The 85th Annual Meeting of the American Association of Physical Anthropologists (2016)

A discriminative model for inferring genome-wide maps of Neandertal and Denisovan ancestry


1Department of Genetics, Harvard Medical School, 2Broad Institute of Harvard and MIT, 3Howard Hughes Medical Institute, Harvard Medical School

April 14, 2016 1:15, Imperial Ballroom B Add to calendar

Understanding the impact of these ancient admixture events on evolution and phenotypes is a central goal in human population genomics. A major step to this end is to infer the archaic local ancestry of modern human populations. I will present a discriminative statistical model called a Conditional Random Field that can combine multiple lines of evidence to infer archaic local ancestry. We have applied the CRF to infer both Neanderthal and Denisovan ancestries in the diverse set of human populations sequenced to high-coverage in the Simons Genome Diversity Project (SGDP) that diverse set of modern human populations.

Our statistical method identifies around 38,000 Neanderthal-derived alleles and around 25,000 Denisovan-derived alleles. Using the confidently inferred Denisovan ancestries across multiple Melanesian individuals, we can reconstruct about 150 Mb of the genome of the introgressing Denisovan. We observe that the proportion of both Denisovan and Neanderthal local ancestry in Melanesian populations is reduced in regions of the genome with strong background selection. This observation is consistent with a model in which Neanderthal and Denisovan alleles are subject to strong purifying selection in the admixed Melanesian populations analogous to the previous observation of strong purifying selection against Neanderthal alleles in non-Africans. In addition, we document a number of regions with elevated proportions of archaic ancestry which represent putative candidates for adaptive introgression.