A computer-driven approach to discover natural product leads for methicillin-resistant staphylococcus aureus infection therapy †

Financial support from Fundacao para a Ciencia e Tecnologia (FCT) Portugal, under Project PTDC/QUIQUI/119116/2010 and grants SFRH/BPD/108237/2015 (F.P.) and IF/00700/2014 (S.P.G.) are greatly appreciated. This work was supported by the LAQV, which is financed by national funds from FCT/MEC (UID/QUI/50006/2013) and co-financed by the ERDF under the PT2020 Partnership Agreement (POCI-01-0145-FEDER-007265). This work was also supported by the UCIBIO, which is financed by national funds from FCT/MEC (UID/Multi/04378/2013) and co-financed by the ERDF under the PT2020 Partnership Agreement (POCI-010145-FEDER-007728). The NMR spectrometers are part of The National NMR Facility, supported by FCT (RECI/BBB-BQB/0230/2012).The risk of methicillin-resistant Staphylococcus aureus (MRSA) infection is increasing in both the developed and developing countries. New approaches to overcome this problem are in need. A ligand-based strategy to discover new inhibiting agents against MRSA infection was built through exploration of machine learning techniques. This strategy is based in two quantitative structure–activity relationship (QSAR) studies, one using molecular descriptors (approach A) and the other using descriptors (approach B). In the approach A, regression models were developed using a total of 6645 molecules that were extracted from the ChEMBL, PubChem and ZINC databases, and recent literature. The performance of the regression models was successfully evaluated by internal and external validation, the best model achieved R 2 of 0.68 and RMSE of 0.59 for the test set. In general natural product (NP) drug discovery is a time-consuming process and several strategies for dereplication have been developed to overcome this inherent limitation. In the approach B, we developed a new NP drug discovery methodology that consists in frontloading samples with 1D NMR descriptors to predict compounds with antibacterial activity prior to bioactivity screening for NPs discovery. The NMR QSAR classification models were built using 1D NMR data ( 1 H and 13 C) as descriptors, from crude extracts, fractions and pure compounds obtained from actinobacteria isolated from marine sediments collected off the Madeira Archipelago. The overall predictability accuracies of the best model exceeded 77% for both training and test sets.publishersversionpublishe