Fig. 1: Genome-wide CRISPR screens identify IAV host dependency factors.
From: Genome-wide CRISPR screen identifies host dependency factors for influenza A virus infection

a Schematics of the genome-wide CRISPR/Cas9 screening strategy. b, c Spatial distribution of CRISPR knockout signals on the genome for the primary and secondary screens. b Gene-level p-values in primary screen. A selection of top hits is highlighted. c Gene-level FDRs for top hits in the secondary screen. The core set of 121 robust hits (FDR < 0.05) is shown highlighted in red. d Representation of shared information content between each data source after MAIC analysis. Size of data source blocks is proportional to the summed information content (MAIC scores) of input list. Lines are colored according to the dominant data source. Data source categories share the same color; the largest categories and data sources are labeled (see supplementary information for full source data). CRISPR screen performed in this study is labeled as CRISPR (viral production). e Example of validation of MAIC screen using synthetic data (see methods). Y-axes show overlap ratio compared to a known truth (see methods; narrow line = average of 100 replicates; shading = 95% confidence interval). MAIC consistently out-performs existing methods (robust rank aggregation, RRA; vote counting; VC) when presented realistic inputs. Example shown is for a mixture of ranked and unranked data sources, and moderate variation in data quality (heterogeneity) between data sources. See supplementary Figs. 2 for full evaluation. f Experimental validation of MAIC (computed without CRISPR data) against an unseen gold standard (CRISPR screen). Plot shows number of overlaps with the top 1000 hits in CRISPR screen for a given position in each ranked dataset. g Contribution of different data categories to predictive value of the MAIC. As in (f), graph shows number of overlaps with the top 1000 CRISPR screen hits. Results are shown with all RNAi data from previous studies removed from the input set, and with all proteomic or protein interaction data removed. The results of the systematic meta-analysis by Tripathi et al.36 are shown for comparison.