Addition of R script to compare CNV caller output #142

cbethell · 2019-10-02T13:37:13Z

Purpose/implementation

The purpose of this draft PR is to produce a copy number plot showing the recurrently amplified/deleted regions detected by CNVkit and Control-FREEC.

The initial plan was to implement the TAGCNA workflow, but I encountered some issues when attempting to format the data to fit their pipeline.

The GenVisR package was used as a temporary alternative to TAGCNA and the plots produced via this package include:
analyses/cnv-comparison/plots/compare_cnv_output_frequency.pdf
analyses/cnv-comparison/plots/compare_cnv_output_proportion.pdf

The ggplot2 package was used to produce the remaining plots found in the plots directory of this module.

Issue

This PR addresses issue #8

Directions for reviewers

Per [the comment made] on the relevant issue #8, I attempted to implement the TAGCNA workflow. However, it was unsuccessful due to the specificity of the input format it expects.

I am still looking into the best way to produce figures like those in the TAGCNA documentation, but felt it necessary to show what has been produced thus far.

Given the purpose of this analysis, I chose to write it in a R script. I used the sample-distribution-analysis module as a guide in this sense. That being said, would anyone argue for the need of separate scripts and a shell script to sequentially execute said scripts?
I also began to include optparse functions and decided to file a draft PR of the script as is so that I can receive a second opinion on whether or not command line options are necessary for this module (again, I used the sample-distribution-analysis module as a reference).

Results

This script produces five plots as follows:
analyses/cnv-comparison/plots/compare_cnv_output_barplot_aberration.pdf
analyses/cnv-comparison/plots/compare_cnv_output_barplot_histology.pdf
analyses/cnv-comparison/plots/compare_cnv_output_boxplot.pdf
analyses/cnv-comparison/plots/compare_cnv_output_frequency.pdf
analyses/cnv-comparison/plots/compare_cnv_output_proportion.pdf

Docker and continuous integration

The dependencies required to run the code in this pull request have been added to the project Dockerfile.
This analysis has been added to continuous integration.

PR Checklist

Run a linter (not done as yet)
Set the seed (if applicable)
Comments and/or documentation up to date
Double check your paths

- Plots were made using GenVisR and ggplot2 - Addition of custom functions script (sourced in plotting script)

- added a `read_in_cnv` function - removed `set.seed()` as it seemed unnecessary

…v-comparison-plotting

`set.seed` no longer seems to be necessary in this script

cansavvy

This looks great! Very-well documented so it was pretty easy to know what was going on. I just have a few suggestions for drying things up and changing your object naming to be more specific. Let me know if you have questions about any of my comments.

Also my take on your two questions:

Given the purpose of this analysis, I chose to write it in a R script. I used the sample-distribution-analysis module as a guide in this sense. That being said, would anyone argue for the need of separate scripts and a shell script to sequentially execute said scripts?

I also began to include optparse functions and decided to file a draft PR of the script as is so that I can receive a second opinion on whether or not command line options are necessary for this module (again, I used the sample-distribution-analysis module as a reference).

I could see either adding more arguments to your functions and then running this from a bash script with some optparse options OR, because this consists of so many plots, making 01-cnv-comparison-plotting into an R Notebook so the plots could be seen all in one spot. I'd say one or the other. We can see what @jashapiro thinks.

cansavvy · 2019-10-08T13:31:46Z

analyses/cnv-comparison/util/cnv-comparison-functions.R

+  # Given the file path of the CNV data, read in the data.
+  #
+  # Arg:
+  #   input_directory: file path of input directory


Is there a reason to designate this separately than the file.path? Can you just make it one file.path to the data? I found this confusing for a second.

Good point, I do see how this can be confusing. I am sure this was better suited for my original, more complicated script, but making it one file.path to the data does seem to make the most sense now.

cansavvy · 2019-10-08T13:33:35Z

analyses/cnv-comparison/util/cnv-comparison-functions.R

+  #   dataframe: the data frame containing the input data
+
+  # Read in cnv data
+  dataframe <-


If this is for sure CNV data, can you name this dataframe something that reflects that like cnv_df or etc?

cansavvy · 2019-10-08T13:33:38Z

analyses/cnv-comparison/util/cnv-comparison-functions.R

+    read.table(gzfile(file.path(input_directory, file_path)), header = TRUE)
+
+  # Rename the columns
+  dataframe <-


Suggested change

dataframe <-

colnames(dataframe) <- c("chrom", "loc.start", "loc.end", "ID", "num.mark", "seg.mean")

IDK if this is really all that better but at least you don't have to say dataframe twice. Could also use dplyr::rename.

It is better from a memory standpoint. No need to copy the data frame.

I now realize that my comment above this line was misleading as this was meant to rearrange the columns, rather than simply rename them. That being said, the function colnames won't be feasible here. I am thinking of using dplyr::select as the replacement here unless either of you may have a better solution.

Yeah, do dplyr::select/rename

cansavvy · 2019-10-08T13:34:12Z

analyses/cnv-comparison/util/cnv-comparison-functions.R

+  # a column with the values 0 and 1 to represent loss and gain, respectively.
+  #
+  # Args:
+  #   dataframe: data.frame containing CNV caller output


I'm going to have the same comment here as above. dataframe doesn't tell us too much about what the object is supposed to be.

cansavvy · 2019-10-08T13:35:03Z

analyses/cnv-comparison/util/cnv-comparison-functions.R

+#
+# Rscript analyses/cnv-comparison/util/cnv-comparison-functions.R
+
+read_in_cnv <- function(input_directory, file_path){


What's the motivation for this being its own function rather than just using these steps as is?

I'm not saying it shouldn't be a function necessarily, just wanna know the motivation. Perhaps as I read the rest of this PR it will make more sense to me.

I made this its own function due to the repetitive nature of the original code I had reading in the original and subset CNV files. In other words, this was my attempt to DRY it up a bit (which I believe can now be achieved when I implement your suggestion of putting the datasets in a list and using lapply to execute this function and subsequent functions).

cansavvy · 2019-10-08T14:16:33Z

analyses/cnv-comparison/util/cnv-comparison-functions.R

+  #
+  # Return:
+  #   barplot: barplot depicting the proportion of aberrations detected across
+  #            chromosomes, annoated by the broad_histology variable in the


Suggested change

# chromosomes, annoated by the broad_histology variable in the

# chromosomes, annotated by the `broad_histology` variable in the

cansavvy · 2019-10-08T14:21:22Z

analyses/cnv-comparison/01-cnv-comparison-plotting.R

+metadata <-
+  readr::read_tsv(file.path(input_directory, "pbta-histologies.tsv"))
+
+# Read in cnvkit data


Since, you've made your functions, you could put all your datasets in a list and then call the function from there so you only have to say it once.

cnv_data <- lapply(<LIST_OF_FILE_PATHS>, read_in_cv)

Now for the rest of your functions, you would only have to run it in one line:

cnv_filtered <- lapply(cnv_data, filter_segmean)

Let me know if this makes sense, if not I can expound on this idea.

cansavvy · 2019-10-08T14:22:17Z

analyses/cnv-comparison/01-cnv-comparison-plotting.R

+# Filter the controlfreec subset data for cutoff in preparation for plotting
+controlfreec_subset <- filter_segmean(controlfreec_subset)
+
+#### Plot frequency and proportion using GenVisR -------------------------------


You could do the same lapply strategy here. But then twice, once for proportion and once for frequency.

cansavvy · 2019-10-08T14:23:56Z

analyses/cnv-comparison/01-cnv-comparison-plotting.R

+# Run `plot_cnFreq` for cnvkit (has to be run with subset because
+# original dataset is too large for function)
+cnvkit_prop_plot <-
+  plot_cnFreq(cnvkit_format, "proportion", "CNVkit proportion")


If you are always planning on running both proportion and frequency, you could make an argument in your function that runs both automatically so you don't have to repeat yourself.

cansavvy · 2019-10-08T14:24:35Z

analyses/cnv-comparison/01-cnv-comparison-plotting.R

+
+##### Plot barplots using ggplot2 ----------------------------------------------
+
+# Run `plot_histology_barplot` on cnvkit


This section could also be lapply-ed and made shorter.

…into cnv-comparison-plotting

cbethell · 2019-10-10T13:51:26Z

@jashapiro I'll wait on your review before changing this script into a R notebook or running it from a bash script with optparse options. I have committed changes that address all of the other comments @cansav09 made.

jashapiro

Looks good. Biggest comment is about fixing the order of the chromosomes by making that column a factor after reading. Just make sure you don't accidentally recode them!

jashapiro · 2019-10-17T14:52:11Z

analyses/cnv-comparison/util/cnv-comparison-functions.R

+
+  # Rearrange the columns
+  cnv_df <- cnv_df %>%
+    dplyr::select("chrom", "loc.start", "loc.end", "ID", "num.mark", "seg.mean")


At this point (I think) you should also recode the chromosome into a factor, with logical order. You want the levels to be:
paste0("chr", c(1:22, "X", "Y"))

Good point, will do.

jashapiro · 2019-10-17T14:56:35Z

analyses/cnv-comparison/util/cnv-comparison-functions.R

+
+}
+
+plot_cn_freq <-


Do you need this function? IT really just calls the GenVisR function, so it seems redundant.

I also see your point here, however, the motivation behind making this a function was to keep the calling of functions in the 01-cnv-comparison-plotting.R script uniformed and the number of lines of code to a minimum.

That being said, would your recommendation here be to call the GenVisR function in the 01 script nonetheless?

I think that would be my rec, yes. You should still be able to use the lapply, though you may need to add named arguments, depending on the order that GenVisR::cnFreq expects the additional options.

- removed the `plot_cnfreq` function from the functions script and implemented `GenVisR::cnFreq` function in the `01` script - reordered the chromosome levels

jashapiro · 2019-10-17T17:29:32Z

analyses/cnv-comparison/util/cnv-comparison-functions.R

+    dplyr::select("chrom", "loc.start", "loc.end", "ID", "num.mark", "seg.mean")
+
+  # Reorder chromosome levels 
+  levels(cnv_df$chrom) <- c(paste0("chr", c(1:22, "X", "Y")))


The above doesn't actually convert to a factor, but makes it a kind of strange hybrid character with a levels attribute:
dplyr::mutate(chrom = factor(chrom, levels = paste0("chr", c(1:22, "X", "Y"))
should work

If you look at the before and after plots, you can see that the order didn't actually change but the labels did. Which would be bad...

Ahh, I thought it was strange when I initially tried to convert to a factor, what R claimed to already be a factor. Thanks, I'll give that a shot now.

oooh... add stringsAsFactors=FALSE to your read.table (or use readr...)

I also made that realization a minute ago. That'll also be in this upcoming commit 👍

- included `stringsAsFactors = FALSE` argument to `read.table` function

jashapiro · 2019-10-17T19:05:06Z

analyses/cnv-comparison/01-cnv-comparison-plotting.R

+
+if (!("cowplot" %in% installed.packages())) {
+  install.packages("BiocManager")
+  BiocManager::install("cowplot")


Is there a reason not to install this from cran?

Good question, I cannot pinpoint my initial motivation here so I can change this to install from cran.

- re-ran plots

jashapiro

This looks pretty good now, though I am very much confused by the results: the two callers are kind of shockingly different, and I find it very strange that ControlFREEC is calling more on the smaller chromosomes. That result makes me somewhat suspicious about what is going on, and I would like to hear more about the data there.

There is one major change that I think would be good for the ggplot-based plots, and that is to combine the two data frames with an extra column designating whether the sample is from CNVkit vs ControlFREEC, then using faceting to split the two, rather than relying on cowplot. This would have the advantage of putting the two plots on the same scale. You can do this pretty easily with dplyr::bind_rows() using the .id option to specify the new column name.

I would also like to see the results for the histology plot separated by aberration type, especially since the two methods have such different proportions of gain/loss calls.

On what I think was one of the original questions, I do think this script might probably be better off converted to a notebook, to give some space to comment a bit on these results, rather than just producing the raw plots.

jashapiro · 2019-10-17T19:09:18Z

analyses/cnv-comparison/util/cnv-comparison-functions.R

+  meta_joined <- filtered_cnv_df %>%
+    dplyr::inner_join(metadata, by = c("sample" = "Kids_First_Biospecimen_ID"))
+
+  # Create barplot where the y-axis represents the size of aberration


On second look, I am pretty confused by these plots. What is actually being plotted here? I had thought it was the number of each type in each chromosome, but the stat = "identity" is confusing me. Would the default stat = "count" make more sense?

On some investigation, it looks like these ultimately produce the same plots as the ones that I was expecting, but if you leave off the y aesthetic, and use the default geom_bar() (no stat) you get a much smaller file size and better y axis labels.

jashapiro · 2019-10-17T19:29:07Z

analyses/cnv-comparison/util/cnv-comparison-functions.R

+  # Create barplot where the y-axis represents the size of aberration
+  barplot <- ggplot2::ggplot(filtered_cnv_df,
+                             ggplot2::aes(x = chromosome,
+                                          y = as.factor(aberration),


See earlier comment about the y axis/stat

jashapiro · 2019-10-17T19:29:49Z

analyses/cnv-comparison/util/cnv-comparison-functions.R

+                                          fill = broad_histology)) +
+    ggplot2::geom_bar(stat = "identity") +
+    ggplot2::theme_bw() +
+    ggplot2::theme(axis.text.y = ggplot2::element_blank())


Add y axis label?

jharenza · 2019-10-18T01:55:54Z

This looks pretty good now, though I am very much confused by the results: the two callers are kind of shockingly different, and I find it very strange that ControlFREEC is calling more on the smaller chromosomes. That result makes me somewhat suspicious about what is going on, and I would like to hear more about the data there.

There is one major change that I think would be good for the ggplot-based plots, and that is to combine the two data frames with an extra column designating whether the sample is from CNVkit vs ControlFREEC, then using faceting to split the two, rather than relying on cowplot. This would have the advantage of putting the two plots on the same scale. You can do this pretty easily with dplyr::bind_rows() using the .id option to specify the new column name.

I would also like to see the results for the histology plot separated by aberration type, especially since the two methods have such different proportions of gain/loss calls.

On what I think was one of the original questions, I do think this script might probably be better off converted to a notebook, to give some space to comment a bit on these results, rather than just producing the raw plots.

@cbethell and @jashapiro - would love to see this in a notebook. Of note, we are still benchmarking (nearing the end) the CN algorithms (and have a caveat on this in the README). We had a few samples show genome-wide gain and after tweaking parameters and adding options for BAF adjustment, we have much better results. Of additional note, I thought I mentioned this, but forgive me if not, I think we would ideally use results from #128 for this ticket, though I think @cgreene and I thought this ticket could still be worked on to get the code ready for a later plug and chug. We have not created a PR for #128 yet because we realized we had to use different CN files from what we had released (containing predicted gain/loss info without the need to threshold LRR) in order to create consensus (these data will be added, hopefully in #146), but the early data is only showing ~50% consensus. We are now working on integrating Manta as a 3rd algorithm and doing a 2/3 consensus and want to toy with degree of overlap based on size of CNV, rather than 2/2. Long story short, I can help you review the results by histology if in a notebook based on what we know about certain diseases and CNV expectations!

jaclyn-taroni · 2019-10-18T12:32:45Z

Of additional note, I thought I mentioned this, but forgive me if not, I think we would ideally use results from #128 for this ticket, though I think @cgreene and I thought this ticket could still be worked on to get the code ready for a later plug and chug.

From #8 (comment):

I imagine the output would still be a SEG file, such that you can start with any of the SEGs we provide.

This appears to have the functionality to read in files in that format: https://github.com/AlexsLemonade/OpenPBTA-analysis/pull/142/files#diff-fd9d53007cdfd11cf0ae23e3eede7eabR12

Long story short, I can help you review the results by histology if in a notebook based on what we know about certain diseases and CNV expectations!

What are you looking for in a notebook @jharenza? Do you want to see the plot here: cnv-comparison/plots/compare_cnv_output_barplot_histology.pdf or are you looking for additional analyses beyond:

I would also like to see the results for the histology plot separated by aberration type, especially since the two methods have such different proportions of gain/loss calls.

My inclination is to, once checked for correctness, get this merged and iterate on it. @jashapiro thoughts?

…into cnv-comparison-plotting

…/OpenPBTA-analysis into cnv-comparison-plotting

- combined data.frames and used faceting to plot ggplots

jharenza · 2019-10-21T14:21:16Z

@jaclyn-taroni ahh thanks for the link, i didn't readily see the plot - the notebook idea was just to see everything gathered together for quick visuals. My question for this analysis - is this more of a QC or do we want to turn this into a manuscript figure? If the latter, I think having this plotted in GISTIC-type format for the different histologies or as an IGV-type figure, grouped by histology, would be most effective for showing recapitulation of previous findings.

cbethell · 2019-10-21T14:28:16Z

@jaclyn-taroni ahh thanks for the link, i didn't readily see the plot - the notebook idea was just to see everything gathered together for quick visuals. My question for this analysis - is this more of a QC or do we want to turn this into a manuscript figure? If the latter, I think having this plotted in GISTIC-type format for the different histologies or as an IGV-type figure, grouped by histology, would be most effective for showing recapitulation of previous findings.

@jharenza I can work on plotting this as an IGV-type figure. My initial attempt to apply GISTIC here was unsuccessful due to compatibility issues with my OS. However, I can give it a more thorough attempt or explore other avenues to produce this figure. I will mention you on this issue for further review once produced.

jharenza · 2019-10-21T15:13:52Z

OK - I can run GISTIC quickly on the v5 data and potentially add this to the data release, will send you the files so you can get started. I also noticed maftools updated some of their CN plotting abilities, so producing those per histology could be nice, too.

cbethell · 2019-10-21T18:29:03Z

@jashapiro I addressed the comments you made above. Does this look ready to merge now? The changes involving GISTIC can be made and filed in a separate PR.

jashapiro

Looks good. One little simplifying tweak. Changing to separate gain and loss really highlights how strange the control-Freec gain calls are.

jashapiro · 2019-10-22T15:32:08Z

analyses/cnv-comparison/01-cnv-comparison-plotting.Rmd

+    file.path(input_directory, "pbta-cnv-cnvkit.seg.gz"),
+    file.path(input_directory, "pbta-cnv-controlfreec.seg.gz")


Name here to save work later.

Suggested change

file.path(input_directory, "pbta-cnv-cnvkit.seg.gz"),

file.path(input_directory, "pbta-cnv-controlfreec.seg.gz")

cnvkit = file.path(input_directory, "pbta-cnv-cnvkit.seg.gz"),

controlfreec = file.path(input_directory, "pbta-cnv-controlfreec.seg.gz")

Use <- assignment though please 🙂

not assignment, naming!

Nope I'm wrong!

jashapiro · 2019-10-22T15:32:36Z

analyses/cnv-comparison/01-cnv-comparison-plotting.Rmd

+combined_cnv_filtered$cnv_caller <- gsub("1", "cnvkit", combined_cnv_filtered$cnv_caller)
+combined_cnv_filtered$cnv_caller <- gsub("2", "controlfreec", combined_cnv_filtered$cnv_caller)


With the named list above, this should be correct automatically, so you should be able to delete these two lines.

jashapiro · 2019-10-22T15:37:04Z

analyses/cnv-comparison/util/cnv-comparison-functions.R

+                                          fill = broad_histology)) +
+    ggplot2::geom_bar() +
+    ggplot2::theme_bw() +
+    ggplot2::facet_grid(cnv_caller~aberration) +


Suggested change

ggplot2::facet_grid(cnv_caller~aberration) +

ggplot2::facet_grid(cnv_caller ~ aberration) +

jashapiro · 2019-10-22T15:37:20Z

analyses/cnv-comparison/util/cnv-comparison-functions.R

+                                          fill = aberration)) +
+    ggplot2::geom_bar() +
+    ggplot2::theme_bw() +
+    ggplot2::facet_grid(cnv_caller~aberration) +


Suggested change

ggplot2::facet_grid(cnv_caller~aberration) +

ggplot2::facet_grid(cnv_caller ~ aberration) +

- fixed spacing around `~` operator - fixed line spacing in `.Rmd` file

jashapiro

Looks good! (Except for ControlFreec...)

cbethell · 2019-10-22T18:08:57Z

Looks good! (Except for ControlFreec...)

Thank you and agreed!

cbethell and others added 2 commits October 2, 2019 09:11

Addition of R script to compare CNV caller output

94fd9d1

- Plots were made using GenVisR and ggplot2 - Addition of custom functions script (sourced in plotting script)

Merge branch 'master' into cnv-comparison-plotting

486d6aa

cbethell changed the title ~~Addition of R script to compare CNV caller output~~ In Progress: Addition of R script to compare CNV caller output Oct 2, 2019

cbethell added 4 commits October 7, 2019 17:01

Attempted to DRY up the code

2e132c8

- added a `read_in_cnv` function - removed `set.seed()` as it seemed unnecessary

Merge remote-tracking branch 'origin/cnv-comparison-plotting' into cn…

6493375

…v-comparison-plotting

Updated branch

d00858d

Removed set.seed

7f5761f

`set.seed` no longer seems to be necessary in this script

cbethell requested review from cansavvy and jashapiro October 7, 2019 23:52

cbethell marked this pull request as ready for review October 7, 2019 23:52

cbethell changed the title ~~In Progress: Addition of R script to compare CNV caller output~~ Addition of R script to compare CNV caller output Oct 7, 2019

cansavvy reviewed Oct 8, 2019

View reviewed changes

cbethell added 2 commits October 8, 2019 16:39

Merge branch 'master' of https://github.com/cbethell/OpenPBTA-analysis …

80fe382

…into cnv-comparison-plotting

@cansav09 PR review suggested changes

c2b2721

Merge branch 'master' into cnv-comparison-plotting

7955075

jashapiro reviewed Oct 17, 2019

View reviewed changes

cbethell and others added 2 commits October 17, 2019 11:51

Merge branch 'master' into cnv-comparison-plotting

e3c00a0

@jashapiro PR review suggested changes

d913351

- removed the `plot_cnfreq` function from the functions script and implemented `GenVisR::cnFreq` function in the `01` script - reordered the chromosome levels

jashapiro reviewed Oct 17, 2019

View reviewed changes

- Converted chrom column vector into a true factor vector

8816983

- included `stringsAsFactors = FALSE` argument to `read.table` function

jashapiro reviewed Oct 17, 2019

View reviewed changes

Revert GenVisR::cnFreq to plot_cn_freq change in 01 script

ea25caa

- re-ran plots

jashapiro reviewed Oct 17, 2019

View reviewed changes

Install cowplot from cran

6222c95

cbethell and others added 3 commits October 21, 2019 09:31

Merge branch 'master' into cnv-comparison-plotting

f9f0d76

Merge branch 'master' of https://github.com/cbethell/OpenPBTA-analysis …

3cf5643

…into cnv-comparison-plotting

Merge branch 'cnv-comparison-plotting' of https://github.com/cbethell…

67869be

…/OpenPBTA-analysis into cnv-comparison-plotting

Converted R script into R notebook

cf889d5

- combined data.frames and used faceting to plot ggplots

Updated .circleci/config.yml

ce84bd2

jashapiro reviewed Oct 22, 2019

View reviewed changes

Renamed input in list in .Rmd file

a5624f8

- fixed spacing around `~` operator - fixed line spacing in `.Rmd` file

jashapiro approved these changes Oct 22, 2019

View reviewed changes

jaclyn-taroni merged commit 7087909 into AlexsLemonade:master Oct 22, 2019

jaclyn-taroni mentioned this pull request Oct 26, 2019

Planned Analysis: Copy number plot showing recurrently amplified/deleted regions in different PBTA cancers #8

Closed

cbethell deleted the cnv-comparison-plotting branch December 18, 2019 19:12

	dataframe <-
	colnames(dataframe) <- c("chrom", "loc.start", "loc.end", "ID", "num.mark", "seg.mean")

	# chromosomes, annoated by the broad_histology variable in the
	# chromosomes, annotated by the `broad_histology` variable in the


		##### Plot barplots using ggplot2 ----------------------------------------------

		# Run `plot_histology_barplot` on cnvkit

		file.path(input_directory, "pbta-cnv-cnvkit.seg.gz"),
		file.path(input_directory, "pbta-cnv-controlfreec.seg.gz")

		combined_cnv_filtered$cnv_caller <- gsub("1", "cnvkit", combined_cnv_filtered$cnv_caller)
		combined_cnv_filtered$cnv_caller <- gsub("2", "controlfreec", combined_cnv_filtered$cnv_caller)

	ggplot2::facet_grid(cnv_caller~aberration) +
	ggplot2::facet_grid(cnv_caller ~ aberration) +

Addition of R script to compare CNV caller output #142

Addition of R script to compare CNV caller output #142

Conversation

cbethell commented Oct 2, 2019 • edited Loading

Purpose/implementation

Issue

Directions for reviewers

Results

Docker and continuous integration

PR Checklist

cansavvy left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

cansavvy Oct 8, 2019 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

cbethell commented Oct 10, 2019

jashapiro left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jashapiro Oct 17, 2019 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jashapiro Oct 17, 2019 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jashapiro left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jharenza commented Oct 18, 2019

jaclyn-taroni commented Oct 18, 2019 • edited Loading

jharenza commented Oct 21, 2019

cbethell commented Oct 21, 2019

jharenza commented Oct 21, 2019

cbethell commented Oct 21, 2019

jashapiro left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jashapiro left a comment

Choose a reason for hiding this comment

cbethell commented Oct 22, 2019

cbethell commented Oct 2, 2019 •

edited

Loading

cansavvy Oct 8, 2019 •

edited

Loading

jashapiro Oct 17, 2019 •

edited

Loading

jashapiro Oct 17, 2019 •

edited

Loading

jaclyn-taroni commented Oct 18, 2019 •

edited

Loading