Browser-based LAMP primer designer keeps target sequences private and flags dimers

A new tool described in Frontiers in Bioinformatics designs complete loop-mediated isothermal amplification (LAMP) primer sets entirely inside a web browser, without installation, registration, or any transmission of the target sequence to a server. The designer, developed by Ruslan Kalendar of the Helsinki Institute of Life Science at the University of Helsinki and the National Laboratory Astana in Kazakhstan, is one module of a broader web platform for PCR and sequence analysis. Because all computation runs as client-side JavaScript, the only network request the page can initiate is an optional retrieval of a sequence from NCBI by accession number, which sends the accession rather than the sequence itself. For clinical samples, agricultural biosecurity work, or unpublished genome targets, that separation between the sequence and the outside world is the tool’s headline property, but the article documents a set of technical features that its author argues matter just as much.

LAMP amplifies a target to detectable levels in under an hour at a single temperature, using a strand-displacing polymerase and no thermal cycler. Its selectivity comes from its primers rather than from temperature cycling: the four core oligonucleotides recognize six separate regions of the target, and two optional loop primers raise that number to eight while roughly halving the reaction time. The six oligonucleotides must be arranged in a specific order along roughly 200 base pairs of template and must fall into two distinct melting-temperature windows. Crucially, none of them may pair with any other, because a LAMP reaction is a single isothermal mixture in which all primers are present simultaneously and no annealing step discriminates against mispriming. Primer-derived mispriming is a documented source of false positives in LAMP and can be traced to individual members of a set, which makes preventing it a central requirement of any design software.

Most published LAMP assays have been designed with PrimerExplorer from Eiken Chemical or the New England Biolabs LAMP Primer Design Tool, both of which are hosted services that upload the target sequence for computation on a third-party machine. Downloadable alternatives such as LAVA, GLAPD, MorphoCatcher, and LAMPrimers iQ each demand a local Perl, C/CUDA, or Python toolchain. Two further client-side implementations now exist: a free calculator page published in April 2026 by a laboratory-equipment vendor, whose documentation describes no screening for dimers between the primers of a set, and primd, an MIT-licensed TypeScript library whose documentation describes no template masking and no complexity filter. What distinguishes the new tool, the author argues, is no longer browser execution itself but the combination of what happens before and after the geometry is satisfied.

Before any candidate is enumerated, the template is masked for dispersed repeats and low-complexity sequence. The repeat mask counts every 15-mer on both strands of the submitted record and marks each position covered by a 15-mer occurring more than once; this is a self-comparison of the record against itself, not a search against a genome or repeat database. The low-complexity mask slides an 18-nucleotide window over the template and counts distinct dinucleotides and trinucleotides on a five-symbol alphabet, initiating a block when a window scores below 14 out of a possible 33. The tool then builds six independent candidate lists, three per strand, growing each oligonucleotide from a minimum length until its melting temperature reaches the user’s window. Outer primers default to 60 to 62 degrees Celsius, while inner and loop primers are drawn from a window 5 degrees higher, close to the 64 to 66 degrees that NEB recommends for loop primers. A 3-prime-end pattern filter admits 40 of the 64 possible terminal trinucleotides, and every surviving candidate must also pass a simple-sequence blacklist, a linguistic-complexity threshold set at 70 percent by default, a self-complementarity test, and a check against any oligonucleotides the user has pasted into a personal stock list.

Linguistic complexity is computed on each candidate as the observed k-word vocabulary over the maximum possible vocabulary for an oligonucleotide of that length, summed over word sizes one to three. The default threshold is strict: a perfect (ACGT)5 repeat scores 31 percent and a 20-mer of a single base scores 7 percent, but even an ordinary-looking 20-mer can fall just under 70 percent. Melting temperatures use unified nearest-neighbor parameters extended with deoxyinosine, LNA, and RNA steps, with salt corrections selected on the ratio of the square root of the magnesium concentration to the monovalent cation concentration. Candidate melting temperatures are evaluated at a fixed condition set of 55 millimolar monovalent cation, 1 millimolar magnesium, and 0.2 micromolar oligonucleotide, shared across the platform so values remain comparable between its tools; the author notes these should be read as relative design targets several degrees below what a LAMP-specific calculator would report, since LAMP chemistry typically runs at 6 to 8 millimolar magnesium with 1.6 micromolar inner primer.

Assembly follows the published LAMP geometry: the 5-prime end of F2 lies at most 20 nucleotides downstream of the 3-prime end of F3, loop-forming intervals span roughly 0 to 40 nucleotides, and the F2-B2 distance must not exceed a user-set maximum of 200 base pairs. Every pair of oligonucleotides is first checked by a fast combinatorial rule that counts a duplex event when an extended match reaches six bases at a 3-prime end or eight bases internally. Assembled assays are then screened thermodynamically: every pair and self-pair among F3, B3, the assembled FIP and BIP, and the loop primers is aligned and scored by a nearest-neighbor engine at the reaction temperature of 65 degrees Celsius and the 349 millimolar sodium equivalent of a LAMP mixture, so the values describe LAMP conditions rather than a standard annealing condition and are not comparable with dimer free energies quoted at 25 or 37 degrees by other software. Assays whose most stable dimer falls below a user-set limit are rejected, and the value and the responsible pair are reported for every assay kept.

Performance is fast. Complete designs took 23 to 245 milliseconds across nine benchmark targets, single-threaded, on one core of an Intel Core i5-4570 under Windows 10, with the design code loaded into a Node.js environment byte-identical to what the browser runs. Run time follows the number of surviving candidates rather than target length: the 1,066-base-pair tandem-repeat control, of which the masks remove 90.4 percent, leaves 24 candidates and designs in 23 milliseconds, while the unmasked 1,026-base-pair hpt record leaves 323 candidates and takes 245 milliseconds. Masking pays only when the submitted record repeats itself. On the tandem-repeat control, the tool proposes 24 candidates with repeat masking on and 310 with it off, a 13-fold difference, with the extra candidates placed on the tandem array itself; across the remaining benchmark, the masks are close to inert, touching only the PrimerExplorer target, whose 419 base pairs carry a genuine internal duplication amounting to 9.1 percent masked, and the P-FMV promoter, whose low-complexity mask covers 7.4 percent.

The defaults are strict, and this is their most important practical consequence: of the nine benchmark targets, four returned no complete assay at default settings, and a fifth returned only two. Candidates were abundant in three of the four failures, so the constraint is not candidate supply but the combination of spacing, amplicon-size, and cross-dimer requirements imposed during assembly. The two parameters that dominate yield are the melting-temperature window and the maximum F2-B2 distance, in that order: widening the window from 60 to 62 degrees to 58 to 64 degrees raised the worked example’s yield from 5 assays to 54, while widening the amplicon limit alone rescued none of the four failing targets. The author also cautions that a greedy thinning rule, which reports a pair only if its F2-B2 fragment is shorter or its summed complexity higher than the previously reported pair, means the reported counts are lower bounds chosen by enumeration order rather than a count of the design space: with thinning disabled, the worked example has 10 admissible dimer-free assays at defaults, and 12,222 once overlapping positions are allowed, of which 2,540 are reported.

A direct comparison with an archived PrimerExplorer V5 design on an identical 419-base-pair target illustrates the stringency gap. PrimerExplorer returned 50 candidate sets that cluster into only 14 genuinely distinct groups when sets sharing four or more of their six primers are collapsed together. The new tool returns nothing on that target at defaults and only reaches 82 four-primer assays at 55 to 66 degrees Celsius, 17 to 28 nucleotides, 50 percent complexity, a 350-base-pair limit, and loop primers disabled. Put the other way around, the default 60 to 62 degree window admits none of the 300 component oligonucleotides of the archived sets and 13 percent of the 24 oligonucleotides of four experimentally validated assays, a difference the author attributes largely to differing salt conventions rather than primer quality.

The dimer screen was calibrated against assays whose fates are known. The four validated assays scored +0.8, +0.3, −2.4, and −4.5 kilocalories per mole; because the tool joins the F1c and B1c segments to F2 and B2 through a TTTTTT linker that the earlier assays lacked, rescoring with the linker shifts values by −1.2 kilocalories per mole on average, and −5 kilocalories per mole emerges as the smallest limit that keeps every assay known to be functional. Four deliberately constructed negative controls, including a self-complementary construct reaching −15.0, are all rejected. That validation, however, covers four four-primer assays on plasmid templates with an endpoint readout, designed with an earlier version of the tool, without loop primers, replicates, or a detection limit; no assay designed by the version described has been tested at the bench. Specificity is assessed only within the submitted sequence, with no genome-wide search, output is not ranked, the bisulphite mode is implemented but not benchmarked, and the design runs in a single thread with a practical ceiling of a few kilobases of unmasked template. The tool is freely available at primerdigital.com/tools/lamp.html, with source modules released under the GNU General Public License v3.0.

Subject of Research: Browser-based design of loop-mediated isothermal amplification primer sets with template repeat masking and dimer screening

Article Title: Browser-based design of loop-mediated isothermal amplification (LAMP) primer sets with template repeat masking and bisulphite support

Article References: Kalendar, R. (2026). Browser-based design of loop-mediated isothermal amplification (LAMP) primer sets with template repeat masking and bisulphite support. Frontiers in Bioinformatics, 6, Article 1992214. https://doi.org/10.3389/fbinf.2026.1992214

Image Credits: AI Generated

DOI: 10.3389/fbinf.2026.1992214

Keywords: LAMP, primer design, isothermal amplification, molecular diagnostics, web tool, repeat masking, primer dimers, linguistic complexity, bisulphite, point-of-care testing, transgene screening, bioinformatics