grur missing data visualization analysis

Thierry Gosselin

2020-10-20

Source: vignettes/vignette_missing_data_analysis.Rmd

Objectives

Learn how to visualize missing genotypes in your genomic dataset with the function grur::missing_visualization (time = 15 min).

Workflow

The function missing_visualization in grur uses various genomic input files and conduct identity-by-missingness analyses (IBM) using Principal Coordinates Analysis (PCoA), also called Multidimensional Scaling (MDS) and RDA (Redundancy Analysis) to highlight missing data patterns. Figures and summary tables of missing information at the marker, individual and population level are generated. Below, the simplest form of the function to get results. More options are available, please see the function documentation.

Prepare your R workspace

Clean your desk

rm(list = ls())

Follow the instruction to install grur

Load the required libraries:

Set your working directory, e.g.:

setwd("~/Documents/test_missing_visualization_vignette")

Note: running codes in chunks inside R Notebook might cause problem, run it outside in the console (the default here).

Download the test data

Dataset: in this example, we use the data in Ferchaud and Hansen (2015 and 2016) paper. The code below gets the vcf from Dryad directly. But you can skip the step if it’s already in the folder.

writeBin(httr::content(httr::GET("http://datadryad.org/bitstream/handle/10255/dryad.97237/sticklebacks_Danish.vcf?sequence=1"), "raw"), "stickleback_ferchaud_2015.vcf")

With a vcf you also need a strata file (indicating population groupings)

  • it’s a tab-delimited file or object, in the global environment.
  • requires a minimum of 2 columns: INDIVIDUALS and STRATA.
  • the STRATA column and the remaining columns can be any hierarchical groupings you like.
  • because in the strickleback vcf the population’s ids are contained in the name of the individuals, we can easily extract this info with string command.
  • here, creating from scratch the strata file is beyond the point of this tutoria (??radiator::read_strata).
  • if you don’t have this kind of individual naming scheme, you can also make the strata file by hand, the old fashion way.
  • if you’ve used for your RADseq pipeline, the strata file is similar to a stacks population map file. Just make sure you have the required column names: INDIVIDUALS and STRATA.

to download the strata for this example: strata link

Run the function

This is the simplest way to run the function:

ibm <- grur::missing_visualization(
  data = "stickleback_ferchaud_2015.vcf", 
  strata = "strata.stickleback.tsv")

The function does a few automatic filters: * Monomorphic markers are removed * Only common markers between strata are kept for the analysis * Individuals and markers statistics are generated automatically

A new object ibm was created in your global environment. It’s a list and to view it’s content use:

names(ibm)

Lots of info in there… Lets focus on just a few. A folder is also created automatically. The function generates by default a large object (list):

  • IBM-PCoA plot
  • heatmap plot showing genotyped/missing data (in black)
  • several plots showing the distribution of missing genotypes by individuals, pop and markers.
  • blacklists of individuals are included in the list ibm and written to the working directory (defaults from 10% to 70%, that automatically stops if no individuals at the threshold is found, can be changed with ind.missing.geno.threshold argument).
  • tables with summary missing information along the vectors of eigenvalues of the principal coordinates analysis are also part of the list output.
  • all these objects are accessed with $.

Visualization

To view the IBM-PCoA plot made with POP_ID grouping:

ibm$ibm.plots$ibm.strata.POP_ID

The dark green bubble from KIB it’s an individual with almost all of his genotypes missing. This one skip the radar of the authors ;)

The heatmap showing missingness:

  • gives a nice global outlook of the missingness
  • the figure is very long to generate (on some computers..), don’t try this with more than 100K markers… or take a couple day off:
heatmap <- ibm$heatmap 
heatmap

The vertical black line highlight the problem in the vcf with the individual missing almost all it’s genotypes.

View the table with summary of missing genotypes per individuals:

table.ind <- ibm$missing.genotypes.ind
table.ind

To view the distribution

ibm$missing.genotypes.ind.plots
ibm$missing.genotypes.ind.histo

Show the helper figure showing how many individuals could potentially be blacklisted based on % on genotypes.

ibm$ind.genotyped.helper.plot

All these figures are combined in the folder…

To view the distribution of missingness per markers

ibm$missing.genotypes.markers.combined.plots

Other figures are created, explore the list of objects and folder. Read the doc.

To view the distribution of FH and missing genotypes per individuals

  • FH : is an identity by descent genomic (IBDg) measure based on the excess in the observed number of homozygous genotypes within an individual relative to the mean number of homozygous genotypes expected under random mating).

  • To learn more about it see (Keller et al., 2011; Kardos et al., 2015; Hedrick & Garcia-Dorado, 2016)

ibm$missing.genotypes.ind.fh.combined.plots

This is weird figure is caused by the outlier individual. To remove this individual, re-run missing_visualization with the argument blacklist.id and one of the several blacklists written to the working directory (e.g. blacklist.id.missing.70.tsv).

Explore the rest by yourself!

Interpretation

Do you see patterns in your plots that provides insight about the relationships that missing values might have with other variables (inspired from r4ds).

If you see a pattern, ask yourself:

  • Is the pattern due to coincidence (i.e. random chance)?

  • Could you describe the relationship in the pattern ?

  • How strong is the relationship implied by the pattern?

  • What other variables might affect the relationship?

  • Does the relationship change if you look at individual subgroups of the data?

  • Do you think the pattern observed in the data could impact the clustering analysis?

Strategies

Arguments

  • use strata.select to select columns from the strata file to generate PCoA-IBM plots. If you have several columns to test e.g. library, sequencer, sequencing lanes, sampling sites, populations, use inside a string to get all of them!
  • use distance.method to explore other distance metric used by the function dist

Filtering

  • explore the impact of different filtering parameters on missing genotypes pattern, with radiator::filter_rad.
  • test your own whitelist of markers with whitelist.markers.
  • use the blacklist of individuals created with the function to manage missing data inside filtering pipelines
  • excluding/including individuals, populations and markers to test the impact of filtering on polymorphism discovery
  • ready to test missing data imputations ? Most stackr functions have built-in imputations arguments.

References

Ferchaud A, Hansen MM (2016) The impact of selection, gene flow and demographic history on heterogeneous genomic divergence: threespine sticklebacks in divergent environments. Molecular Ecology 25(1): 238–259. http://dx.doi.org/10.1111/mec.13399

Ferchaud A, Hansen MM (2015) Data from: The impact of selection, gene flow and demographic history on heterogeneous genomic divergence: threespine sticklebacks in divergent environments. Dryad Digital Repository. http://dx.doi.org/10.5061/dryad.kp11q

Danecek P, Auton A, Abecasis G et al. (2011) The variant call format and VCFtools. Bioinformatics, 27, 2156–2158.

Purcell S, Neale B, Todd-Brown K et al. (2007) PLINK: a tool set for whole-genome association and population-based linkage analyses. American Journal of Human Genetics, 81, 559–575.

Keller MC, Visscher PM, Goddard ME (2011) Quantification of inbreeding due to distant ancestors and its detection using dense single nucleotide polymorphism data. Genetics, 189, 237–249.

Kardos M, Luikart G, Allendorf FW (2015) Measuring individual inbreeding in the age of genomics: marker-based measures are better than pedigrees. Heredity, 115, 63–72.

Hedrick PW, Garcia-Dorado A. (2016) Understanding Inbreeding Depression, Purging, and Genetic Rescue. Trends in Ecology and Evolution. 2016;31: 940-952. doi:10.1016/j.tree.2016.09.005