FILE
DC–2026–07
ENGINE
Bowtie2 + SantaLucia NN
CLASS.
TECHNICAL · PUBLIC
VERDICT
6 levels
BASIS
16 references
10
weighted biophysical factors on the target, returned one by one in the report
3
progressive layers against the exclusions: binding, 3′-end discrimination and amplicon prediction
6
verdict levels, from validated to non-functional, each with its reason
16
published references support the model, cited factor by factor
+
+
+
+
What makes it unique
It is not a homology search with a nice name. It integrates alignment, nearest-neighbour thermodynamics, the biochemical model of 3′-end extension and secondary-structure prediction of the amplicon into a single pipeline — ten weighted biophysical factors and a six-level verdict.
A mismatch in the last three bases of the extending end penalises amplification severely even if the rest of the primer is identical. That is exactly the case a homology search flags as a risk and the tube refutes — and the reverse, the one it waves through that then amplifies what it should not.
01
The state of the art
Resembling a sequence is not amplifying it.
Validating specificity at the bench takes expensive DNA panels, weeks of work and several rounds of optimisation. And the usual alternative — a homology search — answers a different question from the one the lab is asking.
02
What is modelled
Ten factors on the target. Three layers against the rest of the world.
The three off-target layers are progressive: they quickly discard whatever cannot bind, and only what survives reaches the expensive amplicon-prediction analysis.
On target · ten weighted factors
Binding
Melting temperature
3′ stability
Accessibility
Dimers
Amplicon GC
Secondary structure
Length
G-quadruplex
Mispriming
Each enters the reaction success index, from 0 to 1, with its weight, and the report returns them broken down: a candidate is not rejected "for a low score", it is rejected for the factor that sinks it.
Off target · three progressive layers
The verdict, in six levels
03
The search space
What it was searched against, stated in the report itself.
Three scopes, from the declared panel to everything deposited. The report names the one used, because a clean result on a small panel is not a declaration of specificity.
Declared panel
Against the exclusions you define
The default scope: the genomes the assay has to reject, declared by the user or inherited from curated knowledge, matrix-aware where indicated.
Bounded wide
Target clade plus universal contaminants
A fast local search reusing the design specificity pipeline, with forward-by-reverse pairing and 3′-end analysis.
Exhaustive wide
Against everything deposited
A direct query to the public nucleotide database with the same pairing and local analysis. It runs in the background, with progress and an estimated time.
The honesty the module demands
"No off-target findings" only means something when accompanied by the space searched. That is why the report always declares it, and separates what aligns from what primes and amplifies: they are two different lists, and only the second ruins an assay.
The module does not interpret biology either: it validates primers. It consumes the exclusion panels projected by curated knowledge and does not couple to the engine that produces them — which keeps the verdict reproducible even as the knowledge evolves.
04
The contract
What goes in, what comes out, what it chains to.
In
A primer pair and target sequence, standalone or from the institutional repository; the exclusion panel, declared or inherited from curated knowledge; and, if wanted, a whole kit to request a second opinion from an independent engine.
Out
A success index from 0 to 1, a six-level verdict, the breakdown of the ten factors, a list of off-target findings with the reason for each rejection or alert, the predicted melting temperature of the amplicon and optimisation recommendations.
Chains to
It receives from SnapPrime and from the repository; it delivers to FastKit, which only packages primers that have been validated. Matrix mode to cross N primers against M genomes, and batch mode with progress tracking.
05
Technical sheet
Input
A primer pair and target sequence; optionally a probe, an exclusion panel and the working matrix. It also accepts a complete kit from the generator, as a second opinion from an independent engine.
Modes
Single, batch, a matrix of N primers by M genomes, and multiplex with asynchronous progress tracking.
Output
A reaction success index from 0 to 1, a six-level verdict, the breakdown of the ten factors, off-target findings with their reason, the predicted melting temperature of the amplicon and recommendations.
Declared space
Every report names the search space it was issued in, and separates what aligns from what primes and amplifies. Two lists, not one.
Architecture
The module stores no knowledge of its own: it consumes the exclusion panels projected from curated knowledge. That keeps the verdict reproducible and decoupled from the inference engine.
Limit · panels
The freshness of the exclusion panels depends on the publication cycle of curated knowledge. An out-of-date panel narrows the search space without the number showing it: that is why the report declares the date.
Limit · mismatches
The aligner's mismatch policy is fixed in the code and is not yet configurable from the interface. It sits in the module's debt register.
Limit · in silico
It is biophysical prediction, not an assay. It replaces optimisation rounds and discards the unworkable before reagent is spent, but formal validation of a method still requires the laboratory.
Alignment: Bowtie2 in paired mode. Nearest-neighbour thermodynamics: SantaLucia (1998). 3’ end extension model: Kwok (1990) and Stadhouders (2010). Stability and salt correction: Owczarzy (2004). Broad search: local similarity channel and a client for the NCBI public BLAST interface against the nucleotide collection. Sixteen published references in total.
Let the thermocycler confirm, not discover.
BIOTECNO.org · Vitoria-Gasteiz
EU + Codex