Prediction of prime editing insertion efficiencies using sequence features and DNA repair determinants

Abstract

Most short sequences can be precisely written into a selected genomic target using prime editing; however, it remains unclear what factors govern insertion. We design a library of 3,604 sequences of various lengths and measure the frequency of their insertion into four genomic sites in three human cell lines, using different prime editor systems in varying DNA repair contexts. We find that length, nucleotide composition and secondary structure of the insertion sequence all affect insertion rates. We also discover that the 3′ flap nucleases TREX1 and TREX2 suppress the insertion of longer sequences. Combining the sequence and repair features into a machine learning model, we can predict relative frequency of insertions into a site with R = 0.70. Finally, we demonstrate how our accurate prediction and user-friendly software help choose codon variants of common fusion tags that insert at high efficiency, and provide a catalog of empirically determined insertion rates for over a hundred useful sequences.

Chromatin context-dependent regulation and epigenetic manipulation of prime editing

“…Practically, the use of a codon-and structure-optimized prime editing protein or a structurally enhanced pegRNA alone produces limited effects (3-4 fold) 11,13 . In the scenario of installing small protein tags, sequences with high insertion efficiencies predicted by a current machine learning model are only 1.63-fold better than those with low predicted efficiencies 15 . Finally, the strength of the MMR inhibition effects varies depending on the type of edit, with greater effects observed in cases of single-nucleotide substitutions and small indels 11 .…”

“…Fourth, current state-of-the-art technologies for improving prime editing efficiency mainly rely on enhancing the robustness of the prime editing core complex 10,11,[13][14][15] , machine learning-based prediction of target site efficiencies 10,14,15 , and transient suppression of the MMR pathway 11,12 . While these approaches have yielded encouraging results (especially when used in combination), there are notable limitations.…”

“…For the second factor, the impact of the target sequence and the edit sequence has been examined in the context of various kinds of edits (e.g. substitutions and indels), and corresponding machine learning models have been generated to facilitate design 10,14,15 . For the third factor, a rate-limiter for prime editing is DNA mismatch repair (MMR), which detects the intermediate product of prime editing and efficiently repairs the edited strand.…”

Predicting prime editing efficiency across diverse edit types and chromatin contexts with machine learning

“…Our team 1 and other researchers 2–4 have previously developed machine learning models trained on extensive prime editing datasets to predict pegRNA efficiencies. A shared limitation of these models is their specialization in predicting certain edit types: DeepPrime 4 is limited to 1-3 bp edits, MinsePIE 3 is confined to insertions, and PRIDICT 1 primarily focuses on 1 bp replacements as well as short insertions and deletions. Furthermore, each of these models does not account for the potential influence of the local chromatin state on editing rates.…”

Harnessing the power of artificial intelligence to advance cell therapy

“…CRISPR-Cas editing can vary in efficiency and specificity depending on gRNA sequence and Cas protein. The efficiency of prime editing is often low.Numerous research groups have used machine learning approaches and screens of gRNA sequences, Cas proteins, base editors, and DNA inserts to understand how various characteristics of these modules affect specificity and efficiency [129][130][131][132][133][134][135][136][137][138][139][140][141]. These efforts have produced algorithms to predict off-target effects of gRNAs, editing efficiency of different modular base editor designs at different genome loci, and catalogs of insertion rates for various DNA inserts.…”