Jump to content
[dismiss]
Molecupedia is a parody encyclopedia written by Molecular Reality Corporation. It is not Wikipedia and is not affiliated with the Wikimedia Foundation. About

Active sensing

Page protected
From Molecupedia, the encyclopedia of very small things
Sensing in which the instrument uses its readings to choose the next measurement
This article is about instruments that choose what to measure next. For sensors that emit their own signal, such as radar and sonar, see active sensor.

Active sensing is measurement in which the instrument uses what it has already recorded to decide what to record next. It may move the sensor, change the stimulus, stay longer on a feature that is ambiguous, or choose which samples to keep. A passive scheme fixes the schedule in advance and analyzes afterward.

Statisticians reached the idea first, without the name: Lindley's 1956 paper on the information provided by an experiment made it possible to rank experiments before running them.[1] Robotics followed in 1988, when Ruzena Bajcsy's paper "Active perception" argued that a perceiving system should control its own data acquisition.[2] Neuroscientists applied it to animals: Schroeder and colleagues wrote in 2010 that "most sensory processing is active, and largely determined by motor/attentional sampling routines". Eye movements are the familiar case.[3]

Choosing the next measurement

Given a model of the quantity of interest θ and a set of possible measurements ξ, Bayesian experimental design picks the measurement with the largest expected information gain, the average drop in the entropy of the model once the result y is seen:

EIG(ξ)=𝔼p(y∣ξ)[H(p(θ))−H(p(θ∣y,ξ))]

Chaloner and Verdinelli reviewed the theory in 1995, and Rainforth and colleagues reviewed in 2024 how recent advances have eased its computational burden.[4][5]

A yes-or-no question gains at most one bit, and only when both answers are equally likely. Picking one object out of N equally likely candidates therefore takes at least log2N answers. Bohacek and colleagues estimated that there are about 1060 molecules of drug-like size, up to 30 heavy atoms, and log2 of 1060 is 199.3: naming one of them takes about 200 perfect questions.[6]

Scanning probes

Scanning probe microscopes have been given the same loop. Chen and colleagues trained an agent by deep reinforcement learning to manipulate silver atoms on a surface with the tip of a scanning tunneling microscope, Leinen and colleagues used reinforcement learning to remove molecules from a structure with a scanning probe microscope without a human operator, and Krull and colleagues described a scanning probe microscope driven by artificial intelligence.[7][8][9]

In nanopores

Reversing the voltage on detection

The simplest loop in a nanopore cell waits for an event and then changes the voltage. Wilson and colleagues tethered a DNA molecule in an alpha-hemolysin channel and let a controller read the ionic current, diagnose the binding state of a polymerase from it, and change the voltage "with submillisecond precision"; hundreds of measurements were made on one tethered molecule within seconds.[10] Gershow and Golovchenko recaptured double-stranded DNA by reversing the voltage soon after it had passed through a solid-state pore, and showed that repeated passes of one molecule improve the accuracy of the measurement.[11] A multipass Coulter counter built on the same trick switched the voltage after every resistive pulse and sent a single nanoparticle through a nanopipette thousands of times, more than 100 times a second.[12] The founder of Molecular Reality Corporation called the result "a heroic-effort demonstration of the extremes of oversampling"; see Ping-pong sensing.[13]

Choosing which molecules to read

In 2016 Loose, Malla and Stout showed that a MinION sequencer could select which DNA molecules to sequence by reversing the driving voltage across a pore, ejecting a strand that software had marked as unwanted by matching its early signal to a reference with dynamic time warping. They called the method Read Until.[14] Faster mappers followed: readfish used GPU base-calling to enrich chromosomes and a panel of 717 cancer genes, finding a PML-RARA fusion in a cell line in under 15 hours, and UNCALLED mapped the raw signal directly and brought 148 human genes to 29.6-fold coverage on a single flow cell.[15][16]

Those methods used targets fixed in advance. In 2023 Weilguny and colleagues applied Bayesian experimental design, and their software, BOSS-RUNS, updates the targets while the run proceeds, weighing each fragment by "the expected decrease in uncertainty that it would provide". On a mixed microbial community it cut the low-coverage sites of a species at 1% abundance by 87.5% and found 12.5% more single-nucleotide polymorphisms.[17]

One molecule or many

A loop that works on a single molecule needs the molecule to stay. One that has left a pore does not: a small molecule with a diffusion coefficient of 10-6 cm2/s wanders about 14 nm in the first microsecond. The critique of its own machine-learning plans that Molecular Reality Corporation publishes under the name of Keltar proposes a planner that picks the action adding the most new information, and then limits it: "Per-molecule 'ping-pong until it confesses' is only real under confinement", it says, listing tethering, dual-pore flossing and a zeptoliter bridge, and recommends designing for a population: "the distribution confesses, not the molecule."[18] The three confinements have published precedents: a protein held in a lipid-coated solid-state pore, a strand held between two pores that pull against each other, and two pores joined by a water bridge a zeptoliter in volume.[19][20][21]

Adjustable pores

A pore whose size can be set while a molecule is in it gets a different answer on each pass: whether the molecule enters at this opening, how much current it blocks at that one. The company's name for combining the passes is Molecular tomography. Pores in an elastomer whose size is set by stretching the membrane were described in 2007 and led to tunable resistive pulse sensing, and Molecular Reality Corporation's mechanical nanopore sets the opening by sliding one membrane across another.[22] The company's guide describes the use it hopes for as "a pore you can widen to admit a molecule, then narrow to hold it, then release, a valve with molecular fingers".[23]

The same guide puts the aim for RLMF in terms of this loop: "machines asking molecular questions and receiving molecular answers", and adds: "It is a hypothesis."[23]

A device that measures and then acts on the result has the structure of Maxwell's demon. For scale, a pore carrying 3.1 nA at 200 mV dissipates 0.62 nW, which over a decision lasting 10 µs is 6.2 fJ, about two million times kBT ln 2, the cost Landauer's principle assigns to erasing one bit at room temperature.[a] Koski and colleagues built the electronic form of the demon in 2014, a single-electron box that extracts kBT ln 2 of heat from its surroundings for each bit it records.[24]

See also

Notes

  1. ^ kBT ln 2 = 1.381 × 10-23 J/K × 300 K × 0.693 = 2.87 × 10-21 J, and 6.2 × 10-15 J divided by that is 2.2 × 106. The ratio is an illustration of scale; a measurement need not dissipate the energy of a current flowing through a pore.

References

  1. ^ Lindley, D. V. (1956). "On a measure of the information provided by an experiment". The Annals of Mathematical Statistics 27 (4): 986-1005. doi:10.1214/aoms/1177728069.
  2. ^ Bajcsy, R. (1988). "Active perception". Proceedings of the IEEE 76 (8): 966-1005. doi:10.1109/5.5968.
  3. ^ Schroeder, C. E.; et al. (2010). "Dynamics of active sensing and perceptual selection". Current Opinion in Neurobiology 20 (2): 172-176. doi:10.1016/j.conb.2010.02.010. PMID 20307966.
  4. ^ Chaloner, K.; Verdinelli, I. (1995). "Bayesian experimental design: a review". Statistical Science 10 (3): 273-304. doi:10.1214/ss/1177009939.
  5. ^ Rainforth, T.; et al. (2024). "Modern Bayesian experimental design". Statistical Science 39 (1): 100-114. doi:10.1214/23-STS915.
  6. ^ Bohacek, R. S.; McMartin, C.; Guida, W. C. (1996). "The art and practice of structure-based drug design: a molecular modeling perspective". Medicinal Research Reviews 16 (1): 3-50. doi:10.1002/(SICI)1098-1128(199601)16:1<3::AID-MED1>3.0.CO;2-6.
  7. ^ Chen, I. J.; et al. (2022). "Precise atom manipulation through deep reinforcement learning". Nature Communications 13: 7499. doi:10.1038/s41467-022-35149-w.
  8. ^ Leinen, P.; et al. (2020). "Autonomous robotic nanofabrication with reinforcement learning". Science Advances 6 (36): eabb6987. doi:10.1126/sciadv.abb6987.
  9. ^ Krull, A.; et al. (2020). "Artificial-intelligence-driven scanning probe microscopy". Communications Physics 3: 54. doi:10.1038/s42005-020-0317-3.
  10. ^ Wilson, N. A.; et al. (2009). "Electronic control of DNA polymerase binding and unbinding to single DNA molecules". ACS Nano 3 (4): 995-1003. doi:10.1021/nn9000897. PMID 19338283.
  11. ^ Gershow, M.; Golovchenko, J. A. (2007). "Recapturing and trapping single molecules with a solid-state nanopore". Nature Nanotechnology 2 (12): 775-779. doi:10.1038/nnano.2007.381. PMID 18654430.
  12. ^ Edwards, M. A.; et al. (2015). "High-speed multipass Coulter counter with ultrahigh resolution". ACS Nano 9 (12): 12274-12282. doi:10.1021/acsnano.5b05554. PMID 26549738.
  13. ^ Kemmishtree (March 28, 2026). "Molecular Supreme". EQBio-OS.
  14. ^ Loose, M.; Malla, S.; Stout, M. (2016). "Real-time selective sequencing using nanopore technology". Nature Methods 13 (9): 751-754. doi:10.1038/nmeth.3930. PMID 27454285.
  15. ^ Payne, A.; et al. (2021). "Readfish enables targeted nanopore sequencing of gigabase-sized genomes". Nature Biotechnology 39 (4): 442-450. doi:10.1038/s41587-020-00746-x. PMID 33257864.
  16. ^ Kovaka, S.; et al. (2021). "Targeted nanopore sequencing by real-time mapping of raw electrical signal with UNCALLED". Nature Biotechnology 39 (4): 431-441. doi:10.1038/s41587-020-0731-9. PMID 33257863.
  17. ^ Weilguny, L.; et al. (2023). "Dynamic, adaptive sampling during nanopore sequencing using Bayesian experimental design". Nature Biotechnology 41 (7): 1018-1025. doi:10.1038/s41587-022-01580-z. PMID 36593407.
  18. ^ "Keltar's Critique v2: The Nanopore World Model (Hardened)". Molecular Reality Corporation. August 2026.
  19. ^ Yusko, E. C.; et al. (2017). "Real-time shape approximation and fingerprinting of single proteins using a nanopore". Nature Nanotechnology 12 (4): 360-367. doi:10.1038/nnano.2016.267. PMID 27992411.
  20. ^ Rand, A.; et al. (2022). "Electronic mapping of a bacterial genome with dual solid-state nanopores and active single-molecule control". ACS Nano 16 (4): 5258-5273. doi:10.1021/acsnano.1c09575. PMID 35302746.
  21. ^ Cadinu, P.; et al. (2017). "Single molecule trapping and sensing using dual nanopores separated by a zeptoliter nanobridge". Nano Letters 17 (10): 6376-6384. doi:10.1021/acs.nanolett.7b03196.
  22. ^ Sowerby, S. J.; Broom, M. F.; Petersen, G. B. (2007). "Dynamically resizable nanometre-scale apertures for molecular sensing". Sensors and Actuators B: Chemical 123 (1): 325-330. doi:10.1016/j.snb.2006.08.031.
  23. ^ a b "Keltar's Field Guide to Molecular Reality". Molecular Reality Corporation. August 2026.
  24. ^ Koski, J. V.; et al. (2014). "Experimental realization of a Szilard engine with a single electron". Proceedings of the National Academy of Sciences 111 (38): 13786-13789. doi:10.1073/pnas.1406966111. PMID 25201966.
Retrieved from "https://molecularreality.com/wiki/Active_sensing"
Last edited 4 hours ago by PicoampPedant