Fig. 6: Experimental validation of DeepCas13 performance.
From: Modeling CRISPR-Cas13d on-target and off-target effects using machine learning approaches
a A schematic view of the validation screen experiment. The library targets essential protein-coding genes, lncRNAs, and circRNAs. b LFC distribution of top 10% predicted sgRNAs by each tool. n = 632, 246, and 71 guides targeting mRNA, lncRNA and circRNA respectively. The two-sided independent t-test is used for analysis. c LFC distribution of high and low predicted sgRNAs for individual coding gene. H means sgRNA with high predicted score. n = 566, 557, 580, 589, 379, and 579 guides in HDeepCas13, HRFNBT, LDeepCas13, LRFNBT, Hboth, Lboth group respectively. The two-sided independent t-test is used for analysis. d Spearman correlation between LFC and predicted scores. Error band shows the 95% confidence interval for the regression estimate. The p-value is two-sided and calculated by a test of the null hypothesis that the distributions underlying the samples are uncorrelated and normally distributed. e qRT-PCR validation of the top predicted sgRNA from the screen experiment. crNT: non-targeting crRNA. Error bar shows the 95% Confidence Interval (CI). n = 3 independent qPCR experiments. Error bar shows the 95% confidence interval. The two-sided independent t-test is used for analysis. f “Head-to-head” qRT-PCR validation of sgRNAs that target additional coding genes and non-coding RNAs. Error bar shows the 95% CI. n = 3 independent qRT-PCR experiments. Error bar shows the 95% confidence interval. The two-sided independent t-test is used for analysis. The top, mid-line and bottom of the boxplot (b, c) represents the upper quartile (Q3), median, and lower quartile (Q1), respectively. The ends of the whiskers represent the minimum and maximum values in the data set.
