Abstract

PrimerCat does not assign one universal accuracy value to every result. We describe input quality, algorithmic output, database coverage, heuristic assumptions, and experimental confirmation separately so researchers can judge what each layer of evidence supports.

Evidence classes

Outputs with different origins should not be judged on one scale. These classes describe where evidence comes from; they are not confidence percentages.

Class	Evidence type	Typical output	Correct interpretation
C1	Deterministic computation	Sequence length, GC%, coordinates, solution arithmetic, Primer3 fields	PrimerCat should reproduce the value under the same input, parameters, and software version; input error and model assumptions still apply.
D1	Database-dependent screen	RefSeq template, BLAST/Bowtie2 hits, accessions	PrimerCat's statement is valid only for the stated database/assembly, query date, thresholds, and hit cap.
H1	Heuristic rank	qPCR composite score, gRNA activity score, Low/Medium/High labels	Researchers may use it to prioritise candidates, but not as a probability or a value comparable across tools or experiments.
R1	Curated reference	Protocols, recipe notes, chemical-hazard summaries	Researchers may use it for preparation and review, but not instead of primary literature, product instructions, SDS, SOP, or risk assessment.
X0	Experimental confirmation	Efficiency, single product, measured editing, measured off-targets, sample suitability	PrimerCat does not produce this evidence; researchers must establish it in their own experimental system.

Keep three concepts separate

Computation complete

The algorithm returned values or candidates.

Screen passed

No warning rule was triggered in a defined database, threshold, and return set.

Experiment valid

The target sample, reagents, instrument, and controls met pre-specified acceptance criteria.

Claims supported by each tool

PrimerCat states what each module can support and what researchers cannot infer from the result.

Module	Supported interpretation	Unsupported inference	Evidence
qPCR primers	Candidates meet current Primer3 constraints. Human and mouse modes simulate paired amplification on fixed GRCh38.p14 or GRCm39 genomes and their matched RefSeq RNA collections, distinguishing the selected transcript, same-gene isoforms, and other-gene products.	Acceptable efficiency, one melt peak, absence of unannotated transcripts or sample-variant effects, or guaranteed experimental success.	C1 + D1 + H1
Endpoint PCR	Coordinates, product size, and structure fields on the submitted template; paired records in the optional returned BLAST set.	Whole-genome uniqueness, a single band in the sample, or no need to optimise annealing temperature.	C1 + D1
CRISPR gRNA	PAM/strand/position, sequence-feature priority, and near matches found by the current backend.	Measured on-target editing, absence of cellular off-targets, or portability across Cas variants and delivery systems.	C1 + D1 + H1
BLAST	Local-similarity hits and statistics under the selected database and parameters.	Shared function, proven homology, completed species identification, or absence when no hit is returned.	D1
Solutions/protocols/safety	Calculate amounts from stated values and inspect structured references with source links.	Universal recipe suitability, equivalence to an SOP/SDS, or hazard classification across all concentrations and mixtures.	C1 or R1

Fixed references are human GRCh38.p14 with the RefSeq 2025-08 annotation and mouse GRCm39 with the RefSeq 2024-02 annotation; assembly labels in results define the scope of each statement.^13,14,16,17

Major sources of variability

Input sequence, transcript, and genome version
Species filter and database update state
Thresholds, hit cap, timeout, and fallback backend
Polymerase, buffer, template quality, and sample variants
Cell type, chromatin, Cas variant, and delivery method

Algorithms and databases have peer-reviewed foundations, but confidence in this site's combined workflow still depends on implementation details and validation data.^2,3,4,5,8,9

How to read status labels

A PrimerCat status summarises the current screen. Researchers must not convert absence of evidence into a claim of safety or specificity.

Status	What it actually means	Next action
Pass / Low	No additional returned hit met the active warning rule.	Review scope, hit cap, and on-target identification; experimental confirmation is still required.
Review / Medium	Moderately similar extra hits exist, or evidence is insufficient for exclusion.	Compare candidates, inspect genomic context and mismatch positions, and use a more complete specialist tool when needed.
High risk / High	An extra perfect/high-similarity hit, several candidate hits, or failure to anchor the supplied target locus was found.	Usually prefer another candidate; if use is necessary, perform targeted computation and experiments first.
No hits	The backend returned no hit passing its filters.	Check the query, database coverage, and short-query limitations; never treat this as proof of uniqueness.
Not checked / error / skipped	The screen did not run, timed out, or was unavailable.	Do not assign specificity or low risk; retry or use an independent method.

Recommended experimental validation

We provide minimum validation directions, not a universal SOP. Researchers should pre-specify acceptance criteria for their actual experimental system.

RT-qPCR / qPCR

Confirm the accession, target isoform, and amplicon; inspect common variants when relevant.
Include NTC and no-RT controls and confirm a single product by melt curve and/or gel.
Use a standard curve to assess efficiency, linear range, and detection limit; report sequences, concentrations, chemistry, and raw data.
Use validated reference genes and appropriate normalisation; follow MIQE 2.0.

Endpoint PCR

Run an annealing-temperature gradient with positive, negative, and no-template controls.
Confirm product count and size by gel; confirm sequence by Sanger or another method for critical applications.
Check the relevant assembly and sample variation for the real specimen source.

CRISPR

Cross-check candidates in an independent genome-aware design tool and provide an explicit target locus where possible.
Measure on-target editing and indel spectrum in the target cells by amplicon sequencing or an equivalent assay.
Choose targeted-site sequencing or an unbiased method such as GUIDE-seq according to application risk; computational screening cannot replace cellular off-target measurement.

BLAST & references

Verify critical claims against the accession and primary record; add MSA, domain, and phylogenetic analysis when required.
Return from protocols and recipes to the primary source and institutional SOP; for chemicals, check the current label, supplier SDS, and EHS requirements.

Reproducibility and change factors

To let another researcher reconstruct a run, researchers should preserve the computational context rather than only a copied candidate sequence.

Record	Why
Original input or accession + version	The sequence can change when an accession is revised.
Species and reference assembly	Hits and coordinates differ between assemblies.
All submitted parameters	Presets initialise controls; submitted values determine candidates.
Backend, database, and date	Local Bowtie2 and NCBI BLAST have different scope, and remote databases evolve.
Hit caps and exception states	Truncation, timeout, and fallback alter the visible evidence.
PrimerCat and dependency versions	Primer3, Biopython, index, and score-rule changes may alter output.

Current operating basis and verification record

To keep PrimerCat results traceable, we publish the reference versions, index scale, software quality baseline, and production spot checks used by the current service. Researchers should interpret each conclusion within the stated data scope, rules, and check date.

Human annotation-feature records

4,514,782

Human transcript-locus records

202,461

Mouse annotation-feature records

3,023,953

Human reference data

GRCh38.p14 / GCF_000001405.40; GCF_000001405.40-RS_2025_08

Check performed: 17/17 listed runtime files matched by SHA-256 on the build and production hosts.
Scope of conclusion: Confirms that the listed files match across hosts and defines the computational scope for the human genome and accessioned RefSeq RNA.

Mouse reference data

GRCm39 / GCF_000001635.27; GCF_000001635.27-RS_2024_02

Check performed: 3/3 files changed in this update matched by SHA-256 on the build and production hosts.
Scope of conclusion: Confirms the three files involved in this update; it does not claim the same checksum coverage for every mouse runtime file.

Software quality baseline

Git 0f42ebe

Check performed: 283 backend tests passed; Next.js production build passed.
Scope of conclusion: Shows that the baseline meets encoded test contracts; it does not establish biological performance of a candidate.

Production availability spot check

8 public routes; qPCR / PCR / CRISPR / BLAST

Check performed: Production smoke test completed on 2026-09-03.
Scope of conclusion: Shows that pages were reachable and four principal workflows completed at check time; it is not an uptime promise or experimental validation.

NCBI human assembly Human annotation release NCBI mouse assembly Audit-baseline source File checksum manifest Evidence snapshot (JSON) ↓

Pass means only that the stated software or data check completed at the recorded time. Database rows are not unique genes or usable candidates; a computational audit is not a wet-lab success rate; and automated or production smoke tests are not biological or clinical validation.^13,14,16,17

Fixed-reference computational coverage audit v0.6

We selected 200 distinct genes from the pinned mouse RefSeq RNA collection using a public deterministic hash rule, retained one NM_ transcript per gene, and used PrimerCat to regenerate and screen every candidate set. This audit expands pipeline coverage; it is not random sampling, an external accuracy test, or wet-lab validation.^{2,4,9,11,12,13,14,15}

Candidate pairs screened

2000

Joint computational passes

1648/2000

Genes with ≥1 pass

180/200

How this run was performed

The sampling frame is restricted to NM_ transcripts in the pinned RNA FASTA that have a GTF locus, are 250–5,000 bp long, and contain only A/C/G/T. Records are ranked by SHA-256(seed:accession), the first transcript per gene is retained, and the first 200 genes are selected. Primer3 generates ten lowest-penalty pairs per record, which are then screened under the same paired-amplicon rule against GRCm39 and matched RefSeq RNA. A joint pass requires exactly one selected-transcript product, no other-gene or unclassified transcript product, and a compatible genomic result; same-gene isoforms are reported separately.

Metric	Decision rule	Result	Correct interpretation
Candidate pairs screened	Take the lowest-Penalty Primer3 candidates for each of 200 distinct genes	2000	Every candidate used the same assembly, parameters, and decision rule.
Selected-transcript product	Exactly one 50–5000 bp paired product occurs on the selected accession	1992/2000	Confirms the intended product in the fixed RNA collection, not expression in the sample.
Transcript gene-level pass	The target product exists with no other-gene or unclassified transcript product	1749/2000	Other isoforms of the same gene may still be amplified.
Isoform-specific flag	Gene-level pass with no product from another isoform of the same gene	302/2000	Applies only to this fixed RefSeq annotation.
Other same-gene isoform	At least one other RefSeq isoform forms a paired product	1577/2000	Compatible with total-gene measurement but not a single-isoform claim.
Cross-gene product	At least one RefSeq transcript from another gene forms a paired product	215/2000	Fails the joint rule; prefer another candidate.
Joint computational pass	Transcript gene-level pass plus a compatible genomic result	1648/2000	A computational screen pass, not wet-lab success.
Resolved by transcript evidence	No contiguous genomic product, but intended RNA and no cross-gene product are confirmed	320/2000	Handles splice-junction candidates without equating no genomic product with uniqueness.
Transcript hit cap	Either primer exceeds the 128-transcript decision limit	39/2000	Extra products may be hidden, so the conservative rule assigns no pass.
Gene with at least one pass	At least one candidate for the gene meets the joint rule	180/200	A stronger computational option exists; this is not per-gene experimental success.

Download benchmark snapshot ↓View the 100-gene PrimerBank comparator v0.5

These proportions describe computational outcomes under a fixed assembly, fixed RNA collection, software version, and stated rules. They are not accuracy, sensitivity, clinical performance, or wet-lab success. Deterministic hashing makes the cohort reproducible; it does not make the cohort representative of expression abundance, populations, tissues, or real experiments.

To keep evidence types distinct, the 100-gene PrimerBank external comparator remains available as v0.5. PrimerBank experiments validate the published pairs, not newly generated PrimerCat candidates. v0.5 reports 864/1000 joint computational passes and 94/100 genes with at least one pass.

Next evidence layer

Add version-pinned common-variant checks at primer-binding sites. Then pre-sample joint-pass, cross-gene, multi-isoform, and no-contiguous-genomic-product classes for prospective wet-lab testing against efficiency, linearity, single-product, NTC, and no-RT criteria.

Current validation status

We state which evidence PrimerCat does not yet have so researchers do not mistake software availability for academic validation.

We publish per-candidate computational records

v0.6 deterministically selects 200 distinct genes from fixed RefSeq RNA; we also publish the sampling rule, reference-file checksums, runtime parameters, per-candidate screens, and source-code checksums.

PrimerCat has no clinical validation

PrimerCat is for research design only and is not intended for diagnosis, treatment decisions, clinical reporting, or regulatory submission.

Components have peer-reviewed foundations

Primer3, BLAST, Bowtie2, RefSeq, MIQE, and CRISPR studies support components and validation frameworks; they do not endorse PrimerCat's end-to-end performance.

Sample variation and experimental context remain uncovered

Fixed RefSeq RNA can distinguish annotated isoforms, but it is not the sample's expressed transcriptome and does not cover unannotated transcripts, sample SNPs or structural variants, reaction chemistry, or actual amplification. Hit-cap candidates remain indeterminate.

Chemical-safety records use representative sources such as PubChem, but actual work must return to the current product SDS; an aggregated database is not a safety guarantee for a specific product.¹⁰

REF

References

References cover algorithmic foundations, databases, qPCR reporting standards, the PrimerBank reference set, and experimental CRISPR off-target validation.

[1]
Bustin SA, et al. (2025). “MIQE 2.0: Revision of the Minimum Information for Publication of Quantitative Real-Time PCR Experiments Guidelines” Clinical Chemistry. 71:634–651. doi:10.1093/clinchem/hvaf043 ↗
[2]
Untergasser A, et al. (2012). “Primer3—new capabilities and interfaces” Nucleic Acids Research. 40:e115. doi:10.1093/nar/gks596 ↗
[3]
Ye J, et al. (2012). “Primer-BLAST: a tool to design target-specific primers for polymerase chain reaction” BMC Bioinformatics. 13:134. doi:10.1186/1471-2105-13-134 ↗
[4]
Langmead B, Salzberg SL. (2012). “Fast gapped-read alignment with Bowtie 2” Nature Methods. 9:357–359. doi:10.1038/nmeth.1923 ↗
[5]
Doench JG, et al. (2016). “Optimized sgRNA design to maximize activity and minimize off-target effects of CRISPR-Cas9” Nature Biotechnology. 34:184–191. doi:10.1038/nbt.3437 ↗
[6]
Hsu PD, et al. (2013). “DNA targeting specificity of RNA-guided Cas9 nucleases” Nature Biotechnology. 31:827–832. doi:10.1038/nbt.2647 ↗
[7]
Tsai SQ, et al. (2015). “GUIDE-seq enables genome-wide profiling of off-target cleavage by CRISPR-Cas nucleases” Nature Biotechnology. 33:187–197. doi:10.1038/nbt.3117 ↗
[8]
Camacho C, et al. (2009). “BLAST+: architecture and applications” BMC Bioinformatics. 10:421. doi:10.1186/1471-2105-10-421 ↗
[9]
O’Leary NA, et al. (2016). “Reference sequence (RefSeq) database at NCBI: current status, taxonomic expansion, and functional annotation” Nucleic Acids Research. 44:D733–D745. doi:10.1093/nar/gkv1189 ↗
[10]
Kim S, et al. (2025). “PubChem 2025 update” Nucleic Acids Research. 53:D1516–D1525. doi:10.1093/nar/gkae1059 ↗
[11]
Wang X, et al. (2012). “PrimerBank: a PCR primer database for quantitative gene expression analysis, 2012 update” Nucleic Acids Research. 40:D1144–D1149. doi:10.1093/nar/gkr1013 ↗
[12]
Spandidos A, et al. (2008). “A comprehensive collection of experimentally validated primers for Polymerase Chain Reaction quantitation of murine transcript abundance” BMC Genomics. 9:633. doi:10.1186/1471-2164-9-633 ↗
[13]
NCBI RefSeq (2024). “Mus musculus genome assembly GRCm39” NCBI Datasets. RefSeq assembly GCF_000001635.27. Source ↗
[14]
NCBI RefSeq (2024). “Mus musculus Annotation Release GCF_000001635.27-RS_2024_02” NCBI Eukaryotic Genome Annotation. GRCm39 RefSeq annotation report. Source ↗
[15]
NCBI (2026). “Genomes Download FAQ: assembly-directory file content” NCBI Genome. Defines *_rna.fna.gz as accessioned RNA products annotated on a RefSeq assembly. Source ↗
[16]
NCBI RefSeq (2025). “Homo sapiens genome assembly GRCh38.p14” NCBI Datasets. RefSeq assembly GCF_000001405.40. Source ↗
[17]
NCBI RefSeq (2025). “Homo sapiens Annotation Release GCF_000001405.40-RS_2025_08” NCBI Eukaryotic Genome Annotation. GRCh38.p14 RefSeq annotation report. Source ↗

Interpretation principle

Turn confidence into an inspectable record

Researchers should confirm the input and screening backend, inspect hit details, and then validate in the actual experimental system.

Read full methods Start qPCR design