Evaluation of Machine Learning Methods for Prediction of Pharmacological Classes

Pharmaceutical companies had an evolution with machine learning (ML) techniques1,2. This study evaluates six methods of classification based on the analysis of 2D structure. The diuretics drugs structures were obtained from DRUGBANK and ChEMBL to build a dataset of images. Following, Convolution Neural Network3 (CNN) was used to extract features. Next, K-Means (KMN)4, Random Forest (RF), Decision Tree (DT), Gaussian Process (GP), Adaboost (Adb) and Naive Bayes (NB) were applied to evaluate their performance in the classification process. As a result, their accuracies were 0.52, 0.56, 0.49, 0.42, 0.51, and 0.46 for KMN, RF, DT, GP, Adb and NB, respectively. In conclusion, the ML methods had 50% of accuracy. Thus, simulation using Deep Learning methods are in progress to improve the accuracy of classification to reach more reliable results.