Fig. 8

A The process and results of high-throughput quantification of gRNA efficiency in cells. In panel (a), a graphic illustrates the sequence of actions involved, which includes employing a lentiviral surrogate vector, synthesizing an oligo pool, performing PCR amplification, using golden-gate assembly, packing the genetic material into lentiviruses, and then introducing it. Panel (b) showcases the editing efficiency of gRNA at all surrogate locations, assessed through targeted amplicon sequencing. The results are presented for HEK293T-SpCas9 cells at 2, 8, and 10 days following the introduction. Panel (c) displays the correlation between gRNA editing efficiency on days 8 and 10 post-transduction. Panel (d) presents the patterns of indels (deletions ranging from 1–30 bp and insertions ranging from 1–10 bp) introduced by SpCas9 in HEK293T-SpCas9 cells at 2, 8, and 10 days after the transduction. Panel (e) depicts the agreement between the observed indel patterns in cells and those predicted by inDelphi, visualized as a violin plot with medians and quartiles. In panel (f), a scatter plot portrays the frequency of 1-bp insertion indels (mean ± 95% confidence interval), categorized based on the nucleotide at position N17 of the protospacer and the type of inserted nucleotide. Lastly, panel (g) exhibits the association between gRNA editing efficiencies in this study and those from other significant research, with a particular emphasis on common gRNA + PAM (23 nt) cases, presented using a Venn diagram. B The CRISPR on model and its ability to generalize on independent test sets. Panel a displays a visual depiction of the input DNA sequence for CRISPRon, including the prediction algorithm. The deep learning network receives inputs in the form of a one-hot encoded 30mer and the binding energy (ΔGB). It's worth noting that only the filtering (convolutional) layers and the three fully connected layers are explicitly depicted, with the thin vertical bars representing the output of one layer, serving as the input for the next layer. In panel b, a performance evaluation comparing CRISPRon to other existing models is presented, specifically focusing on independent test sets containing over 1000 gRNAs. Reprinted from [223] with permission from Springer Nature