Advances in SNP based phylogenomics

Kunming Institute of Botany, 2019

Deren Eaton, Columbia University


The goal of phylogenomics

Characterize evolutionary relationships from a subset of sampled genomes.

WGS vs. RAD-seq genomic sampling

Characterize whole genomes from a subset of sequenced markers.


Coalescent variation

Different genomic regions have different genealocial histories.

Can sparse SNP sampling reconstruct
genome-wide patterns?

Filtering and formatting to deal with missing data...

Outline: RAD-seq phylogenomics in ipyrad



1. ipyrad-analysis toolkit.

2. Gene tree extraction: concatenation.

3. Gene tree distributions: sliding window consensus.

4. Sticking with SNPs: genome-wide inference.

ipyrad-analysis toolkit (and toytree) and jupyter

ipyrad-analysis toolkit

Filter or impute missing data; easily distribute massively parallel jobs.

 
  import ipyrad.analysis as ipa

  # initiate an analysis tool with arguments
  tool = ipa.pca(data=data, ...)

  # run job (distribute in parallel)
  tool.run()

  # examine results
  ...
    

Quercus section Virentes

35 samples, 7 species plus outgroups.

ipyrad (ref): 58K loci, 51% missing, 484K SNPs. (30 min., 40 cores).


Hipp et al. (2014); Eaton et al. (2015); Cavender-Bares et al. (2015)

PCA: very sensitive to missing data

PCA: very sensitive to missing data


PCA: clear delimitation when (some) data are imputed


Outline: RAD-seq phylogenomics in ipyrad



1. ipyrad-analysis toolkit.

2. Gene tree extraction: concatenation.

3. Gene tree distributions: sliding window consensus.

4. Sticking with SNPs: genome-wide inference.

Missing data in phylogenetics

Missing data in phylogenetics

Missing data in phylogenetics

Window_extracter: extract, filter, format.

Reference mapped RAD loci can be "spatially binned" to form larger loci.

 
  import ipyrad.analysis as ipa

  # initiate an analysis tool with arguments
  tool = ipa.window_extacter(
      data=data,
      scaffold_idx=0, 
      start=0, 
      end=1000000,
  )

  # writes a phylip file
  tool.run()
    

Window_extracter: extract, filter, format.

Reference mapped RAD loci can be "spatially binned" to form larger loci.

Window_extracter: extract, filter, format.

Reference mapped RAD loci can be "spatially binned" to form larger loci.

Herbicide resistance among Amaranthus species.

Missing data in phylogenetics

Missing data in phylogenetics

Goal: A distribution of gene trees representing every species.

Missing data: Consensus sampling

Represent species by the consensus genotype across sampled individuals

treeslider: extract windows across chromosomes.

Runs raxml on windows and parses results into a "tree_table"

 
  # define population groups
  imap = {
      "sp1": ["a0", "a1", "a2", "a3"],
      "sp2": ["b0", "b1", "b2", "b3"],
      "sp3": ["c0", "c1", "c2", "c3"],
      "sp4": ["d0", "d1", "d2", "d3"],
  }

  # initiate an analysis tool with arguments
  tool = ipa.treeslider(
      data=data,
      window_size=1e6, 
      slide_size=1e6,
      imap=imap,
  )

  # distributes raxml jobs across all 1M windows in data set
  tool.run()
    

Consensus sampling yields 3X as many fully sampled loci.

One sample of each species: 12K/57K loci

Consensus for each species: 32K/57K loci

Hipp et al. (2014); Eaton et al. (2015); Cavender-Bares et al. (2015)

Sliding windows

How well do concatenated RAD windows represent gene tree variation?

RAxML gene trees.

Sliding windows

How well do concatenated RAD windows represent gene tree variation?

Astral species trees inferred from gene trees.

Clade weights (sensu Martin et al. 2017)

Chrom 1 weighted support for a (Cuba, Florida) vs (Cuba, Mexico)

Missing data in phylogenetics

SNP-based species trees


SNAPP: joint inference of gene trees and species trees.

(Bryant et al. 2012)


SVDquartets: infers quartet trees from SNPs and joins these into a species tree.

(Chifman and Kubatko 2014)

Advantages to SVDquartets-based methods

Each quartet is inferred independently: missing data has almost no effect.

It is fast! My software tool, tetrad is even much faster.

e.g., Guo Cen applied tetrad to infer a 218 species tree for Arundinarieae

What are the next frontiers in SNP based inference?

1. Faster methods for integrating over SNPs to infer species trees (i.e., faster-SNAPP)

2. Network inference by integrating over SNPs (i.e., SNP-based-phylonet)

3. Network inference from SNP

Patrick McKenzie

Background of SVDquartets


SVDquartets orders SNP patterns among quartet samples into 3 matrices.

One contains more patterns matching the tree (e.g., AABB)

Two contain patterns discordant with the tree (e.g., BABA, ABBA)

The matrix matching the tree aligns along rows/cols.

SVD decomposition measures matrix filling, and quickly finds best tree.

However...

SVDquartets discards the other two subtrees which contain
information relevant to inferring admixture (e.g., ABBA-BABA imbalance)
(Durand et al. 2011)

In addition...

SVDquartets only examines individual quartets at a time

Because quartets are not independent of each other, introgression of one
may affect multiple other quartets (Eaton et al. 2012, 2015).

Ideally, all taxa, or all quartets, would be examined simultaneously.

Joint inference of all quartet matrices for network inference

SVDquartets reduces SNP matrices to categorical results

SVDquartets reduces SNP matrices to categorical results

SVDquartets reduces SNP matrices to categorical results

SVDquartets reduces SNP matrices to categorical results

SVDquartets reduces SNP matrices to categorical results

Stacked count matrices

Unique fingerprint for different admixture scenarios

Simulate SNP data on a tree, train ML model on admixture scenario

ExtraTrees Classifier (scikit-learn)

Conclusions


1. With ipyrad-analysis it is easy to run dozens of analyses optimized for RAD missing-ness with a few lines of code.


2. Concatenating RAD loci in scaffold windows, and consensus or imputation sampling, dramatically improve the utility of RAD.


3. SNP based methods are in their infancy, but work well with RAD data.

Acknowledgements


Oaks data set:
Jeannine Cavender-Bares
Andrew Hipp
Antonio Gonzalez-Rodriguez

ipyrad development:
Isaac Overcast

Eaton lab members

Funding:
NSF DEB 1557059
Columbia University

Questions?


RADcamp wetlab and bioinformatics workshop in New York City Oct. 2019
Bring your DNA samples. Library preparation and sequencing will be free.
(sponsored by SSB, Columbia, CCNY).

https://radcamp.github.io/NYC2019/