Frequently Asked Questions

Analyses

How can I reproduce an environment to explore JUMP data?

The easiest way to set things up will be installing from pip in your environment of choice:

pip install jump-deps

Data

Does JUMP contain my compound/gene of interest?

The easiest way to find out is to follow the instructions on the perturbation availability section. Alternatively, you can explore the metadata tables on the datasets repository, which are used to generate the explorable table.

Where are the datasets’ specifications?

The main resource to understand the technicalities of the JUMP datasets collection and assembly is on this repo.

Why do some samples have images but no downstream analysis?

Some plates failed Quality Control (QC) but we kept them because they may be useful for developing QC methods.

Why do some perturbations have so many replicates?

Some genetic and chemical perturbations are positive or negative controls (see next question) and thus appear frequently in the dataset.

What are the identifiers for positive and negative controls?

You can find all positive and negative controls for JUMP-ORF, -CRISPR, and -compound here. For the compounds dataset the only negative control is ‘JCP2022_033924’ (DMSO). Most chemical compound plates contain 16 negative control wells, while some have as many as 28 wells. In the ORF dataset, replicates are positioned in wells O23, O24, P23 and P24. The remaining wells contain ORF treatments, with a single replicate of each per plate map and with five replicate plates produced per plate map (private link | html).

What are the CellProfiler pipelines used to process images and output raw single cell level profiles for JUMP?

You can find the CellProfiler pipelines that perform illumination correction, segmentation and feature extraction here.

What are the data analysis pipelines that convert raw profiles into well-level profiles?

There are two pipelines used: One converts single cell raw profiles into aggregated well level profiles, the second transforms these aggregated profiles to improve signal quality. We call the first pipeline profiling-recipe and it is located in this permalink. For the second pipeline, the profiles_index.json file contains links to the specific version of the jump profiling recipe used and its configuration. To clarify, the profiling-recipe and jump-profiling-recipe are not the same, they correspond to the first and second pipelines, respectively.

What is the JUMP production profile, and how does it differ from the original public compound profile?

The JUMP production profile means the compound-only CellProfiler profile named compound_no_source7, which was used for the JUMP production paper. It is not a new version of the whole JUMP dataset: it has no ORF, CRISPR, or gallery counterpart.

Compound profile	What it contains	When to use it
Original public compound	All compound sources, including source_7.	The default for general JUMP analyses and for consistency with work based on the original public release.
JUMP production compound_no_source7	The final paper profile, excluding source_7 because it made batch correction difficult.	Reproducing the production paper or aligning a compound analysis with it. Removing source_7 changes the data, so activity values and other statistics must be calculated from this profile rather than copied from the original public profile.

The production profile does not supersede the original public compound profile; both remain useful for their respective purposes. Its Harmony output retains 758 feature-aligned dimensions, so the same production profile supports similarity and CellProfiler feature analyses by restoring the original feature names positionally.

The frozen paper-consistent tables are preserved independently in Zenodo record 21515641. You can explore the historical compound matches and CellProfiler feature rankings in Datasette Lite. These historical links are separate from the living jump_rr tables, which continue to track recommended profiles.

Background discussion: jump_hub PR #91 and monorepo issue #101.

Do we expect one gene’s JCP ID (JUMP Cell Painting ID) to be associated with multiple targets?

Yes, many genes are associated with multiple targets and are correctly annotated as such. For instance, JCP2022_050797 (quinidine/quinine) has the targets KCNK1 and KCNN4. Other genes are annotated as targeting genes in disparate families.

Do JCP IDs within the compound dataset refer to the same compound?

Sometimes, two compounds were given separate JCP IDs because they had different names and broad_sample names. But after all of the data cleanup steps, they ended up being the same. Hence two different entries.

Do JCP IDs within either the CRISPR or ORF datasets refer to the same gene?

In CRISPR, each JCP ID corresponds to a different gene. But in ORF there are frequently multiple reagents representing the same gene. In this case, we compute consensus profiles at the gene level by aggregating profiles by Metadata_NCBI_Gene_ID rather than by Metadata_JCP2022. This approach was selected after testing six different consensus strategies and evaluating their performance using phenotypic activity metrics (private link | html).