Predicting base editing outcomes with an attention-based deep learning algorithm trained on high-throughput target library screens

Abstract

<jats:title>Abstract</jats:title><jats:p>Base editors are chimeric ribonucleoprotein complexes consisting of a DNA-targeting CRISPR-Cas module and a single-stranded DNA deaminase. They enable conversion of C•G into T•A base pairs and vice versa on genomic DNA. While base editors have vast potential as genome editing tools for basic research and gene therapy, their application has been hampered by a broad variation in editing efficiencies on different genomic loci. Here we perform an extensive analysis of adenine- and cytosine base editors on thousands of lentivirally integrated genetic sequences and establish BE-DICT, an attention-based deep learning algorithm capable of predicting base editing outcomes with high accuracy. BE-DICT is a versatile tool that in principle can be trained on any novel base editor variant, facilitating the application of base editing for research and therapy.</jats:p>

Genome dependent Cas9/gRNA search time underlies sequence dependent gRNA activity

“…5 for a similar analysis for E. coli datasets. i We applied the same prediction method to a dCas9 dataset from Horlbeck et al 2016 39 and base editing datasets from Marquart et al 2020 40 . Shown are the Pearson correlations between Activity and Predicted Activity.…”

“…While the datasets discussed above are of wild-type Cas9 or Cas9 variants with nuclease activity, we would expect to see similar sequence preference in nuclease null Cas9 variants, including deactivated Cas9 (dCas9) and base editors, if our hypothesis is correct. Using datasets from Horlbeck et al 2016 39 (dCas9) and Marquart et al 2020 40 (base editors) in human cells, we followed the same analysis of the PAM proximal 2 bp sequence that we did for other Cas9 datasets. We found that these datasets shared PAM proximal sequence preference with nuclease active Cas9 datasets in humans but had weaker or negative correlations in other species, despite not inducing double strand breaks (Fig.…”

“…Both dCas9 and base editing have additional requirements to achieve high activity, such as correct positioning relative to the gene being inhibited for dCas9 or containing the correct sequence in the editing window for base editors, compared to wild-type Cas9 39,40 . This would be expected to lead to generally lower activity that is independent of gRNA sequence.…”

In vivo adenine base editing of PCSK9 in macaques reduces LDL cholesterol levels

“…For the normalized absolute number and relative amount of six substitution types, the samples were classified based on the injected chemicals; for each group, the mean and standard deviation were calculated and plotted. To illustrate the potential TadA activity in the samples, the identified TadA motif 39 was used as TadA signature for cosine similarity comparison. The 96-trinuclueotide frequencies were pooled from the two signatures and normalized so that the frequencies add up to 1.…”

“…6b,c). In addition, cosine similarity analysis revealed that clones derived from AAV-or LNP-treated mice showed no enrichment in TadA trinucleotide signatures 39 compared to negative control clones (derived from untreated and CBE-treated mice) (Fig. 3c and Supplementary Fig.…”