Supplementary figures. A hybrid resampling algorithms SMOTE and ENN based deep learning models for identification of Marburg virus inhibitors

Supplementary Figure 1 A: The loss value of the training and test dataset are plotted against the number of epoch on our proposed ANN modelusing resampling algorithms of (a) Undersampling, (b) Oversampling, and (c) SMOTE-ENN, where the SMOTE-ENN+ANN hybrid model shows lowest loss value (0.12).

Supplementary Figure 1B: The loss value of the training and test dataset are plotted against the number of epoch on our proposed CNN model using resampling algorithms of (a) Undersampling, (b) Oversampling, and (c) SMOTE-ENN, where the SMOTE-ENN+CNN hybrid model shows lowest loss value (0.16).

Supplementary Figure 2: Bar chart showing specificity and sensitivity for the deep learning models using resampling algorithm of (a) Undersampling, (b) Oversampling, and (c) SMOTE-ENN, where the SMOTE-ENN+ANN hybrid model shows maximum sensitivity (0.95)  and SMOTE-ENN+ CNN shows a maximum specificity (1.00) of the anti-MARV test dataset.

Supplementary Figure 3: Confusion matrix of hybrid models (a) SMOTE-ENN+ANN and (b) SMOTE-ENN+CNN showing the proportion of each predicted class (x-axis) for molecules in each true class (y-axis); “0” represents inactive molecules and “1” represents active molecules.