Dataset contents

This DepMap release contains data from CRISPR knockout screens from project Achilles, as well as genomic characterization data from the CCLE project.

Achilles data

This Achilles dataset contains the results of genome-scale CRISPR knockout screens for 17,634 genes in 563 cell lines. It was processed using the following steps:

Sum raw readcounts by replicate and guide
Remove the list of guides with suspected off-target activity
Remove guides with pDNA counts less than one millionth of the pDNA pool
Remove replicates that fail fingerprinting match to parent or derivative lines
Remove replicates with total reads less than 15 million
Remove replicates that do not have a Pearson coefficient > .7 with at least one other replicate for the line
Calculate log2-fold-change from pDNA counts for each replicate
Calculate the SSMD for each cell line using guides targeting the Hart reference essentials and non-essentials, and remove those with values more positive than -0.5. See Hart et al., Mol. Syst. Biol, 2014.
Calculate the SSMD for each cell line after averaging remaining replicates, and remove those more positive than -0.5
Run CERES to generate gene-level scores.
"Batch-correct" CERES data to remove known sources of bias.
Identify pan-dependent genes as those for whom 90% of cell lines rank the gene above a given dependency cutoff. The cutoff is determined from the central minimum in a histogram of gene ranks in their 90th percentile least dependent line.
For each CERES gene score, infer the probability that the score represents a true dependency or not. This is done using an EM step until convergence independently in each cell line. The dependent distribution is determined empirically from the scores of the pan-dependent genes. The null distribution is determined from unexpressed gene scores in those cell lines that have expression data available, and from the Hart non-essential gene list in the remainder.

The source for copy number data varies by cell line. Copy number data indicated as "Sanger WES" are based on the Sanger Institute whole exome sequencing data (COSMIC: http://cancer.sanger.ac.uk/cell_lines, EGA accession number: EGAD00001001039) reprocessed using CCLE pipelines. Copy number source was chosen according to the following logic: - Broad WES for lines where available - Broad SNP when Broad WES is not available and Sanger WES not available, or Sanger WES copy number has less correlation with logfold change than Broad SNP - Sanger WES in all other cases

CCLE data

Expression

CCLE expression data is quantified from RNAseq files using the GTEx pipelines. A detailed description of the pipelines and tool versions can be found here: https://github.com/broadinstitute/gtex-pipeline/blob/v9/TOPMed_RNAseq_pipeline.md. We provide a subset of the data files outputted from this pipeline available on FireCloud. These are aligned to hg38.

Copy number

The CN is called using Broad WES as top priority. If Broad WES is not yet available for a line, then Sanger WES data is used as long as the Sanger WES CN data correlates well with the Achilles log-fold change data. If the correlation is too low and the Sanger WES CN profile differs from the Broad Affy CN, then we used Broad Affy CN. If no WES data is available for a cell line, Broad Affy data is used. We plan to complete WES for all lines that do not have WES soon so we will phase out using Affy CN data. For the WES data, CN is called using a panel of normals that is female only. Therefore X CN is relative to 2 copies of X for WES data. We do not use the X chromosome CN data for cell lines where CN was derived from Affy arrays.

CCLE WES copy number data is generated by running the GATK copy number pipeline aligned to hg38. Tutorials and descriptions of this method can be found here https://software.broadinstitute.org/gatk/documentation/article?id=11682, https://software.broadinstitute.org/gatk/documentation/article?id=11683. WES samples have been realigned to hg38 and run through this pipeline.

Post processing of the segmented level data addressed gaps and the ends of the called chromosomal regions.

Mutations

CCLE mutation calls are aggregated from several different sources and sequencing technologies.

Currently, WES-based calls are generated on a quarterly basis using a cell line implementation of the CGA WES Characterization Pipeline (more info here: https://docs.google.com/document/d/1VO2kX_fgfUd0x3mBS9NjLUWGZu794WbTepBel3cBg08/edit). The cell line implementation of this pipeline has one key difference: because CCLE lines do not have a matched control sample, we use the same normal sample as a matched normal input for all samples using the same baits (Agilent vs. ICE). In addition, mutation calls still use hg19 alignment.

Quarterly WES-based mutation calls are added to the existing mutation calls from previous releases. WGS, Sanger WES, RNAseq, HC, and RD based calls from the 2nd phase of the CCLE project are also included. More details on how these mutations were called and filtered can be found in the manuscript “Next generation characterization of the Cancer Cell Line Encyclopedia” in Nature.

Fusions

CCLE generates RNAseq based fusion calls using the STAR-Fusion pipeline. A comprehensive overview of how the STAR-Fusion pipeline works can be found here: https://github.com/STAR-Fusion/STAR-Fusion/wiki. We run STAR-Fusion version 1.6.0 using the plug-n-play resources available in the STAR-Fusion docs for gencode v29. We run the fusion calling with default parameters except we add the --no_annotation_filter and --min_FFPM 0 arguments to prevent filtering.