Virus detection with nanopores

Virus detection with nanopores is the measurement of whole virus particles as they pass through a pore in a membrane. Each particle displaces conducting liquid and so changes the ionic current for a moment, the principle of the Coulter counter, and the size, shape and surface charge of the particle shape the pulse. Two kinds of result have come from it: counts and sizes of virions, available since the 1970s, and identification of the type of virus from the pulse, shown in the 2010s with the help of machine learning. Neither reads the viral genome. Reading genomes is the work of nanopore sequencing, which is used in outbreaks and is described in Biosurveillance.

The method is as old as the resistive-pulse technique. In 1970 DeBlois and Bean counted and sized submicrometer particles in pores of less than a micrometer, and in 1977 DeBlois and Wesley applied the method to viruses. They passed type C oncornaviruses and the bacteriophage T2 through single pores made by the Nuclepore process and sized viruses above about 60 nm to within a few nanometers in their hydrated state.^[1][2]

The same paper gave a volume of 5.10 × 10⁻¹⁶ cm³ for T2. Concentrations of 10⁹ to 10¹¹ particles per milliliter that were reasonably clear of debris could be measured in a few minutes to about 15%, and the practical lower limit was near 5 × 10⁷ per milliliter.^[2] A sphere of diameter d in a cylindrical pore of diameter D and length L raises the resistance, to leading order, by d³ / (D² L) of its baseline. For a 100 nm particle in a pore 300 nm wide and 1 µm long that is about 1%.

A microfluidic analyzer of 2011 could handle 500,000 particles a second, and it sized and titered bacteriophage T7 in salt solution and in mouse blood plasma from about a microliter.^[3] Hepatitis B capsids assembled in vitro, of 31 nm and 36 nm, were told apart in conical pores about 40 nm across at the tip, and the larger capsids took longer to cross, which suggested that a capsid must overcome an entropic barrier to enter a pore.^[4] The filamentous bacteriophage fd is too stiff to fold, so it crosses a solid-state pore lengthwise, with a mobility that does not depend on the voltage.^[5] A review of 2016 covers the quantification of virus particles with these techniques.^[6]

Size alone does not separate viruses of one family. In 2018 a group at Osaka University took three strains of influenza grown in chicken eggs, A/Puerto Rico/8/34 (H1N1), A/Hong Kong/8/68 (H3N2) and B/Lee/40, which are of common size and spherical shape and differ in their surface proteins. The pore was a hole 300 nm across in a silicon nitride membrane 50 nm thick. Its wall carries a negative charge (a zeta potential of −42 mV at pH 7.6), and the resulting electroosmotic flow filters weakly charged contaminants at the orifice, so that the negatively charged virions are sensed selectively. A classifier trained on the ionic-current signatures assigned the strain with 68% accuracy for a single virion, and with more than 95% accuracy when the pulses of more than 20 virions were pooled.^[7]

In 2021 the same laboratory described a platform of three parts: machine-learning software on a server, a portable instrument for measuring current at high speed and precision, and a pore module made by semiconductor processes. The pore chips are 5 mm squares cut from 12-inch wafers, each with a hole about 300 nm across in silicon nitride 50 nm thick, against coronaviruses of 80 to 120 nm, and 15 µL of buffer and of specimen go into a module 25 mm across. The instrument applied 0.1 V, amplified the current by 10⁷ and sampled at 1 MHz. A random forest algorithm classified the pulses by features such as peak current and duration.^[8]

It identified four coronaviruses of similar size (HCoV-229E, SARS-CoV, MERS-CoV and SARS-CoV-2). After training on 40 saliva samples that were positive by PCR and 40 that were negative, it detected SARS-CoV-2 in saliva with a sensitivity of 90% and a specificity of 96% in a measurement of five minutes. Unlike the usual reverse-transcription PCR test, it needed no extraction of RNA. The authors point out that a sensitivity of 90% means a false-negative rate of 10%, which could suit screening tests that handle many people.^[8] The company that supplied the software, Aipore, concluded its activities in 2025.^[9]

A pulse reports physical features of a particle, and its molecular identity has to be inferred from them. A classifier recognizes the viruses it was trained on, under the conditions of the training, and the published demonstrations cover a handful of types.

Concentration is the other limit, and the 2021 paper measured it. Cultured HCoV-229E at 100 plaque-forming units per microliter gave 14.2 pulses a minute at −0.1 V and, with the bias reversed, one waveform in 15 minutes. At 2.5 plaque-forming units per microliter a measurement of 15 minutes gave an average of three waveforms, and at 0.25 it gave none, which put the detection limit at 2.5 per microliter, or 2,500 per milliliter.^[8] That is eight to forty-four times the 303 and 57 copies per milliliter of SARS-CoV-2 RNA found in the sewage of one Massachusetts plant in March 2020, with the caveat that plaque-forming units and RNA copies are different units; the sewage figures are in Wastewater surveillance. The count limit of the 1977 instrument, near 5 × 10⁷ per milliliter, was about 20,000 times higher.

Virus identification on solid-state pores is the work of others. Molecular Reality Corporation credits the Osaka paper of 2018 as "a key paper that gave us confidence" in January 2020, when the company tried to launch its Molecular Aquarium because of COVID-19; the effort left a badge called CoronaPore.^[10] The company's own machine-learning result concerns DNA barcodes, not viruses: in April 2026 a Conformer classified barcode events from a public data set better than a retrained convolutional network, a result that is unpublished.^[11] Its long-range goal, Utility-scale molecular sensing, and the Planetary Bioshield of molecular sentinels that it is meant to supply, are stated as aims.^[12]

• Resistive pulse sensing
• Machine learning on nanopore signals
• Aipore
• Wastewater surveillance