MalariaSED: a deep learning framework to decipher the regulatory contributions of noncoding variants in malaria parasites

Malaria remains one of the deadliest infectious diseases. Transcriptional regulation effects of noncoding variants in this unusual genome of malaria parasites remain elusive. We developed a sequence-based, ab initio deep learning framework, MalariaSED, for predicting chromatin profiles in malaria parasites. The MalariaSED performance was validated by published ChIP-qPCR and TF motifs results. Applying MalariaSED to similar to 1.3 million variants shows that geographically differentiated noncoding variants are associated with parasite invasion and drug resistance. Further analysis reveals chromatin accessibility changes at Plasmodium falciparum rings are partly associated with artemisinin resistance. MalariaSED illuminates the potential functional roles of noncoding variants in malaria parasites.

Automated summary

This study developed a deep learning framework, called MalariaSED, to predict chromatin profiles in malaria parasites. Analysis of approximately 1.3 million variants showed that geographically differentiated noncoding variants are associated with parasite invasion and drug resistance. Furthermore, changes in chromatin accessibility in Plasmodium falciparum rings were found to be partly related to artemisinin resistance.