Development of machine learning models for the discovery of new antiviral compounds against yellow fever virus

Abstract

Yellow fever virus (YFV) is an acute viral hemorrhagic disease transmitted by infected mosquitoes. The YFV is endemic in tropical areas of Africa and Central and South America. Large epidemics of yellow fever occur when infected people introduce the virus into heavily populated areas with high mosquito density and where most people have little or no immunity, due to lack of vaccination. In the last two years there was a resurgence of yellow fever (YF) outbreaks close to large urban centers. In Brazil, a mass vaccine campaign was conducted in order to protect the population and control viral presence in those regions, since the only prophylactic method available is the 17D attenuated virus vaccine. The lack of treatment for YF, as well as for homologous infections, e.g. zika and dengue, highlights the importance of those flaviviruses as a public health concern. The increase in drug approval costs followed by a decrease in R&D productivity raised questions on how industry and academia could improve its efficiency. With the advancement in hardware and data availability, new tools based on machine learning methods have being developed for predicting drug candidates. Machine learning methods were introduced in drug discovery as means to reduce costs and increase accuracy in drug development. The use of predictive machine learning models using previous data from HTS campaigns or public databases in lead discovery can increase success rate and save resources by picking compounds with good activity and ADME profiles. This project aims to develop predictive machine learning models using public and in-house datasets of YFV assays. The model will then be used to virtually screen chemical libraries in order to select compounds with good activity and ADME profiles against YFV. These compounds will then ultimately be tested by our collaborators or in our own laboratories. (AU)