Defining a Cancer Dependency Map: which genes each cancer cell line cannot live without
Genome-scale RNAi screens across 501 cancer cell lines, analysed with the DEMETER algorithm to remove off-target noise, identified 769 genes on which subsets of cancers depend and showed that most dependencies can be predicted from the cell's molecular features.
The Broad Institute's Project Achilles knocked down 17,098 genes in 501 cell lines spanning many cancer types using pooled shRNA libraries. A new computational method, DEMETER, separated on-target from seed-sequence off-target effects, a longstanding problem of RNAi screens.
The analysis identified 769 genes with strong differential dependency across lines; more than 90% of lines depended on at least one such gene, and 426 of the 769 dependencies could be predicted from mutation, copy number or expression features. Predictive markers were often not the gene itself but paralog loss, lineage or pathway activity, pointing to synthetic-lethal and lineage-specific targets.
The paper defined the goal of the Cancer Dependency Map (DepMap), which has since moved to genome-wide CRISPR screens (Meyers 2017, Behan 2019) in over 1,000 lines with matched multi-omics, and is the most-used resource for target discovery and biomarker hypothesis generation.
- 501 cell lines screened with 17,098-gene shRNA library; DEMETER algorithm removed seed-based off-target effects
- 769 genes with differential dependency; over 90% of lines depended on at least one
- 426 dependencies (55%) predictable from genomic or expression features, often via paralogs or lineage factors
- Dependencies on WRN in MSI-high lines and on paralogs (for example SMARCA2 in SMARCA4-mutant lines) emerged from these and follow-on screens
DepMap is the lookup table drug hunters use to ask: which cancers would die if we blocked this gene, and how would we recognise them? It generated targets such as WRN and PRMT5-MTAP now in clinical trials, and it is public.
- Cell lines lack microenvironment, immune context and drug pharmacology; many in vitro dependencies do not translate
- RNAi knockdown is partial; CRISPR knockout can give different results for essential genes
- Lineages and ancestries are unevenly represented; some cancers have few lines
- Dependency does not equal druggability; most dependencies are transcription factors or lineage genes
Pages like this
not linked directly; found by shared links- IdeaA drug screen that only rewards killing sleeping cancer cells
Shares DepMap (Cancer Dependency Map), Functional (ex vivo) drug testing, CRISPR functional genomics, Preclinical models that do not predict people.
- IdeaAttack the backup copy when a tumour has lost the original gene
Shares Synthetic lethality, Synthetic lethality approaches, Broad Institute of MIT and Harvard, CRISPR functional genomics.
- IdeaTurn chromosomal chaos into a weakness with KIF18A inhibitors
Shares Synthetic lethality, Synthetic lethality approaches, CRISPR functional genomics, The undruggable drivers.
- IdeaWRN inhibitors: a second synthetic-lethal win for mismatch-repair cancers
Shares Synthetic lethality, Synthetic lethality approaches, CRISPR functional genomics, The undruggable drivers.
- IdeaEvery drug screen includes standard reference compounds whose performance is published
Shares DepMap (Cancer Dependency Map), Functional (ex vivo) drug testing, CRISPR functional genomics, Preclinical models that do not predict people.
- PersonFrancisca Vazquez
Shares Synthetic lethality approaches, Broad Institute of MIT and Harvard, CRISPR functional genomics.
- IdeaA synthetic lethality map for every cancer driver in every tissue context
Shares Synthetic lethality, Synthetic lethality approaches, CRISPR functional genomics, The undruggable drivers.
- PathwaySynthetic lethality: paired dependencies
Shares WRN helicase (MSI-high cancers), PRMT5 (MTAP-deleted cancers), Synthetic lethality, Synthetic lethality approaches.