Deep Learning‐Enhanced Nanopore Sensing of Single‐Nanoparticle Translocation Dynamics

Abstract

More specifically, whereas the noise in nanopore systems was elucidated to stem from several sources such as surface charge fluctuations and dielectric loss, coupling between the device capacitance and the voltage noise in current amplifiers was found to be the most significant factor giving rise to high-frequency noise above 10 kHz. [25][26][27] Previous nanopore measurements often used low-pass filters to cut-off this fast noise for detecting resistive pulse signals, which has proven useful in studying translocation dynamics of small molecules. [28,29] Nonetheless, as the sensitivity of the nanosensor was improved by employing ultrathin membranes such as graphene [30,31] and MoS 2 , [19] it started to become possible to probe not only the size of objects but their shapes, surface charge densities, and even mass from the ionic current signals. [8,32] In this context, it turns out to be an important issue to reduce the high frequency noise without any pre-filters as it generally involves signal blunting that obscures the fine yet important ionic current profiles. One of the effective strategies was to employ highly insulating materials as low-capacitance substrates, for example quartz and polymers. [33] Later, it was also found that covering the membrane surface with a thick polymer layer can serve to reduce the noise. [34] These works indeed provided nanopore chip designs for diminishing the current fluctuations by more than an order of magnitude through tailoring surface reactions and capacitive coupling. [27,[34][35][36][37] Besides the material and device engineering for controlling the physical/chemical phenomena relevant to ionic current fluctuations, digital post processing has been utilized to mitigate the noise via band-pass filtering and wave transformations. [38,39] However, previous studies have found it to be a non-trivial task since the computation in frequency domains inevitably entails signal distortions thereby obscured the small yet important features occurring at variable time scales due to the stochastic and random nature of the translocation dynamics. [39] Here we report on a novel concept for post-denoising of ionic current in solid-state nanopores. Our method is based on a deep learning algorithm formulated to extract noise floor from ionic current (I ion ) versus time (t) curves without clean data (Figure 1). This strategy is known as Noise2Noise [40] proven to be useful in processing noisy digital images. [41] In the present study, we exploited it to demonstrate clarification of fine signatures in a resistive pulse signal that may otherwise be immersed in noise or completely smeared out under the conventional signal processing.Noise is ubiquitous in real space that hinders detection of minute yet important signals in electrical sensors. Here, the authors report on a deep learning approach for denoising ionic current in resistive pulse sensing. Electrophoretically-driven translocation motions of single-nanoparticles in a nanocorrugated nanopore are detected. The noise is reduced by a convo...

A Guide to Signal Processing Algorithms for Nanopore Sensors

“…A Deep Neural Network (DNN) can be adopted to filter out the noise from the signals generated by carboxylated polystyrene nanoparticles translocating a 5-μm-long nanochannel. 62 In such an algorithm, the time sequence traces as signals are first sent to a convolutional autoencoding Neural Network (NN) that repeats the convolution of input and passes on the features to the next stage, i.e., an activation function of either rectified linear unit (ReLU) or LeakyReLU. This operation converts current traces into vectors in a high-dimensional feature-enhanced space by keeping the features and dropping the time resolution.…”

“…This solution can considerably improve the performance of system tasks. By taking such a strategy, DL has been applied to solve analyte classification with automatically extracted features, , translocation waveform regression and identification, , and noise recognition and elimination , in nanopore sensing. Yet, DL has its own drawbacks that render it difficult to implement in some scenarios.…”

AutoNanopore: An Automated Adaptive and Robust Method to Locate Translocation Events in Solid-State Nanopore Current Traces

“…Following the event detection results, machine-learning models have been proposed to analyze the detected events: e.g., Carral et al developed a deep-learning method to distinguish single nucleotides at high accuracies and Xia et al recently proposed a machine-learning-based method to classify signals generated by four synthetic glycosaminoglycans through solid-state nanopores . There is also another study where a deep-learning method was developed for postdenoising of ionic current in a nanofluidic channel having five pairs of nanoprotrusions …”

“… 23 There is also another study where a deep-learning method was developed for postdenoising of ionic current in a nanofluidic channel having five pairs of nanoprotrusions. 24 …”

Translocation of Specific DNA Nanocarrier through an Ultrasmall Nanopipette: Toward Single-Protein-Molecule Detection with Superior Signal-to-Noise Ratio

“…Nanopore technology offers bright prospects for single-protein-molecule detection (SPMD) by measuring the resistive pulse sensing (RPS) signal produced from the change in the circuit resistance as proteins pass through a single nanopore under an applied potential. , Although plentiful electrical signals can be collected through nanopore measurements, the lack of selectivity limits the identification of protein molecules in complex samples . Loading proteins with functional carriers can promote the measurement selectivity. − Among various carriers, nucleic acid nanostructures have been shown to be promising nanocarriers to ship proteins in SPMD. Actis et al have recently pioneered that specific hollow DNA origami can capture and identify the human C-reactive protein with high selectivity.…”