guntingAI
Terminology, cutoffs, scores, and result fields
Reference guide
Understand every value shown in guntingAI
Search parameters, biological terminology, model-distance rules, guide-design scores, Multiplex values, annotations, coordinate-map symbols, account limits, and result fields.
Browse all definitions or enter a term.
No matching help entry
Try a shorter term or search for the exact field shown in the result table.
Biological and coordinate foundations
Core concepts and abbreviations
Spacer
The guide-recognition DNA sequence submitted to guntingAI. Current reference models use strict 20-nucleotide spacers.
Also: guide spacer, query spacerMatched spacer
The 20-base fragment retrieved from the selected model and compared with the query spacer. Only a row explicitly tagged Match is highlighted blue across the entire row and shown as a blue 5′→3′ arrow in the coordinate viewer.
Result field: fragment_seqPAM
Protospacer-adjacent motif. A short sequence beside the spacer that is required by the selected Cas system. The observed PAM is drawn on the strand-correct side of the spacer arrow.
Examples: NGG, NAGCas / SpCas9
CRISPR-associated nuclease. SpCas9 refers to Cas9 from Streptococcus pyogenes, commonly associated with an NGG PAM pattern.
Cas = CRISPR-associated proteinIUPAC PAM symbols
Ambiguity symbols accepted in PAM patterns: R=AG, Y=CT, S=GC, W=AT, K=GT, M=AC, B=CGT, D=AGT, H=ACT, V=ACG, and N=ACGT.
N means any DNA base5′ and 3′
Chemical directionality of a nucleic-acid strand. The blue arrow points in the displayed spacer’s 5′→3′ direction.
Read: five-prime and three-primeStrand: + or −
The orientation of the matched fragment relative to the stored reference sequence. A minus-strand arrow points toward decreasing displayed genomic coordinates.
Also: forward and reverse orientationReverse complement
The opposite-strand representation obtained by reversing a DNA sequence and replacing A↔T and C↔G.
GC fraction
The fraction of spacer bases that are G or C. A value of 0.50 means 50% GC. Design and Multiplex defaults accept 0.35–0.75.
Result field: gc_fractionHomopolymer
A run of the same base. The current Multiplex sequence check flags runs at or above the configured homopolymer length, commonly four repeated bases.
Example: AAAAHamming distance / mismatches
The number of base positions that differ between equal-length sequences. Zero means an exact spacer match.
Result field: hamming_distanceLocus
A genomic coordinate location. The same spacer may appear at more than one locus.
Nearest-fragment retrieval
Search parameters and result values
Species model / selected model
The reference or user-owned private model searched for nearest fragments. A private model is visible only to its owner unless publication is approved.
Parameter: model_keyTop matches
Maximum number of ranked rows returned for each query spacer after enabled filters are applied.
Parameter: top_kModel distance
Distance between the query and a fragment in the model’s learned PCA representation. Lower means closer in that representation. It is not a cleavage probability, biological-risk probability, or percent identity.
Result field: pca_distanceModel distance cutoff
A strict hard filter. With the default value 0.75, only rows with
model distance < 0.75 are returned. A row equal to
0.75 is excluded.
Maximum mismatches
Largest allowed Hamming distance for a positive Search call. The default is three mismatched positions.
Parameter: max_hammingRequire PAM match
When enabled, a retrieved row must contain a PAM compatible with the submitted pattern to be called positive.
Parameter: require_pamShow nearest candidates
Includes nearest retrieved rows even when they fail the positive mismatch or PAM rules. The model-distance hard ceiling still applies.
Parameter: include_nearest_even_if_not_positiveSearch depth
Multiplier controlling how many nearest candidates are initially retrieved from each model partition before ranking and trimming. Higher values can improve recall but increase work.
Parameter: search_multiplierModel partition limit
Optional limit on how many stored model partitions are searched. Leaving it blank searches all available partitions.
Parameter: max_shardsModel version
Optional specific checkpoint version. Leaving it blank uses the model’s configured default version.
Parameter: checkpoint_versionPositive / Match
A row that passes enabled PAM and mismatch rules after the strict model-distance filter. A positive call can still contain mismatches when the allowed mismatch value is greater than zero.
Result field: positive_callExact spacer match
Query spacer and matched spacer are identical at every position. This is stricter than a positive call.
Internal/result field: exact_spacer_matchPAM match
Whether the observed PAM satisfies at least one submitted IUPAC PAM pattern.
Result field: pam_passRank
Position of a row within one query spacer’s returned results. Positive and exact rows are prioritized before mismatch, PAM, and model-distance tie-breaking.
Query spacer versus matched spacer
Query spacer is the user-supplied sequence. Matched spacer is the retrieved model fragment. Their positional differences determine the mismatch count.
Model partition
A stored chromosome, contig, or bounded shard of a larger model. Partitioning limits memory requirements and enables resumable reference construction.
Result field: shard_keyCandidate generation and screening
Design parameters and scores
Target type
How the target is interpreted: gene name, gene ID, genomic interval, or pasted DNA sequence.
Parameter: target_typeTarget value
The gene, identifier, coordinate interval, or DNA sequence used to generate candidate spacers.
Parameter: target_valuePreferred region
Optional annotation filter such as CDS or gene. “Any” permits all matching annotated feature types.
Parameter: design_regionReturn top candidates
Number of final ranked Design rows returned after candidate generation and nearest-hit screening.
Parameter: top_nCandidates to screen
Maximum deduplicated candidate spacers sent through the approximate target/off-target Search stage. Increasing it expands screening but requires more computation.
Parameter: max_candidates_to_screenMinimum and maximum GC
Accepted GC-fraction range. Defaults are 0.35 and 0.75. A candidate outside the range loses Design preference and affects risk classification.
Parameters: min_gc, max_gcOff-target mismatches
Largest mismatch count considered when counting nearby target/off-target candidates during Design screening.
Parameter: max_hamming_offtargetDesign score
A ranking heuristic. The initial score rewards GC near 50%, being inside the accepted GC range, and absence of a homopolymer. The screening stage subtracts 10 points per nearby hit up to a 50-point penalty and 15 points per additional exact target. Higher is preferred; the score is not a calibrated experimental probability and can fall below zero.
Result field: design_scoreDesign risk
Low: no nearby hits, at most one exact hit, acceptable GC, and no homopolymer. Moderate: at most three nearby hits with acceptable GC. Other screened candidates are marked high.
Result field: risk_levelExact targets in screen
Number of retrieved PAM-compatible exact spacer targets in the approximate screening set.
Result field: exact_target_count_in_screenNearby targets in screen
PAM-compatible retrieved loci within the allowed mismatch cutoff, excluding the candidate’s source fragment.
Result field: near_target_count_in_screenObserved PAM
PAM stored beside the generated candidate at its source coordinate.
Result field: observed_pamGuide-panel compatibility
Multiplex parameters, cutoffs, and scores
Panel size
Number of guides included in a recommended combination. Leaving it blank evaluates the full submitted set as the main panel.
Parameter/result field: panel_sizeTop panel combinations
Maximum number of ranked guide combinations returned.
Parameter: top_combinationsHits per guide
Number of nearest Search rows retained for each guide when estimating exact and nearby target burden.
Parameter: top_k_per_guideGuide score
A 0–100 heuristic composed of: exact-target recovery up to 35 points, uniqueness up to 15, specificity up to 25, GC preference up to 15, and 10 points when no homopolymer is detected.
Result field: guide_scoreTarget recovered
Whether at least one PAM-compatible exact target was retrieved for the guide.
Result field: on_target_recoveredAlternative exact count
Number of additional exact targets after subtracting the first exact target.
Result field: alternative_exact_countNearby target / off-target count
PAM-compatible retrieved rows with one or more mismatches up to the configured cutoff.
Result field: near_offtarget_countPair compatibility score
Starts at 100 and subtracts penalties for duplicate or similar spacers, long complementarity runs, shared exact targets, and shared target/off-target loci. It is clipped to 0–100.
Result field: pair_compatibility_scoreSpacer Hamming distance
Number of positions that differ between two guide spacers. A small value indicates highly similar guide sequences.
Result field: spacer_hamming_distanceComplementarity run
Longest contiguous sequence-complementarity proxy found while aligning one spacer against the reverse complement of another. It is not a full RNA-folding calculation.
Result field: longest_complementarity_runShared exact targets / shared loci
Genomic locations retrieved by both guides. Shared loci can include exact targets and mismatch-near target/off-target locations.
Multiplex score
A 0–100 panel heuristic: 55% mean guide score, 30% mean pair compatibility, up to 10 points for target recovery, and up to 5 points for distinct exact-locus diversity.
Result field: multiplex_scoreMultiplex risk thresholds
Score ≥80 is low risk, score ≥55 and below 80 is moderate, and score below 55 is high. An individual guide with no recovered exact target is marked unresolved.
Target recovery fraction
Fraction of guides in a recommended panel with a recovered best exact target. A value of 1.0 means all included guides recovered an exact target in the screened results.
Result field: on_target_recovery_fractionDistinct loci and genes
Counts of unique exact target coordinates and unique annotated genes represented by a panel.
Coordinates and supplied feature data
Visual maps, chromosomes, and annotation
Blue spacer arrow
A blue shaft represents a Search row explicitly tagged Match, while other returned targets use gray shafts. Cyan-green arrowheads mark the 5′→3′ direction clearly in both light and dark themes.
PAM block
The small yellow block beside the arrow. It appears after a plus-strand spacer and before a minus-strand spacer in genomic coordinate orientation.
Chromosome / contig track
One coordinate lane for each returned chromosome, scaffold, plasmid, contig, or submitted FASTA record.
Displayed coordinate span
The map scales to the returned loci plus padding. It does not claim to show an entire chromosome unless full chromosome-length metadata is available.
Start and end
Stored genomic interval of the matched 20-base spacer. Annotation TSV, GFF3, and GTF use 1-based inclusive coordinates. BED input is interpreted using its standard 0-based, half-open convention and converted internally.
Annotation
Optional feature information supplied with a private genome model, such as genes, transcripts, exons, CDS regions, or custom intervals.
GFF3 / GTF / BED / TSV
Supported annotation formats. GFF3 is recommended. For guntingAI TSV, required columns are seqid, start, and end; optional columns include strand, feature_type, gene_id, gene_name, and annotation.
Sequence-ID matching
Annotation seqid values must match the first token after “>” in the corresponding FASTA header.
Feature type / region
Biological or custom interval class such as gene, mRNA, exon, CDS, intergenic, or uploaded_sequence.
Result field: feature_typeIntergenic
A guide window that does not overlap any supplied annotation interval.
Gene ID and gene name
Gene ID is an identifier from the supplied/reference annotation. Gene name is a human-readable symbol or label. Their accuracy depends on the annotation source.
Ownership, limits, and processing
Accounts, jobs, queues, and private models
Private model
A bounded model trained from a registered user’s FASTA sequence. It is restricted to its owner unless public contribution is requested and approved.
User-contributed model
A formerly private model approved by an administrator for inclusion in the shared model catalog. It is visibly marked as contributed by a user.
Genome-training cap
Maximum total FASTA bases a user may submit for self-service private training. Administrators may assign an individual cap up to the service safety ceiling.
One active job
A guest session or registered account may have one active Search, Design, Multiplex, or private-training job at a time.
Queued
The job has been accepted and is waiting for the relevant service slot.
Running
The job has been dispatched and processing is underway. Workspace remains the persistent source of truth if browser polling is interrupted.
Completed / done
Processing finished and the saved result is available to the authorized account.
Failed
The service reported a terminal processing or validation error. This differs from a temporary browser polling interruption.
Account provenance
Name, institution, affiliation, position, and email associated with model ownership, publication review, and saved records.
Light and dark themes
Use the appearance button in the header or mobile menu. The selected theme is saved in the browser; first visits follow the operating-system preference.
Appearance preferenceApproximate screening
Search, Design, and Multiplex inspect nearest model fragments rather than proving exhaustive genome-wide biological specificity. Experimental and application-specific validation remain necessary.
Interpretation boundary
Scores rank computational evidence; they do not prove editing performance.
guntingAI does not directly model chromatin accessibility, delivery, expression, guide folding, nuclease loading, cellular competition, repair outcomes, toxicity, phenotype, or clinical safety.
Workflow guidance