Frequently asked questions

Model distance measures how close a query is to a retrieved fragment in the selected model’s learned coordinate space. The Search default returns rows where model distance is strictly less than 0.75. It is a ranking and filtering signal, not a direct probability of cleavage or editing.

The PAM, mismatch limit, selected model, model-distance ceiling, or guide sequence may be too restrictive. Confirm that the spacer contains exactly 20 DNA bases, verify the PAM pattern, and increase the distance or mismatch allowance cautiously.

The computation may still be active even when a browser request is interrupted. Workspace is the persistent source of truth for registered users. Reopen the saved job there rather than submitting a duplicate job.

The arrow points in the spacer’s 5′→3′ direction. The PAM block is placed on the strand-correct side of the spacer. Each chromosome or contig is shown on a separate coordinate track.

No. It summarizes guide recovery, nearby target burden, sequence complementarity, shared loci, and panel compatibility. It does not model delivery, expression, chromatin, RNA processing, toxicity, repair, or phenotype.

Use FASTA for the sequence. Self-service training is intended for bounded genomes, plasmids, contigs, constructs, or targeted reference regions within the account’s administrator-controlled base-pair ceiling.

GFF3 is recommended. GTF, BED, and guntingAI TSV are also supported. The annotation sequence ID must match the first token after the corresponding FASTA header’s greater-than sign.

The required columns are seqid, start, and end. Optional columns include strand, feature_type, gene_id, gene_name, and annotation.

An administrator may increase your individual ceiling up to the platform safety limit. Institution-scale references should use a capacity-reviewed build and institutional quote.

Only its owner can select a private model. You may request publication. After administrator approval, it becomes visible in the model catalog and is marked user contributed.

Yes. The owner may request deletion from the Workspace. Deletion removes the account record and requests removal of the corresponding private artifact.

The one-job rule prevents a single guest session or account from monopolizing shared analysis and training capacity. Analysis and private training use separate service queues, but the account-level lock applies across both.

Registered-user job metadata and encrypted result records are managed by the private account workspace. Guest jobs are session-bound and do not provide the same persistent record history.

These fields provide provenance for private-model ownership, public-contribution review, account administration, and institution-scale capacity decisions.

Yes. Use Recent Jobs in Workspace to remove records you no longer need. Active jobs may need to finish or be cancelled before their record can be removed safely.

Your authenticated session ends. Private models, saved results, and account settings remain associated with your account and require a new sign-in to access.

No. guntingAI supports computational target exploration. Experimental validation is required for cleavage, editing efficiency, specificity, delivery, phenotype, and safety.

No. Model distance only reflects proximity in the model representation. Biological risk requires broader off-target assessment, genome context, experimental validation, and application-specific safety review.

User-provided annotations are mapped according to their supplied coordinates and sequence IDs. guntingAI does not independently verify biological curation, assembly identity, gene nomenclature, or annotation completeness.

A gene or annotated interval may contain multiple candidate spacer positions. Opposite-strand guide windows can also overlap the same feature while having different PAM positions and 5′→3′ orientations.