In andrew-edwards/eDNAfalse: Methods for Detecting Suspected False Positives in eDNA Samples

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.width = 6
)
# Switch for landscape tables in pdf:
output_html = ifelse(rmarkdown::metadata$output == "rmarkdown::html_vignette", TRUE, FALSE)

Small example

The object small_example has one mock sample and three non-mock samples, with four control species in the mock sample and three non-control species. The first column identifies the sample, and remaining columns are a1, a2, a3 and a4 for the four control species, and 1, 2 and 3 for the non-control species. Corresponding entries are the number of reads of each species in each sample.

library(eDNAcutoff)

small_example

# kable() is to display the results for vignette, but the chunk above is what you
#  would run in command line. Thus most chunks here are repeated to format the output
#  better in the vignette.
knitr::kable(small_example)

To remove the false positives use the function remove_false_pos():

small_removed <- remove_false_pos(small_example)
small_removed

small_removed <- remove_false_pos(small_example)
knitr::kable(small_removed)

To see which reads were set to zero use which_set_to_zero():

small_set_to_zero <- which_set_to_zero(small_example, small_removed)
small_set_to_zero

small_set_to_zero <- which_set_to_zero(small_example, small_removed)
knitr::kable(small_set_to_zero)

That used the default value of 0.2 for the parameter $\alpha$, which is the tolerance for ...add text

To test sensitivity to $\alpha$ use the function change_alpha() for a vector of $\alpha$ values:

alpha_values <- c(0.01, seq(0.1, 1, 0.1))
small_sens_alpha <- change_alpha(small_example, alpha_vec = alpha_values)

which returns a list containing:

• the number of reads set to zero (for each value of alpha_values)

small_sens_alpha$num_zeroed
plot(alpha_values, small_sens_alpha$num_zeroed,
     type = "o",
     col = "red",
     xlab = "Tolerance level",
     ylab = "Number of reads set to zero",
     ylim = c(0, max(small_sens_alpha$num_zeroed))
    )

• the number of control species left in non-mock samples (check)

small_sens_alpha$num_mock_sp_left

• the number of non-mock (check**?) samples remaining with non-control species

small_sens_alpha$num_samp_rem_with_mock_sp

• a data frame (tibble) for each value of alpha_values, for example

alpha_values[6]

small_sens_alpha$out_list[[6]]

knitr::kable(small_sens_alpha$out_list[[6]])

and

alpha_values[7]

small_sens_alpha$out_list[[7]]

knitr::kable(small_sens_alpha$out_list[[7]])

Note how the read of 10 for sample 2 and species 2 gets kept for $\alpha=0.5$ but zeroed for $\alpha=0.6$ (corresponding to the red line increasing up at $\alpha=0.6$ in the figure).

Big example

The more realistic big_example has XXX mock samples and XXX non-mock samples, with ...

big_example

if(output_html){
  knitr::kable(big_example) } else {
                              library(kableExtra)
                              kableExtra::kable(big_example, longtable = TRUE,
                                                booktabs = TRUE) %>%
                                kable_styling(font_size = 7) %>%
                                landscape()
                              }

To remove the false positives:

big_removed <- remove_false_pos(big_example)
big_removed

big_removed <- remove_false_pos(big_example)
if(output_html){
  knitr::kable(big_removed) } else {
                              kableExtra::kable(big_removed, longtable = TRUE,
                                                booktabs = TRUE) %>%
                                kable_styling(font_size = 7) %>%
                                landscape()
                              }

And again, to see which reads were set to zero use which_set_to_zero():

big_set_to_zero <- which_set_to_zero(big_example, big_removed)
big_set_to_zero

big_set_to_zero <- which_set_to_zero(big_example, big_removed)
if(output_html){
  knitr::kable(big_set_to_zero) } else {
                                kableExtra::kable(big_set_to_zero, longtable = TRUE,
                                                booktabs = TRUE) %>%
                                kable_styling(font_size = 7) %>%
                                landscape()
                              }

To test sensitivity to $\alpha$, do as above (but just plot the results here)

As above, to test sensitivity to $\alpha$:

big_sens_alpha <- change_alpha(big_example, alpha_vec = alpha_values)

which returns a list containing:

• the number of reads set to zero (for each value of alpha_values)

big_sens_alpha$num_zeroed
plot(alpha_values, big_sens_alpha$num_zeroed,
     type = "o",
     col = "red",
     xlab = "Tolerance level",
     ylab = "Number of reads set to zero",
     ylim = c(400, max(big_sens_alpha$num_zeroed))
    )

• the number of control species left in non-mock samples (check)

big_sens_alpha$num_mock_sp_left

• the number of non-mock samples remaining with non-control species

big_sens_alpha$num_samp_rem_with_mock_sp

• a data frame (tibble) for each value of alpha_values available as big_sens_alpha$out_list[[6]] etc. (not shown).

TODO Seems surprising that the big data set is fairly insensitive to $\alpha$ - look into somewhat manually and check. And set up a test().

andrew-edwards/eDNAfalse documentation built on Dec. 19, 2020, 1:08 a.m.