Deep learning analyses of DNA sequences resolve the retention of the Duffy-null resistance to Plasmodium vivax malaria in Africa

P. vivax, the most geographically widespread human malaria parasite with millions of clinical cases per year, is however quasi absent in sub-Saharan Africa. Positive selection targeting the rs2814778 protective mutation, also known as the Duffy-null allele, may explain the absence (or quasi absence) of vivax in sub-Saharan Africa by a progressive purge of the pathogen due to a quasi-fixation of the Duffy-null allele and the resulting high rates of protected carriers in western, central and eastern populations. Yet, while positive selection has been clearly evidenced in admixed populations coexisting with vivax, the selection model currently admitted poorly explains the lack of the Duffy-null allele in Europe, or in Asia where the pathogen is mainly observed. In this article, several validated Deep Learning methods applied to high coverage sequence data obtained in 589 African individuals resolved this retention of the Duffy-null resistance to vivax in Africa. The CNN and GAN algorithms implemented in this study also predict a rise in frequency of the Duffy-null mutation due to selection 25-35 kya years ago in the western part of Africa, a geographical region and a time frame overlapping with the rise of another protective mutation, βS, the sickle-cell mutation protective at heterozygous state against the malaria caused by P. falciparum. In addition, the pattern of Duffy-null haplotypes highlights a quick spread of the Duffy-null allele in sub-Saharan Africa due to post-admixture selection events following the road of the recent Bantu expansion. Independent lines of evidence describing malaria as a life-threatening disease in West Africa from ∼30 kya, together with a rise in frequency followed by recent disseminations of the Duffy-null resistance, open new perspectives about both the history of malaria as a major human disease and the history of the main protective mutations in Africa.