In CCDM-Programming-Club/codonfriend: Analyse Codon Usage With Your Pals

knitr::opts_chunk$set(echo = TRUE)

In the basics vignette we saw some groups of genomes with common central statistics. In this notebook we'll look at grouping genomes based on the CAI distributions.

library("tibble")
library("magrittr")
library("ggplot2")
library("dplyr")
library("mclust")

library("codonfriend")

We'll mostly be using the tables prepared in the basics notebook, so make sure you've run that first.

basics_dir <- "basics_results"
out_dir <- "clustering_results"
dir.create(out_dir, showWarnings = FALSE)

First let's load the pre-summarised data.

summaries <- readr::read_tsv(file.path(basics_dir, "summarised_cais.tsv"))
head(summaries)

And we'll try clustering the genomes into groups by fitting multiple gaussian distributions to the medians.

gmm <- Mclust(summaries$median)
summary(gmm)

Cool, so mclust finds support for 4 clusters in the data. Let's have a look at those distributions.

summaries$classif <- gmm$classification
readr::write_tsv(summaries, file.path(out_dir, "summarised_cais_classif.tsv"))

ggplot(summaries, aes(x = median, fill = as.factor(classif))) + geom_density(alpha = 0.5)

They aren't quite perfect normal distributions, but they are the groups that we saw in our data. The narrow tall group might be from some group that is over-represented in the database. Possibly a yeast.

Let's add the classifications to the big table containing all CAI values from all genomes.

cais <- readr::read_tsv(file.path(basics_dir, "joined_cais.tsv")) %>%
  left_join(summaries[, c("file", "classif")], by="file") %>%
  readr::write_tsv(file.path(out_dir, "joined_cais_classif.tsv"))

head(cais)

And now we can plot our 4 groups with the whole cai distributions.

ggplot(cais, aes(x = cai, colour = file)) +
  geom_density() +
  facet_wrap(~ as.factor(cais$classif), ncol = 1) +
  theme(legend.position="none")

Great!

That's a bit cleaner. I'm seeing that generally the majority fall neatly into a "main" distribution. Groups with higher median CAIs tend to have tighter distributions, while those with lower medians are generally broader.

In anycase it is very clear that these distributions can't be fit using the same models, and they probably aren't normally distributed.

CCDM-Programming-Club/codonfriend documentation built on May 9, 2019, 3:20 a.m.