SeqBench

Validation

Each row runs a SeqBench tool on inputs frozen for the purpose and compares it with the answers an independent implementation, the oracle, gave for the same inputs. The figures are computed when the site is built, by running the tools themselves, so they cannot drift from the code and they move when an engine does. Agreement is rounded down and deviations up; where we differ from the oracle, the table says so.

Primers and oligos

Tool and quantityChecked againstCasesAgreementMean deviationMax deviationOracle date
melting_temperature
Nearest-neighbour Tm
Agreement: Tm within 0.02 °C
Primer3 (primer3-py 2.4.0, libprimer3 2.6.1)
340100%0.0025 °C0.0050 °C2026-11-21
annealing_temperature
Annealing temperature across NEB polymerases
Agreement: Annealing temperature identical to the calculator's, both Tms within its display rounding
NEB Tm Calculator v1.17.0
35100%0 °C0 °C2026-09-25
oligo_analysis · hairpin
Hairpin ΔG against Primer3 at 1 M Na⁺
Agreement: Same call as Primer3 at −6 kcal/mol, the tools' own line: flagged, or not
Primer3 (primer3-py 2.4.0, libprimer3 2.6.1)
36098.6%
1 missed · 4 extra
0.214 kcal/mol3.32 kcal/mol2026-11-21
oligo_analysis · self-dimer
Self-dimer ΔG against Primer3 at 1 M Na⁺
Agreement: Same call as Primer3 at −6 kcal/mol, the tools' own line: flagged, or not
Primer3 (primer3-py 2.4.0, libprimer3 2.6.1)
36097.2%
0 missed · 10 extra
0.212 kcal/mol1.77 kcal/mol2026-11-21
cross_dimer
Cross-dimer ΔG against Primer3 at 1 M Na⁺
Agreement: Same call as Primer3 at −6 kcal/mol, the tools' own line: flagged, or not
Primer3 (primer3-py 2.4.0, libprimer3 2.6.1)
38098.6%
1 missed · 4 extra
0.194 kcal/mol1.79 kcal/mol2026-11-21
cross_dimer · perfect duplexes
Dimer ΔG of perfectly complementary pairs against Primer3 at 1 M Na⁺
Agreement: ΔG within 0.1 kcal/mol
Primer3 (primer3-py 2.4.0, libprimer3 2.6.1)
40100%0.032 kcal/mol0.036 kcal/mol2026-11-21
in_silico_pcr
Predicted products, no mismatches allowed
Agreement: Same products (start, end, length, sequence) and the same count held back by the length window
Exact-match reference (Python, written for this check)
156100%——2026-11-21

Restriction digests

Tool and quantityChecked againstCasesAgreementMean deviationMax deviationOracle date
restriction_sites
Cut positions, 542 enzymes on 5 templates, linear and circular
Agreement: Same 1-based top-strand cut positions
Biopython 1.87
5,420100%——2026-11-21
virtual_gel
Digest fragment sizes, one to three enzymes
Agreement: Same fragment sizes
Biopython 1.87
160100%——2026-11-21

Sequence

Tool and quantityChecked againstCasesAgreementMean deviationMax deviationOracle date
translate
Protein, each of six reading frames, with ambiguity codes
Agreement: Same protein string, including B (Asx), Z (Glx) and J (Xle)
Biopython 1.87
930100%——2026-11-21
codon_adaptation_index
CAI and GC3, 8 host tables
Agreement: CAI within 0.0006 and GC3 within 0.06 points
Biopython 1.87
96100%0.00030.00052026-11-21

Alignment

Tool and quantityChecked againstCasesAgreementMean deviationMax deviationOracle date
multiple_sequence_alignment · two sequences
Alignment score of a pair, DNA and protein
Agreement: Score equals the optimum
Biopython 1.87
60100%002026-11-21
multiple_sequence_alignment
DNA families of 4 to 9 sequences: true aligned residue pairs missed
Agreement: Deviation is the share of the simulation's true aligned residue pairs (1 − Q) that the alignment does not reproduce
Simulated true alignments (seeded substitution and indel evolution on a random tree)
MAFFT 7.526 (--auto) on the same families: mean 6.63%
30—4.50%13.59%2026-11-21
multiple_sequence_alignment
Protein families of 4 to 9 sequences: true aligned residue pairs missed
Agreement: Deviation is the share of the simulation's true aligned residue pairs (1 − Q) that the alignment does not reproduce
Simulated true alignments (seeded substitution and indel evolution on a random tree)
MAFFT 7.526 (--auto) on the same families: mean 1.17%
30—1.37%4.67%2026-11-21

Files

Tool and quantityChecked againstCasesAgreementMean deviationMax deviationOracle date
parse_genbank
46 records, GenBank and EMBL: locus, topology, sequence, every feature
Agreement: Same locus, length, topology and sequence per record; same type, strand, spans and qualifiers per feature
Biopython 1.87
353100%——2026-11-21
parse_snapgene
30 SnapGene .dna files: read against Biopython, and written and read back
Agreement: Same sequence, topology and per-feature type, strand, spans and qualifiers, for the frozen bytes and after write and read back
Biopython 1.87
508100%——2026-11-21

Hairpin and dimer ΔG against Primer3 at its default salt

The same cases as above, against Primer3 at 50 mM Na⁺. Only the salt Primer3 is run at differs from the first table.

Tool and quantityChecked againstCasesAgreementMean deviationMax deviationOracle date
oligo_analysis · hairpin
Hairpin ΔG against Primer3 at 50 mM Na⁺, its default
Agreement: Same call as Primer3 at −6 kcal/mol, the tools' own line: flagged, or not
Primer3 (primer3-py 2.4.0, libprimer3 2.6.1)
36099.1%
0 missed · 3 extra
0.213 kcal/mol2.63 kcal/mol2026-11-21
oligo_analysis · self-dimer
Self-dimer ΔG against Primer3 at 50 mM Na⁺, its default
Agreement: Same call as Primer3 at −6 kcal/mol, the tools' own line: flagged, or not
Primer3 (primer3-py 2.4.0, libprimer3 2.6.1)
36098.0%
1 missed · 6 extra
0.253 kcal/mol2.88 kcal/mol2026-11-21
cross_dimer
Cross-dimer ΔG against Primer3 at 50 mM Na⁺, its default
Agreement: Same call as Primer3 at −6 kcal/mol, the tools' own line: flagged, or not
Primer3 (primer3-py 2.4.0, libprimer3 2.6.1)
38098.4%
1 missed · 5 extra
0.242 kcal/mol2.13 kcal/mol2026-11-21

Where we differ, and what is not covered

Hairpin and dimer ΔG.
Compared as a yes or no call at the line the tools use to flag a structure, and as a distance in kcal/mol. “Missed” counts structures Primer3 flags that the tool does not, “extra” the reverse. The structure rows in the first tables are against Primer3 at 1 M Na⁺; the table headed “default salt” holds the same cases against its default 50 mM. Primer3 flags few structures in oligo_analysis · hairpin, so there the shared call says little and the deviation is the figure to read. The basis each tool states for the figure it reports: oligo_analysis, “Tm and the structure ΔG (hairpinDG, selfDimerDG) are at the stated conditions; hairpinDG1M and selfDimerDG1M are the 1 M reference, and dG37 is a 1 M nearest-neighbour figure”; cross_dimer, “A nearest-neighbour alignment (stacks, bulges and internal loops, terminal mismatches, dangling ends and the two terminal initiation terms) at the stated buffer's sodium equivalent, with a SantaLucia 1998 entropy salt term. crossDimerDG1M is the same structure at the 1 M reference; compare the -6 kcal/mol line against whichever basis is on screen.”
Multiple alignment.
MAFFT 7.526 (--auto) aligned the same families and misses 6.63% of the true pairs on the DNA ones and 1.17% on the protein ones, on average, against the tool's 4.50% and 1.37%. The server-side alignment modes are not part of this table.
in_silico_pcr.
The reference is a few lines of code written from the definition of a product, not a mature program, and it covers priming with no mismatches only.
codon_adaptation_index.
The oracle is given the tool's own host tables, so the arithmetic is checked and the tables, which are data, are not.
annealing_temperature.
The oracle is what NEB's calculator displayed through its own form. Its code is not used.
Not in this table.
Tools that return a modelled estimate, which state their own validation on their pages; lookups of external records; and the NEB buffer advice, which has no independent implementation to compare with.
Reading it.
Agreement with an oracle on these cases is not proof of correctness: a case set can miss a corner, and an oracle can be wrong.