Signal noise reduction processing method and system

Abstract

The invention relates to the technical field of biology, in particular to a signal noise reduction processing method and system, and the method comprises the steps: collecting an original ion current signal, and constructing an original ion current signal segment and a training set associated with a pure ion current signal segment corresponding to the original ion current signal segment; constructing a noise reduction neural network model, and defining a loss function for the noise reduction neural network model; training trainable parameters of the noise reduction neural network model through the training set and the loss function so as to complete training of the noise reduction neural network; inputting all to-be-detected original ion current signal segments of to-be-detected original ion current signals into the trained noise reduction neural network model, and outputting corresponding to-be-detected noise reduction ion current signal segments; and splicing all the to-be-detected noise reduction ion current signal segments to obtain a complete to-be-detected noise reduction ion current signal. According to the invention, the noise in the original ion current signal can be effectively removed, and the distortion of the ion current signal is avoided, so that the recognition effect of the basic group is ensured.

Description

Technical Field

[0001] This application relates to the field of biotechnology, and in particular to a signal noise reduction processing method and system. Background Technology

[0002] Nanopore sequencing technology identifies bases by measuring changes in ionic current caused by a single DNA or RNA molecule passing through a nanopore. However, the original ionic current signal is susceptible to interference from background noise, baseline drift, and complex interferences, resulting in an extremely low signal-to-noise ratio. Background noise originates from electronic noise caused by the thermal motion of electrolyte ions and the measurement circuitry. Baseline drift is low-frequency noise caused by environmental disturbances and slow changes in the pore state. Complex interferences are structural noise caused by the synergistic effect of multiple bases, molecular oscillation, and non-specific adsorption in the pore.

[0003] Currently, traditional filtering methods (such as low-pass filtering and wavelet transform) are commonly used to process the original ion current signal. Although this can filter out some high-frequency noise, it is difficult to distinguish noise that overlaps with the frequency band of the useful ion current signal, and it is very easy to cause distortion of the ion current signal and loss of key feature information, thereby affecting the recognition effect of bases.

[0004] Therefore, how to effectively remove noise from the original ion current signal while avoiding distortion of the ion current signal to ensure the recognition effect of bases is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] This application provides a signal noise reduction processing method and system to effectively remove noise from the original ion current signal while avoiding distortion of the ion current signal, so as to ensure the recognition effect of bases.

[0006] To solve the above-mentioned technical problems, this application provides the following technical solution: A signal denoising processing method includes the following steps: Step T10: Collect raw ion current signals and construct a training set associated with raw ion current signal segments and their corresponding pure ion current signal segments; Step T20: Construct a denoising neural network model and define a loss function for the denoising neural network model; Step T30: Train the trainable parameters of the denoising neural network model using the training set and the loss function to complete the training of the denoising neural network; Step T40: Input all raw ion current signal segments to be tested into the trained denoising neural network model and output the corresponding denoised ion current signal segments to be tested; Step T50: Concatenate all denoised ion current signal segments to be tested to obtain a complete denoised ion current signal to be tested.

[0007] In the signal denoising processing method described above, preferably, the architecture of the constructed denoising neural network model is an encoder-decoder architecture.

[0008] In the signal denoising processing method described above, the preferred feature is the loss function of the denoising neural network model: ; in, The output of the loss function, This represents all trainable parameters in the denoising neural network model; The first output of the decoder Noise reduction ion current signal segment; For the first in the training set A segment of pure ion current signal; The number of signal pairs in the training set.

[0009] In the signal denoising processing method described above, preferably, multiple segments of the original ion current signal to be tested are extracted from the original ion current signal corresponding to a DNA / RNA molecule according to the base shift events, and all the extracted original ion current signal segments are arranged in chronological order; all the original ion current signal segments are sequentially input into the trained denoising neural network model, and all the output denoised ion current signal segments are arranged in chronological order.

[0010] The signal denoising processing method described above preferably involves splicing all the denoised ion current signal segments to obtain a complete denoised ion current signal, including the following sub-steps: ① Arranging all the denoised ion current signal segments to be tested on a unified time axis in chronological order, and filling the gaps between two adjacent denoised ion current signal segments; ② Extracting the key features of the denoised ion current signal segments to be tested, and using the key features of two adjacent denoised ion current signal segments to be tested as splicing anchor points to obtain a complete denoised ion current signal; ③ Using the slope difference between adjacent sampling points before and after the splicing point, determining whether all splicing points of the complete denoised ion current signal to be tested abruptly change, in order to eliminate splicing errors.

[0011] A signal denoising system includes: a signal collection unit, a model building unit, a model training unit, a denoising processing unit, and a signal splicing unit. The signal collection unit collects raw ion current signals and constructs a training set associated with raw ion current signal segments and their corresponding pure ion current signal segments. The model building unit constructs a denoising neural network model and defines a loss function for the model. The model training unit trains the trainable parameters of the denoising neural network model using the training set and the loss function to complete the training of the denoising neural network. The denoising processing unit inputs all raw ion current signal segments to be measured into the trained denoising neural network model and outputs corresponding denoised ion current signal segments. The signal splicing unit splices all denoised ion current signal segments to be measured to obtain a complete denoised ion current signal.

[0012] In the signal denoising processing system described above, preferably, the architecture of the constructed denoising neural network model is an encoder-decoder architecture.

[0013] In the signal denoising processing system described above, the loss function of the denoising neural network model is preferably: ; in, The output of the loss function, This represents all trainable parameters in the denoising neural network model; The first output of the decoder Noise reduction ion current signal segment; For the first in the training set A segment of pure ion current signal; The number of signal pairs in the training set.

[0014] In the signal denoising system described above, preferably, the denoising unit extracts multiple segments of the original ion current signal to be measured from the original ion current signal corresponding to a DNA / RNA molecule according to base shift events, and all the extracted segments are arranged in chronological order; the denoising unit sequentially inputs all segments of the original ion current signal to be measured into the trained denoising neural network model, and outputs all denoised ion current signal segments in chronological order.

[0015] In the signal denoising processing system described above, preferably, the signal splicing unit includes: a signal arrangement subunit, a signal splicing subunit, and an error elimination subunit; the signal arrangement subunit arranges all the noise-reduced ion current signal segments to be tested on a unified time axis in chronological order and fills the gaps between two adjacent noise-reduced ion current signal segments to be tested; the signal splicing subunit extracts the key features of the noise-reduced ion current signal segments to be tested, and splices them using the key features of two adjacent noise-reduced ion current signal segments to be tested as splicing anchor points to obtain a complete noise-reduced ion current signal to be tested; the error elimination subunit uses the slope difference between adjacent sampling points before and after the splicing point to determine whether all splicing points of the complete noise-reduced ion current signal to be tested have abrupt changes, so as to eliminate splicing errors.

[0016] Compared with the above-mentioned background technology, the signal denoising processing method and system provided in this application construct a large-scale, high-fidelity training set and train a denoising neural network model to learn the essential characteristics of ion current signals under noise-free conditions, thereby achieving high-precision denoising and reconstruction of the original ion current signals. At the same time, it can avoid ion current signal distortion and significantly improve the accuracy and reliability of subsequent base identification. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0018] Figure 1 This is a flowchart of the signal noise reduction processing method provided in the embodiments of this application;

[0019] Figure 2 This is a schematic diagram of the signal noise reduction processing system provided in the embodiments of this application. Detailed Implementation

[0020] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. Example 1

[0021] like Figure 1 As shown, this application provides a signal noise reduction processing method, including the following steps: Step T10: Collect the raw ion current signal and construct a training set that associates the raw ion current signal fragments with the corresponding pure ion current signal fragments; TB-level raw ion current signals were collected from public databases or internal experiments. These raw ion current signals were cleaned to correct identifiable errors. Then, signal fragments corresponding to each base shift event were extracted from the cleaned raw ion current signals. A base shift event is an event in which a DNA molecule moves one base in a nanopore. The signal fragment corresponding to a base shift event is a fragment of the ion current signal generated during the entry, occupation, and exit of a k-mer (a fixed-length nucleotide string) into the nanopore. Each extracted signal fragment contains information on the changes in all ion current signals generated during the residence of a k-mer in the nanopore, including average blockage level, signal shape, duration, and fluctuation pattern.

[0022] Each extracted signal fragment was correlated with the pure ion current signal corresponding to the k-mer in the reference sequence to form a "raw signal-pure signal" signal pair. The reference sequence refers to the standard DNA sequence used to describe base shift events. For example, the 10nt DNA sequence ACTCGATGCT underwent six base shift events in the nanopore. The six base shift events correspond to six k-mers, namely ACTCG, CTCGA, TCGAT, CGATG, GATGC, and ATGCT. These six k-mers all belong to the reference sequence. The pure ion current signals corresponding to the k-mers in the reference sequence are obtained from high-quality annotation or theoretical generation.

[0023] All the "raw signal - clean signal" signal pairs are grouped together to form a training set. .in, For the first For signal pairs; For the first A segment of the original ion current signal. For the first A segment of pure ion current signal, the first Segment of original ion current signal and the Segment of pure ion current signal Combination No. For signal pairs; The number of signal pairs in the training set; ,in, The first time step generated the first time step The first sub-segment of the original ion current signal segment. The second time step generated the first The second sub-segment of the original ion current signal segment. For the first The first time step generated by the first time step The first segment of the original ion current signal Each segment.

[0024] Step T20: Construct a noise reduction neural network model and define a loss function for the noise reduction neural network model; Construct a noise reduction neural network model with an encoder-decoder architecture. This model can capture long-term dependencies in signals.

[0025] For each time step The encoder expression is as follows: ; in, For the first The first time step generated by the first time step The first segment of the original ion current signal Each segment represents the encoder's input value; The encoder is for the first The first segment of the original ion current signal Sub-segment The resulting hidden state, The encoder is for the first The first segment of the original ion current signal Sub-segment The resulting cell state, The encoder is for the first The first segment of the original ion current signal Sub-segment The resulting hidden state, The encoder is for the first The first segment of the original ion current signal Sub-segment The generated cell state, the encoder at the initial time (i.e. The hidden state and cell state are both vectors of all zeros, meaning that at the initial time step... and All are vectors containing only zeros; These are the parameters that the encoder can learn; This is the encoder function.

[0026] After the encoder finishes processing the first time step, proceed to the next... The entire original ion current signal segment at each time step Then, the encoder output is for the first... The first segment of the original ion current signal Hidden state generated by each sub-segment and targeting the The first segment of the original ion current signal Cellular state generated by each sub-segment This hidden state and cell state It is believed to contain the entire segment of the original ion current signal. The summary information will be the hidden state output by the encoder. and cell state as context vector It is passed to the decoder as part of the decoder's input value.

[0027] The decoder expression is as follows: ; in, The first output of the decoder A segment of the denoised ion current signal is used as part of the input value of the decoder for calculation. ; The first output of the decoder Noise reduction ion current signal segment; The decoder is for the first Segment of original ion current signal The resulting hidden state, The decoder is for the first Segment of original ion current signal The resulting cell state; the decoder in hour, A vector whose values ​​are all zeros; These are the parameters that the decoder can learn; For decoder functions; These are the weights of the fully connected layer; It is the bias of the fully connected layer; It is an activation function, and the Sigmoid function can be used.

[0028] The goal of the denoising neural network model is to make the predicted denoised ion current signal segment... A segment of signal that is infinitely close to pure ion current Therefore, the following loss function is defined: ; in, The output of the loss function, Represents all trainable parameters in the denoising neural network model. .

[0029] Step T30: Train the trainable parameters of the denoising neural network model using the training set and loss function to complete the training of the denoising neural network. training set The original ion current signal segment As input to the denoising neural network model, the original ion current signal fragment is processed by the denoising neural network model. Output noise-reduced ion current signal segment after noise reduction The noisy ion current signal segment and training set Pure ion current signal segment The parameters are input to the aforementioned loss function. Then, the loss function is calculated for all trainable parameters. gradient Then, an optimizer (e.g., Adam) is used to update all trainable parameters based on the gradients. This completes the training of the noise reduction neural network.

[0030] Step T40: Input all the original ion current signal segments to be tested into the trained denoising neural network model, and output the corresponding denoised ion current signal segments to be tested. During sequencing, DNA / RNA molecules are captured under the influence of an electric field, and the bases (A / T / C / G) on their strands enter the nanopore at a predetermined speed. The entry of the bases partially blocks the ion channels provided by the nanopore for charged ions, thereby generating the ion current signal to be measured. In order to sequence the bases of DNA / RNA molecules, the ion current signal to be measured is obtained.

[0031] Due to limitations in sequencing technology, the base sequence of a single DNA / RNA molecule cannot be determined in one go. Therefore, it is necessary to extract multiple segments of the original ion current signal corresponding to a single DNA / RNA molecule based on base shift events, and then arrange all the extracted segments in chronological order. ,in, The first segment of the original ion current signal to be measured. The second segment of the original ion current signal to be measured. For the first A segment of the original ion current signal to be measured.

[0032] All segments of the original ion current signal to be measured. All inputs are used as inputs to the trained denoising neural network model, which then denoises each original ion current segment to be measured, outputting the corresponding denoised ion current signal segment. All output denoised ion current signal segments are arranged in chronological order. ,in This is the first segment of the noise-reduced ion current signal to be tested. This is the second segment of the noise-reduced ion current signal to be tested. For the first A segment of the noise-reduced ion current signal to be tested.

[0033] Step T50: Segment all the noise-reduced ion current signal segments to be tested to obtain the complete noise-reduced ion current signal to be tested;

[0034] The noise reduction neural network model outputs all noise reduction ion current signal segments to be measured. Then, these noise-reduced ion current signal segments to be tested... The signals are spliced ​​together to obtain a complete noise-reduced ion current signal to be tested. This complete noise-reduced ion current signal corresponds to a DNA / RNA molecule. Subsequently, the base sequence of the DNA / RNA molecule can be obtained by performing base detection on this complete noise-reduced ion current signal.

[0035] Specifically, all segments of the noise-reduced ion current signal to be tested are spliced ​​together to obtain the complete noise-reduced ion current signal to be tested, including the following sub-steps:

[0036] ① Arrange all the noise-reduced ion current signal segments to be tested on a unified time axis in the order of time flow, and fill the gaps between two adjacent noise-reduced ion current signal segments to be tested; Since all the noise-reduced ion current signal segments output from the noise-reducing neural network model are arranged in chronological order, they can be directly arranged on a unified time axis in this chronological order. Typically, adjacent noise-reduced ion current signal segments overlap; therefore, arranging these segments on a unified time axis allows for a direct identification of which adjacent segments have gaps. If gaps exist between adjacent noise-reduced ion current signal segments, they are filled using a baseline completion method to avoid abrupt changes.

[0037] ② Extract the key features of the noise-reduced ion current signal segment to be tested, and use the key features of two adjacent noise-reduced ion current signal segments to be tested as splicing anchor points to obtain the complete noise-reduced ion current signal to be tested. Key features of all noise-reduced ion current signal segments to be tested are extracted, such as signal peak position, rise / fall time, and baseline level. These features are then used as anchor points for splicing between adjacent segments to obtain the complete noise-reduced ion current signal. Using these anchor points avoids the slight time shifts caused by noise reduction that could disrupt the signal's timing, ensuring the integrity of the spliced ​​noise-reduced ion current signal.

[0038] ③ By using the slope difference between adjacent sampling points before and after the splicing point, determine whether there are abrupt changes in all splicing points of the complete noise-reduced ion current signal to be measured, so as to eliminate splicing errors; After splicing all the noise-reduced ion current signal segments to obtain the complete noise-reduced ion current signal, the points at predetermined times before each splicing point on the complete noise-reduced ion current signal are set as sampling points, and the points at predetermined times after each splicing point are also set as sampling points. By using the slope difference between two adjacent sampling points before and after the splicing point, it is determined whether there is an abrupt change in the splicing point between these two sampling points, that is, whether it violates the physiological continuity, thereby eliminating splicing abnormalities.

[0039] Specifically, the expression for the slope between the splicing point and the preceding sampling point is as follows: ; in, For the first The slope between each splicing point and the preceding sampling point; For the first The noise-reduced ion current signal value at each splicing point; For the first The noise-reduced ion current signal values ​​of the sampling points before the splicing point; This is the time interval between the sampling point and the splicing point.

[0040] The expression for the slope between the splicing point and the subsequent sampling points is as follows: ; in, For the first The slope between each splicing point and the subsequent sampling points; For the first The noise-reduced ion current signal value of the sampling point after splicing.

[0041] The expression for the slope difference between two adjacent sampling points is as follows: ; in, For the first The slope difference between two adjacent sampling points before and after a splicing point; if Then the first If the first splicing point is normal, then the second one is not. One splicing point is abnormal; Slope difference threshold The threshold is set according to the signal type. For example, the slope difference threshold for ECG signals is usually less than 0.5mV / ms.

[0042] If the percentage of abnormal splicing points is greater than a predetermined value (e.g., 30%), the noise reduction processing of the original ion current signal to be measured fails, and the process returns to step T40 to perform noise reduction processing again; if the percentage of abnormal splicing points is not greater than the predetermined value, the noise reduction processing of the original ion current signal to be measured succeeds, and the process ends.

[0043] Example 2 like Figure 2 As shown, this application provides a signal noise reduction processing system 200, including: a signal collection unit 210, a model building unit 220, a model training unit 230, a noise reduction processing unit 240, and a signal splicing unit 250.

[0044] The signal collection unit 210 collects the original ion current signal and constructs a training set that associates the original ion current signal segment with the corresponding pure ion current signal segment.

[0045] The signal collection unit 210 collects terabytes of raw ion current signals from public databases or internal experiments. These raw ion current signals are cleaned to correct identifiable errors. Then, the signal collection unit 210 extracts the signal fragment corresponding to each base shift event from the cleaned raw ion current signals. A base shift event is an event in which a DNA molecule moves one base in a nanopore. The signal fragment corresponding to a base shift event is a fragment of the ion current signal generated during the process of a k-mer (fixed-length nucleotide string) entering, occupying, and leaving the nanopore. Each extracted signal fragment contains all the changes in the ion current signals generated during the residence of a k-mer in the nanopore, including the average blocking level, signal shape, duration, and fluctuation pattern.

[0046] The signal collection unit 210 associates each extracted signal fragment with the pure ion current signal corresponding to the k-mer in the reference sequence, forming a "raw signal-pure signal" signal pair. The reference sequence refers to the standard DNA sequence used to describe base shift events. For example, the 10nt DNA sequence ACTCGATGCT underwent six base shift events in the nanopore. The six base shift events correspond to six k-mers, namely ACTCG, CTCGA, TCGAT, CGATG, GATGC, and ATGCT. These six k-mers all belong to the reference sequence. The pure ion current signal corresponding to the k-mer in the reference sequence comes from high-quality annotation or theoretical generation.

[0047] Signal collection unit 210 gathers all the "raw signal - clean signal" signal pairs together to form a training set. .in, For the first For signal pairs; For the first A segment of the original ion current signal. For the first A segment of pure ion current signal, the first Segment of original ion current signal and the Segment of pure ion current signal Combination No. For signal pairs; The number of signal pairs in the training set; ,in, The first time step generated the first time step The first sub-segment of the original ion current signal segment. The second time step generated the first The second sub-segment of the original ion current signal segment. For the first The first time step generated by the first time step The first segment of the original ion current signal Each segment.

[0048] Model building unit 220 constructs a denoising neural network model and defines a loss function for the denoising neural network model.

[0049] Model building unit 220 constructs a noise reduction neural network model. The architecture of this noise reduction neural network model is an encoder-decoder architecture. This noise reduction neural network model can capture long-term dependencies in the signal.

[0050] For each time step The encoder expression is as follows: ; in, For the first The first time step generated by the first time step The first segment of the original ion current signal Each segment represents the encoder's input value; The encoder is for the first The first segment of the original ion current signal Sub-segment The resulting hidden state, The encoder is for the first The first segment of the original ion current signal Sub-segment The resulting cell state, The encoder is for the first The first segment of the original ion current signal Sub-segment The resulting hidden state, The encoder is for the first The first segment of the original ion current signal Sub-segment The generated cell state, the encoder at the initial time (i.e. The hidden state and cell state are both vectors of all zeros, meaning that at the initial time step... and All are vectors containing only zeros; These are the parameters that the encoder can learn; This is the encoder function.

[0051] After the encoder finishes processing the first time step, proceed to the next... The entire original ion current signal segment at each time step Then, the encoder output is for the first... The first segment of the original ion current signal Hidden state generated by each sub-segment and targeting the The first segment of the original ion current signal Cellular state generated by each sub-segment This hidden state and cell state It is believed to contain the entire segment of the original ion current signal. The summary information will be the hidden state output by the encoder. and cell state as context vector It is passed to the decoder as part of the decoder's input value.

[0052] The decoder expression is as follows: ; in, The first output of the decoder A segment of the denoised ion current signal is used as part of the input value of the decoder for calculation. ; The first output of the decoder Noise reduction ion current signal segment; The decoder is for the first Segment of original ion current signal The resulting hidden state, The decoder is for the first Segment of original ion current signal The resulting cell state; the decoder in hour, A vector whose values ​​are all zeros; These are the parameters that the decoder can learn; For decoder functions; These are the weights of the fully connected layer; It is the bias of the fully connected layer; It is an activation function, and the Sigmoid function can be used.

[0053] The goal of the denoising neural network model is to make the predicted denoised ion current signal segment... A segment of signal that is infinitely close to pure ion current Therefore, the loss function for model building unit 220 is defined as follows: ; in, The output of the loss function, Represents all trainable parameters in the denoising neural network model. .

[0054] The model training unit 230 trains the trainable parameters of the denoising neural network model using the training set and loss function to complete the training of the denoising neural network.

[0055] Model training unit 230 will train the training set The original ion current signal segment As input to the denoising neural network model, the original ion current signal fragment is processed by the denoising neural network model. Output noise-reduced ion current signal segment after noise reduction The noisy ion current signal segment and training set Pure ion current signal segment As input to the aforementioned loss function, the parameters are fed into it. Then, the model training unit 230 calculates the loss function for all trainable parameters. gradient Then, an optimizer (e.g., Adam) is used to update all trainable parameters based on the gradients. This completes the training of the noise reduction neural network.

[0056] The noise reduction processing unit 240 inputs all the original ion current signal segments to be measured into the trained noise reduction neural network model and outputs the corresponding noise reduction ion current signal segments to be measured.

[0057] During sequencing, DNA / RNA molecules are captured under the influence of an electric field, and the bases (A / T / C / G) on their strands enter the nanopore at a predetermined speed. The entry of the bases partially blocks the ion channels provided by the nanopore for charged ions, thereby generating the ion current signal to be measured. In order to sequence the bases of DNA / RNA molecules, the ion current signal to be measured is obtained.

[0058] Due to limitations in sequencing technology, the base sequence of a single DNA / RNA molecule cannot be determined in one go. Therefore, the noise reduction processing unit 240 needs to extract multiple segments of the original ion current signal corresponding to a single DNA / RNA molecule based on base shift events, and arrange all extracted segments in chronological order. ,in, The first segment of the original ion current signal to be measured. The second segment of the original ion current signal to be measured. For the first A segment of the original ion current signal to be measured.

[0059] The noise reduction processing unit 240 processes all segments of the original ion current signal to be measured. All inputs are used as inputs to the trained denoising neural network model, which then denoises each original ion current segment to be measured, outputting the corresponding denoised ion current signal segment. All output denoised ion current signal segments are arranged in chronological order. ,in This is the first segment of the noise-reduced ion current signal to be tested. This is the second segment of the noise-reduced ion current signal to be tested. For the first A segment of the noise-reduced ion current signal to be tested.

[0060] The signal splicing unit 250 splices together all the noise-reduced ion current signal segments to be tested, and obtains the complete noise-reduced ion current signal to be tested.

[0061] The noise reduction neural network model outputs all noise reduction ion current signal segments to be measured. Then, the signal splicing unit 250 splices these noise-reduced ion current signal segments to be measured. The signals are spliced ​​together to obtain a complete noise-reduced ion current signal to be tested. This complete noise-reduced ion current signal corresponds to a DNA / RNA molecule. Subsequently, the base sequence of the DNA / RNA molecule can be obtained by performing base detection on this complete noise-reduced ion current signal.

[0062] Specifically, the signal splicing unit 250 includes: a signal arrangement subunit 251, a signal splicing subunit 252, and an error elimination subunit 253.

[0063] The signal arrangement subunit 251 arranges all the noise-reduced ion current signal segments to be tested on a unified time axis in the order of time flow, and fills the gap between two adjacent noise-reduced ion current signal segments to be tested.

[0064] Since all the noise-reduced ion current signal segments output from the noise-reducing neural network model are arranged in chronological order, they can be directly arranged on a unified time axis in this chronological order. Typically, adjacent noise-reduced ion current signal segments overlap; therefore, arranging these segments on a unified time axis allows for a direct identification of which adjacent segments have gaps. If gaps exist between adjacent noise-reduced ion current signal segments, they are filled using a baseline completion method to avoid abrupt changes.

[0065] The signal splicing subunit 252 extracts the key features of the noise-reduced ion current signal segment to be tested, and splices them together using the key features of two adjacent noise-reduced ion current signal segments to be tested as splicing anchor points to obtain the complete noise-reduced ion current signal to be tested.

[0066] Key features of all noise-reduced ion current signal segments to be tested are extracted, such as signal peak position, rise / fall time, and baseline level. These features are then used as anchor points for splicing between adjacent segments to obtain the complete noise-reduced ion current signal. Using these anchor points avoids the slight time shifts caused by noise reduction that could disrupt the signal's timing, ensuring the integrity of the spliced ​​noise-reduced ion current signal.

[0067] Error elimination subunit 253 uses the slope difference between adjacent sampling points before and after the splicing point to determine whether there is a sudden change in all splicing points of the complete noise-reduced ion current signal to be measured, so as to eliminate splicing errors.

[0068] After splicing all the noise-reduced ion current signal segments to obtain the complete noise-reduced ion current signal, the points at predetermined times before each splicing point on the complete noise-reduced ion current signal are set as sampling points, and the points at predetermined times after each splicing point are also set as sampling points. By using the slope difference between two adjacent sampling points before and after the splicing point, it is determined whether there is an abrupt change in the splicing point between these two sampling points, that is, whether it violates the physiological continuity, thereby eliminating splicing abnormalities.

[0069] Specifically, the expression for the slope between the splicing point and the preceding sampling point is as follows: ; in, For the first The slope between each splicing point and the preceding sampling point; For the first The noise-reduced ion current signal value at each splicing point; For the first The noise-reduced ion current signal values ​​of the sampling points before the splicing point; This is the time interval between the sampling point and the splicing point.

[0070] The expression for the slope between the splicing point and the subsequent sampling points is as follows: ; in, For the first The slope between each splicing point and the subsequent sampling points; For the first The noise-reduced ion current signal value of the sampling point after splicing.

[0071] The expression for the slope difference between two adjacent sampling points is as follows: ; in, For the first The slope difference between two adjacent sampling points before and after a splicing point; if Then the first If the first splicing point is normal, then the second one is not. One splicing point is abnormal; Slope difference threshold The threshold is set according to the signal type. For example, the slope difference threshold for ECG signals is usually less than 0.5mV / ms.

[0072] If the percentage of abnormal splicing points is greater than a predetermined value (e.g., 30%), the noise reduction processing of the original ion current signal to be measured fails, and the process returns to step T40 to perform noise reduction processing again; if the percentage of abnormal splicing points is not greater than the predetermined value, the noise reduction processing of the original ion current signal to be measured succeeds, and the process ends.

[0073] This application constructs a large-scale, high-fidelity training set and trains a denoising neural network model to learn the essential characteristics of ion current signals under noise-free conditions, thereby achieving high-precision denoising and reconstruction of the original ion current signals. At the same time, it can avoid ion current signal distortion and significantly improve the accuracy and reliability of subsequent base identification.

[0074] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0075] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A signal noise reduction processing method, characterized in that, Includes the following steps: Step T10: Collect the raw ion current signal and construct a training set that associates the raw ion current signal fragments with the corresponding pure ion current signal fragments; Step T20: Construct a noise reduction neural network model and define a loss function for the noise reduction neural network model; Step T30: Train the trainable parameters of the denoising neural network model using the training set and loss function to complete the training of the denoising neural network. Step T40: Input all the original ion current signal segments to be tested into the trained denoising neural network model, and output the corresponding denoised ion current signal segments to be tested. Step T50: Segment all the noise-reduced ion current signal segments to be tested to obtain the complete noise-reduced ion current signal to be tested.

2. The signal noise reduction processing method according to claim 1, characterized in that, The architecture of the constructed noise reduction neural network model is an encoder-decoder architecture.

3. The signal noise reduction processing method according to claim 2, characterized in that, Loss function of denoising neural network model: ； in, The output of the loss function, This represents all trainable parameters in the denoising neural network model; The first output of the decoder Noise reduction ion current signal segment; For the first in the training set A segment of pure ion current signal; The number of signal pairs in the training set.

4. The signal noise reduction processing method according to any one of claims 1 to 3, characterized in that, Multiple segments of the original ion current signal corresponding to a DNA / RNA molecule are extracted according to the base shift event, and all the extracted segments of the original ion current signal are arranged in the order of time. All the original ion current signal segments to be measured are sequentially input into the trained denoising neural network model, and all the output denoising ion current signal segments to be measured are arranged in the order of time.

5. The signal noise reduction processing method according to any one of claims 1 to 3, characterized in that, All segments of the noise-reduced ion current signal to be tested are spliced ​​together to obtain the complete noise-reduced ion current signal to be tested, including the following sub-steps: ① Arrange all the noise-reduced ion current signal segments to be tested on a unified time axis in the order of time flow, and fill the gaps between two adjacent noise-reduced ion current signal segments to be tested; ② Extract the key features of the noise-reduced ion current signal segment to be tested, and use the key features of two adjacent noise-reduced ion current signal segments to be tested as splicing anchor points to obtain the complete noise-reduced ion current signal to be tested. ③ By using the slope difference between adjacent sampling points before and after the splicing point, determine whether there are abrupt changes in all splicing points of the complete noise-reduced ion current signal to be measured, so as to eliminate splicing errors.

6. A signal noise reduction processing system, characterized in that, include: The system comprises a signal acquisition unit, a model building unit, a model training unit, a noise reduction unit, and a signal splicing unit. The signal collection unit collects the raw ion current signal and constructs a training set that associates the raw ion current signal segments with the corresponding pure ion current signal segments; The model building unit constructs a denoising neural network model and defines a loss function for the denoising neural network model. The model training unit trains the trainable parameters of the denoising neural network model using the training set and loss function to complete the training of the denoising neural network. The noise reduction processing unit inputs all the original ion current signal segments to be measured into the trained noise reduction neural network model and outputs the corresponding noise-reduced ion current signal segments to be measured. The signal splicing unit splices together all the noise-reduced ion current signal segments to be tested, and obtains the complete noise-reduced ion current signal to be tested.

7. The signal noise reduction processing system according to claim 6, characterized in that, The architecture of the constructed noise reduction neural network model is an encoder-decoder architecture.

8. The signal noise reduction processing system according to claim 7, characterized in that, Loss function of denoising neural network model: ； in, The output of the loss function, This represents all trainable parameters in the denoising neural network model; The first output of the decoder Noise reduction ion current signal segment; For the first in the training set A segment of pure ion current signal; The number of signal pairs in the training set.

9. The signal noise reduction processing system according to any one of claims 6 to 8, characterized in that, The noise reduction processing unit extracts multiple segments of the original ion current signal corresponding to a DNA / RNA molecule according to the base shift event, and all the extracted segments of the original ion current signal are arranged in the order of time. The noise reduction processing unit sequentially inputs all the original ion current signal segments to be measured into the trained noise reduction neural network model, and the output of all the noise reduction ion current signal segments to be measured is arranged in the order of time.

10. The signal noise reduction processing system according to any one of claims 6 to 8, characterized in that, The signal splicing unit includes: a signal arrangement subunit, a signal splicing subunit, and an error elimination subunit; The signal arrangement subunit arranges all the noise-reduced ion current signal segments to be tested on a unified time axis in the order of time flow, and fills the gaps between two adjacent noise-reduced ion current signal segments to be tested. The signal splicing subunit extracts the key features of the noise-reduced ion current signal segment to be tested, and splices it using the key features of two adjacent noise-reduced ion current signal segments to be tested as splicing anchor points to obtain the complete noise-reduced ion current signal to be tested. The error elimination subunit uses the slope difference between adjacent sampling points before and after the splicing point to determine whether there are abrupt changes in all splicing points of the complete noise-reduced ion current signal to be measured, so as to eliminate splicing errors.

Images