TargIDe: a machine-learning workflow for target identification of molecules with antibiofilm activity against Pseudomonas aeruginosa

Abstract

Bacterial biofilms are a source of infectious human diseases and are heavily linked to antibiotic resistance. Pseudomonas aeruginosa is a multidrug-resistant bacterium widely present and implicated in several hospital-acquired infections. Over the last years, the development of new drugs able to inhibit Pseudomonas aeruginosa by interfering with its ability to form biofilms has become a promising strategy in drug discovery. Identifying molecules able to interfere with biofilm formation is difficult, but further developing these molecules by rationally improving their activity is particularly challenging, as it requires knowledge of the specific protein target that is inhibited. This work describes the development of a machine learning multitechnique consensus workflow to predict the protein targets of molecules with confirmed inhibitory activity against biofilm formation by Pseudomonas aeruginosa. It uses a specialized database containing all the known targets implicated in biofilm formation by Pseudomonas aeruginosa. The experimentally confirmed inhibitors available on ChEMBL, together with chemical descriptors, were used as the input features for a combination of nine different classification models, yielding a consensus method to predict the most likely target of a ligand. The implemented algorithm is freely available at https://github.com/BioSIM-Research-Group/TargIDe under licence GNU General Public Licence (GPL) version 3 and can easily be improved as more data become available.

Keywords: Biofilms; Ligand targets; Machine learning; Pseudomonas aeruginosa.

Fig. 1

Molecular representation of the protein targets included in this study and the number of ligands used per target

Fig. 2

Curated training database 10 most relevant features using RF and selected after the recursive feature elimination and cross-validation process

Fig. 3

Bar graphs of the evaluation of several classification model predictions on the test (10%) and training (90%) datasets using the 10 most relevant features. The cross-validation values are also shown. Bar values for train/test/cross-validation represent, from left to right, recall, precision, F1 score, AUC. and classification accuracy (CA).

Fig. 4

Applicability domain calculated for the train and test dataset calculated by PCA bounding box. The first principal component (PCA1) is the direction in which the data varies the most. The second principal component (PCA2) is orthogonal to the first and represents the direction of maximum variance that is not captured by the first principal component

Fig. 5

ROC One-vs-rest curve for the 5 target (gene) classes with higher number of samples considering the best performing model (gradient boosting) for the training SigMol dataset. The graph shows the mean of the true positive rate (TP rate) and the false positive rate (FP rate)

Fig. 6

Workflow of machine learning models implementation in Orange software