The Load Example button is able to replicate the example inputs shown in this help page for quick testing of the DeepBE web server. Simply click the Load Example button then press Submit Job to obtain an example results page.
In the mode that scores one base editor against one single guide RNA (sgRNA) — Single base editor and sgRNA mode — Load Example fills the target sequence and the nickase; press ENTER in the target sequence box to generate the candidate guides, then pick one. The candidates a nickase can cut differ between the nine, so they are generated from the nickase you have selected rather than stored in advance.
The user may select Single base editor and sgRNA mode, which generates results for a single base editor and sgRNA, or Comparison mode, which comprehensively generates results for all base editors in base editor categories or all Cas9 variants.
Users may select the specific deamination domain and Cas9 nickase domain.
Users should enter the target sequence under Enter target sequence (length >= 30) and press ENTER to generate the guide sequence.
Then, users should select the guide RNA sequence from the dropdown menu Select guide sequence (press ENTER above to generate) and submit the job.
Users are able to view the Base edited outcomes (%) and the guide RNA sequence. Result download is available below.
Adenine Base Editor (ABE): A → G or T → C
Cytosine Base Editor (CBE): C → T or G → A
C-to-G Base Editor (CGBE): C → G or G → C
All Edits: Runs all base editor models (ABE + CBE + CGBE) in a single job.
Comparison mode is split into two parts; selecting the base editor model or Cas9 variant model, and supplying the input sequence(s) in any of the nine input formats.
After selecting the base editor or Cas9 variant model, users should choose an Input Format and fill in the field it shows. Nine formats are accepted, and each one resolves to the target sequences the models score: FASTA, Gene Symbol, NMID, Ensembl ID, HGNC ID, Genomic Position, Edit Sequence (an unedited/edited pair), ClinVar ID and COSMIC ID. The dropdown belongs to comparison mode; single base editor and sgRNA mode takes one target sequence directly.
For FASTA, you can select Enter FASTA Text (sequence length >= 30) or Upload a FASTA file (.txt, .fa, .fasta, sequence length >= 30).
When uploading files, press the [+] button to add files. Uploaded files can be cleared by pressing either of the checkmark buttons.
[+]
Result downloads are available at the bottom of all results pages.
These three images were captured with DeepBE-ABE selected rather than the default All Edits, so that the editor columns stay readable at this size. Leaving the default in place gives the same table with the CBE and CGBE editors beside these, roughly three times as wide.
Users are able to select the row of the results for details of the target sequences.
One additional visualization is the capability to press any editing-efficiency score, which will visualize the single base editor and sgRNA mode result(s) for the specific target sequence. Colour ranks a score against the rest of its own column in this table -- green for the top quarter, red for the bottom -- so it says how a candidate compares with its neighbours rather than whether it can be opened.
A gene identifier is looked up in Human (GENCODE v50 on GRCh38) or Mouse (GENCODE M39 on GRCm39). The selector appears with the four gene formats, and it is offered wherever the genome-wide off-target search offers Mouse, so a run that screens against the mouse genome can design in mouse as well.
A mouse symbol from an older nomenclature resolves to the current one and the job log states the substitution — Olfr1 designs against Or1e16. Whichever of the four gene formats is chosen, Mouse replaces the identifier box with Enter Mouse Gene Name or ID, which takes a mouse symbol or a mouse Ensembl gene ID. ClinVar ID and COSMIC ID carry no species selector: both are human variant databases, and both resolve against the human reference.
Olfr1
Or1e16
Cloning Vector decides whether the results carry an ordering-ready oligo pair, and it is set to SpCas9 / lentiGuide-Puro (BbsI or BsmBI) by default. Every candidate then gains Oligo (forward) and Oligo (reverse), in the results table and in the downloadable files alike: the two strands that anneal into the digested vector, each presenting its own 4 nt 5' overhang — CACC on the forward strand, AAAC on the reverse. The guide between them is written (G/g)N19, position 1 replaced by the G the U6 promoter transcribes from.
Oligo (forward)
Oligo (reverse)
CACC
AAAC
Setting the vector to None (No Oligo Output) leaves both columns out rather than blank. The four overhang fields — Forward 5′, Reverse 5′, Forward 3′ and Reverse 3′ — override what the vector supplies; left blank, each takes the vector's own. A guide carrying anything outside A, C, G and T gives an empty pair, because there is nothing orderable to write.
The Off-target score (higher = less specific) column is a mismatch-tolerance score, not a count of sites in a genome: for this guide, how much activity would survive a single mismatch, averaged over every single-mismatch variant of the protospacer. A guide that tolerates mismatches well is more likely to engage a site it was not aimed at — and an engaged base editor deaminates there — which is why a higher number is the worse one.
It comes from the SpCas9 cutting frequency determination (CFD) table of Doench, Fusi et al. (2016). The table is indexed over the 20 nt protospacer and takes no PAM, so it describes the RNA-DNA duplex readout every nickase offered here shares; what separates them sits in the PAM-interacting domain, which the table does not read. It needs no genome, and it cannot tell you where an off-target is.
For that, set Genome-wide off-target search to Human or Mouse. That search adds one Genome-wide specificity (higher = more specific) column, which runs the other way round: there, higher is better. Every site it finds is condensed into that single number; the individual loci are not listed, so no chromosome, coordinate or strand for them reaches the table. The sites that search finds are scored on the same table, so the two columns differ only in the question they answer -- one guide against every single-mismatch variant of itself, one guide against the sites a genome really holds -- and never in how a mismatch is weighed.
Genome-search scoring is where that last clause stops holding. The CFD table knows which base was swapped for which, and was fitted on SpCas9 with an NGG PAM; the MIT position weights (Hsu et al. 2013) know only how tolerant each position is, and were fitted across nucleases rather than on one. Neither is a superset of the other. Leaving the setting on CFD only is the behaviour described above; choosing CFD and MIT adds a second Genome-wide specificity, MIT (higher = more specific) column, aggregated from the same search over the same sites, so the two readings can be compared directly on the guides you are choosing between. The choice applies to the genome search alone -- averaged over a complete single-mismatch panel the MIT weights return one number for every guide of a given length, which is a constant rather than a measurement, so the genome-free column is always the table's.
The table this web server returns is the one file it writes, so what the grid shows and what a download contains are the same columns.
Both modes report, per candidate: ID, Location, Input Sequence (30bp), Target Sequence (30bp), Guide RNA Sequence (20bp), GC Contents (%), and the derived columns described above, among them Off-target score (higher = less specific) and the Oligo (forward) / Oligo (reverse) pair. Cut site (1-based in the target sequence) is the position three bases from the PAM, which is where the Cas9 domain each editor is built on acts. A base editor carries a nickase rather than a nuclease, so it nicks there rather than cutting both strands, and the base it changes is inside the deamination window rather than at that position; the column is there so a candidate can be placed against the sequence it came from. Single base editor and sgRNA mode also adds a PAM class column: primary where the row's PAM is the selected nickase's published PAM, secondary where the row belongs to a wider PAM class the model was trained to score but the nickase is not documented to cut. That column belongs to the seven nickases whose published PAM is narrower than the set searched here; SpRY and Sc++ carry a single PAM list, so their rows have one class and nothing to mark. Clicking a row opens the detailed view for that target.
ID
Location
Input Sequence (30bp)
Target Sequence (30bp)
Guide RNA Sequence (20bp)
GC Contents (%)
Off-target score (higher = less specific)
Cut site (1-based in the target sequence)
PAM class
primary
secondary
Single base editor and sgRNA mode adds Outcome Sequence (30bp) and one efficiency column named for the editor you chose, since one editor is being asked about. That file is DeepBE_Single_Result.txt.
Outcome Sequence (30bp)
DeepBE_Single_Result.txt
Comparison mode on a base editor category replaces that single column with one efficiency column per editor for every nickase compared — the editors of the category chosen, all seven of them under All Edits, across the nine nickases — so the header is three rows deep: the model (DeepBE or DeepNG-BE), the nickase under it, and the editor name under that. Read a score by following its column up through both. The whole comparison is DeepBE_Results.txt, and each nickase-editor pair is also written on its own as <model>_<nickase>_<editor>_Results.txt for anyone who wants one of them without the rest. Choosing DeepCas9variants instead scores the nine variants themselves, one column each, and writes DeepCas9variants_Results.txt. All of them are in the results ZIP.
DeepBE
DeepNG-BE
DeepBE_Results.txt
<model>_<nickase>_<editor>_Results.txt
DeepCas9variants_Results.txt
A column belongs to the table rather than to a nickase once the header's own grouping ends, which is where the derived columns sit — they describe the guide, not one editor's opinion of it.
The computational models shown here (i.e., DeepBE and DeepCas9variants) were generated based on the measured activities of these base editors and Cas9 variants when HEK293T cells were transduced with a lentiviral vector that expresses a base editor or Cas9 variant at a multiplicity of infection of 0.15, so that almost every transduced cell had only one copy of the Cas9 or base editor variant-encoding sequence. Western blotting showed that the levels of most, but not all, Cas9 and base editor variant proteins were comparable.
Western blotting showed that the levels of most Cas9 and base editor variant proteins were comparable except that NG-ABE8e(V106W) showed statistically significant higher protein levels than three YE1-BE4max variants (NG-YE1-BE4max, SpRY-YE1-BE4max, NRCH- YE1-BE4max) and two APOBEC-nCas9-Ung variants (NG-APOBEC-nCas9-Ung, NRCH-APOBEC-nCas9-Ung) as shown below.
Western blot analysis of Cas9 protein levels in cells expressing Cas9 variants, base editor variants with different deaminases, and base editor variants containing Cas9 variants. The results from three independent western blot analyses are presented (n = 3). Error bars show mean ± s.e.m. Subsets of variants without statistically significant differences (P > 0.05, analysis of variance (ANOVA) followed by Tukey’s post hoc test) in their protein levels are indicated by letters, such as a and b.
The following example interactive outputs are available for reference:
If you have any questions, bug reports, or suggestions, please do not hesitate to contact us through the Support page.
