In eclarke/metaglomr: Tools for tidy metagenomics

knitr::opts_chunk$set(echo = TRUE)

This is a package to combine counts, sample metadata, and taxonomic annotations into one long-form tidy dataframe for further analysis. It offers functions to switch between phyloseq objects and these 'agglomerated' dataframes.

Installation

devtools::install_github("eclarke/metaglomr")

Short usage overview

suppressPackageStartupMessages(library(tidyverse))
library(metaglomr)

data(features)  # A counts matrix, with rows as samples and columns as features
data(samples)   # A dataframe with sample metadata
data(taxa)      # A dataframe or matrix with taxonomic annotations for features

# Combine all of these datasets into one long-form dataframe
(agg <- agglomerated(features, samples, taxa, "sample_id"))

Motivation

In metagenomics, we often work with three distinct datasets: a feature count table, a sample table, and a taxonomic annotation table.

Feature count table

The feature count table is is a matrix of features $\times$ samples, and the cells of the matrix are the times that feature was seen in that sample. Features can be OTUs, ASVs, species, genomes, or whatever. Often these come from a biom file.

data(features)
features[1:5, 1:15]

Sample metadata

The sample metadata can be anything about the samples, but frequently contains at least a description of the sample type and study group of each sample. This is frequently referred to as a mapping file in QIIME.

data(samples)
as_tibble(samples)

Taxonomic annotations

This contains taxonomic annotations for all the features (OTUs or otherwise), split by rank (i.e. Kingdom, Phylum, etc).

data(taxa)
taxa[1:5, ]

Agglomerating the three datasets

For users of the tidyverse, it's frequently easiest to work with long-form melted datasets, where each row is a unique observation or data point. The unique reference in these three datasets is the count of a feature in a particular sample. Therefore, we can create a dataframe where each row is this unique combination of feature + sample + count, with additional columns describing the sample and feature further.

(agg <- agglomerated(features, samples, taxa, "sample_id"))

While this may seem overly repetitive (as the metadata is duplicated in lots of rows), R and dplyr actually handle this pretty well. Things only start breaking down with feature tables that have more than > 100,000,000 cells. What this buys you is the ability to use standard tidyverse verbs and operations easily.

Examples

Subsetting

Subsetting is easy through the use of the filter verb:

filter(agg, study_group == "case")
filter(agg, Phylum == "Bacteroidetes")

Adding new columns

Here's how to calculate proportions and add it to your dataframe:

agg <- agg %>%
  group_by(sample_id) %>%
  mutate(proportion = count/sum(count))

# Showing just a subset of the data:
select(agg, sample_id, otu_id, count, proportion)

Aggregating based on metadata:

Aggregate based on taxonomic rank:

agg %>% 
  group_by(sample_id, Phylum) %>%
  summarize(count = sum(count)) %>%
  ungroup() %>%
  # re-add sample data that got lost in the summarizing
  left_join(get_samples(agg, sample_id, study_group, sample_type))

Summarizing

Or find the most prevalent phyla in your study groups:

agg %>% 
  ungroup() %>%
  group_by(study_group, Phylum) %>%
  summarize(mean_proportion = mean(proportion)) %>%
  top_n(1, mean_proportion)

eclarke/metaglomr documentation built on May 30, 2019, 12:46 p.m.