Identifying genomic regions responsible for recent adaptation is a major challenge in population genetics. Particularly in humans, the task of confidently detecting the action of recent adaptive natural selection (or positive selection) has proved troublesome. Indeed there is considerable controversy over whether recent positive selection has a substantial impact on human genetic variation. The work proposed here will address this problem by creating a more complete map of positive selection across many human populations, identifying selection on de novo mutations as well as selection on previously standing variation. Specifically, the proposed research seeks to construct a scan for positives election that is more robust and accurate than any currently existing methods (Aim 1). This tool will utilize supervised machine learning techniques allowing it combine information from a number of existing tests for natural selection, and will be tested extensively on a large suite of population genetic simulations presenting a wide range of potentially confounding scenarios. This tool will then be released to the public. Next, it will be applied to 26 human populations in which a large sample of genomes have been sequenced by the 1000 Genomes Project (Aim 2), revealing similarities and differences in the tempo, mode, and targets of adaptive evolution across human populations. Finally, because selection on both beneficial and deleterious mutations skews genetic variation, our method will be used to identify regions of the genome least affected by natural selection, which will in turn be used to produce more accurate inferences of human demographic histories (Aim 3). The mentored phase of this work will be performed within the Department of Genetics at Rutgers University. This is an intellectually stimulating environment with numerous journal clubs, an excellent seminar series, and several other research groups using computational techniques. The project will be performed under the stewardship of Dr. Andrew Kern, from whom the candidate will also receive training in machine learning and population genetics. Dr. Schrider will also receive training in population genetics and guidance from Dr. Jody Hey (Co-mentor) at nearby Temple University. This training will help Dr. Schrider acquire skills that will aid not only in the completion of the proposed work but also his transition to principle investigator of an internationally recognized independent research program studying the evolutionary forces driving patterns of human genetic variation.
Detecting genes underpinning recent human adaptation remains a major challenge, and such genes are often associated with human disease. The work proposed here seeks to use supervised machine learning techniques to detect genomic regions responsible for recent adaptation across 26 different human populations. This work will also clarify human population size and migration histories, information that has implications for the prevalence of disease-causing mutations and efforts to identify them.
| Schrider, Daniel R; Kern, Andrew D (2018) Supervised Machine Learning for Population Genetics: A New Paradigm. Trends Genet 34:301-312 |
| Kern, Andrew D; Schrider, Daniel R (2018) diploS/HIC: An Updated Approach to Classifying Selective Sweeps. G3 (Bethesda) 8:1959-1970 |
| Schrider, Daniel R; Kern, Andrew D (2017) Soft Sweeps Are the Dominant Mode of Adaptation in the Human Genome. Mol Biol Evol 34:1863-1877 |
