Voting-based ensemble method for prediction of bioactive molecules

Abstract

Machine learning has its tentacles spread over all major areas of science. The current rise in the amount of data being generated as necessitated its adoption in virtually all aspects including chemoinformatics. Several machine learning methods have been applied to the drug discovery process due to the importance of prediction of bioactivity before the release of drug into the market. The need for the most accurate method is hence evident. Majority voting ensemble is a method whose application is rare in predicting bioactive molecules. This study applies the method using different combination of commonly used classifiers as the base classifier on a chemical dataset of 8294 instances and 1024 attributes retrieved from the MDL Drug Data Report (MDDR). The accuracy of majority voting with the best combination of classifiers is found to be higher than the accuracy of the commonly used classifiers in the field, and makes it suitable for large chemical datasets.