Surabhi Dixit and Deepak Singla* Pages 303 - 310 ( 8 )
Background: Discovery of apicoplast as a drug target offers a new direction in the development of novel anti-malarial compounds, especially against the drug-resistant strains. Drugs such as azithromycin were reported to block the apicoplast development that leads to unusual phenotypes affecting the parasite. This phenomenon suggests that identification of new apicoplast inhibitors will aid in the anti-malarial drug discovery. Therefore, in this study, we developed a computational model to predict apicoplast inhibitors by applying state-of-the-art machine learning techniques.
Methods: We have used two high-throughput chemical screening data (AID-504850, AID-504848) from PubChem BioAssay database and applied machine learning techniques. The performance of the models were assessed on various types of binary fingerprints.
Results: In this study, we developed a robust computational algorithm for the prediction of apicoplast inhibition. We observed 73.7% sensitivity and 84% specificity along with 81.4% accuracy rate only on 41 PubChem fingerprints on 48 hrs dataset. Similarly, an accuracy rate of 75.8% was observed for 96 hrs dataset. Additionally, we observed that our model has ~70% positive prediction rate on the independent dataset obtained from ChEMBL-NTD database. Furthermore, the fingerprint analysis suggested that compounds with at least one heteroatom containing hexagonal ring would most likely belong to the antimalarial category as compared to simple aliphatic compounds. We also observed that aromatic compounds with oxygen and chlorine atoms were preferred in inhibitors class as compared to sulphur. Additionally, the compounds with average molecular weight >380Da and XlogP>4 were most likely to belong to the inhibitor category.
Conclusion: This study highlighted the significance of simple interpretable molecular properties along with some preferred substructure in designing the novel anti-malarial compounds. In addition to that, robustness and accuracy of models developed in the present work could be utilized to screen a large chemical library. Based on this study, we developed freely available software at http://deepaklab. com/capi. This study would provide the best alternative for searching the novel apicoplast inhibitors against Plasmodium.
Plasmodium, QSAR, machine learning, classification, inhibitor, virtual screening, apicoplast.
Infectious Diseases Laboratory, National Institute of Immunology, New Delhi, Centre for Agricultural Bioinformatics, Indian Agricultural Statistics Research Institute, Library Avenue, PUSA, New Delhi