receptor. The predictive accuracy for training and test sets was 0.942 and 0.865, respectively. For compounds with probability estimate higher than 0.7, the predictive accuracy of the model for training and test sets was 0.954 and 0.927, respectively. To further validate our model, the receiver operating characteristic (ROC) curve was plotted, and the Area-Under-the-ROC- Curve (AUC) value was calculated to be 0.883 for training set and 0.906 for test set.
Conclusion: A reliable SVM model was successfully developed that could effectively distinguish agonists and antagonists among the ligands of the 5-HT1A receptor. To our knowledge, this is the first effort for the classification of 5-HT1A receptor agonists and antagonists based on a diverse dataset. This method may be used to classify the ligands of other members of the GPCR family.
Keywords:
Methods: Support vector machine (SVM), a well-known machine learning method, was employed to build a prediction model, and genetic algorithm (GA) was used to select the most relevant descriptors and to optimize two important parameters, C and r of the SVM model. The overall dataset used in this study comprised 284 ligands of the 5-HT1A receptor with diverse structures reported in the literatures.
Results: A SVM model was successfully developed that could be used to predict the probability of a ligand being an agonist or antagonist of the 5-HT1A receptor. The predictive accuracy for training and test sets was 0.942 and 0.865, respectively. For compounds with probability estimate higher than 0.7, the predictive accuracy of the model for training and test sets was 0.954 and 0.927, respectively. To further validate our model, the receiver operating characteristic (ROC) curve was plotted, and the Area-Under-the-ROC- Curve (AUC) value was calculated to be 0.883 for training set and 0.906 for test set.
Conclusion: A reliable SVM model was successfully developed that could effectively distinguish agonists and antagonists among the ligands of the 5-HT1A receptor. To our knowledge, this is the first effort for the classification of 5-HT1A receptor agonists and antagonists based on a diverse dataset. This method may be used to classify the ligands of other members of the GPCR family.