Compound Prioritization through Meta-Analysis Enhances the Discovery of Antimicrobial Hits against Bacterial Pathogens

Abstract

The development of informatic tools to improve the identification of novel antimicrobials would significantly reduce the cost and time of drug discovery. We previously screened several plant (Xanthomonas sp., Clavibacter sp., Acidovorax sp., and Erwinia sp.), animal (Avian pathogenic Escherichia coli and Mycoplasma sp.), and human (Salmonella sp. and Campylobacter sp.) pathogens against a pre-selected small molecule library (n = 4182 SM) to identify novel SM (hits) that completely inhibited the bacterial growth or attenuated at least 75% of the virulence (quorum sensing or biofilm). Our meta-analysis of the primary screens (n = 11) using the pre-selected library (approx. 10.2 ± 9.3% hit rate per screen) demonstrated that the antimicrobial activity and spectrum of activity, and type of inhibition (growth versus virulence inhibitors) correlated with several physico-chemical properties (PCP; e.g., molecular weight, molar refraction, Zagreb group indexes, Kiers shape, lipophilicity, and hydrogen bond donors and acceptors). Based on these correlations, we build an in silico model that accurately classified 80.8% of the hits (n = 1676/2073). Therefore, the pre-selected SM library of 4182 SM was narrowed down to 1676 active SM with predictable PCP. Further, 926 hits affected only one species and 1254 hits were active against specific type of pathogens; however, no correlation was detected between PCP and the type of pathogen (29%, 34%, and 46% were specific for animal, human foodborne and plant pathogens, respectively). In conclusion, our in silico model allowed rational identification of SM with potential antimicrobial activity against bacterial pathogens. Therefore, the model developed in this study may facilitate future drug discovery efforts by accelerating the identification of uncharacterized antimicrobial molecules and predict their spectrum of activity.