Wilby Labs

Fold it in software first.

Wilby Labs designs DNA nanostructures and is building a simulator that predicts their shape, motion and assembly yield before a single strand is ordered.

Get in touch Open the 3D viewer
A sheet of twelve DNA double helices, twisted end to end into a bow-tie shape. Staple strands are pastel teal, blue, pink and amber over a pale scaffold.
Our 12-helix test sheet, rendered from its design file with the 100° end-to-end twist the elastic model predicts. The twist is spread evenly along the sheet for display.
Status
In active development, not yet public
Reads
cadnano and scadnano designs
Predicts
Assembly yield, shape and dynamics, with checks for failure modes
Exports
Pre-relaxed oxDNA configurations and staple order sheets

Today you find out at the end. A DNA origami folds when dozens to a few hundred short staple strands pin a long scaffold strand into shape, over an anneal that can run for hours. If the design misfolds, the gel tells you afterward. A good prediction moves that answer to the start. Design tools draw the shape and oxDNA checks the physics, but no general tool we know of takes a design file and predicts how well it will fold. That is what we are building.

The simulator has four layers, from milliseconds to hours.

The fast layers are built to answer most design questions in minutes. The slower ones check them, with oxDNA as the reference at the end. Alongside them, a strand-displacement module predicts toehold kinetics for devices such as walkers and gates.

Sequence

Nearest-neighbor thermodynamics and crosstalk screens

See the design checks

Assembly

Domain-level stochastic folding with loop-closure cooperativity

See the folding report

Mechanics

Base-pair elastic network and Brownian dynamics

See the predicted twist

Nucleotide

oxDNA2 molecular dynamics, on a GPU where possible

See the oxDNA export

The folding model follows every staple through the anneal.

It works at the level of binding domains1,2. Staples bind and unbind one domain at a time, and each bound domain makes the next loop easier to close. A staple can also bind a wrong scaffold site that matches an unbroken run of eight or more of its bases, where it blocks the staple that belongs there. The report gives the assembly yield, the share of simulated anneals that end with every staple bound (for now an upper bound, since staple defects and edge effects are not yet modeled), along with where in the anneal the structure folds and how often each staple ends up bound.

rectangle_12h 12 helices, 67 staples, 2,256 base pairs

Model output

Staple map of the test sheetTwelve helices drawn as rows. The scaffold runs back and forth through all of them. 39 staples have a domain that melts at 55 °C or above, 25 have their strongest domain between 45 and 55 °C, and 3 have every domain below 45 °C: ST2[187]1[187], ST6[187]5[187], ST7[25]5[24]. Those three are circled. Staple map of the test sheetTwelve helices drawn as columns. The scaffold runs back and forth through all of them. 39 staples have a domain that melts at 55 °C or above, 25 have their strongest domain between 45 and 55 °C, and 3 have every domain below 45 °C: ST2[187]1[187], ST6[187]5[187], ST7[25]5[24]. Those three are circled.
  • Strongest domain melts at 55 °C or above
  • 45 to 55 °C
  • Every domain melts below 45 °C
  • Scaffold

Share of staples fully bound as the anneal cools

After 5 minutes at 80 °C, cooled to 65 °C at 1 minute per degree, then to 25 °C at 10 minutes per degree, in 40 mM Tris with 12.5 mM magnesium. The line joins readings taken once per degree.

What the report found

  • All 16 simulated anneals ended with every staple bound. Read that as an upper bound. The model assumes every staple is made perfectly and binds as well at the edges as in the middle. Measured on real origami, staple incorporation averages about 84 percent per site and 48 percent at the edges3. At no recorded temperature was any staple on a wrong scaffold site.
  • Three staples bind weakly. In the simulation they still bind in every anneal. The flag comes from their melting temperatures instead: none of their domains melts at 45 °C or above, so the report suggests re-cutting them. ST2[187]1[187]ST6[187]5[187]ST7[25]5[24]
  • Five strong off-target matches. Each is at least 70 percent as strong as its staple’s strongest intended domain. In the simulated anneals no staple stayed on a wrong site, so the report ranks them as low priority. See the design checks
Predicted for the test sheet. Temperatures are uncalibrated: the model has been compared with one published measurement, on a similar sheet under a different anneal and buffer, and not yet with our own lab data. Read the shape of the curve and the ranking between designs before the absolute numbers.
Watch an anneal in 3D

The test sheet twists because of how it was drawn.

A base-pair elastic network, built on the physics of CanDo4, relaxes the design in 3D and reports its size, global twist and how much each part fluctuates. This sheet is laid out at 10.67 base pairs per turn, looser than the 10.5 that DNA prefers, so it winds up along its length. Deleting three base pairs from each inner helix, inside the stretch its crossovers hold, brings it to 10.47, and the sheet relaxes nearly flat.

The test sheet as laid out in the design file. It is flat. The sheet at the predicted twist of 155 degrees per 100 nanometers, right-handed. It curls strongly from one end to the other. The sheet with the deletions, at the predicted 17 degrees per 100 nanometers, left-handed. It lies nearly flat.

FlatAs laid out in the design file, before any mechanics.

155° per 100 nmFor the sheet as drawn: about 100° from one end to the other, right-handed.

−17° per 100 nmNearly flat, with a slight left-handed twist.

Turn the sheet in 3D
Global twist from the elastic model, rendered from the design file at each value and from the same angle. Each render spreads the twist evenly for display. Positive values are right-handed.
Close-up of five neighboring helices in the test sheet at nucleotide resolution. Each nucleotide is a bead, as oxDNA represents it, and staples cross between helices.

Dynamics run at two levels of detail.

Brownian dynamics of the elastic model shows how much each part of a structure moves, and a faster harmonic analysis of the same model gives how far a chosen site wobbles and how often two sites meet. When a question needs single-nucleotide detail, the simulator writes a pre-relaxed configuration for oxDNA25 and runs it. Starting relaxed avoids the large initial strains of a raw export: the largest bond is 1.9 oxDNA units, against 5.4 in scadnano’s own export.

Watch the sheet move in 3D

Design checks cover every staple.

Off-target matches on the scaffold and staple dimers are scored with the same nearest-neighbor energies6 as the intended binding, so the two can be compared directly. The strongest scaffold matches also take part in the folding simulation as wrong sites a staple can bind. Hairpins are folded with the DNA parameters in ViennaRNA, and runs of four or more G are flagged from the sequence. The report names the strands involved and, for weak staples and off-targets, what to change.

Excerpt from the report for the test sheet. Free energies at 50 °C.

Weak staples
Three staples have no domain that melts at 45 °C or above. The report suggests re-cutting them so that each has one domain stable enough to start binding on its own.ST2[187]1[187] · ST6[187]5[187] · ST7[25]5[24]
Scaffold off-targets
Five matches are at least 70 percent as strong as the staple’s strongest intended domain. The strongest, on ST2[168]4[169], binds at −9.2 kcal/mol. That is stronger than two of the staple’s own three domains and 71 percent of its strongest (−12.9).ST2[168]4[169] ACGACGGCC → scaffold 713
Hairpins
Nine staples form a hairpin of −1 kcal/mol or stronger at 55 °C, the cool end of the folding window, and three still do at 70 °C. The strongest, in ST4[40]6[41], is −4.7 kcal/mol at 55 °C.ST4[40]6[41] .((((((((.........))))))))......
Runs of G
Eight runs of four or more G, each a G-quadruplex risk.ST6[187]5[187] GGGG · ST5[25]3[24] GGGGG · ST10[40]10[57] GGGGG

Every layer is checked against a reference, but not yet against our own lab data.

The simulator is in active development. A reference can be an exact result, a second implementation, a published measurement or oxDNA. The folding model has been compared with one published folding temperature. The open rows are the next work, starting with an annealing curve we measure ourselves.

CheckCompared withResult
Duplex melting temperatures, four sequences in three buffersAn independent implementation of the same nearest-neighbor6 and salt-correction7 parameters (Biopython)Agrees within 0.05 °C
Folding model, detailed balance, with and without off-target bindingThe exact Boltzmann distribution of each staple’s bound statesAgrees to one part in a million
Folding temperature of a single-layer origami under a published annealStaple incorporation measured by Dunn and colleagues2, cooling at 0.3 °C per minute in 12.5 mM magnesium65.0 °C predicted, against 63.9 to 64.0 °C measured for staples in the body of the tile, with default parameters and no fitting. The two published magnesium salt models the simulator can use differ by 2.5 °C here and bracket the measurement. The model folded a similar 24-helix rectangle, not their exact design.
Strand-displacement rate against toehold lengthThe trend measured by Zhang and Winfree8, whose three-step model the simulator usesFollows the steep rise for short toeholds, about tenfold per added base. Not yet a validation: the plateau near is an input, and the rate at each toehold length has not been compared.
Twist and bend of honeycomb bundlesElectron microscopy by Dietz and colleagues9Twist at 10, 10.5 and 11 base pairs per turn: −65.1, +3.6 and +69.4° per 100 nm, against −76.6 ± 10.4, about 0 and +62.9 ± 10.6 measured. A bundle designed to bend 120° relaxes to 111°, against 121 ± 8.4° measured. The crossover stiffness was fitted to all four bundles.
Twist of an earlier version of the test sheet, with and without deletionsoxDNA2, 2.7 ns of dynamics, started from shapes relaxed by an earlier version of the modelWithout the deletions, 150 against 133° per 100 nm; with them, 39 against 55. The oxDNA twist was still falling when the runs ended. Both sheets and the crossover stiffness have changed since, and the comparison has not been repeated.
Site wobble from the harmonic analysisBrownian dynamics of the same elastic model, on an 8-helix bundleWithin 2 percent on average, over 24 staple ends. On a block with the crossover spacing used in the lab, Brownian dynamics gives about a fifth less wobble in the middle and over a third less near helix ends, so there the harmonic numbers are an upper estimate.
Relaxed size of a separate, earlier test design, before the crossover stiffness was fittedoxDNA2, same design, same starting structureLength within 1 percent. Width 9 percent narrower: oxDNA spaces helices about 10 percent wider at 0.5 M salt. 61.8 × 26.9 × 6.6 nm, against 62.4 × 29.7 × 6.3 nm
oxDNA exportscadnano’s own oxDNA exportSame conventions, with the largest bond at 1.9 oxDNA units against 5.4
Folding temperature of our own designs, in our bufferAn annealing curve we measureOpen
Design-dependent yield changesPublished folding experiments by Dunn and colleagues2Open
Predicted yieldGel yields of our own designsOpen
Duplex melting temperaturesMeasured melting tables, rather than another implementationOpen
Bundle stiffness and site wobblePublished stiffness and positioner measurementsOpen

Next

  • Calibrate folding against annealing curves and gel yields from our own designs
  • Staple synthesis defects and edge effects, so that predicted incorporation can match what is measured on real origami
  • Hinges, rotors and walkers, combining mechanics with strand displacement
  • Partial misfolds and staple dimers inside the folding simulation, which already handles whole staples on wrong sites
  • Scaffold-free assemblies such as DNA bricks10 and tiles
  • Staple breakpoints and scaffold rotation chosen for predicted yield

Known limits

  • Absolute folding temperatures are checked against one published measurement only. Other buffers, anneal rates and 3D designs are unchecked, so until we measure an annealing curve of our own, rankings between designs are more reliable than single numbers.
  • Staple dimers and partial misfolds, where one domain sits on its site and another on a wrong one, are screened but not simulated.
  • Aggregation, scaffold secondary structure and errors in staple synthesis are not modeled yet.

Each layer starts from a published model.

We keep an annotated library of more than 60 papers, from DNA thermodynamics to molecular machines. We surveyed the tools the field already relies on, and the simulator’s architecture follows from that survey.

What existing tools cover

Design tools such as cadnano11, scadnano12, DAEDALUS, PERDIX, ATHENA and MagicDNA turn a target shape into strand routings and sequences. They don’t say whether the design will fold. Shape predictors such as CanDo4 and mrDNA13 give fast shape and flexibility, checked against cryo-EM, but not whether the structure assembles at all.

Nearest-neighbor tables, NUPACK14 and ViennaRNA handle duplexes and small complexes, without the 3D context that makes origami folding cooperative. oxDNA5 is the best-validated model at the level of single nucleotides, but folding a whole origami with it is orders of magnitude too slow. Domain-level research code1,2 predicted design-dependent yield changes that experiments then confirmed, but it was written for specific designs.

We haven’t found a tool that connects these, from a design file to a yield that is checked against the lab. That is the gap the simulator is built for. oxDNA stays the reference: we export to it rather than rebuild it.

Papers cited on this page

  1. Modelling DNA origami self-assembly at the domain levelDannenberg et al., Journal of Chemical Physics, 2015
  2. Guiding the folding pathway of DNA origamiDunn et al., Nature, 2015
  3. Quantifying absolute addressability in DNA origami with molecular resolutionStrauss et al., Nature Communications, 2018
  4. A primer to scaffolded DNA origamiCastro et al., Nature Methods, 2011
  5. Introducing improved structural properties and salt dependence into a coarse-grained model of DNASnodin et al., Journal of Chemical Physics, 2015
  6. The thermodynamics of DNA structural motifsSantaLucia and Hicks, Annual Review of Biophysics and Biomolecular Structure, 2004
  7. Predicting stability of DNA duplexes in solutions containing magnesium and monovalent cationsOwczarzy et al., Biochemistry, 2008
  8. Control of DNA strand displacement kinetics using toehold exchangeZhang and Winfree, Journal of the American Chemical Society, 2009
  9. Folding DNA into twisted and curved nanoscale shapesDietz, Douglas and Shih, Science, 2009
  10. Three-dimensional structures self-assembled from DNA bricksKe et al., Science, 2012
  11. Rapid prototyping of 3D DNA-origami shapes with caDNAnoDouglas et al., Nucleic Acids Research, 2009
  12. scadnano: a browser-based, scriptable tool for designing DNA nanostructuresDoty et al., DNA 26, LIPIcs, 2020
  13. MrDNA: a multi-resolution model for predicting the structure and dynamics of DNA systemsMaffeo and Aksimentiev, Nucleic Acids Research, 2020
  14. NUPACK: analysis and design of nucleic acid systemsZadeh et al., Journal of Computational Chemistry, 2011
  15. Folding DNA to create nanoscale shapes and patternsRothemund, Nature, 2006
  16. A synthetic DNA walker for molecular transportShin and Pierce, Journal of the American Chemical Society, 2004
  17. A proximity-based programmable DNA nanoscale assembly lineGu et al., Nature, 2010
  18. A DNA origami rotary ratchet motorPumm et al., Nature, 2022
  19. Molecular engineering: an approach to the development of general capabilities for molecular manipulationDrexler, PNAS, 1981
  20. Nucleic acid junctions and latticesSeeman, Journal of Theoretical Biology, 1982

The long-term aim is molecular machines.

A close view along the twisted test sheet. Pastel strands wind around each helix and the sheet curls away into the dark.

DNA is a material you can program. Sequence decides what binds to what, so a long scaffold strand and a couple of hundred short staples can fold themselves into a chosen shape15. The same rules have already produced walkers16, a molecular assembly line17 and a rotary motor18.

Wilby Labs is working toward molecular assembly in the spirit of Drexler’s assemblers19, built from DNA parts in the tradition Seeman started20 rather than from diamond. The simulator comes first, so that every part is designed with a prediction in hand.

Get in touch.

The simulator isn’t public yet, but we’d like to hear from labs that fold DNA origami, from people who build design tools, and from anyone working toward molecular assembly. If you have folding data we could test our predictions against, we’d especially like to talk.