QuipuNet
| QuipuNet | |
|---|---|
| Developer(s) | Molecular Reality Corporation |
| Initial release | July 2026 |
| Written in | JavaScript, Python |
| Operating system | Web browser |
| Type | Viewer for nanopore current traces |
| Website | molecularreality.com/Q |
QuipuNet is a full-screen web viewer for the electrical traces of single DNA molecules passing through nanopores, published by Molecular Reality Corporation in July 2026. It shows events from a public dataset of DNA carriers marked with three-bit barcodes, 84,156 events in eight classes, drawn one at a time like the screen of an oscilloscope.[1][2] The viewer takes its name from QuipuNet, a neural network published in 2018 to classify the same events, whose authors compared the barcodes to "the knots on a string used in the Inca Quipu system".[3]
Alongside the company's live molecular stream and its Streams archive, the two halves of what it calls molecular streaming, the viewer is one of its public windows on nanopore data, and the only one that shows another laboratory's measurements.
Background
The dataset
The events come from a study published in Nature Nanotechnology in 2016, in which a long DNA carrier was built with a barcode in its first half and a site that binds a target protein in its second.[4] Each bit of the barcode is a group of DNA hairpins that is either present or absent. Where it is present, the carrier is thicker, and as the molecule threads through the pore the blockade deepens briefly into a spike. Three bits give eight designs, numbered 000 to 111. In the company's register of the designs, some carriers hold biotin, which catches the protein streptavidin, and some digoxigenin, which catches an antibody.[2] The design is a nanopore DNA barcode.
The network and the name
In 2018 a convolutional network called QuipuNet learned to read the barcodes, with a precision of 0.946 on all the data, and to recognize the bound protein with a precision of 0.971. Its authors reported that it raised the number of usable events about fivefold. The paper's one reference to the Andes is its comparison of the barcode to a quipu, the knotted cords on which the Inca state kept its records.[3] A later network trained on the same data, reported in 2024 with an accuracy of 95.8 percent, kept the theme: it is called YupanaNet, after the yupana, an Inca counting board.[5]
The company uses the dataset as its public test bed. In unpublished work by its AI lead, Jade O'Neill, a Conformer classifier read the barcodes with 85.4 percent accuracy, against 81.0 percent for a retrained copy of the QuipuNet network, on a test that held out whole experiments.[2][6] Testing on public data before collecting one's own is the first step proposed in the company's essay The Nanopore World Model.[7]
The company had already read a literal string of knots. In March 2026 its founder described pulling a nylon filament with knots tied in it through a hole about 2 mm wide and recording "downward (resistive) spikes as the knots we tied in a nylon filament are drawn through" it.[8] The demonstration is described in knot sequencing.
Interface
The toolbar carries the name "QuipuNet" in teal, then one button for each barcode with its count of events, and arrows that step through the events of the chosen barcode while a counter shows the position. Arrow keys do the same, and the keys Alt+0 to Alt+7 jump between barcodes. A box marked "Σ" lays over the trace a band showing the mean and standard deviation of every event with that barcode, each stretched to a common length. "Flip" reverses the trace, for molecules that entered the pore tail first. The scale can be raw or normalized three ways, "Z-Score" being the default. A button switches the horizontal axis between samples and milliseconds, labelled "(ms @ 100 kHz)": the viewer assumes ten microseconds per sample.[1]
The mouse wheel zooms around the cursor, dragging pans, and two fingers zoom on a phone. The page's address records the barcode, the event and the normalization, so that a view can be shared. Pointing at the trace reads out the sample number and value. The trace is drawn in bright aqua with a soft glow.
Regions
With "Regions" switched on, the screen is divided into nine colored bands with codes along the top: ENT for the entry, B1, G1, B2, G2 and B3 for the three bits and the gaps between them, LNK for the linker, SEN for the sensing region and EXT for the exit. The bands are fixed fractions of each event's length. The entry is the first 8 percent, bit 1 runs from 10 to 18 percent, the sensing region from 65 to 85 percent and the exit is the last 8 percent, on every event, whatever the molecule did.[a] Small marks labelled "hp" show where hairpins are expected.[1]
Status
In late October 2026 the viewer opened without its dataset. The barcode buttons did not appear, the screen stayed dark and the status bar read "Ready".[9]
See also
Notes
- ^ Nineteen percent of each event lies in no band: the gaps from 8 to 10 percent, from 55 to 65 percent and from 85 to 92 percent of its length.
References
- ^ a b c "QuipuNet". Molecular Reality Corporation. July 2026.
- ^ a b c "Company records". Molecular Reality Corporation. October 2026.
- ^ a b "QuipuNet: convolutional neural network for single-molecule nanopore sensing". Nano Letters 18 (6): 4040-4045. 2018. doi:10.1021/acs.nanolett.8b01709.
- ^ "Digitally encoded DNA nanostructures for multiplexed, single-molecule protein sensing with nanopores". Nature Nanotechnology 11: 645-651. 2016. doi:10.1038/nnano.2016.50.
- ^ "Brownian motion data augmentation: a method to push neural network performance on nanopore sensors". bioRxiv. 2024. doi:10.1101/2024.09.10.612270.
- ^ "ai-experiments". GitHub. Retrieved October 25, 2026.
- ^ "Keltar's Critique v2: The Nanopore World Model (Hardened)". Molecular Reality Corporation. June 2026.
- ^ Kemmishtree (March 29, 2026). "Sequencing a string". EQBio-OS.
- ^ "QuipuNet". Molecular Reality Corporation. Retrieved October 25, 2026.