tidypopgen
In tidypopgen: Tidy Population Genetics

knitr::opts_chunk$set(echo = TRUE)

This vignette gives a short introduction on how to use the tidypopgen package to perform basic data cleaning and a PCA on a dataset of European lobsters (Homarus gammarus). The original data are available at https://datadryad.org/dataset/doi:10.5061/dryad.2v1kr38, but for the purpose of this vignette we will use a subset of the data stored as a .bed file, available in the inst/extdata/lobster directory of the package. To install tidypopgen from r-universe, use:

install.packages("tidypopgen",
  repos = c(
    "https://evolecolgroup.r-universe.dev",
    "https://cloud.r-project.org"
  )
)

Next, load the tidypopgen package and the ggplot2 package for plotting.

library(tidypopgen)
library(ggplot2)

Creating a `gen_tibble`

In tidypopgen, we represent data as a gen_tibble, a subclass of tibble containing the columns id and genotypes for each individual. Genotypes are stored in a compressed format as a File-Backed Matrix that can be easily accessed by functions in tidypopgen. The genotypes column contains a vector of the row indices of each individual in the File-Backed Matrix, and when printed, the genotypes column shows the first two genotypes for each individual.

Additionally, if data are loaded from a .bed file with information in the FID column, this is treated as population information and is automatically added to the gen_tibble as column population. As with a normal tibble, this information can be changed, updated, or removed from the gen_tibble if needed. tidypopgen can also read data form packedancestry files and vcf.

Let's start by creating a gen_tibble from the lobster.bed file.

lobsters <- gen_tibble(
  x = system.file("extdata/lobster/lobster.bed", package = "tidypopgen"),
  quiet = TRUE, backingfile = tempfile()
)
head(lobsters)

We can see the structure of the gen_tibble above, with the expected columns. In this case, our .bed file contains population information corresponding to the sampling site of each lobster in the dataset.

If we want to take a look at the genotypes of our lobsters, we can use the show_genotypes function to return a matrix of genotypes, where rows are individuals and columns are loci. This is a big table, so we will just look at the first ten loci for the first 5 individuals:

lobsters %>% show_genotypes(indiv_indices = 1:5, loci_indices = 1:10)

And, similarly, if we want to see information about the loci we can use the show_loci function, which returns a tibble with information about each locus. Again this is a big table, so we will use head() to only look at the first few:

head(lobsters %>% show_loci())

Now we have a gen_tibble to work with, we can start to clean the data.

Quality control

Let's start by checking the quality of the data for each individual lobster in our dataset. We can use the qc_report_indiv function to generate a report which contains information about missingness and heterozygosity for each individual.

indiv_qc_lobsters <- lobsters %>% qc_report_indiv()

We can take a look at this data using the autoplot function:

autoplot(indiv_qc_lobsters, type = "scatter")

We can see that most individuals have low missingness and heterozygosity, but there are a few individuals missing >20% of their genotypes. We can remove these individuals using the filter function, specifying to keep only individuals with missingness under 20%.

lobsters <- lobsters %>% filter(indiv_missingness(genotypes) < 0.2)

Now lets check our loci. We can use the qc_report_loci function to generate a report of the loci quality. This function will return another qc_report object, which contains information about missingness, minor allele frequency, and Hardy-Weinberg Equilibrium for each locus.

loci_qc_lobsters <- lobsters %>% qc_report_loci()

Here, we get a message because the qc_report_loci function calculates Hardy-Weinberg equilibrium assuming that all individuals are part of a single population. As our dataset contains multiple lobster populations, we should group our data by population first:

lobsters <- lobsters %>% group_by(population)
loci_qc_lobsters <- lobsters %>% qc_report_loci()

That's better. Now, lets take a look at minor allele frequency for all loci:

autoplot(loci_qc_lobsters, type = "maf")

And we can see that we don't have any monomorphic SNPs.

Now let's look at missingness.

autoplot(loci_qc_lobsters, type = "missing")

Our data mostly have low missingness, but we can see that some loci have > 5% missingness across individuals, and we want to remove these individuals.

In tidypopgen, there are two functions to subset the loci in a gen_tibble object: select_loci and select_loci_if. The function select_loci operates on information about the loci (e.g filtering by chromosome or by rsID), while select_loci_if operates on the genotypes at those loci (e.g filtering by minor allele frequency or missingness). In this case, we want to remove loci with >5% missingness, so we can use select_loci_if with the loci_missingness function, operating on the genotypes column of our gen_tibble.

lobsters <- lobsters %>% select_loci_if(loci_missingness(genotypes) < 0.05)

Through our filtering, we have removed a few individuals and loci. We should now update the file backing matrix to reflect these changes, using the function gt_update_backingfile:

lobsters <- gt_update_backingfile(lobsters, backingfile = tempfile())

Now our data are clean and the backingfile is updated, we are ready to create a PCA.

Impute

First, we need to impute any remaining missing data using the gt_impute_simple function.

lobsters <- gt_impute_simple(lobsters, method = "random")

PCA

Then we can run a PCA. There are a number of PCA algorithms, here, we will use the gt_pca_partialSVD function:

partial_pca <- gt_pca_partialSVD(lobsters)

And we can create a simple plot using autoplot:

autoplot(partial_pca, type = "scores")

That was easy! The autoplot gives us a quick idea of the explained variance and rough distribution of the samples, but we need to see the different populations within our dataset.

For a quick overview, we could add an aesthetic to our plot:

autoplot(partial_pca, type = "scores") +
  aes(color = lobsters$population) +
  labs(color = "population")

However, if we want to fully customise our plot, we can wrangle the data directly and use ggplot2.

Plot with ggplot2

For a customised plot, we can extract the information on the scores of each individual using the augment method for gt_pca.

pcs <- augment(x = partial_pca, data = lobsters)

Then we can extract the eigenvalues for each principal component with the tidy function, using the "eigenvalues" argument:

eigenvalues <- tidy(partial_pca, "eigenvalues")

xlab <- paste("Axis 1 (", round(eigenvalues[1, 3], 1), " %)",
  sep = ""
)
ylab <- paste("Axis 2 (", round(eigenvalues[2, 3], 1), " %)",
  sep = ""
)

And finally plot:

ggplot(data = pcs, aes(x = .fittedPC1, y = .fittedPC2)) +
  geom_hline(yintercept = 0) +
  geom_vline(xintercept = 0) +
  geom_point(aes(fill = population),
    shape = 21, size = 3, show.legend = FALSE
  ) +
  scale_fill_distruct() +
  labs(x = xlab, y = ylab) +
  ggtitle("Lobster PCA")

Or we could create a labelled version of our PCA by determining the centre of each group to place the labels

# Calculate centre for each population
centroid <- aggregate(cbind(.fittedPC1, .fittedPC2, .fittedPC3) ~ population,
  data = pcs, FUN = mean
)

# Add these coordinates to our augmented pca object
pcs <- left_join(pcs, centroid, by = "population", suffix = c("", ".cen"))

And then add labels to the plot:

ggplot(data = pcs, aes(x = .fittedPC1, y = .fittedPC2)) +
  geom_hline(yintercept = 0) +
  geom_vline(xintercept = 0) +
  geom_segment(aes(xend = .fittedPC1.cen, yend = .fittedPC2.cen),
    show.legend = FALSE
  ) +
  geom_point(aes(fill = population),
    shape = 21, size = 3, show.legend = FALSE
  ) +
  geom_label(
    data = centroid,
    aes(label = population, fill = population),
    size = 4, show.legend = FALSE
  ) +
  scale_fill_distruct() +
  labs(x = xlab, y = ylab) +
  ggtitle("Lobster PCA")

That's it! There are more functions to run different types of analyses (DAPC, ADMIXTURE, f statistics with admixtools2, etc.); each resulting object has autoplot, tidy, and augment to explore the results and integrated into the information from the gen_tibble.