In microbiome/miaTime: Microbiome Time Series Analysis

library(Cairo)
knitr::opts_chunk$set(cache = FALSE,
                      fig.width = 9,
                      dpi=300,
                      dev = "png",
                      dev.args = list(type = "cairo-png"),
                      message = FALSE,
                      warning = FALSE)

library(mia)
library(miaTime)
library(dplyr)
library(lubridate)
library(SummarizedExperiment)

Introduction

miaTime implements tools for time series manipulation based on the TreeSummarizedExperiment [@TSE] data container. Much of the functionality is also applicable to the SummarizedExperiment [@SE] data objects. This tutorial shows how to use miaTime methods as well as the broader R/Bioconductor ecosystem to manipulate time series data.

Check also the related package TimeSeriesExperiment.

Installation

Installing the latest development version in R.

library(devtools)
devtools::install_github("microbiome/miaTime")

Loading the package:

library(miaTime)

Sorting samples

The tidySingleCellExperiment package provides handy functions for (Tree)SE manipulation.

library("tidySingleCellExperiment")
data(hitchip1006)
tse <- hitchip1006
tse2 <- tse %>% tidySingleCellExperiment::arrange(subject, time)

Storing time information with Period class

miaTime utilizes the functions available in the package lubridate to convert time series field to "Period" class object. This gives access to a number of readily available time series manipulation tools.

Load example data:

# Load packages
library(miaTime)
library(lubridate)
library(SummarizedExperiment)

# Load demo data
data(hitchip1006)
tse <- hitchip1006

# Time is given in days in the demo data.
# Convert days to seconds
time_in_seconds <- 60*60*24*colData(tse)[,"time"]
# Convert the time data to period class
Seconds <- as.period(time_in_seconds, unit="sec")
# Check the output
Seconds[1140:1151]

Conversion between time units

The time field in days is now shown in seconds. It can then be converted to many different units using the lubridate package.

Hours <- as.period(Seconds, unit = "hour")
Hours[1140:1151]

The updated time information can then be added to the SummarizedExperiment data object as a new colData (sample data) field.

colData(tse)$timeSec <- Seconds
colData(tse)

Calculating time differences

The \link[lubridate]{as.duration} function helps to specify time points as duration.

Duration <- as.duration(Seconds)
Duration[1140:1151]

The difference between subsequent time points can then be calculated.

Timediff <- diff(Duration)
Timediff <- c(NA, Timediff)
Timediff[1140:1151]

The time difference from a selected point to the other time points can be calculated as follows.

base <- Hours - Hours[1] #distance from starting point
base[1140:1151]

base_1140 <- Seconds - Seconds[1140]
base_1140[1140:1151]

Time point rank

Rank of the time points can be calculated by rank function provided in base R.

colData(tse)$rank <- rank(colData(tse)$time)
colData(tse)

Operations per unit

Sometimes we need to operate on time series per unit (subject, reaction chamber, sampling location, ...).

Add time point rank per subject.

library(dplyr)
colData(tse) <- colData(tse) %>%
   as.data.frame() %>%
   group_by(subject) %>%
   mutate(rank = rank(time, ties.method="average")) %>%
   DataFrame()

Subset to baseline samples

library(tidySummarizedExperiment)

# Pick samples with time point 0
tse <- hitchip1006 |> filter(time == 0)
# Or: tse <- tse[, tse$time==0]

# Sample with the smallest time point within each subject
colData(tse) <- colData(tse) %>%
   as.data.frame() %>%
   group_by(subject) %>%
   mutate(rank = rank(time, ties.method="average")) %>%
   DataFrame

# Pick the subset including first time point per subject
tse1 <- tse[, tse$rank == 1]