A gene analysis method and device based on optical pulse neural network

Abstract

The present invention proposes a gene analysis method and device based on optical pulse neural network, wherein optical pulse neurons are simulated by an optical device VCSEL-SA, so that signals are transmitted in the optical pulse neural network in the form of optical pulses, thereby constructing the VCSEL-SA optical pulse neural network, and then according to the one-to-one correspondence between sub-data sets and the VCSEL-SA optical pulse neural network, the VCSEL-SA optical pulse neural network is iteratively trained using input samples corresponding to the sub-data sets to obtain a plurality of trained VCSEL-SA optical pulse neural networks; since optical pulse neurons can process nanosecond or even picosecond level information, the processing speed is fast and the calculation efficiency is high, and since the voting principle is used to determine the sample results in multiple prediction classifications, the efficiency and accuracy of predicting and classifying the current HIV protease sample can be improved, and the prediction and classification results are provided to doctors for reference and assistance, thereby reducing the workload of doctors.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer simulation and method optimization, and specifically relates to a gene analysis method and device based on optical pulse neural network. Background Art

[0002] In people's long-term research, it is found that light has many excellent properties, such as high speed, high bandwidth, low power consumption, low crosstalk, etc., which is very suitable for ultrafast information processing. Therefore, the photonic neuromorphic system that integrates photonic technology and neuromorphic computing came into being, stimulating the research enthusiasm of more and more experts and scholars. In the photonic neuromorphic system, photon synapses and photon pulse neurons are realized by optical devices, and signals are transmitted in the form of light pulses. Photon pulse neurons can process information at the nanosecond or even picosecond level. Since many optical devices in the photonic neuromorphic system are passive devices, the energy consumption of the system can also be reduced to a certain extent. Considering the continuous development of microelectronics technology, the photonic neuromorphic system is moving towards integration and chipization, which can promote the realization of long-term goals such as medical diagnosis, computer vision, etc. In the future, there may be more complex and more general all-optical brain-like neural systems, that is, the real "optical brain", which will contribute greatly to the development of national science and technology and the improvement of people's quality of life.

[0003] HIV, which is generally considered to be the pathogen of acquired immunodeficiency syndrome (AIDS), encodes an aspartic protease called HIV protease, whose function is essential for HIV replication. Understanding the polyprotein cleavage sites of HIV protease will deepen our understanding of its specificity and help design specific and effective HIV protease inhibitors. If a genetic method based on optical pulse neural network is found to accurately, reliably and quickly predict the cleavage sites of HIV protease in proteins, it will quickly and accurately determine the positive and negative of sample data in HIV-1 datasets, which has certain practical application significance.

[0004] At present, the research on HIV-1 protease cleavage sites in gene analysis tasks is mainly based on machine learning methods. For example, the paper "Rognvaldson T, You L, Garwicz D. State of the Art Prediction of HIV-1 Protease Cleavage Sites [J]. Bioinformatics, 2015, 31 (8): 1204-1210." uses the support vector machine method to predict the sample data in the HIV-1 data set. "Fan Guangpeng, Sun Rencheng, Shao Fengjing. Research on the Prediction of HIV-1 Protease Cleavage Sites [J]. Journal of Qingdao University (Engineering Technology Edition)" uses a long short-term memory recurrent neural network to classify and predict HIV-1 protease cleavage sites.

[0005] Existing technical solutions all implement genetic analysis tasks on computers through machine learning or deep learning. This solution relies on computer hardware devices, and electrical signals are transmitted during the calculation process, which makes it difficult to use on hardware devices with limited resources. Summary of the invention

[0006] In order to solve the above problems existing in the prior art, the present invention provides a gene analysis method and device based on optical pulse neural network. The technical problem to be solved by the present invention is achieved by the following technical solutions:

[0007] In a first aspect, the present invention provides a gene analysis method based on an optical pulse neural network, comprising:

[0008] Step 1: Obtain a data set consisting of positive samples cleaved by HIV protease and negative samples that cannot be cleaved by HIV protease;

[0009] Among them, positive samples and negative samples carry their corresponding sample labels;

[0010] Step 2: Divide the data set into multiple sub-data sets, and preprocess all samples in each sub-data set to convert the real-valued features of the samples into optical signal features, and obtain the input samples corresponding to each sub-data set;

[0011] Step 3: Use VCSEL-SA devices to simulate PRE neurons and POST neurons, and construct multiple VCSEL-SA optical pulse neural networks with the same number as the sub-data sets;

[0012] Each VCSEL-SA optical pulse neural network consists of an input layer, a hidden layer and an output layer, wherein the input layer neurons have PRE neuron functions, and the hidden layer and output layer neurons have POST neuron functions;

[0013] Step 4: According to the one-to-one correspondence between the sub-datasets and the VCSEL-SA optical pulse neural network, the VCSEL-SA optical pulse neural network is iteratively trained using the input samples corresponding to the sub-datasets to obtain a plurality of trained VCSEL-SA optical pulse neural networks;

[0014] Step 5: Use the trained multiple VCSEL-SA optical pulse neural networks to predict and classify the current HIV protease sample, obtain the reference result of whether the current HIV protease sample is negative or positive, and use the voting method to select the reference result with the most votes and provide it to the doctor as an auxiliary reference.

[0015] In a second aspect, the present invention provides a gene analysis device based on an optical pulse neural network, comprising:

[0016] An acquisition module is configured to acquire a data set formed by positive samples cleaved by HIV protease and negative samples that cannot be cleaved by HIV protease;

[0017] Among them, positive samples and negative samples carry their corresponding sample labels;

[0018] A preprocessing module is configured to divide the data set into a plurality of sub-data sets, and preprocess all samples in each sub-data set to convert the real-valued features of the samples into optical signal features, thereby obtaining input samples corresponding to each sub-data set;

[0019] A construction module is configured to use VCSEL-SA devices to simulate PRE neurons and POST neurons, and respectively construct a plurality of VCSEL-SA optical pulse neural networks having the same number as the sub-data sets;

[0020] Each VCSEL-SA optical pulse neural network consists of an input layer, a hidden layer and an output layer, wherein the input layer neurons have PRE neuron functions, and the hidden layer and output layer neurons have POST neuron functions;

[0021] A training module is configured to iteratively train the VCSEL-SA optical pulse neural network using input samples corresponding to the sub-data sets according to a one-to-one correspondence between the sub-data sets and the VCSEL-SA optical pulse neural network, to obtain a plurality of trained VCSEL-SA optical pulse neural networks;

[0022] The auxiliary module is configured to use multiple trained VCSEL-SA optical pulse neural networks to predict and classify the current HIV protease sample, obtain a reference result of whether the current HIV protease sample is negative or positive, and use a voting method to select the reference result with the most votes to provide it to the doctor as an auxiliary reference.

[0023] Beneficial effects of the present invention:

[0024] The present invention proposes a gene analysis method and device based on optical pulse neural network. The optical pulse neuron is simulated by an optical device VCSEL-SA, so that the signal is transmitted in the optical pulse neural network in the form of optical pulses, thereby constructing the VCSEL-SA optical pulse neural network. Then, according to the one-to-one correspondence between the sub-data set and the VCSEL-SA optical pulse neural network, the VCSEL-SA optical pulse neural network is iteratively trained using input samples corresponding to the sub-data set to obtain a plurality of trained VCSEL-SA optical pulse neural networks. Since the optical pulse neuron can process nanosecond or even picosecond information, the processing speed is fast and the calculation efficiency is high. Since the voting principle is used to determine the sample results in multiple prediction classifications, the efficiency and accuracy of the prediction classification of the current HIV protease sample can be improved, and the prediction classification results are provided to doctors for reference and assistance, thereby reducing the workload of doctors.

[0025] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a schematic diagram of a process of a gene analysis method based on an optical pulse neural network provided by an embodiment of the present invention;

[0027] Figure 2 Schematic diagram of the structure of a VCSEL-SA optical pulse neural network provided by an embodiment of the present invention;

[0028] Figure 3 It is a schematic diagram of the process of training a VCSEL-SA optical pulse neural network provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0029] The present invention is further described in detail below with reference to specific embodiments, but the embodiments of the present invention are not limited thereto.

[0030] like Figure 1 As shown, the present invention provides a gene analysis method based on optical pulse neural network, which includes:

[0031] Step 1: Obtain a data set consisting of positive samples cleaved by HIV protease and negative samples that cannot be cleaved by HIV protease;

[0032] Among them, positive samples and negative samples carry their corresponding sample labels;

[0033] Step 2: Divide the data set into multiple sub-data sets, and preprocess all samples in each sub-data set to convert the real-valued features of the samples into optical signal features, and obtain the input samples corresponding to each sub-data set;

[0034] The present invention can use m Gaussian functions to perform Gaussian encoding on positive samples and negative samples, so that both the positive samples and the negative samples are mapped into pulse time information; the pulse time information after encoding the positive samples and the negative samples is generated into a rectangular wave signal with a height of Kp and a width of tpst; and the rectangular wave signal is used as an input sample.

[0035] The input samples are mapped into pulse timing information through Gaussian coding. Gaussian coding uses m Gaussian functions, and the center μ of each Gaussian function is:

[0036]

[0037] Among them, I min ,I max , represents the minimum and maximum values ​​of the center positions of these Gaussian functions;

[0038] The standard deviation of the Gaussian function satisfies the following formula:

[0039]

[0040] Among them, β represents the spontaneous radiation coupling factor, and each real-valued feature will produce m intersection points with m Gaussian functions as the result of sample data encoding.

[0041] Step 3: Use VCSEL-SA devices to simulate PRE neurons and POST neurons, and construct multiple VCSEL-SA optical pulse neural networks with the same number as the sub-data sets;

[0042] refer to Figure 2 Each VCSEL-SA optical pulse neural network consists of an input layer, a hidden layer and an output layer. The input layer neurons have PRE neuron functions, and the hidden layer and output layer neurons have POST neuron functions. VCSEL-SA is a vertical cavity surface semiconductor emitting laser with LIF neuron-like dynamic characteristics. In the present invention, VCSEL-SA neurons are used to change the cavity of VCSEL-SA. When the neuron threshold is reached, the semiconductor emitting laser will emit pulses.

[0043] refer to Figure 2 In this optical pulse neural network, the external signal is first precoded to obtain a precoded signal, which is generally in the form of a rectangular wave. The center time of the signal is recorded as t inEach input layer neuron will receive a pre-coded signal and generate a pulse in response. The time when the pulse is generated is recorded as t i , so t i Mainly affected by in The influence of the pre-coding signal and the reasonable setting of the waveform parameters of the pre-coding signal can ensure that each neuron in the input layer outputs at most one pulse. The connection weight between the input layer neuron and the hidden layer neuron is denoted by ω ih , so ω ih is a size of N i ×N h The time when the hidden layer neuron generates a pulse is recorded as t h The connection weight between the hidden layer neurons and the output layer neurons is denoted by ω ho The time when the output layer neuron generates a pulse is t o , which represents the final classification result.

[0044] Step 4: According to the one-to-one correspondence between the sub-datasets and the VCSEL-SA optical pulse neural network, the VCSEL-SA optical pulse neural network is iteratively trained using the input samples corresponding to the sub-datasets to obtain a plurality of trained VCSEL-SA optical pulse neural networks;

[0045] Before training, the present invention needs to perform parameter initialization, including setting network training parameters, VCSEL parameters and randomly initializing network weights. The network training parameters mainly include data set division, training times and the number of training samples each time.

[0046] refer to Figure 3 The process of iteratively training the VCSEL-SA optical pulse neural network in step 4 of the present invention is as follows:

[0047] Repeatedly input each input sample in the sub-data set to each neuron of the input layer of the corresponding VCSEL-SA optical pulse neural network, so that each neuron generates a spike pulse signal to output to the neurons in the hidden layer; the neurons in the hidden layer try to generate a pulse signal and transmit the generated result to the output layer; the output layer predicts and classifies the input sample according to the result of whether the pulse signal is generated; if the predicted classification of the input sample is inconsistent with the sample label, the weight of the VCSEL-SA optical pulse neural network is adjusted according to the preset weight adjustment rule until the training cutoff condition is met.

[0048] The present invention first passes the spike pulse signal generated by the input sample through the POST neuron of the fully connected layer. If the POST neuron generates a pulse, the classification result is 1, otherwise the classification result is 0. Then, if the classification result of the output neuron is incorrect, the weight will be adjusted under the guidance of the supervised learning algorithm.

[0049] In the present invention, VCSEL-SA is used to simulate neurons. The input layer neurons are suitable for PRE neuron model, and its neuron model can be expressed by formula (1-1) and formula (1-3); the hidden layer and output layer neurons are suitable for POST neuron model, and its neuron model can be expressed by formula (1-2) and formula (1-3). The number of neurons in each layer is represented by Ni, Nh and No, respectively, where No=1.

[0050]

[0051]

[0052]

[0053]

[0054]

[0055]

[0056] Wherein, subscript a represents the gain region of the VCSEL-SA device, subscript s represents the absorption region of the VCSEL-SA device, subscripts pre and post represent PRE neurons and POST neurons respectively, and n a represents the carrier density in the gain region, n s represents the carrier density in the absorption region, represents the intensity of the signal, represents the central time of the signal, Δτ represents the pulse duration of the signal, ω i represents the weight between the pulse neurons, T is the simulation time of each layer of neurons, V a Represents the cavity volume in the gain region, V s Absorption zone cavity volume, Γ a represents the gain region limiting factor, Γ s represents the absorption region limitation factor, τ a represents the carrier lifetime in the gain region, τ s represents the carrier lifetime in the absorption region, g a represents the differential gain or loss in the gain region, g s is the differential gain / loss in the absorption region, n 0s represents the initial carrier density in the absorption region, n 0a represents the initial carrier density in the gain region, I a Represents the gain region bias current, I s is the absorption region bias current, λ pre,post represents the wavelength of PRE neuron, the wavelength of POST neuron, λ i represents the laser wavelength, B rrepresents the bimolecular complex term, β represents the spontaneous radiation coupling factor, η c represents the output power coupling coefficient, c represents the speed of light, h represents the Planck constant, τ ph represents the photon lifetime, S represents the photon density in the cavity, e represents the elementary charge, , and Δτ represent the intensity of the signal, the central moment of the signal and the pulse duration of the signal respectively, and Pi represents the output power.

[0057] For the VCSEL-SA pulse neuron model, the pump current I a Change the carrier density so that the pulse amplitude of the pulse neuron reaches the threshold. Formula (1-4) is only applicable to PRE neurons and can be used to calculate the response of the input layer neurons. In addition, , and Δτ in formula (1-4) represent the signal intensity, the central moment of the signal and the pulse duration of the signal, respectively, and Pi represents the output power. It is dimensionless. For the sake of unification, it is set that Δτ = 2ns. Formula (1-5) is applicable to the weighted summation of POST neurons. Represents the weight between pulse neurons. T is the simulation time of each layer of neurons. Formula (1-6) is applicable to calculating the output optical power of PRE neurons or POST neurons.

[0058] If the classification result of the output neuron is incorrect, the weight will be adjusted under the guidance of the supervised learning algorithm. The specific weight adjustment rules meet the following requirements:

[0059]

[0060] ω(x+1)=ω(x)+Δω (1-8)

[0061] Among them, η represents the learning rate, t o represents the pulse time of the output neuron, represents the pulse time of the input neuron, d represents the time delay between neurons, represents the state of the output neuron, and represents that the corresponding output neuron has a pulse, n o =0 means that the corresponding output neuron has no pulse, n indicates the output neuron state expected by the sample label, and n indicates that the corresponding output neuron is expected to have a pulse. label =1 means no pulse is expected; for output neurons that are expected to have pulses but actually have no pulses, the weight is adjusted by the STDP process; for output neurons that are expected to have no pulses but actually have pulses, the weight is adjusted by the anti-STDP process.

[0062] Step 5: Use the trained multiple VCSEL-SA optical pulse neural networks to predict and classify the current HIV protease sample, obtain the reference result of whether the current HIV protease sample is negative or positive, and use the voting method to select the reference result with the most votes and provide it to the doctor as an auxiliary reference.

[0063] The training set is divided into different sample sets to train different classifiers. In actual operation, the present invention trains N optical pulse neural networks according to step 4. In the process of gene analysis, inference and prediction, after the input sample enters each optical pulse neural network, the optical pulse neural network obtains its own results and uses the voting method to make a comprehensive decision to obtain the final classification result, that is, the classification result with the largest number is taken as the final result.

[0064] The present invention proposes a gene analysis method based on an optical pulse neural network. An optical device VCSEL-SA is used to simulate an optical pulse neuron, so that a signal is transmitted in the optical pulse neural network in the form of an optical pulse, thereby constructing the VCSEL-SA optical pulse neural network. Then, according to a one-to-one correspondence between a sub-data set and the VCSEL-SA optical pulse neural network, the VCSEL-SA optical pulse neural network is iteratively trained using input samples corresponding to the sub-data set to obtain a plurality of trained VCSEL-SA optical pulse neural networks. Since the optical pulse neuron can process nanosecond or even picosecond information, the processing speed is fast and the calculation efficiency is high. Since the voting principle is used to determine the sample results in multiple prediction classifications, the efficiency and accuracy of the prediction classification of the current HIV protease sample can be improved, and the prediction classification results are provided to doctors for reference and assistance, thereby reducing the workload of doctors.

[0065] Another advantage is that the present invention can quickly assist doctors during emergency treatment and alert them to confirm the patient's condition through further medical examinations, which can greatly reduce the risk of doctors being exposed to surgery.

[0066] The present invention provides a gene analysis device based on an optical pulse neural network, comprising:

[0067] An acquisition module is configured to acquire a data set formed by positive samples cleaved by HIV protease and negative samples that cannot be cleaved by HIV protease;

[0068] Among them, positive samples and negative samples carry their corresponding sample labels;

[0069] A preprocessing module is configured to divide the data set into a plurality of sub-data sets, and preprocess all samples in each sub-data set to convert the real-valued features of the samples into optical signal features, thereby obtaining input samples corresponding to each sub-data set;

[0070] A construction module is configured to use VCSEL-SA devices to simulate PRE neurons and POST neurons, and respectively construct a plurality of VCSEL-SA optical pulse neural networks having the same number as the sub-data sets;

[0071] Each VCSEL-SA optical pulse neural network consists of an input layer, a hidden layer and an output layer, wherein the input layer neurons have PRE neuron functions, and the hidden layer and output layer neurons have POST neuron functions;

[0072] A training module is configured to iteratively train the VCSEL-SA optical pulse neural network using input samples corresponding to the sub-data sets according to a one-to-one correspondence between the sub-data sets and the VCSEL-SA optical pulse neural network, to obtain a plurality of trained VCSEL-SA optical pulse neural networks;

[0073] The auxiliary module is configured to use multiple trained VCSEL-SA optical pulse neural networks to predict and classify the current HIV protease sample, obtain a reference result of whether the current HIV protease sample is negative or positive, and use a voting method to select the reference result with the most votes to provide it to the doctor as an auxiliary reference.

[0074] Optionally, the preprocessing module is configured to:

[0075] Step 3-1: Use m Gaussian functions to perform Gaussian encoding on positive samples and negative samples, so that both positive samples and negative samples are mapped into pulse time information;

[0076] Step 3-2: Generate a rectangular wave signal with a height of Kp and a width of tpst from the encoded pulse time information of the positive sample and the negative sample;

[0077] Step 3-4: Take the rectangular wave signal as input sample.

[0078] Optionally, the center μ of each Gaussian function in step 3-1 is:

[0079]

[0080] Among them, I min ,I max , represents the minimum and maximum values ​​of the center positions of these Gaussian functions;

[0081] The standard deviation of the Gaussian function satisfies the following formula:

[0082]

[0083] Where β is the spontaneous emission coupling factor.

[0084] Optionally, the process of iteratively training the VCSEL-SA optical pulse neural network by the training module is:

[0085] Repeatedly input each input sample in the sub-data set to each neuron of the input layer of the corresponding VCSEL-SA optical pulse neural network, so that each neuron generates a spike pulse signal to output to the neurons in the hidden layer; the neurons in the hidden layer try to generate a pulse signal and transmit the generated result to the output layer; the output layer predicts and classifies the input sample according to the result of whether the pulse signal is generated; if the predicted classification of the input sample is inconsistent with the sample label, the weight of the VCSEL-SA optical pulse neural network is adjusted according to the preset weight adjustment rule until the training cutoff condition is met.

[0086] The specific advantages of the present invention are explained below through experiments.

[0087] Step a: Parameter initialization, including setting network training parameters and VCSEL-SA parameters. This gene analysis dataset contains a total of 746 samples, each of which is a polypeptide composed of 8 amino acids, represented by eight capital letters. The sample labels of the dataset are divided into two types, which can be cleaved by HIV protease (positive) and cannot be cleaved by HIV protease (negative), represented by 1 and 0 respectively. The HIV-1 dataset contains a total of 401 positive samples and 345 negative samples.

[0088] The ratio of training set to test set is set to 4:1, the default number of samples for one training is 150, the number of iterations is 400, and the number of samples for one test is 100. The main parameter settings of VCSEL-SA are shown in Table 1:

[0089] Table 1 Parameter setting table

[0090]

[0091]

[0092] Step b: Sample preprocessing. The optical pulse neural network requires optical signals as input, so the samples in the data set need to be pre-encoded to convert the real-valued features of the samples into signals that can be recognized by the pulse neurons. The specific process is as follows:

[0093] The input sample is mapped into pulse time information through Gaussian coding. The time information after sample coding generates a high A rectangular wave signal with a width of tpst=2ns is sent to the PRE neuron to generate a spike pulse signal.

[0094] Step c: VCSEL-SA-based optical pulse neural network training process.

[0095] c.1 Send the spike pulse signal generated by the input sample to the POST neuron in the hidden layer. If the POST neuron generates a pulse, the classification result is 1, otherwise the classification result is 0.

[0096] c.2 If the classification result of the output neuron is incorrect, the weights will be adjusted under the guidance of the supervised learning algorithm.

[0097] Step d: Genetic analysis prediction based on optical pulse neural network of VCSEL-SA. The training set is divided into different sample sets to train different classifiers. This patent trains N=3 optical pulse neural network classifiers according to step 3. In the process of genetic analysis reasoning prediction, after the input sample enters each classifier, the classifiers obtain their respective results and use the voting method to make a comprehensive decision to obtain the final classification result, that is, the classification result with the largest number is taken as the final result.

[0098] Compared with other algorithms, the gene analysis algorithm based on optical pulse neural network adopted in the present invention has the advantages of novel network structure and high computational efficiency, and can quickly, accurately and reliably predict the negative and positive of HIV-1 test set sample data.

[0099] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0100] Although the present application is described herein in conjunction with various embodiments, in the process of implementing the claimed application, those skilled in the art may understand and implement other variations of the disclosed embodiments by reviewing the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality of components or steps.

[0101] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A gene analysis method based on optical pulse neural network, characterized in that: include: Step 1: Obtain a data set consisting of positive samples cleaved by HIV protease and negative samples that cannot be cleaved by HIV protease; Among them, positive samples and negative samples carry their corresponding sample labels; Step 2: Divide the data set into multiple sub-data sets, and preprocess all samples in each sub-data set to convert the real-valued features of the samples into optical signal features, and obtain the input samples corresponding to each sub-data set; Step 3: Use VCSEL-SA devices to simulate PRE neurons and POST neurons, and construct multiple VCSEL-SA optical pulse neural networks with the same number as the sub-data sets; Each VCSEL-SA optical pulse neural network consists of an input layer, a hidden layer and an output layer, wherein the input layer neurons have PRE neuron functions, and the hidden layer and output layer neurons have POST neuron functions; Step 4: According to the one-to-one correspondence between the sub-datasets and the VCSEL-SA optical pulse neural network, the VCSEL-SA optical pulse neural network is iteratively trained using the input samples corresponding to the sub-datasets to obtain a plurality of trained VCSEL-SA optical pulse neural networks; Step 5: Use the trained multiple VCSEL-SA optical pulse neural networks to predict and classify the current HIV protease sample, obtain the reference result of whether the current HIV protease sample is negative or positive, and use the voting method to select the reference result with the most votes and provide it to the doctor as an auxiliary reference.

2. A gene analysis method based on optical pulse neural network according to claim 1, characterized in that: The step 3 comprises: Step 3-1: Use m Gaussian functions to perform Gaussian encoding on positive samples and negative samples, so that both positive samples and negative samples are mapped into pulse time information; Step 3-2: Generate a rectangular wave signal with a height of Kp and a width of tpst from the encoded pulse time information of the positive sample and the negative sample; Step 3-4: Take the rectangular wave signal as input sample.

3. A gene analysis method based on optical pulse neural network according to claim 2, characterized in that: The center μ of each Gaussian function in step 3-1 is: Among them, I min ,I max , represents the minimum and maximum values ​​of the center positions of these Gaussian functions; The standard deviation of the Gaussian function satisfies the following formula: Where β is the spontaneous emission coupling factor.

4. A gene analysis method based on optical pulse neural network according to claim 1, characterized in that: The process of iteratively training the VCSEL-SA optical pulse neural network in step 4 is as follows: Repeatedly input each input sample in the sub-data set to each neuron of the input layer of the corresponding VCSEL-SA optical pulse neural network, so that each neuron generates a spike pulse signal to output to the neurons in the hidden layer; the neurons in the hidden layer try to generate a pulse signal and transmit the generated result to the output layer; the output layer predicts and classifies the input sample according to the result of whether the pulse signal is generated; if the predicted classification of the input sample is inconsistent with the sample label, the weight of the VCSEL-SA optical pulse neural network is adjusted according to the preset weight adjustment rule until the training cutoff condition is met.

5. A gene analysis method based on optical pulse neural network according to claim 4, characterized in that: The number of neurons in the input layer, hidden layer, and output layer are represented by Ni, Nh, and No, respectively; The neurons in the input layer are represented as: The neurons in the hidden layer and the output layer are represented as: Wherein, subscript a represents the gain region of the VCSEL-SA device, subscript s represents the absorption region of the VCSEL-SA device, subscripts pre and post represent PRE neurons and POST neurons respectively, and n a represents the carrier density in the gain region, n s represents the carrier density in the absorption region, represents the intensity of the signal, represents the central time of the signal, Δτ represents the pulse duration of the signal, ω i represents the weight between the pulse neurons, T is the simulation time of each layer of neurons, V a Represents the cavity volume in the gain region, V s Absorption zone cavity volume, Γ a represents the gain region limiting factor, Γ s represents the absorption region limitation factor, τ a represents the carrier lifetime in the gain region, τ s represents the carrier lifetime in the absorption region, g a represents the differential gain or loss in the gain region, g s is the differential gain / loss in the absorption region, n 0s represents the initial carrier density in the absorption region, n 0a represents the initial carrier density in the gain region, I a represents the gain region bias current, I s is the absorption region bias current, λ pre,post represents the wavelength of PRE neuron, the wavelength of POST neuron, λ i represents the laser wavelength, B r represents the bimolecular complex term, β represents the spontaneous radiation coupling factor, η c represents the output power coupling coefficient, c represents the speed of light, represents Planck's constant, τ ph represents the photon lifetime, S represents the photon density in the cavity, e represents the elementary charge, k ei , τ i and Δτ represent the signal strength, the central moment of the signal and the pulse duration of the signal respectively, and Pi represents the output power.

6. A gene analysis method based on optical pulse neural network according to claim 5, characterized in that: The weight adjustment rule is: ω(x+1)=ω(x)+Δω Where η represents the learning rate, t o represents the pulse time of the output neuron, represents the pulse time of the input neuron, d represents the time delay between neurons, represents the state of the output neuron, and represents that the corresponding output neuron has a pulse, n o =0 means that the corresponding output neuron has no pulse, n indicates the output neuron state expected by the sample label, and n indicates that the corresponding output neuron is expected to have a pulse. label =1 means no pulse is expected; for output neurons that are expected to have pulses but actually have no pulses, the weight is adjusted by the STDP process; for output neurons that are expected to have no pulses but actually have pulses, the weight is adjusted by the anti-STDP process.

7. A gene analysis device based on optical pulse neural network, characterized in that: include: An acquisition module is configured to acquire a data set formed by positive samples cleaved by HIV protease and negative samples that cannot be cleaved by HIV protease; Among them, positive samples and negative samples carry their corresponding sample labels; A preprocessing module is configured to divide the data set into a plurality of sub-data sets, and preprocess all samples in each sub-data set to convert the real-valued features of the samples into optical signal features, thereby obtaining input samples corresponding to each sub-data set; A construction module is configured to use VCSEL-SA devices to simulate PRE neurons and POST neurons, and respectively construct a plurality of VCSEL-SA optical pulse neural networks having the same number as the sub-data sets; Each VCSEL-SA optical pulse neural network consists of an input layer, a hidden layer and an output layer, wherein the input layer neurons have PRE neuron functions, and the hidden layer and output layer neurons have POST neuron functions; A training module is configured to iteratively train the VCSEL-SA optical pulse neural network using input samples corresponding to the sub-data sets according to a one-to-one correspondence between the sub-data sets and the VCSEL-SA optical pulse neural network, to obtain a plurality of trained VCSEL-SA optical pulse neural networks; The auxiliary module is configured to use multiple trained VCSEL-SA optical pulse neural networks to predict and classify the current HIV protease sample, obtain a reference result of whether the current HIV protease sample is negative or positive, and use a voting method to select the reference result with the most votes to provide it to the doctor as an auxiliary reference.

8. The gene analysis device based on optical pulse neural network according to claim 7, characterized in that: The preprocessing module is configured as follows: Step 3-1: Use m Gaussian functions to perform Gaussian encoding on positive samples and negative samples, so that both positive samples and negative samples are mapped into pulse time information; Step 3-2: Generate a rectangular wave signal with a height of Kp and a width of tpst from the encoded pulse time information of the positive sample and the negative sample; Step 3-4: Take the rectangular wave signal as input sample.

9. The gene analysis device based on optical pulse neural network according to claim 8, characterized in that: The center μ of each Gaussian function in step 3-1 is: Among them, I min ,I max , represents the minimum and maximum values ​​of the center positions of these Gaussian functions; The standard deviation of the Gaussian function satisfies the following formula: Where β is the spontaneous emission coupling factor.

10. The gene analysis device based on optical pulse neural network according to claim 7, characterized in that: The process of iterative training of the VCSEL-SA optical pulse neural network by the training module is as follows: Repeatedly input each input sample in the sub-data set to each neuron of the input layer of the corresponding VCSEL-SA optical pulse neural network, so that each neuron generates a spike pulse signal to output to the neurons in the hidden layer; the neurons in the hidden layer try to generate a pulse signal and transmit the generated result to the output layer; the output layer predicts and classifies the input sample according to the result of whether the pulse signal is generated; if the predicted classification of the input sample is inconsistent with the sample label, the weight of the VCSEL-SA optical pulse neural network is adjusted according to the preset weight adjustment rule until the training cutoff condition is met.

Images