Product Documentation

Each dataset is stored as a RO-Crate package. Datasets contain the data for a gene set measured in a specific cell line under a specific treatment. (Including “untreated”)

The RO-Crate format

Research Object Crate (RO-Crate) is a lightweight standard for packaging research data with their metadata. It contains schema.org and other standard structured vocabulary annotations in JSON-LD, of all included or referenced data objects, and aims to make best-practice in formal metadata description accessible and practical for use in a wide variety of situations. Each RO-Crate is a folder that contains, along with one or more data files, a ro-crate-metadata.json file, containing structured information about the data (metadata) in JSON-LD format. It is good practice to also include a readme.md text file describing the contents.

Image Datasets

Each RO-Crate contains:

• A cell line – antibody – treatment table describing the experiment
• Four channel data folders: red, green, blue yellow
  • Each folder contains images in .jpeg format.
• ro-crate-metadata.json
  • Structured metadata for the dataset
  • README

Example README file:

This data set has been generated in the Emma Lundberg lab (https://ell-core.stanford.edu/) in July-August 2023 as part of the CM4AI Research Program (https://digitaltumors.org/).

This data set displays the spatial localization of 95 proteins in the breast cancer cell line MDA-MB-468. MDA-MB-468 cells were seeded into 96-well glass bottom plates and treated either with the chemotherapeutic drug Paclitaxel or with the chemotherapeutic drug Vorinostat (SAHA) or not treated. For drug treatment, higher cell numbers were seeded than for the cells that were not treated to reach a sufficient amount of remaining cells at the end of the experiment (treatment with Paclitaxel or Vorinostat disturbs cell growth). After fixation of the cells with PFA, for each protein, the subcellular distribution of the protein was investigated by immunofluorescence-based staining (ICC-IF) and confocal microscopy. Each well was stained with DAPI (to label nuclei, “blue” channel), a Calreticulin antibody (to label the endoplasmic reticulum, “yellow” channel), a Tubulin antibody (to label Microtubuli, “red” channel), and an antibody against a protein of interest (a different antibody / protein for each well, “green” channel). In the green channel, for each well, one of 95 proteins was stained with an antibody from the proteinatlas.org project. 

The .tsv file describes each image in the data set. Each row represents one image. The columns describe the staining from which the image was taken:

“Antibody ID” describes the antibody ID for the antibody applied to stain the protein visible in the “green” channel. The antibody ID can be looked up at proteinatlas.org to find out more information about the antibody.

“ENSEMBL ID” indicates the ENSEMBL ID(s) of the gene(s) of the proteins visualized in the “green” channel.

Treatment refers to how the cells that are depicted in the image were treated (with Paclitaxel, Vorinostat, or untreated)

“Well” refers to the well coordinate on the 96-well plate

“Region” is a unique identifier for the position in the well, where the cells were acquired.

 

AP-MS Datasets

Each RO-Crate contains three files:

• readme.md
  • Documentation for the dataset
• apms.tsv
  • Processed AP-MS data as a table of bait-prey interactions
• ro-crate-metadata.json
  • Structured metadata for the dataset

Columns of the apms.tsv interaction data

• Bait: Name of the pull downed protein
• Prey: Uniprot ID number of identified proteins by MS in pull down (putative bait interactor).
• PreyGene.x: Uniprot protein name of identified protein by MS in pull down (putative bait interactor).
• Spec: Number of spectral count in each test biological replicates (separated by | ).
• SpecSum: Sum of Spectral counts in test samples.
• AvgSpec: Average Spectral counts across replicates in test samples.
• NumReplicates.x: Number of replicates in test samples.
• ctrlCounts: Number of spectral count in each control replicates (separated by | ).
• AvgP.x: Average probability that an interaction is true, measure of the likelihood that a given interaction is a true positive rather than a random or non-specific interaction. A lower AvgP indicates higher confidence in the interaction being genuine.
• MaxP.x: maximum probability associated with a protein interaction in the context of its prey-bait pair. Similar to AvgP, a lower MaxP suggests a higher likelihood of the interaction being true.
• TopoAvgP.x: extension of the AvgP score that also takes into consideration the topology of the interaction network. It incorporates information about the hierarchical structure of the interaction data to provide a refined assessment of the interactions.
• TopoMaxP.x: topology-aware score that considers the maximum probability of an interaction in the context of the interaction network’s topology.
• SaintScore.x: composite score that integrates multiple aspects of the interaction data, including spectral counts and probability estimates. It’s designed to prioritize interactions based on their strength and reliability. Higher SaintScores indicate interactions that are more likely to be true.
• logOddsScore: Logarithm of the odds ratio between test and control conditions for each prey as a measure of interaction significance. The LogOddsScore is a statistical score that represents the logarithm of the odds ratio for a protein-protein interaction. It’s used to quantify the strength and significance of the association between two proteins in an interaction network. The odds ratio compares the likelihood of the interaction occurring to the likelihood of it not occurring. Taking the logarithm of the odds ratio often helps to transform the score into a more symmetric and interpretable form, making it easier to compare and analyze the interactions. Higher LogOddsScores typically indicate stronger evidence for the interaction.
• FoldChange.x: represents the ratio of the abundance of a protein or interaction in one experimental condition (Test) compared to another (control). It helps assess whether the abundance of a protein changes significantly between different conditions.
• BFDR.x: Bayesian False Discovery Rate

 

CRISPR perturbation scRNA-Seq Datasets

Each RO-Crate contains:

h5ad file ———————————————————– This file is a .h5ad file. Details about the data structure of this format can be found here: https://anndata.readthedocs.io/en/latest/. This data is a cell x gene matrix, with all pre-procesing steps done in accordance to single cell best practices guide https://www.sc-best-practices.org/conditions/perturbation_modeling.html#analysing-single-pooled-crispr-screens) up to section 19.4.5. The .X layer is set to the ‘X_pert’ output of the mixscape pipeline.