Add CDRP dataset as BBBC047

[shntnu]

The description can be copied from the abstract of
https://academic.oup.com/gigascience/article/6/12/giw014/2865213
and then link to that page

Additionally, add these notes

ftp://parrot.genomics.cn/gigadb/pub/10.5524/100001_101000/100351/profiles.tar.gz
comprises per-well averages of each extracted morphological feature computed across the cells.

This dataset is a superset of BBBC036 (which has only known bioactive compounds)

[shntnu]

@N3llz can you let @MarziehHaghighi know the BBBC dataset id for this dataset? I know it isn't going to be up for a while, but we just need a number for now so that she can use it in her Rosetta report.

[N3llz]

Yes, let's go with BBBC047. @MarziehHaghighi @shntnu

[shntnu]

Question from Gabriel Musso

• @shntnu reply to Gabe when we have figured this out. There might be some clues in https://github.com/broadinstitute/2015_Bray_GigaScience

During our investigation, we wanted to look for any systemic biases and evaluate options to normalize accordingly. We first began looking at plate specific effects, and then progressed towards the plate groups identified (Metadata_Plate_Map_Name in the processed features data set). Sorting these plate groups alphabetically, we noticed there were larger trends present across plate groups. The groups obtained by splitting the plate group names by hyphens (e.g. C-2113-01-D39-002 or H-BIOA-005-3) were highly correlated to plate similarity (see attached figure). However, we weren’t sure what the meaning was behind these labels.

We were hoping that you or someone on your team might be able to help us with the following questions:

• What is the meaning between C and H in the first portion of the label? The C plates seem to group strongly together.
• What is the meaning behind the remaining portions of these plate groups? Are they freezer locations? Were they run during the same day?
• Why do some plate groups contain more than 4 plates (e.g. see plates ~130-175 in figure, column 5)?

barcode_platemap_cdrp.txt

[shntnu]

https://broadinstitute.atlassian.net/wiki/spaces/IP/pages/114638720/2017-04-19+CDP2+data+show+decent+quality+both+for+bioactive+and+DOS+compounds might have relevant info

[shntnu]

Here's a summary of the prefixes of all the platemaps

# https://raw.githubusercontent.com/broadinstitute/2015_Bray_GigaScience/master/barcode_platemap_25412.csv
plateid_platemap <- 
  read_csv("barcode_platemap_25412.csv") %>%
  rename(platemap = Plate_Map_Name,
         plateid = Assay_Plate_Barcode)

plateid_platemap %>% mutate(platemap_prefix = str_extract(platemap, "(^[A-Z]-[A-Z0-9]+)")) %>% count(platemap_prefix, name = "num_platemaps", sort = TRUE) %>% knitr::kable()

platemap_prefix	num_platemaps
C-2113	130
H-BIOA	55
H-CBLE	39
H-CBLG	39
H-CBLC	27
H-CBLH	24
H-CBLB	20
H-CBLD	19
H-CBLF	12
H-CBLA	10
H-CBLJ	9
H-CBLO	9
H-CBLN	8
H-CBLP	8
H-CBLK	4

• H-BIOA is a designation for a designation for Broad's bioactives library.
• H-CBLx are DOS compounds (Diversity Oriented Synthesis, Stuart Schreiber lab)
• C-2113 plates were created for this project, and most likely come from the MLSMR library described below.

Here's a summary of the categories from the data resource paper

10,080 compounds came from the Molecular Libraries Small Molecule Repository (MLSMR), 2260 were drugs, natural products, and small-molecule probes that are part of the Broad Institute known bioactive compound collection, 269 were confirmed screening hits from the Molecular Libraries Program (MLP), and 18,051 were novel compounds derived from diversity-oriented synthesis.

Also available is the file cdrp_runs.txt contains the run id of each assay plate and the date on which it was run

The data is in this format:

run_id	platemap	barcode	date_started
2113-01-W01-01-12	C-2113-01-D39-013	AU00024329	2011-05-24
2113-01-W01-01-12	C-2113-01-D39-013	AU00024330	2011-05-24
2113-01-W01-01-12	C-2113-01-D39-013	AU00024331	2011-05-24
2113-01-W01-01-12	C-2113-01-D39-013	AU00024332	2011-05-24
2113-01-W01-01-12	C-2113-01-D39-012	AU00024333	2011-05-24
2113-01-W01-01-12	C-2113-01-D39-012	AU00024334	2011-05-24

This info was obtained today from Broad's CBIP database via this link (project = 2113 CDRP Cell painting & GE-HTS)

[nishanthmerwin]

Thank you @shntnu for looking into this, and further, for finding the date information. It seems like beyond the batch ID alone, the date is highly correlated with the average similarity between controls across plates. For clarity, here’s a brief summary of steps to generate the figure posted below:

1. Download pre-processed morphological features summarized per well from here: ftp://parrot.genomics.cn/gigadb/pub/10.5524/100001_101000/100351/profiles.tar.gz
2. Subset only the controls and calculate the centroid point across all features per plate.
3. Sort according to batch ID (Metadata_Plate_Map_Name)
4. Compute the manhattan pairwise similarity between all plates
5. Plot in heatmap and alongside outer bars that alternate colors every time there is a change according to covariate of interest.

[shntnu]

@nishanthmerwin Thank you for posting your results on this!

One thing worth testing is how severe this effect is in the treatment signatures. Systematic effects of this sort (plate-to-plate variation or well position effects) affect DMSO signatures much more than treatment because the former is typically weaker.

Here’s what I would test if I suspected that date was driving the effect:
Compute similarities between wells of 

1. different compounds and 
2. on different plate maps and 
3. of the same category eg DOS compounds 

And then test whether the difference in dates between pairs of wells correlates with/predicts the similarity between the pairs. 

(There are other ways to do this e.g. using random/fixed effects models)

[bledford87]

@shntnu @MarziehHaghighi Are the data complete for this dataset? If so, I'll just need access to the files and for someone to fill out the info on this form.

[shntnu]

@MarziehHaghighi – please go ahead and sort this out over the next couple of weeks because you reference this dataset in the NeurIPS paper

[MarziehHaghighi]

@MarziehHaghighi – please go ahead and sort this out over the next couple of weeks because you reference this dataset in the NeurIPS paper

@shntnu Could you please clarify what should I sort out here?

[shntnu]

Ah, whatever @bledford87 said

#61 (comment)

[bledford87]

@MarziehHaghighi I saw you submitted the form for this, so now I just need a zip file with the images and ground truth.

[shntnu]

The images actually live elsewhere: no actual day will be uploaded for this dataset. The form has the details

On Wed, Jun 16, 2021 at 9:17 AM Becki Ledford ***@***.***> wrote: @MarziehHaghighi <https://github.com/MarziehHaghighi> I saw you submitted the form for this, so now I just need a zip file with the images and ground truth. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#61 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAJHQPD3EEMZXOZVY3LWGS3TTCP75ANCNFSM4GDKDXAQ> .

-- -Shantanu

[bledford87]

Oh interesting, okay thanks! I don't think we've done a BBBC entry like this in the past, linking to the images in another spot, but I'll sort it out next week.

[shntnu]

Indeed, this dataset is unique in that sense.

…

On Wed, Jun 16, 2021 at 9:34 AM Becki Ledford ***@***.***> wrote: Oh interesting, okay thanks! I don't think we've done a BBBC entry like this in the past, linking to the images in another spot, but I'll sort it out next week. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#61 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAJHQPA4FANIH37BF7VBC5LTTCR6PANCNFSM4GDKDXAQ> .

[shntnu]

We might reprocess this dataset sometime this year, to make the profiles compatible with the feature set being used in https://jump-cellpainting.broadinstitute.org/

If/when we do, we will use this issue to document our progress

cc @jccaicedo @bethac07

Internal ref: https://broadinstitute.slack.com/archives/C01AF25CQLT/p1643838553629459?thread_ts=1634213872.002500&cid=C01AF25CQLT

[AnneCarpenter]

Some notes from the slack msg:
Notes from reprocessing the data back in 2017 https://broadinstitute.atlassian.net/wiki/spaces/IP/pages/114638720/2017-04-19+CDP2+data+show+decent+quality+both+for+bioactive+and+DOS+compounds
The images now live in s3://cellpainting-gallery/cpg0012-wawer-bioactivecompoundprofiling

[shntnu]

@AnneCarpenter Per our new protocol, we will make notes here broadinstitute/cellpainting-gallery#13 (comment)

[shntnu]

In case some is wondering: won't it be confusing that this dataset has an entry in both, BBBC and Cell Painting Gallery?

Yes! :)

But thankfully our Airtable
https://airtable.com/appctUGldmRNkVS19/tblXX3mTxhCR9Bxbq/viwNJfGOOJot7Wr3x/recIRNKWanTVXppE1?blocks=hide (private) will link up everything

(and Erin/Beth will decide how to make that info public)

Going forward, new profiling datasets will exist only in cellpainting-gallery. See #52 (comment)

[bethac07]

I think this is done, yes?

[shntnu]

I think this is done, yes?
Yes