Activity Detection of Estrogen Receptor α Ligand Binding Domain using Proposed Ensemble Learning

To know whether specific chemical compounds can interfere with bodily functions that might have a negative impact on health, scientists have developed an effective bagging-based ensemble model for better activity detection. Here, authors are utilizing a variety of physicochemical characteristics (molecular descriptors), where an in-silico method has been developed for the detection of the activity of estrogen receptor alpha ligand binding domain pharmaceuticals. To extract features, one uses the PaDEL-Descriptor. The 7741 drug molecules in the ER-LBD dataset have a total of 1444 features, with 438 of them being active and 7303 of them being inactive. This dataset has a lot of features and is very unbalanced. First, a problem with class imbalance is fixed by Smote algorithm and selecting important features with the CFS algorithm. An ensemble learning-based prediction model is created, where categorization was done using four existing classifiers i.e., SVM, decision tree, AdaBoost, and random forest. With the help of this strategy, the proposed bagging- based detection model provides outperformed accuracy. To assess the model's consistency across all target classes, the K-fold cross validation is used.