our new @CancerDepMap @DepMapSanger collaboration: data harmonisation and methods benchmarking yielding integrated inter-study datasets of cancer dependencies in ~800 cell lines - led by @clarepacini w @CancerDataSci @emanuelvgo @Garnettlab @targetvalidate biorxiv.org/content/10.110… 
1/8 Last year we showed that @broadinstitute and @sangerinstitute independent #CRISPRcas9 cancer dependency datasets are concordant and they yield consistent core-fitness/context-specific essential genes and biomarkers, opening to their integration
nature.com/articles/s4146…
nature.com/articles/s4146…
2/8 we have now consolidated a data-harmonisation/method-benchmarking strategy yielding integrated datasets of ~800 cell lines, expanding the possibility of performing tissue and cancer type specific analyses and new targets' identification to a much larger set of diseases
3/8 we have assessed and selected the best batch correction pipeline based on the ability to match dependency profiles of cell lines screened in both original studies, yet preserving heterogeneity of gene essentiality across tissues and gene expression based addiction biomarkers
4/8 we have also compared methods for correcting CNA biases, accounting for heterogenous sgRNA on-target efficiency, based on post-processed TPRs of known essential genes (and their fitness effect strengths), and preservation of cell-line/context specific essential genes
5/8 another comparison criteria was assessing the extent to which each processing method preserves correlations between a drug response profile across cell lines and the essentiality profile of the gene coding for the drug target
6/8 Our final CERES/CRISPRcleanR Broad/Sanger integrated cancer dependency datasets yield more reliable sets of core-fitness genes (regardless the method used to compute them) and better FDRs compared to the two individual datasets
7/8 The Broad/Sanger integrated cancer dependency dataset also allows detecting more biomarkers of gene essentiality and with increased statistical power, when compared with the two individual datasets
8/8 Full story at depmap.org/broad-sanger/; integrated datasets available at score.depmap.sanger.ac.uk/downloads
