Note: if you use frenchFISH in published research, please cite:
Macintyre, G., Piskorz, A.M., Ross, E., Morse, D.B., Yuan, K., Ennis, D., Pike, J.A., Goranova, T., McNeish, I.A., Brenton, J.D., Markowetz, F. (2018) FrenchFISH: Poisson models for quantifying DNA copy-number from fluorescence in situ hybridisation of tissue sections bioRxiv 10.1101/487926
To provide a real-world use case for frenchFISH, let’s take some FISH probe counts which were generated using FISHalyseR, an existing Bioconductor package which uses computer vision to automatically count spots visible in FISH images. These data, saved in SampleFISH.jpg_data.csv, are the output of FISHalyseR’s vignette, and includes counts for two color probes (a red one and a green one) that were generated for 44 cell nuclei. First, we must use frenchFISH’s built-in function to convert the CSV output of FISHalyseR to a probe count matrix formatted correctly for input to frenchFISH’s getAutomaticCountsEstimates function:
fishalyserCsvPath <- system.file("extdata",
"SampleFISH.jpg_data.csv",
package="frenchFISH")
automatic_counts <- convertFishalyserCsvToCountMatrix(fishalyserCsvPath)Once the count matrix has been created, we can run frenchFISH to quantify the copy number of each probed region by both correcting for nuclear volume correction and implementing a homogenous Poisson Point Process model. In addition to our count matrix, we include the nuclear radius of the cells we are using and the height of our tissue section (which should be modified as appropriate depending on the cell type and section height you are using). Note that before running the Poisson model, we set the seed to insure our results will be reproducible:
set.seed(366)
nuclear_radius <- 40
section_height <- 5
corrected_automatic_counts <- getAutomaticCountsEstimates(automatic_counts,
nuclear_radius,
section_height)
#>
#> Number of observations not used in the estimation process: 0
#> Total number of time observations: 46
#> Number of events: 70
#>
#> Convergence code: 0
#> Convergence attained
#> Loglikelihood: -41.805
#>
#> Estimated coefficients:
#> b0
#> 0.376
#> Full coefficients:
#> b0
#> 0.376
#> attr(,"TypeCoeff")
#> [1] "Fixed: No fixed parameters"
#>
#>
#> Number of observations not used in the estimation process: 0
#> Total number of time observations: 46
#> Number of events: 74
#>
#> Convergence code: 0
#> Convergence attained
#> Loglikelihood: -40.183
#>
#> Estimated coefficients:
#> b0
#> 0.434
#> Full coefficients:
#> b0
#> 0.434
#> attr(,"TypeCoeff")
#> [1] "Fixed: No fixed parameters"We can now see our corrected spot counts for each probe (including high and low confidence intervals and the median):
Next, let’s simulate a matrix of spot counts that have been manually counted for three color probes (a red one, a green one, and a blue one) across 20 cell nuclei. Note that this time, no area column is included as a first column:
manual_counts <- cbind(red = round(runif(20,0,5), 0),
green = round(runif(20,0,5), 0),
blue = round(runif(20,0,5), 0))To adjust these spot counts using frenchFISH, we simply input this
matrix, along with the nuclear radius of our cell type and section
height into the
getManualCountsEstimates function:
nuclear_radius <- 8
section_height <- 4
corrected_manual_counts <- getManualCountsEstimates(manual_counts,
nuclear_radius,
section_height)Here are the corrected spot counts as computed by Markov chain Monte Carlo modelling at five different percentiles per probe: