DeepGE: Unified web genome editing platform for CRISPR–Cas variants, base editors, and prime editors

DeepGEUnified web genome editing platform for CRISPR–Cas variants, base editors, and prime editorsNot sure what you need? Obtain results for all genome editor types.
DeepGE designs and predicts guide RNAs for CRISPR–Cas variants, base editors, and prime editors from one input.Which web server do you need?

Knockout: Large deletions, screens, donor insertion? DeepSpCas9variants is the default recommendation. DeepCas9variants, a mode of DeepBE, adds 6 Cas9 variants; DeepSmallCas9 covers 17 small Cas9 orthologs; DeepCpf1 serves CRISPR–Cas12a for A/T-rich regions. Recommended: DeepSpCas9variants, DeepCas9variants, DeepSmallCas9, DeepCpf1.

Precise edit: Substitution or indel, no donor? To substitute specific bases, use base editors (use the DeepBE web server) or prime editors (use the SynDesign or DeepPrime web servers); to insert or delete bases, use prime editors. CRISPR–Cas variants on their own are generally not effective for these tasks. Recommended: DeepBE, SynDesign, DeepPrime.

Saturation mutagenesis: Every variant of a gene? Prime editing is likely the most convenient option, providing all 12 single-base substitutions and small multi-base insertions/deletions wherever a PAM lies within reach. SynDesign provides extensive support for saturation genome editing (SGE). Recommended: SynDesign.

Size limit: Compact editor for an AAV vector? DeepSmallCas9 is highly recommended; you can design and predict 17 small Cas9 orthologs. Recommended: DeepSmallCas9.

Three kinds of genome editorsCRISPR–Cas variants. An RNA-guided nuclease that makes a double-strand break at a target defined by a protospacer adjacent motif (PAM). Disrupting genes, large deletions, pooled screens, and inserting sequences with a donor. Not for precise single-base changes without a template.Base editors. A Cas nickase fused to a deaminase rewrites a single base within a small editing window, with a single-strand nick, no double-strand break and no donor DNA. Installing or correcting point mutations, and gene disruption through stop-codon or splice-site edits. Bystander edits can occur inside the window.Prime editors. A Cas9 nickase paired with a reverse transcriptase, guided by a prime editing guide RNA (pegRNA) that both specifies the target and templates the edit. Arbitrary precise edits and saturation genome editing: all 12 single-base substitutions and small insertions and deletions, with no double-strand break or donor DNA. Efficiency is lower than that of base editors.What every design comes back withThis is what arrives in the results table whichever web server you pick: the same derived columns, computed the same way, on every web server that can support them. Every column is also in the downloaded file, so what you read on the page is what you keep.Cloning oligos (15 web servers — everywhere a guide is designed). A forward and reverse oligo pair, ready to order, with the overhangs of a named vector: SpCas9/lentiGuide-Puro (BbsI or BsmBI), Cas12a/pY108 or pU6-crRNA, Cas12a/pRDA_052 (Broad GPP), small-Cas9/Lenti-gRNA-Puro #84752, and a GG acceptor for pegRNA extensions. Every overhang stays editable under Show Advanced Cloning Vector Settings, because an overhang belongs to your vector rather than to the nuclease.HDR donor template (9 web servers, when the input carries an edit). A donor built around each guide's own cut site: 100 nt homology arms by default (or 50 nt, or 25–1000 nt of your choosing), sense, antisense or both strands, as an ssODN, a long single-stranded fragment or a double-stranded duplex, with optional silent PAM-blocking mutations so the nuclease cannot re-cut the repaired allele. Donors for one target are ranked.Off-target score (Mismatch-tolerance on 9 web servers; whole-genome search on 11). Two different questions, and the column headers say which. The mismatch-tolerance score needs no genome: it asks how much activity survives a single mismatch, scored on a published substitution table, and a higher number is the worse guide. The genome-wide specificity score comes from a real Cas-OFFinder pass over GRCh38 or GRCm39 and runs the other way — higher is better. On 7 web servers you may aggregate that search with the CFD table, the MIT/Hsu position weights, or both side by side. DeepSniper and DeepSmallCas9 score searched sites with their nuclease's own published off-target model instead.Position figure (13 web servers). Every candidate drawn where it sits in the sequence you submitted, above the table. Click a guide to filter the table to it; filter the table and the figure redraws to match. A results table states a position only as text like 59:132 — this is where that becomes a place.Cut site and location (Wherever a guide can be placed in your input). The cut position, 1-based in your own sequence, and the guide's span. For an input that names a locus — a gene, a variant, a coordinate — the genomic position is given as well.epegRNA motifs (The prime-editing web servers). A 3' structured motif appended to the pegRNA extension, from a chosen panel, with its own orderable oligo pair. It is never the default, because an epegRNA is a different construct to order and to publish than the pegRNA beside it.Which of these a particular web server offers depends on what its model can support: a prime editor installs its own edit, so it is offered no HDR donor.Citing DeepGECiting DeepGE in your publication sustains the maintenance and improvement of this web server and keeps the infrastructure available to everyone.

Available in the near future, please stay tuned.

Moreover, please cite the specific paper(s) for the tool(s) you used. Citations are shown on each web server page.

BEvariants – Evaluations of functional effects of cancer-related transition mutations using base editors and single guide RNA (sgRNA) libraries, published in Kim et al. Nature Biotechnology (2022).

CINDEL – Prediction of CRISPR–Cas12a/Cpf1 (AsCpf1) guide activity from sequence context, published in Kim et al. Nature Methods (2017).

DeepBE – Deep learning-based base editor and Cas9 variant guide RNA design and efficiency prediction tool, published in Kim et al. Nature Biotechnology (2024).

DeepBaseEditor – Deep learning-based guide RNA design and prediction of Adenine Base Editor (ABE) and Cytosine Base Editor (CBE) activities, published in Song et al. Nature Biotechnology (2020).

DeepCpf1 – Deep learning-based CRISPR–Cas12a (AsCas12a, Cpf1, or AsCpf1) guide RNA design and efficiency prediction, published in Kim et al. Nature Biotechnology (2018).

DeepGE – Comprehensive web server for predicting and designing guide RNAs for CRISPR–Cas9 Variants, Base Editors, and Prime Editors.

DeepPE – Deep learning-based prime editing guide RNA design and efficiency prediction, published in Kim et al. Nature Biotechnology (2021).

DeepPrime – Deep learning-based model predicting prime editing efficiencies for all possible pegRNAs of a given target sequence, published in Yu et al. Cell (2023).

DeepSmallCas9 – Deep learning-based guide RNA design and efficiency prediction for 17 small Cas9 orthologs and variants, published in Seo et al. Nature Methods (2023).

DeepSniper – Deep learning-based guide RNA design and efficiency prediction for Sniper2L and Sniper1 (Sniper-Cas9), published in Kim et al. Nature Chemical Biology (2023).

DeepSpCas9 – Deep learning-based CRISPR–Cas9 (SpCas9) guide RNA design and efficiency prediction tool with conventional scaffold, published in Kim et al. Science Advances (2019).

DeepSpCas9-NG – Deep learning-based guide RNA design and efficiency prediction of SpCas9-NG efficiency with conventional scaffold, published in Kim et al. Nature Biomedical Engineering (2020).

DeepSpCas9variants – Deep learning-based CRISPR–Cas9 (SpCas9) and variant guide RNA design and efficiency prediction with optimized scaffold, published in Kim et al. Nature Biotechnology (2020).

DeepTrunc-cell – DeepTrunc-cell predicts cellular CRISPR/Cas indel frequencies of matched, mismatched, and bulge-mutated target–sgRNA pairs.

DeepxCas9 – Deep learning-based guide RNA design and efficiency prediction of xCas9 efficiency with conventional scaffold, published in Kim et al. Nature Biomedical Engineering (2020).

EGFRvariants – Resistance profiling of EGFR kinase domain variants in EGFR-mutant NSCLC cells, screened by prime editing, published in Kim et al. Nature Biotechnology (2025) and Lee et al. npj Precision Oncology (2026).

PEselector – Machine learning-based tool for selecting and predicting hyPE2 prime editors from Target, Reverse transcription template (RTT), and Primer binding site (PBS) sequences, published in Song et al. Nature Communications (2021).

SynDesign – Web-based prime editing guide RNA design and evaluation tool with saturation genome editing (SGE) capabilities, published in Park et al. Nucleic Acids Research (2024).

Your browser does not support ES modules.
Please use a modern browser.