dpcR is an R package designed to perform an analysis of digital PCR experiments.
dpcR does not have built-in tools for reading specific data types. Instead, we advise to use other packages belonging do the pcRuniversum such as RDML or dedicated packages such as ReadqPCR (available from bioconductor.org). Further information can be found in Pabinger et al. 2014 (A survey of tools for the analysis of quantitative PCR (qPCR) data. Biomolecular Detection and Quantification 1, 23–33. doi:10.1016/j.bdq.2014.08.002).
Before analysis data should be saved using one of the three classes - adpcr (end-point array digital PCR experiments), ddpcr (droplet digital PCR experiments) and rtadpcr (real-time array digital PCR experiments). To do so, use create_dpcr function.
library(dpcR)
#generate some data from 15x16 array. Let's presume, that we have results from two plates
sample_runs <- matrix(rpois(480, lambda = 1.5), ncol = 2)
#check its class - it's a typical R structure
class(sample_runs)
## [1] "matrix"
#save it to adpcr object
adpcr_experiments <- create_dpcr(sample_runs, n = 240L, type = "nm", adpcr = TRUE)
class(adpcr_experiments)
## [1] "adpcr"
## attr(,"package")
## [1] "dpcR"
Data can be easily visualized using plot_panel function.
#remember, you can plot only single panel at once
plot_panel(extract_dpcr(adpcr_experiments, 1), nx_a = 15, ny_a = 16, main = "Experiment 1")
The same data can be visualized easily in binarized form (positive/negative partitions).
#remember, you can plot only single panel at once
plot_panel(binarize(extract_dpcr(adpcr_experiments, 1)), nx_a = 15, ny_a = 16, main = "Experiment 1")
Generalized Linear Models (GLM) are linear models for data, where the response variables may have non-normal distributions (as for example binomially distributed positive partitions in digital PCR experiments). Using GLM we can describe relationships between results of digital PCR:
\[\log{Y} = \beta^T X\]
where \(Y\) are counts, \(X\) are experiments names (categorical data) and \(\beta\) are linear model coefficients for every experiment. Moreover, \(\exp{\beta} = \lambda\).
Estimated means copies per partitions obtained from the model are compared each other using multiple t-test.
#comparison of the experiments using GLM
#1. Simulate results of three different experiments with two repetitions each
#experiment 2 differs significantly from others
adpcr1 <- sim_adpcr(m = 20, n = 765, times = 1e5, pos_sums = FALSE, n_panels = 2)
adpcr2 <- sim_adpcr(m = 100, n = 765, times = 1e5, pos_sums = FALSE, n_panels = 2)
adpcr3 <- sim_adpcr(m = 20, n = 765, times = 1e5, pos_sums = FALSE, n_panels = 2)
#2. Join results for convenience
adpcrs <- bind_dpcr(adpcr1, adpcr2, adpcr3)
#3. Perform test.
comp <- test_counts(adpcrs)
## Warning in chkdots(...): Argument(s) 'level' passed to '...' are ignored
## Warning in RET$pfunction("adjusted", ...): lower == upper
## Warning in RET$pfunction("adjusted", ...): lower == upper
#4. See summary of the test
summary(comp)
## group lambda lambda.low lambda.up
## 1 a 0.02649297 0.02119064 0.03314573
## 2 b 0.12890717 0.11620816 0.14309751
#5. Plot results of the test
plot(comp, aggregate = FALSE)
plot(comp, aggregate = TRUE)
#extract coefficients for the further usage
coef(comp)
## group lambda lambda.low lambda.up
## 1.1 a 0.02917979 0.02358095 0.03613218
## 1.2 a 0.02515023 0.01999742 0.03165195
## 2.1 b 0.12072855 0.10847707 0.13445923
## 2.2 b 0.13708579 0.12393925 0.15173578
## 3.1 a 0.02515023 0.01999742 0.03165195
## 3.2 a 0.02649162 0.02118678 0.03314687
The Poisson regression on binary data (positive/negative partition) can be used only when the concentration of template molecules in samples is small (positive partitions contain very rarely more than 1 template particle). Higher concentrations requires binomial regression.