Fast virtual screening of sesquiterpene lactones from Asteraceae with potential antileishmanial activity

Leishmanioses is a human tropical parasitic disease that causes approximately 50,000 death cases annually. Secondary metabolites play an important role to propose new active lead structures, and studies highlights antiprotozoal activities of sesquiterpene lactones (SLs)¹. In view of this, we performed a virtual screening (VS) in an in-house databank of 1328 SLs of Asteraceae, corresponding to 2323 botanical occurrences (B.O. – number of times that a compound appears in different species) using fragment descriptors and Randon Forest (RF). We select from CHEMBL database a diversity set of 269 compounds, which were screened against L. donovani amastigotes (MHOM/ET/67/L82) to generate a RF model and were classified using -logIC₅₀ (mol/L)= pIC₅₀ values, being active (> 5) and inactive (< 4.5) and, between 4.5 and 5 were excluded to reduce border effect between both classes. DRAGON v. 6.0 generated descriptors, which with constant and near constant values, standard deviation < 10^-4, and pair correlation ≥0.90 were excluded. The 148 remaining descriptors and class variable were exported to Knime 2.7.2 that was used to perform all analysis process described hereinafter. Data were divided in train and test set and only 7 variables were selected by backward feature elimination method. RFs were generated using WEKA nodes. Table 1 summarizes match rates of RF of 269 compounds of CHEMBL database. RF selected 476 SLs (658 B.O.) with potential activity which are present mainly in Senecioneae and Eupatorieae tribes. The VS, that is part of the activities of ResNetNPND (http://www.uni-muenster.de/ResNetNPND/), is rapid and can be applied to larger natural products databases.

Tab. 1: Summary of match rates.

	Train			Validation a		Test
	Samples	Match	%Match	Match	%Match	Samples	Match	%Match
Active	97	95	97.8	78	80.4	24	22	91.7
Inactive	118	116	98.3	101	85.6	30	23	76.7
Overall	215	211	98.1	179	83.3	54	45	83.3
a – cross-validation (10 stratified groups)

References:

[1] Schmidt J, et al. Curr. Top. Med., 2012, 19, 2128 – 2175.