Registering data

In greatR: Gene Registration from Expression and Time-Courses in R

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

This article will show users how to register data using different types of input, as illustrated below.

knitr::include_graphics("figures/registration_opt_diagram.png")

Using data frame input

Loading sample data

greatR provides an example of data frame containing two different species A. thaliana and B. rapa with two and three different replicates, respectively. This data frame can be read as follows:

# Load the package
library(greatR)
library(data.table)

# Load a data frame from the sample data
b_rapa_data <- system.file("extdata/brapa_arabidopsis_data.csv", package = "greatR") |>
  data.table::fread()

Note that the data has all of five columns required by the package:

b_rapa_data |>
  knitr::kable()

b_rapa_data[, .SD[1:6], by = accession][, .(gene_id, accession, timepoint, expression_value, replicate)] |>
  knitr::kable()

Registering the data

To align gene expression time-course between Arabidopsis Col-0 and B. rapa Ro18, we can use the function register(). When using the default use_optimisation = TRUE, greatR will find the best stretch and shift parameters through optimisation. For more details on the other function arguments, go to register().

registration_results <- register(
  b_rapa_data,
  reference = "Ro18",
  query = "Col0",
  scaling_method = "z-score"
)
#> ── Validating input data ────────────────────────────────────────────────────────
#> ℹ Will process 10 genes.
#> ℹ Using estimated standard deviation, as no `exp_sd` was provided.
#> ℹ Using `scaling_method` = "z-score".
#>
#> ── Starting registration with optimisation ──────────────────────────────────────
#> ℹ Using L-BFGS-B optimisation method.
#> ℹ Using computed stretches and shifts search space limits.
#> ℹ Using `overlapping_percent` = 50% as a registration criterion.
#> ✔ Optimising registration parameters for genes (10/10) [2s]

When dealing with thousands of genes, users may speed up the registration process by using the argument num_cores to run the registration using parallel processing.

parallel::detectCores()
#> 8

registration_results <- register(
  b_rapa_data,
  reference = "Ro18",
  query = "Col0",
  scaling_method = "z-score",
  num_cores = 8
)

Registration results

The function register() returns a list with S3 class res_greatR containing three different objects:

• data is a data frame containing the expression data and an additional timepoint_reg column which is a result of registered time points by applying the registration parameters to the query data.
• model_comparison is a data frame containing (a) the optimal stretch and shift for each gene_id and (b) the difference between Bayesian Information Criterion for the separate model and for the combined model (BIC_diff) after applying optimal registration parameters for each gene. If the value of BIC_diff < 0, then expression dynamics between reference and query data can be registered (registered = TRUE). (Default S3 print).
• fun_args is a list of arguments used when calling the function (reference, query, scaling_method, ...).

To check whether a gene is registered or not, we can get the summary results by accessing the model_comparison table from the registration result.

# Load a data frame from the sample data
registration_results <- system.file("extdata/brapa_arabidopsis_registration.rds", package = "greatR") |>
  readRDS()

registration_results$model_comparison |>
  knitr::kable()

From the sample data above, we can see that for nine out of ten genes, registered = TRUE, meaning that reference and query data between those nine genes can be aligned or registered. These data frame outputs can further be summarised and visualised; see the documentation on the processing registration results article.

Using other inputs

Loading sample data

As noted in the input data requirements article, register() also accepts a list of data frames or a list of reference and query vectors as input:

# Define expression value vectors
ref_expressions <- c(1.9, 3.1, 7.8, 31.6, 33.7, 31.5, 131.4, 107.5, 116.7, 112.5, 109.7, 57.4, 50.9)
query_expressions <- c(14, 12.1, 15.9, 47, 30.9, 50.5, 80.1, 67.4, 72.9, 61.7)

list_vector <- list(
  reference = ref_expressions,
  query = query_expressions
)

Registering the data

As previously shown, to register the input list, users can run the function register():

registration_results_vectors <- register(
  list_vector,
  reference = "Ref",
  query = "Query",
  scaling_method = "z-score"
)
#> ── Validating input data ───────────────────────────────────────────────────────
#> ℹ Will process 1 gene.
#> ℹ Using estimated standard deviation, as no `exp_sd` was provided.
#> ℹ Using `scaling_method` = "z-score".
#> 
#> ── Starting registration with optimisation ─────────────────────────────────────
#> ℹ Using L-BFGS-B optimisation method.
#> ℹ Using computed stretches and shifts search space limits.
#> ℹ Using `overlapping_percent` = 50% as a registration criterion.
#> ✔ Optimising registration parameters for genes (1/1) [170ms]

Registration results

The registration result object will have the same structure as when using a data frame as an input. Since no ID is explicitly defined in the input vector list, a unique gene_id will be automatically generated for the reference and query pair.

registration_results_vectors$model_comparison |>
  knitr::kable()

greatR documentation built on May 29, 2024, 12:18 p.m.