Abstract Detail

Phylogenomic Perspectives on Reproductive Isolation and Introgression

McKenzie, Patrick [1], Eaton, Deren [2].

Inferring historical introgression using machine learning on phylogenetic invariants.

Detecting historical introgression is a common goal of plant evolutionary studies. Current methods include: (1) likelihood-based inference from parameter-rich demographic models (e.g., IMa, G-PhoCS); (2) pseudo-likelihood based inference of phylogenetic networks from inferred gene trees (e.g., SNAQ); and (3) non-parametric inference from the distribution of discordant SNP patterns under different phylogenetic hypotheses (e.g., ABBA-BABA tests). Each method has its limitations: the first does not scale to large trees and requires a priori placement of admixture edges; the second relies on resolved gene trees and cannot use SNP data; and the third only applies to subsets of four or five taxa at a time, making it difficult to interpret results of many interdependent tests. To address this, we’ve developed a non-parametric method for inferring placement of admixture edges on large phylogenetic trees from large SNP data sets. Our method (simcat) analyzes SNPs segregating among quartet samples in a data set (similar to ABBA-BABA tests), but examines all quartet information simultaneously (similar to the pseudolikelihood methods) as a large multi-dimensional matrix. Coalescent simulations are used to train a machine learning algorithm to discriminate multi-dimensional SNP matrix patterns that result under different admixture scenarios. Using dimensionality reduction we demonstrate the ability for this method to classify admixture events from simulated SNP data and we demonstrate its effectiveness on simulated and empirical data sets. Further research will focus on evaluating the robustness of the method to missing data and uncertainty in species tree estimation.

1 - Columbia University, Ecology, Evolution, and Environmental Biology, 1200 Amsterdam Ave., 10th Floor Schermerhorn Ext., New York, NY, 10027, USA
2 - Columbia University, Department of Ecology, Evolution, and Environmental Biology, Schermerhorn Ext. Office 1007, 1200 Amsterdam Avenue, New York, NY, 10027, USA

machine learning.

Presentation Type: Symposium Presentation
Number: 0003
Abstract ID:884
Candidate for Awards:None

Copyright © 2000-2019, Botanical Society of America. All rights reserved