Machine learning for artemisinin resistance in malaria treatment across in vivo-in vitro platforms.

Drug resistance has been rapidly evolving with regard to the first-line malaria treatment, artemisinin-based combination therapies. It has been an open question whether predictive models for this drug resistance status can be generalized across in vivo-in vitro transcriptomic measurements. In this study, we present a model that predicts artemisinin treatment resistance developed with transcriptomic information of Plasmodium falciparum. We demonstrated the robustness of this model across in vivo clearance rate and in vitro IC50 measurement and based on different microarray and data processing modalities. The validity of the algorithm is further supported by its first placement in the DREAM Malaria challenge. We identified transcription biomarkers to artemisinin treatment resistance that can predict artemisinin resistance and are conserved in their expression modules. This is a critical step in the research of malaria treatment, as it demonstrated the potential of a platform-robust, personalized model for artemisinin resistance using molecular biomarkers.