Using R: 10 years with R

Yesterday, 29 Feburary 2020, was the 20th anniversary of the release R 1.0.0. Jozef Hajnala’s blog has a cute anniversary post with some trivia. I realised that it is also (not to the day, but to the year) my R anniversary.

I started using R in 2010, during my MSc project in Linköping. Daniel Nätt, who was a PhD student there at the time, was using it for gene expression and DNA methylation work. I think that was the reason he was pulled into R; he needed the Bioconductor packages for microarrays. He introduced me. Thanks, Daniel!

I think I must first have used it to do something with qPCR melting curves. I remember that I wrote some function to reshape/pivot data between long and wide format. It was probably an atrocity of nested loops and hard bracket indexing. Coming right from an undergraduate programme with courses using Ada and C++, even if we had also used Minitab for statistics and Matlab for engineering, I spoke R with a strong accent. At any rate, I was primed to think that doing my data analysis with code was a good idea, and jumped at the opportunity to learn a tool for it. Thanks, undergraduate programme!

I think the easiest thing to love about R is the package system. You can certainly end up in dependency hell with R and metaphorically shoot your own foot, especially on a shared high performance computing system. But I wouldn’t run into any of that until after several years. I was, and still am, impressed by how packages just worked, and could do almost anything. So, the Bioconductor packages were probably, indirectly, why I was introduced to R, and after that, my R story can be told in a series of packages. Thanks, CRAN!

The next package was R/qtl, that I relied on for my PhD. I had my own copy of the R/qtl book. For a period, I probably wrote thing every day:

library(qtl)

cross <- read.cross(file = "F8_geno_trim.csv", format = "csv")

R/qtl is one of my favourite pieces or research software, relatively friendly and with lots of documentation. Thanks, R/qtl developers!

Of course it was Dom Wright, who was my PhD supervisor, who introduced me to R/qtl, and I think it was also he who introduced me to ggplot2. At least he used it, and at some point we were together trying to fix the formatting of a graph, probably with some ugly hack. I decided to use ggplot2 as much as possible, and as it is wont to, ggplot2 made me care about rearranging data, thus leading to reshape2 and plyr. ”The magic is not in plotting the data but in tidying and rearranging the data for plotting.” After a while, most everything I wrote used the ddply function in some way. Thank you, Hadley Wickham!

Then came the contemporary tidyverse. For the longest time, I was uneasy with tidyr, and I’m still not a regular purrr user, but one can’t avoid loving dplyr. How much? My talk at the Swedish Bioinformatics Workshop in 2016 had a slide expressing my love of the filter function. It did not receive the cheers that the function deserves. Maybe the audience were Python users. With new file reading functions, new data frames and functions to manipulate data frames, modern R has become smoother and friendlier. Thanks, tidyverse developers!

The history of R on this blog started in 2011, originally as a way to make notes for myself or, ”a fellow user who’s trying to google his or her way to a solution”. This turned into a series of things to help teach R to biologists around me.

There was the Slightly different introduction to R series of blog posts. It used packages that feel somewhat outdated, and today, I don’t think there’s anything even slightly different about advocating RStudio, and teaching ggplot2 from the beginning.

This spawned a couple of seminars in course for PhD students, which were updated for the Wright lab computation lunches, and eventually turned into a course of its own given in 2017. It would be fun to update it and give it again.

The last few years, I’ve been using R for reasonably large genome datasets in a HPC environment, and gotten back to the beginnings, I guess, by using Bioconducor a lot more. However, the package that I think epitomises the last years of my R use is AlphaSimR, developed by colleagues in Edinburgh. It’s great to be able throw together a quick simulation to check how some feature of genetics behaves. AlphaSimR itself is also an example of how far the R/C++ integration has come with RCpp and RCppArmadillo. Thanks, Chris!

In summary, R is my tool of choice for almost anything. I hope we’ll still be using it, in new and interesting ways, in another ten years. Thank you, R core team!

Using R: Animal model with hglm and Stan (with Cholesky trick)

A few weeks ago I posted about fitting the quantitative genetic animal model with MCMCglmm and R-INLA. Since then, I listened to a talk by Lars Rönnegård, one of the creators of the hglm package, and this paper was published in GSE about animal models in Stan.

hglm

The hglm package fits hierarchical generalised linear models. That includes the animal model with pedigree or genomic relatedness. Hierarchical generalised linear models also allow you to model the dispersion of random effects, which lets you do tricks like variance QTL mapping (Rönnegård & Valdar 2011), breeding values for variances (Rönnegård et al. 2010) or genomic prediction models with predictors of marker variance (Mouresan, Selle & Rönnegård 2019). But let’s not get ahead of ourselves. How do we fit an animal model?

Here is the matrix formulation of the animal model that we skim through in every paper. It’s in this post because we will use the design matrix interface to hglm, which needs us to give it these matrices (this is not a paper, so we’re not legally obliged to include it):

\mathbf{y} = \mu + \mathbf{X} \mathbf{b} + \mathbf{Z} \mathbf{a} + \mathbf{e}

The terms are the the trait value, intercept, fixed coefficients and their design matrix, genetic coefficients and their design matrix, and the residual. The design matrix Z will contain one row and column for each individual, with a 1 to indicate its position in the phenotype table and pedigree and the rest zeros. If we sort our files, it’s an identity matrix.

The trick with the genetic coefficients is that they’re correlated, with a specific known correlation structure that we know from the pedigree (or in genomic models, from markers). It turns out (Lee, Nelder & Pawitan 2017, chapter 8) that you can change the Z matrix around so that it lets you fit the model with an identity covariance matrix, while still accounting for the correlations between relatives. You replace the random effects for relatedness with some transformed random effects that capture the same structure. One way to do this is with Cholesky decomposition.

\mathbf{Z_{fudged}} = \mathbf{Z_0} \mathbf{L}

As an example of what the Cholesky decomposition does, here is slice of the additive relationship matrix of 100 simulated individuals (the last generation of one replicate of these simulations) and the resulting matrix from Cholesky decomposition.

So instead of

\mathbf{a} \sim N(0, \mathbf{A} \sigma)

We can fit

\mathbf{a_{fudged}} \sim N(0, \mathbf{I} \sigma)

This lets us fit the animal model with hglm, by putting in a modified Z matrix.

Assuming we have data frames with a pedigree and a phenotype (like, again, from these simulations):

library(AGHmatrix)
library(hglm)

A  <- Amatrix(ped)

Z0  <- diag(1000)
L <- t(chol(A))
Z  <- Z0 %*% L
X <- model.matrix(~1, pheno)

model <- hglm(y = pheno$pheno,
              X = X,
              Z = Z,
              conv = 1e-8)

est_h2  <- model$varRanef / (model$varRanef + model$varFix)

(I found the recommendation to decrease the convergence criterion from the default for animal models in a YouTube video by Xia Chen.)

Stan

When we turn to Stan, we will meet the Cholesky trick again. Stan is a software for Markov Chain Monte Carlo, built to fit hierarchical linear models, and related high-dimensional models, more effectively than other sampling strategies (like Gibbs). rstan is a helpful package for running Stan from within R.

Nishio & Arakawa (2019) recently published a Stan script to fit an animal model, comparing Stan to a Gibbs sampler (and a related MCMC sampler that they also didn’t publish the code for). If we look into their Stan model code, they also do a Cholesky decomposition to be able to use an identity matrix for the variance.

First, they decompose the additive relationship matrix that the program takes in:

transformed data{
  matrix[K,K] LA;
  LA = cholesky_decompose(A);
}

And then, they express the model like this:

vector[N] mu;
vector[K] a;
a_decompose ~ normal(0, 1);
a = sigma_G * (LA * a_decompose);
mu = X * b + Z * a;
Y ~ normal(mu, sigma_R);

We can add this line to the generated quantities block of the Stan program to get heritability estimates directly:

real h2;
h2 = sigma_U / (sigma_U + sigma_E)

Here, we’ve saved their model to a stan file, and now we can run it from R:

pheno$scaled_pheno <- as.vector(scale(pheno$pheno))

model_stan <- stan(file = "nishio_arakawa.stan",
                   data = list(Y = pheno$scaled_pheno,
                               X = X,
                               A = A,
                               Z = Z0,
                               J = 1,
                               K = 1000,
                               N = 1000))

est_h2_stan <- summary(model_stan, pars = "h2")$summary

Important note that I always forget: It's important to scale your traits before you run this model. If not, the priors might be all wrong.

The last line pulls out the summary for the heritability parameter (that we added above). This gives us an estimate and an interval.

The paper also contains this entertaining passage about performance, which reads as if it was a response to a comment, actual or anticipated:

R language is highly extensible and provides a myriad of statistical and graphical techniques. However, R language has poor computation time compared to Fortran, which is especially well suited to numeric computation and scientific computing. In the present study, we developed the programs for GS and HMC in R but did not examine computation time; instead, we focused on examining the performance of estimating genetic parameters and breeding values.

Yes, two of their samplers (Gibbs and HMC) were written in R, but the one they end up advocating (and the one used above), is in Stan. Stan code gets translated into C++ and then compiled to machine code.

Stan with brms

If rstan lets us run Stan code from R and examine the output, brms lets us write down models in relatively straightforward R syntax. It’s like the MCMCglmm of the Stan world. We can fit an animal model with brms too, by directly plugging in the relationship matrix:

model_brms <- brm(scaled_pheno ~ 1 + (1|animal),
                  data = pheno,
                  family = gaussian(),
                  cov_ranef = list(animal = A),
                  chains = 4,
                  cores = 1,
                  iter = 2000)

Then, we can pull out the posterior samples for the variability, here expressed as standard errors, compute the heritability and then get the estimates (and interval, if we want):

posterior_brms <- posterior_samples(model_brms,
                                    pars = c("sd_animal", "sigma"))

h2_brms  <- posterior_brms[,1]^2 /
    (posterior_brms[,1]^2 + posterior_brms[,2]^2)

est_h2_brms <- mean(h2_brms)

(Code is on GitHub: both for the graphs above, and the models.)

Using R: Animal model with simulated data

Last week’s post just happened to use MCMCglmm as an example of an R package that can get confused by tibble-style data frames. To make that example, I simulated some pedigree and trait data. Just for fun, let’s look at the simulation code, and use MCMCglmm and AnimalINLA to get heritability estimates.

First, here is some AlphaSimR code that creates a small random mating population, and collects trait and pedigree:

library(AlphaSimR)

## Founder population
FOUNDERPOP <- runMacs(nInd = 100,
                      nChr = 20,
                      inbred = FALSE,
                      species = "GENERIC")

## Simulation parameters 
SIMPARAM <- SimParam$new(FOUNDERPOP)
SIMPARAM$addTraitA(nQtlPerChr = 100,
                   mean = 100,
                   var = 10)
SIMPARAM$setGender("yes_sys")
SIMPARAM$setVarE(h2 = 0.3)
 
## Random mating for 9 more generations
generations <- vector(mode = "list", length = 10) 
generations[[1]] <- newPop(FOUNDERPOP,
                           simParam = SIMPARAM)


for (gen in 2:10) {

    generations[[gen]] <- randCross(generations[[gen - 1]],
                                    nCrosses = 10,
                                    nProgeny = 10,
                                    simParam = SIMPARAM)

}

## Put them all together
combined <- Reduce(c, generations)


## Extract phentoypes
pheno <- data.frame(animal = combined@id,
                    pheno = combined@pheno[,1])

## Extract pedigree
ped <- data.frame(id = combined@id,
                  dam = combined@mother,
                  sire =combined@father)
ped$dam[ped$dam == 0] <- NA
ped$sire[ped$sire == 0] <- NA

## Write out the files
write.csv(pheno,
          file = "sim_pheno.csv",
          row.names = FALSE,
          quote = FALSE)

write.csv(ped,
          file = "sim_ped.csv",
          row.names = FALSE,
          quote = FALSE)

In turn, we:

  1. Set up a founder population with a AlphaSimR’s generic livestock-like population history, and 20 chromosomes.
  2. Choose simulation parameters: we have an organism with separate sexes, a quantitative trait with an additive polygenic architecture, and we want an environmental variance to give us a heritability of 0.3.
  3. We store away the founders as the first generation, then run a loop to give us nine additional generations of random mating.
  4. Combine the resulting generations into one population.
  5. Extract phenotypes and pedigree into their own data frames.
  6. Optionally, save the latter data frames to files (for the last post).

Now that we have some data, we can fit a quantitative genetic pedigree model (”animal model”) to estimate genetic parameters. We’re going to try two methods to fit it: Markov Chain Monte Carlo and (the unfortunately named) Integrated Nested Laplace Approximation. MCMC explores the posterior distribution by sampling; I’m not sure where I heard it described as ”exploring a mountain by random teleportation”. INLA makes approximations to the posterior that can be integrated numerically; I guess it’s more like building a sculpture of the mountain.

First, a Gaussian animal model in MCMCglmm:

library(MCMCglmm)

## Gamma priors for variances
prior_gamma <- list(R = list(V = 1, nu = 1),
                    G = list(G1 = list(V = 1, nu = 1)))
    
## Fit the model
model_mcmc  <- MCMCglmm(scaled ~ 1,
                        random = ~ animal,
                        family = "gaussian",
                        prior = prior_gamma,
                        pedigree = ped,
                        data = pheno,
                        nitt = 100000,
                        burnin = 10000,
                        thin = 10)

## Calculate heritability for heritability from variance components
h2_mcmc_object  <- model_mcmc$VCV[, "animal"] /
    (model_mcmc$VCV[, "animal"] + model_mcmc$VCV[, "units"])
 
## Summarise results from that posterior
h2_mcmc  <- data.frame(mean = mean(h2_mcmc_object),
                       lower = quantile(h2_mcmc_object, 0.025),
                       upper = quantile(h2_mcmc_object, 0.975),
                       method = "MCMC",
                       stringsAsFactors = FALSE)

And here is a similar animal model in AnimalINLA:

library(AnimalINLA)

## Format pedigree to AnimalINLA's tastes
ped_inla <- ped
ped_inla$id  <- as.numeric(ped_inla$id)
ped_inla$dam  <- as.numeric(ped_inla$dam)
ped_inla$dam[is.na(ped_inla$dam)] <- 0
ped_inla$sire  <- as.numeric(ped_inla$sire)
ped_inla$sire[is.na(ped_inla$sire)] <- 0
    
## Turn to relationship matrix
A_inv <- compute.Ainverse(ped_inla)
    
## Fit the model
model_inla  <- animal.inla(response = scaled,
                           genetic = "animal",
                           Ainverse = A_inv,
                           type.data = "gaussian",
                           data = pheno,
                           verbose = TRUE)

## Pull out summaries from the model object
summary_inla  <- summary(model_inla)

## Summarise results
h2_inla  <- data.frame(mean = summary_inla$summary.hyperparam["Heritability", "mean"],
                       lower = summary_inla$summary.hyperparam["Heritability", "0.025quant"],
                       upper = summary_inla$summary.hyperparam["Heritability", "0.975quant"],
                       method = "INLA",
                       stringsAsFactors = FALSE)

If we wrap this all in a loop, we can see how the estimation methods do on replicate data (full script on GitHub). Here are estimates and intervals from ten replicates (black dots show the actual heritability in the first generation):

As you can see, the MCMC and INLA estimates agree pretty well and mostly hit the mark. In the one replicate dataset where they falter, they falter together.

‘Simulating genetic data with R: an example with deleterious variants (and a pun)’

A few weeks ago, I gave a talk at the Edinburgh R users group EdinbR on the RAGE paper. Since this is an R meetup, the talk concentrated on the mechanics of genetic data simulation and with the paper as a case study. I showed off some of what Chris Gaynor’s AlphaSimR can do, and how we built on that to make the specifics of this simulation study. The slides are on the EdinbR Github.

Genetic simulation is useful for all kinds of things. Sure, they’re only as good as the theory that underpins them, but the willingness to try things out in simulations is one of the things I always liked about breeding research.

This is my description of the logic of genetic simulation: we think of the genome as a large table of genotypes, drawn from some distribution of allele frequencies.

To make an utterly minimal simulation, we could draw allele frequencies from some distribution (like a Beta distribution), and then draw the genotypes from a binomial distribution. Done!

However, there is a ton of nuance we would like to have: chromosomes, linkage between variants, sexes, mating, selection …

AlphaSimR addresses all of this, and allows you to throw individuals and populations around to build pretty complicated designs. Here is the small example simulation I used in the talk.


library(AlphaSimR)
library(ggplot2)

## Generate founder chromsomes

FOUNDERPOP <- runMacs(nInd = 1000,
                      nChr = 10,
                      segSites = 5000,
                      inbred = FALSE,
                      species = "GENERIC")


## Simulation parameters

SIMPARAM <- SimParam$new(FOUNDERPOP)
SIMPARAM$addTraitA(nQtlPerChr = 100,
                   mean = 100,
                   var = 10)
SIMPARAM$addSnpChip(nSnpPerChr = 1000)
SIMPARAM$setGender("yes_sys")


## Founding population

pop <- newPop(FOUNDERPOP,
              simParam = SIMPARAM)

pop <- setPheno(pop,
                varE = 20,
                simParam = SIMPARAM)


## Breeding

print("Breeding")
breeding <- vector(length = 11, mode = "list")
breeding[[1]] <- pop

for (i in 2:11) {
    print(i)
    sires <- selectInd(pop = breeding[[i - 1]],
                       gender = "M",
                       nInd = 25,
                       trait = 1,
                       use = "pheno",
                       simParam = SIMPARAM)

    dams <- selectInd(pop = breeding[[i - 1]],
                      nInd = 500,
                      gender = "F",
                      trait = 1,
                      use = "pheno",
                      simParam = SIMPARAM)

    breeding[[i]] <- randCross2(males = sires,
                                females = dams,
                                nCrosses = 500,
                                nProgeny = 10,
                                simParam = SIMPARAM)
    breeding[[i]] <- setPheno(breeding[[i]],
                              varE = 20,
                              simParam = SIMPARAM)
}



## Look at genetic gain and shift in causative variant allele frequency

mean_g <- unlist(lapply(breeding, meanG))
sd_g <- sqrt(unlist(lapply(breeding, varG)))

plot_gain <- qplot(x = 1:11,
                   y = mean_g,
                   ymin = mean_g - sd_g,
                   ymax = mean_g + sd_g,
                   geom = "pointrange",
                   main = "Genetic mean and standard deviation",
                   xlab = "Generation", ylab = "Genetic mean")

start_geno <- pullQtlGeno(breeding[[1]], simParam = SIMPARAM)
start_freq <- colSums(start_geno)/(2 * nrow(start_geno))

end_geno <- pullQtlGeno(breeding[[11]], simParam = SIMPARAM)
end_freq <- colSums(end_geno)/(2 * nrow(end_geno))

plot_freq_before <- qplot(start_freq, main = "Causative variant frequency before") 
plot_freq_after <- qplot(end_freq, main = "Causative variant frequency after") 

This code builds a small livestock population, breeds it for ten generations, and looks at the resulting selection response in the form of a shift of the genetic mean, and the changes in the underlying distribution of causative variants. Here are the resulting plots:

What single step does with relationship

We had a journal club about the single step GBLUP method for genomic evaluation a few weeks ago. In this post, we’ll make a few graphs of how the single step method models relatedness between individuals.

Imagine you want to use genomic selection in a breeding program that already has a bunch of historical pedigree and trait information. You could use some so-called multistep evaluation that uses one model for the classical pedigree + trait quantitative genetics and one model for the genotype + trait genomic evaluation, and then mix the predictions from them together. Or you could use the single-step method, which combines pedigree, genotypes and traits into one model. It does this by combining the relationship estimates from pedigree and genotypes. That matrix can then go into your mixed model.

We’ll illustrate this with a tiny simulated population: five generations of 100 individuals per generation, where ten random pairings produce the next generation, with families of ten individuals. (The R code is on Github and uses AlphaSimR for simulation and AGHmatrix for matrices). Here is a heatmap of the pedigree-based additive relationship matrix for the population:

What do we see? In the lower left corner are the founders, and not knowing anything about their heritage, the matrix has them down as unrelated. The squares of high relatedness along the diagonal are the families in each generation. As we go upwards and to the right, relationship is building up.

Now, imagine the last generation of the population also has been genotyped with a SNP chip. Here is a heatmap of their genomic relationship matrix:

Genomic relationship is more detailed. We can still discern the ten families within the last generation, but no longer are all the siblings equally related to each other and to their ancestors. The genotyping helps track segregation within families, pointing out to us when relatives are more or less related than the average that we get from the pedigree.

Enter the single-step relationship matrix. The idea is to put in the genomic relationships for the genotyped individuals into the big pedigree-based relationship matrix, and then adjust the rest of the matrix to propagate that extra information we now have from the genotyped individuals to their ungenotyped relatives. Here is the resulting heatmap:

You can find the matrix equations in Legarra, Aguilar & Misztal (2009). The matrix, called H, is broken down into four partitions called H11, H12, H21, and H22. H22 is the part that pertains to the genotyped animals, and it’s equal to the genomic relationship matrix G (after some rescaling). The others are transformations of G and the corresponding parts of the additive relationship matrix, spreading G onto A.

To show what is going on, here is the difference between the additive relationship matrix and the single-step relationship matrix, with lines delineating the genotyped animals and breaking the matrix into the four partitions:

What do we see? In the top right corner, we have a lot of difference, where the genomic relationship matrix has been plugged in. Then, fading as we go from top to bottom and from right to left, we see the influence of the genomic relationship on relatives, diminishing the further we get from the genotyped individuals.

Literature

Legarra, Andres, I. Aguilar, and I. Misztal. ”A relationship matrix including full pedigree and genomic information.” Journal of dairy science 92.9 (2009): 4656-4663.