Fig. 2

Empirically selected sgRNAs in the HD CRISPR library. a Distribution of log2 fold changes for sgRNAs targeting core essential reference genes. The blue curve represents sequences with strong on-target phenotypes. The red curve represents sgRNAs with weak on-target phenotypes (see “Methods”). b Phenotypes of sgRNAs targeting the core essential gene NOP2 for 4 screens performed with the library described in Wang et al. [27]. Compared to other NOP2-targeting sgRNAs, sgRNA 9 showed unexpectedly weak depletion in these experiments and was thus labeled “ineffective”. c Distribution of log2 fold changes for sgRNAs targeting nonessential reference genes. The blue curve represents sequences with low off-target activity. The red curve represents sgRNAs that led to unexpectedly strong phenotypes. d Phenotypes of sgRNAs targeting the nonessential gene KRT35 in 4 screens performed with the library described in Wang et al. [27]. Unlike other KRT35-targeting sgRNAs, sgRNA 8 consistently displayed toxic phenotypes in these experiments and was therefore marked as “toxic”. e Percentage of sgRNA sequences in sub-libraries A (left) and B (right) that could be selected based on empirical evidence from published screening experiments. Empirical essential sgRNAs were selected based on inducing a viability phenotype, empirical nonessential sgRNAs based on the absence of a toxic phenotype. f–h Calculated sequence scores applying either the rule set 2 [19], the DeepHF [20] or the Hart et al. [17] algorithms. Score performance of the HD CRISPR sub-libraries A and B was benchmarked against the libraries whose design is based on respective scores (Brunello for rule set 2, TKOv3 for Hart et al.) if available as well as the GeCKOv2 library and a random sample of sgRNAs from published libraries. The DeepHF score was used as an independent measure none of the investigated libraries was designed on. i–k Comparison of sgRNA scores for empirically and de novo-designed sgRNAs within the HD CRISPR sub-libraries