Parallel processing synchronization optimization method and system for high-throughput gene sequencing

Abstract

The present invention relates to a parallel processing synchronous optimization method and system for high-throughput gene sequencing. The method: channel allocation and signal acquisition initialization are performed on multiple DNA samples to obtain a signal acquisition parameter matrix; feature extraction and cluster analysis are performed on the fluorescent signals collected by each channel to obtain a DNA base feature sequence and a signal classification probability matrix; the short-term parameters of the sequencing cycle are dynamically adjusted by a convex optimization algorithm, and the long-term sequencing performance is optimized by an integer programming algorithm to obtain a sequencing parameter optimization set and a sequencing cycle allocation strategy; the sequencing parameter optimization set and the sequencing cycle allocation strategy are input into a double-layer nested control model, and the upper resource allocation network and the lower parameter optimization network are jointly solved to obtain a global optimization control parameter set. The implementation of the present invention improves the sequencing accuracy and realizes the improvement of computing resource utilization.

Description

Technical Field

[0001] The present invention relates to the technical field of gene sequencing, and in particular to a parallel processing synchronization optimization method and system for high-throughput gene sequencing. Background Art

[0002] When processing multiple DNA samples, traditional gene sequencing methods lack an effective parallel processing mechanism, which easily leads to signal interference between channels during the sequencing process, resulting in unstable sequencing quality and affecting the accuracy of sequencing results.

[0003] In the sequencing process, the collection and processing of fluorescence signals are key links, but the existing technology lacks a systematic optimization solution for feature extraction and classification analysis of fluorescence signals. Especially when multiple samples are sequenced in parallel, the reliability of sequencing data is seriously affected due to signal cross-contamination between samples and the influence of background noise, and the dynamic adjustment of sequencing parameters lacks theoretical support. In addition, the current sequencing system has obvious deficiencies in computing resource allocation and parameter optimization. The lack of an effective coordination mechanism between short-term parameter adjustment and long-term performance optimization during the sequencing process results in inefficient utilization of system resources and is unable to meet the performance requirements of large-scale sequencing tasks. At the same time, the sequencing system has limited real-time monitoring and dynamic optimization capabilities for sequencing quality, making it difficult to cope with complex and changing sequencing environments. Summary of the invention

[0004] The main purpose of the present invention is to provide a parallel processing synchronization optimization method and system for high-throughput gene sequencing, so as to improve the sequencing accuracy and improve the utilization rate of computing resources.

[0005] To achieve the above object, the present invention provides a method for parallel processing synchronization optimization of high-throughput gene sequencing, comprising the following steps:

[0006] Perform channel allocation and signal acquisition initialization for multiple DNA samples to obtain a signal acquisition parameter matrix;

[0007] Performing feature extraction and cluster analysis on the fluorescence signals collected by each channel according to the signal collection parameter matrix to obtain a DNA base feature sequence and a signal classification probability matrix;

[0008] Based on the DNA base feature sequence and the signal classification probability matrix, the short-term parameters of the sequencing cycle are dynamically adjusted by a convex optimization algorithm, and the long-term sequencing performance is optimized by an integer programming algorithm to obtain a sequencing parameter optimization set and a sequencing cycle allocation strategy;

[0009] The sequencing parameter optimization set and the sequencing cycle allocation strategy are input into a double-layer nested control model, and are jointly solved by an upper-layer resource allocation network and a lower-layer parameter optimization network to obtain a global optimization control parameter set.

[0010] The present invention also provides a parallel processing synchronization optimization system for high-throughput gene sequencing, comprising:

[0011] An initialization module is used to perform channel allocation and signal acquisition initialization for multiple DNA samples to obtain a signal acquisition parameter matrix;

[0012] An analysis module, used to perform feature extraction and cluster analysis on the fluorescence signals collected by each channel according to the signal collection parameter matrix, to obtain a DNA base feature sequence and a signal classification probability matrix;

[0013] An optimization module, which is used to dynamically adjust the short-term parameters of the sequencing cycle through a convex optimization algorithm based on the DNA base feature sequence and the signal classification probability matrix, and optimize the long-term sequencing performance through an integer programming algorithm to obtain a sequencing parameter optimization set and a sequencing cycle allocation strategy;

[0014] The solution module is used to input the sequencing parameter optimization set and the sequencing cycle allocation strategy into the double-layer nested control model, and jointly solve them through the upper resource allocation network and the lower parameter optimization network to obtain a global optimization control parameter set.

[0015] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0016] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above-mentioned methods are implemented.

[0017] In summary, the technical solution provided by the present invention uses a multi-channel DNA sample parallel loading and signal acquisition parameter initialization mechanism, combined with a Markov chain state transition model for dynamic parameter adjustment, which effectively reduces signal interference between channels and improves sequencing throughput; a triple recognition network and multiple sequence alignment technology are used to extract and match DNA base features, and a signal classification algorithm based on maximum likelihood is used to improve sequencing accuracy; a double-layer nested control model is introduced, and the utilization rate of computing resources is improved through the synergy of an upper-layer resource allocation network and a lower-layer parameter optimization network; through the combined application of a convex optimization algorithm and an integer programming algorithm, an effective combination of short-term dynamic adjustment of sequencing parameters and long-term performance optimization is achieved, thereby improving the optimization efficiency of the sequencing cycle; based on a hierarchical control architecture, precise regulation of the sequencing system is achieved, and the overall performance and stability of the system are improved through multi-objective optimization and collaborative iterative calculation. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1It is a schematic diagram of the steps of a method for parallel processing and synchronization optimization of high-throughput gene sequencing in one embodiment of the present invention;

[0019] Figure 2 It is a structural block diagram of a parallel processing synchronization optimization system for high-throughput gene sequencing in one embodiment of the present invention;

[0020] Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0021] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0023] Reference Figure 1 This embodiment provides a method for parallel processing synchronization optimization of high-throughput gene sequencing, comprising the following steps:

[0024] S1, performing channel allocation and signal acquisition initialization for multiple DNA samples to obtain a signal acquisition parameter matrix;

[0025] Among them, the channel spatial distribution of multiple DNA samples is calculated to ensure reasonable resource allocation under limited channel capacity. By analyzing the matching relationship between the number of samples and the channel capacity, a channel allocation table is constructed to clearly define the specific channel allocation scheme for each DNA sample. Based on this channel allocation table, the concentration of each DNA sample is gradient measured to determine the specific concentration distribution of the sample in each channel, and a sample concentration distribution matrix is ​​generated, which reflects the distribution characteristics of the concentration of all DNA samples in the channel space. The sample concentration distribution matrix is ​​used to analyze the signal interference and obtain the signal baseline data of each channel. Based on the channel signal baseline data, the excitation light intensity is quantitatively calculated to ensure that the excitation efficiency of the fluorescence signal is optimal under different concentration conditions, and the light intensity control matrix is ​​obtained. Based on the light intensity control matrix, the time window for signal acquisition is optimized. The design of the time window directly affects the effective acquisition and dynamic response capability of the fluorescence signal. Through optimization calculation, a time window parameter set is obtained, which defines the time range and sampling frequency of the acquisition signal. Based on the time window parameter set, the gain coefficient of the signal is calculated to adjust the signal amplitude and sensitivity to ensure that weak signals can be fully amplified and strong signals will not be over-saturated. The gain coefficient matrix is ​​obtained to provide specific signal gain settings for different channels. The state probability calculation is performed on the gain coefficient matrix. By modeling the signal state, the state transition probability matrix is ​​calculated using mathematical statistical methods. The matrix reflects the possibility of signal transition between different states and provides a quantitative description of the dynamic characteristics of the signal. Based on the state transition probability matrix, multiple parameters are coupled and analyzed to fully consider the mutual influence between channels and the correlation between parameters. Through coupling analysis, a comprehensive and optimized signal acquisition parameter matrix is ​​obtained, which integrates information such as channel allocation, light intensity control, time window and gain adjustment.

[0026] S2, performing feature extraction and cluster analysis on the fluorescence signals collected by each channel according to the signal acquisition parameter matrix to obtain the DNA base feature sequence and signal classification probability matrix;

[0027] Specifically, the fluorescence signal of each channel is processed according to the signal acquisition parameter matrix, and the pure fluorescence signal data is extracted by eliminating the background noise. The background noise elimination is based on the filtering algorithm or the baseline correction technology to ensure that the random interference components are eliminated, so that the characteristics of the fluorescence signal are more prominent and clear. The pure fluorescence signal data after noise processing is mapped to the multidimensional feature space, and the multidimensional feature data set is constructed by quantitatively analyzing the feature dimensions such as the intensity, frequency, and waveform of the fluorescence signal. The multidimensional feature data set is normalized, and the numerical range of different feature dimensions is adjusted to the standardized range to eliminate the analysis bias caused by the scale difference, and a standardized feature matrix is ​​generated. The standardized feature matrix is ​​input into the maximum likelihood detection model, which classifies each fluorescence signal through the probability density function to calculate the probability of the sample category to which it belongs, and generates an initial classification probability matrix. The initial classification probability matrix is ​​optimized by the iterative clustering method. Iterative clustering uses methods such as Gaussian mixture model (GMM) or expectation maximization algorithm (EM) to gradually refine the boundaries of signal classification and improve classification accuracy. In the iterative process, clustering label sequences with clear attribution are generated by continuously adjusting the cluster center and category probability distribution. The clustering label sequence labels each fluorescent signal by category. Based on the clustering label sequence, DNA base recognition and sequence matching calculations are performed. By combining the temporal characteristics of the fluorescent signal and the clustering results, the sequences are aligned using known DNA base feature templates to generate a preliminary DNA base feature sequence. The initially generated feature sequence is corrected. The correction process uses an error model to correct the existing base misjudgments, reduce the sequencing error rate, and generate a corrected feature sequence. The classification probability of the corrected feature sequence is optimized, and the classification probability distribution is optimized by combining context information with multidimensional data such as signal intensity to obtain a signal classification probability matrix.

[0028] The multidimensional feature data set is processed to extract the fluorescence intensity feature, wavelength feature and time series feature, and the numerical range of each feature is calculated to construct the feature range matrix. On this basis, the linear transformation method is used to preprocess the data of each feature dimension, and the feature value is mapped to a unified range to generate preprocessed feature data. The preprocessed feature data is statistically calculated to eliminate the offset and scale differences between features. By calculating the mean of each feature dimension and performing centralization, all feature data are adjusted to a zero mean state to obtain a zero mean feature matrix. The variance of the zero mean feature matrix is ​​calculated to generate a feature variance vector, which describes the fluctuation range of each feature dimension. Using the feature variance vector, the zero mean feature matrix is ​​scaled to adjust the variance of each feature dimension to a unit value to generate a standardized feature matrix. The standardized feature matrix is ​​statistically analyzed to extract the distribution characteristics of the feature. Through analysis, the distribution function of each feature is obtained, and the maximum likelihood detection model is input based on the feature distribution function. In the model, the probability density function of each feature is calculated by Gaussian distribution fitting, and these probability density functions describe the distribution pattern of the features in each channel. According to the feature probability density function, a joint probability distribution model is constructed to comprehensively analyze the correlation between multidimensional features. The joint probability distribution model is an extension based on Gaussian distribution and is used to process the coordinated change relationship of multidimensional data. In order to optimize the classification performance, a log-likelihood function is constructed based on the joint probability distribution. The log-likelihood function is the logarithmic form of the joint probability distribution and is used to measure the degree of fit between the observed data and the model parameters. The log-likelihood function is maximized by the gradient descent algorithm, and the model parameters are iteratively updated to obtain the optimal parameter set. Through the optimal parameters, the log-likelihood values ​​of each sample category in each channel are calculated and organized into a likelihood value matrix. The likelihood value matrix is ​​subjected to probability normalization operation to convert the log-likelihood value into an interpretable probability value to obtain a category probability vector. These probability values ​​represent the possibility that each signal belongs to each sample category. In order to improve the reliability of classification, confidence threshold filtering is performed based on the category probability vector to screen out high-confidence probability data and exclude noise and misjudgment. The high confidence probability data are rearranged according to the channel identifier to construct the initial classification probability matrix.

[0029] The initial classification probability matrix is ​​subjected to feature dimensionality reduction. The first feature dimensionality reduction method, i.e., linear discriminant analysis, is used to extract the main classification feature set, which can maximize the discrimination between categories and thus enhance the classification effect. At the same time, the second feature dimensionality reduction adopts the local preservation projection technology to retain the local neighborhood structure of the data and generate a local feature set. The local preservation projection can effectively capture subtle changes in the feature space and provide a more accurate representation of complex signal patterns. The main classification feature set and the local feature set are fused to obtain a fused feature matrix. Through weighted combination or deep feature embedding, the advantages of the two features are combined to provide more comprehensive information for subsequent analysis. A distance metric function is constructed based on the fused feature matrix to measure the similarity between different data points. A mixed distance matrix is ​​generated by weighted combination of multiple distance metric functions. An iterative clustering algorithm is performed on the mixed distance matrix, and an improved K-means algorithm, a density clustering algorithm, or a hierarchical clustering algorithm is used to gradually optimize the cluster centers and classification results to obtain a cluster label sequence. The cluster label sequence is input into the triple recognition network for deep processing. The first recognition subnetwork is used to extract base features, and the potential patterns in the clustering label sequence are mined through convolutional neural networks and other technologies to extract features with biological significance. The second recognition subnetwork focuses on sequence pattern matching, and uses recurrent neural networks or long short-term memory networks to analyze the time dependence of the sequence, thereby identifying the order and pattern of DNA bases. The third recognition subnetwork performs quality scoring, comprehensively analyzes the credibility of the extracted features and matching patterns, generates quality scores for each base, and constructs a base candidate sequence set. Multiple sequence alignment calculations are performed on the generated base candidate sequence set, and methods such as Markov chain Monte Carlo or dynamic programming are used to generate an alignment sequence matrix. The alignment sequence matrix reveals the consistency characteristics between each candidate base by accurately evaluating the matching degree of the candidate sequences. Based on the alignment sequence matrix, sequence consistency analysis is performed to evaluate the credibility of each base in different candidate sequences to obtain a sequence credibility matrix. As an important basis for measuring base quality, the sequence credibility matrix effectively eliminates low-confidence base noise by screening and optimizing bases with higher credibility. Based on the sequence credibility matrix, the candidate sequence is optimized and reconstructed to generate the final DNA base feature sequence. During the reconstruction process, high-confidence regions in the candidate sequence set are comprehensively utilized, and the results of quality scoring and consistency analysis are combined to ensure the accuracy and completeness of the output sequence.

[0030] S3, based on the DNA base feature sequence and signal classification probability matrix, dynamically adjust the short-term parameters of the sequencing cycle through the convex optimization algorithm, optimize the long-term sequencing performance through the integer programming algorithm, and obtain the sequencing parameter optimization set and sequencing cycle allocation strategy;

[0031] It should be noted that the DNA base feature sequences are aligned to obtain a quality score matrix, which reflects the accuracy and consistency of the base sequence. Combined with the signal classification probability matrix, a short-term optimization model is constructed, which aims to optimize the dynamic parameters related to real-time data in the sequencing process. Based on the short-term optimization model, the constraints of quadratic programming are constructed, and the stability and accuracy of signal acquisition are comprehensively considered. The convex optimization algorithm is used to solve the operation and obtain the short-term parameter adjustment. The solution process uses the fast convergence characteristics of convex optimization to find the optimal solution for parameter adjustment under multiple constraints. The short-term parameter adjustment is constrained to ensure that the adjustment results meet the limitations of the actual sequencing equipment and system, and a short-term sequencing parameter set is generated. Based on the short-term parameter adjustment, the time series prediction model is used to speculate on the future sequencing performance. The performance prediction data is generated based on historical data and the trend of short-term parameters, which can reflect the potential changes of the sequencing system in long-term operation. Combined with the performance prediction data, a long-term optimization model is constructed, which focuses on the overall allocation efficiency of sequencing resources and the global optimization of long-term performance. The constraints of integer programming are established through the long-term optimization model, which include equipment resource allocation constraints, cycle time constraints, and performance targets. The integer programming problem is solved by the branch-and-bound algorithm. By partitioning and pruning the solution space, the exponential complexity of the calculation is effectively avoided, and the sequencing cycle allocation scheme is obtained. In order to ensure the practical feasibility of the scheme, the sequencing cycle allocation scheme is verified by resource constraints. The verification process includes a comprehensive analysis of the equipment capacity, the number of sample channels and the signal processing capability, and an executable target cycle scheme is generated. After the target cycle scheme is generated, the sensitivity analysis of the key parameters of the scheme is performed. By constructing a parameter influence matrix, the influence of each sequencing parameter on the overall performance is quantified, the importance of the parameters is ranked, and the parameter priority sequence is generated. The parameter priority sequence and the short-term sequencing parameter set are subjected to multi-objective optimization calculations, and a comprehensive parameter scheme is generated by balancing the short-term and long-term optimization goals. The comprehensive parameter scheme comprehensively considers multiple goals of performance improvement, resource utilization and dynamic response. To verify the effectiveness of the scheme, a performance evaluation is performed, including indicators such as sequencing accuracy, throughput efficiency and resource consumption, and a sequencing parameter optimization set is generated. Based on the sequencing parameter optimization set, the target cycle scheme is dynamically adjusted. The adjustment process includes real-time optimization of resource allocation strategy and parameter configuration, and a sequencing cycle allocation strategy is generated. This strategy is able to dynamically adapt to changes in sequencing conditions while maximizing system efficiency and data accuracy.

[0032] S4, the sequencing parameter optimization set and sequencing cycle allocation strategy are input into the two-layer nested control model, and the upper-layer resource allocation network and the lower-layer parameter optimization network are jointly solved to obtain the global optimization control parameter set.

[0033] Specifically, the sequencing parameter optimization set is hierarchically decoupled, and the complex optimization problem is divided into two relatively independent control dimensions. Through decoupling, a computing resource control parameter subset and a sequencing performance control parameter subset are obtained, wherein the computing resource control parameter subset mainly involves the allocation and scheduling of hardware resources, while the sequencing performance control parameter subset focuses on optimizing the key performance parameters in the sequencing process. The computing resource control parameter subset is input into the upper-layer resource allocation network, which consists of a resource scheduling layer and a task allocation layer. In the resource scheduling layer, an optimization function is constructed based on the sequencing cycle allocation strategy to achieve efficient allocation of computing resources and storage resources. The resource scheduling layer includes four scheduling units, each of which is internally configured with a computing resource allocation module and a storage resource allocation module to manage the scheduling of processing power and data storage capacity respectively. Through optimized calculation, the resource scheduling layer generates an initial resource allocation plan, which clarifies the scheduling strategy of each resource. The initial resource allocation plan is input into the task allocation layer, in which the task decomposition unit divides the sequencing task into subtasks. The division is based on the complexity and dependency of the task, and the sequencing task is disassembled into multiple independent or partially related subtasks. At the same time, the priority allocation unit determines the execution order of tasks according to the timeliness and resource requirements of subtasks and generates a task allocation sequence. This sequence optimizes the order of task execution, thereby improving the overall operating efficiency of the system. At the same time, the subset of sequencing performance control parameters is input into the lower parameter optimization network, which consists of a parameter update layer and a performance evaluation layer. In the parameter update layer, multi-objective optimization is solved to balance the conflicts between sequencing accuracy, throughput and resource consumption. The parameter update layer contains six optimization units, each of which has an embedded parameter generation module and a constraint verification module. The parameter generation module generates new parameter settings through an optimization algorithm, while the constraint verification module ensures that the generated parameters meet the constraints of the device and system. After multiple rounds of optimization calculations, the parameter update layer outputs a parameter update plan. The parameter update plan is input into the performance evaluation layer. In the performance evaluation layer, the performance indicator calculation unit evaluates the optimized parameters and quantifies the optimization effect by calculating indicators such as sequencing accuracy, signal quality and resource utilization. At the same time, the error compensation unit dynamically adjusts the deviations in the optimization process to improve the accuracy of the parameter plan. Through comprehensive calculation and correction, the performance evaluation layer finally outputs the optimized parameter plan. The task allocation sequence and optimization parameter scheme are input into the two-layer collaborative optimizer, and the generation of the global optimization control parameter set is realized through interactive iterative operations between the upper and lower layers. In this stage, the upper-layer resource allocation network and the lower-layer parameter optimization network are dynamically coordinated through bidirectional data exchange. The resource allocation results of the upper layer provide support for the parameter optimization of the lower layer, and the performance optimization results of the lower layer are fed back to the upper layer to adjust the resource scheduling strategy. Through multiple rounds of interactive iterations, the two-layer collaborative optimizer finally converges to a global optimal solution and generates a global optimization control parameter set.

[0034] In one example, channel allocation and signal acquisition initialization are performed on multiple DNA samples to obtain a signal acquisition parameter matrix, including:

[0035] Calculate the channel spatial distribution of multiple DNA samples, establish a channel allocation table based on the matching relationship between the number of samples and the channel capacity, and determine the concentration gradient of each DNA sample based on the channel allocation table to obtain a sample concentration distribution matrix;

[0036] Perform signal interference analysis on the sample concentration distribution matrix to obtain channel signal baseline data, and perform quantitative calculation on the excitation light intensity based on the channel signal baseline data to obtain a light intensity control matrix;

[0037] The signal acquisition window is optimized for the light intensity control matrix to obtain a time window parameter set, and the signal gain coefficient is calculated for the time window parameter set to obtain a gain coefficient matrix;

[0038] The state probability of the gain coefficient matrix is ​​calculated to obtain the state transition probability matrix, and multi-parameter coupling analysis is performed based on the state transition probability matrix to obtain the signal acquisition parameter matrix.

[0039] In this example, in the sample allocation stage, a channel allocation table is established based on the matching relationship between the number of samples and the channel capacity. Suppose the number of samples is And the capacity of a single channel is , number of channels pass Calculate. Assign each sample to a specific channel, the allocation table It is represented by a binary matrix, where Representation sample Is it assigned to a channel? The allocation table must not only meet the channel capacity limit, but also try to balance the load of each channel. After completing the sample allocation, the concentration gradient of each sample is measured according to the channel allocation table. The measurement process is modeled by the linear relationship between the fluorescence signal intensity and the sample concentration. With signal strength The relationship is expressed as ,in and is the coefficient obtained through calibration experiment, and the sample concentration distribution matrix is ​​obtained , describes the specific distribution of sample concentrations in all channels. For the sample concentration distribution matrix Signal interference analysis is performed to quantify the signal superposition effect between different channels. The interference effect is calculated using a baseline shift model, assuming that the total channel signal is , baseline shift It is expressed as ,in is the interference coefficient. By calculating the baseline offset of all channels, the channel signal baseline data matrix is ​​formed. The baseline data reflects the background signal level of different channels and provides a basis for adjusting the excitation light intensity. Based on the signal baseline data, the excitation light intensity of each channel is adjusted. The excitation light intensity is inversely proportional to the baseline shift, which can be expressed as ,in is the adjustment coefficient. By calculating the light intensity of all channels, the light intensity control matrix is ​​generated . Light intensity control matrix Optimize the time window. With light intensity It is a linear relationship, expressed as ,in is the time adjustment coefficient. Optimizing the time window can not only ensure the complete acquisition of the signal, but also improve the dynamic response capability of the system. After that, it is analyzed to calculate the signal gain coefficient matrix . Gain factor Defined as ,in is the gain adjustment parameter. The gain coefficient is used to adjust the signal strength to adapt it to the dynamic range of the system and avoid acquisition errors caused by too weak or too strong signals. Through the Markov chain model, based on the gain coefficient matrix Calculate the state transition probability matrix Assumptions Indicates that the signal is from the state Transfer to state The probability of state transition is expressed in matrix form as The modeling of state probability helps to describe the dynamic changes of signal characteristics and provide a basis for subsequent multi-parameter coupling analysis. Based on this, multi-parameter coupling analysis is performed. By constructing the coupling matrix , considering the combined effects of signal gain, acquisition time window and baseline interference, the signal acquisition parameter matrix Expressed as ,in Reflection gain factor and state probability The final signal acquisition parameter matrix is Integrates all information from signal distribution to acquisition optimization.

[0040] In one example, feature extraction and cluster analysis are performed on the fluorescence signals collected by each channel according to the signal collection parameter matrix to obtain a DNA base feature sequence and a signal classification probability matrix, including:

[0041] Eliminating background noise of the fluorescence signal according to the signal acquisition parameter matrix to obtain pure fluorescence signal data, and performing multi-dimensional feature space mapping on the pure fluorescence signal data to obtain a multi-dimensional feature data set;

[0042] Normalize the multidimensional feature data set to obtain a standardized feature matrix, and input the standardized feature matrix into the maximum likelihood detection model. The probability of the sample category to which each signal belongs is calculated through the probability density function to obtain the initial classification probability matrix.

[0043] Iteratively cluster the initial classification probability matrix to obtain a clustering label sequence, and perform base recognition and sequence matching calculations on the clustering label sequence to obtain a DNA base feature sequence;

[0044] The DNA base feature sequence is corrected to obtain a corrected feature sequence, and the classification probability of the corrected feature sequence is optimized to obtain a signal classification probability matrix.

[0045] In this example, according to the signal acquisition parameter matrix , the background noise of the fluorescence signal is eliminated. The collected data of the fluorescence signal is expressed as ,in Indicates the target signal, Represents background noise. Using the noise estimation model ,in and are the noise amplitude factor and baseline offset, The white noise component can be removed by filtering technology (such as low-pass filtering) or by signal decomposition technology (such as wavelet transform) to separate the target signal and the noise component to obtain pure fluorescence signal data. . Perform multi-dimensional feature space mapping on the pure fluorescence signal data and construct a multi-dimensional feature data set. Contains sampling values ​​at multiple time points, through feature extraction function , converting the signal into a feature vector containing fluorescence intensity, time gradient, and frequency domain components ,in is the characteristic dimension, Indicates The signal is By extracting features from all signal samples, we can get a multi-dimensional feature dataset. ,in is the number of signal samples. The multidimensional feature data set is normalized to standardize the range of the features. Normalization is done by zero mean and variance scaling. The normalization formula is:

[0046] ;

[0047] in and Respectively The mean and standard deviation of the features, is the normalized eigenvalue. The normalized eigenvalue matrix Ensure that the mean of each feature is 0 and the variance is 1, thereby eliminating the impact of different feature scales on subsequent analysis. Input into the maximum likelihood detection model, and calculate the sample category probability of each signal through the probability density function. Assume that the signal category It obeys Gaussian distribution, and its probability density function It is expressed as:

[0048] ;

[0049] in and Respectively The maximum likelihood detection model is based on each sample The maximum a posteriori probability Determine its category probability, the calculation formula is:

[0050] ;

[0051] in Yes Category The prior probability of . The above formula is used to calculate all signals to obtain the initial classification probability matrix ,in Each row of corresponds to the category probability distribution of a signal. For the initial classification probability matrix Perform iterative clustering to generate cluster label sequences. Use Gaussian mixture model (GMM) or expectation maximization (EM) algorithm to optimize cluster centers and category probability distribution. Assume that each category The cluster center is , by iterative calculation to maximize the following log-likelihood function:

[0052] ;

[0053] in is the set of all model parameters, Yes Category Mixing coefficient. Iterate until convergence and output cluster label sequence ,in It is a sample The category to which it belongs. Cluster the label sequence It is converted into base recognition and sequence matching, and a sequence alignment algorithm (such as Needleman-Wunsch or Smith-Waterman algorithm) is used to match the known base reference template to obtain the DNA base feature sequence. The DNA base feature sequence is corrected and the possible errors are adjusted through the model. The corrected sequence is optimized through classification probability, the signal classification probability is updated, and the final signal classification probability matrix is ​​generated.

[0054] In one example, a multidimensional feature data set is normalized to obtain a standardized feature matrix, and the standardized feature matrix is ​​input into a maximum likelihood detection model, and the probability of the sample category to which each signal belongs is calculated by a probability density function to obtain an initial classification probability matrix, including:

[0055] The fluorescence intensity feature, wavelength feature and time series feature in the multidimensional feature data set are respectively calculated in numerical range to obtain a feature range matrix, and each feature dimension is linearly transformed based on the feature range matrix to obtain preprocessed feature data;

[0056] Based on the preprocessed feature data, the features of each dimension are averaged and centralized to obtain a zero-mean feature matrix, and the variance of the zero-mean feature matrix is ​​calculated to obtain a feature variance vector;

[0057] The zero-mean feature matrix is ​​scaled according to the feature variance vector to obtain a standardized feature matrix, and a statistical distribution analysis is performed on the standardized feature matrix to obtain a feature distribution function;

[0058] The standardized feature matrix and feature distribution function are input into the maximum likelihood detection model, the feature probability density function is obtained by Gaussian distribution fitting calculation, and the joint probability distribution is constructed according to the feature probability density function;

[0059] Based on the joint probability distribution, a log-likelihood function is constructed, and the log-likelihood function is maximized by the gradient descent algorithm to obtain the optimal parameter set. The log-likelihood value of the sample category of each channel is calculated according to the optimal parameter set to obtain the likelihood value matrix.

[0060] The likelihood value matrix is ​​subjected to probability normalization operation to obtain a category probability vector, and confidence threshold filtering is performed based on the category probability vector to obtain high-confidence probability data. The high-confidence probability data is rearranged according to the channel identifier to obtain an initial classification probability matrix.

[0061] In this example, for the multidimensional feature dataset, we start with three key dimensions: fluorescence intensity, wavelength, and time series. Suppose the original feature dataset is ,in Indicates The sample in The value of feature dimensions. For each feature dimension , calculate its value range for:

[0062] ;

[0063] in and Respectively represent The maximum and minimum values ​​of the features. The numerical ranges of all feature dimensions constitute the feature range matrix ,in is the total number of feature dimensions. Based on the feature range matrix, the value of each feature dimension is linearly transformed to obtain the preprocessed feature data , the linear transformation formula is:

[0064] ;

[0065] Through this transformation, all eigenvalues ​​are normalized to the interval [0,1], eliminating the scale differences between different feature dimensions. The preprocessed feature data is averaged and centered to generate a zero-mean feature matrix. . No. The mean of the feature dimensions Calculated as:

[0066] ;

[0067] in is the total number of samples. The centering is done by subtracting the mean from each eigenvalue. The eigenvalues ​​after centering are It is expressed as:

[0068] ;

[0069] After obtaining the zero-mean feature matrix, calculate the variance of each feature dimension , the characteristic variance vector Each component of is calculated by the following formula:

[0070] ;

[0071] Use the feature variance vector to scale the zero-mean feature matrix to generate a standardized feature matrix , the standardized formula is:

[0072] ;

[0073] The standardization process ensures that the mean of each feature dimension is 0 and the variance is 1, thereby balancing the contribution of different features. Perform statistical distribution analysis to construct a probability distribution function for each feature Assuming that the eigenvalues ​​follow a Gaussian distribution, the probability density function is:

[0074] ;

[0075] in and Respectively The mean and standard deviation of the features (set to 0 and 1 through standardization). By combining the probability distributions of multiple features, a joint probability distribution is constructed , for the sample The joint probability density function is expressed as:

[0076] ;

[0077] Based on the joint probability distribution, construct the log-likelihood function , whose expression is:

[0078] ;

[0079] in Represents the set of model parameters. The log-likelihood function is maximized by the gradient descent algorithm, and the optimization objective is:

[0080] ;

[0081] In order to obtain the optimal parameter set Finally, these parameters are used to calculate the class log-likelihood value of the samples in each channel ,in is the category index, sample Belongs to category The log-likelihood value calculation formula is:

[0082] ;

[0083] in Representation sample In category The joint probability density value under . All log-likelihood values ​​constitute the likelihood value matrix . For the likelihood matrix Perform probability normalization operations to calculate the probability vector of each sample belonging to each category ,in Representation sample Belongs to category The probability is calculated as:

[0084] ;

[0085] By filtering through confidence thresholds, only high-confidence probability data is retained, and the confidence threshold is set , keep satisfied The probability value of is set to 0, and the rest are set to 0. Rearrange the high confidence probability data according to the channel identification to generate the initial classification probability matrix .

[0086] In one example, the initial classification probability matrix is ​​iteratively clustered to obtain a clustering label sequence, and base recognition and sequence matching calculations are performed on the clustering label sequence to obtain a DNA base feature sequence, including:

[0087] Performing first feature dimensionality reduction and second feature dimensionality reduction on the initial classification probability matrix, wherein the first feature dimensionality reduction obtains a main classification feature set through linear discriminant analysis, and the second feature dimensionality reduction obtains a local feature set through local preservation projection, and performing feature fusion on the main classification feature set and the local feature set to obtain a fused feature matrix;

[0088] Based on the fusion feature matrix, a distance metric function is constructed and weighted combination is performed to obtain a mixed distance matrix. The mixed distance matrix is ​​iteratively clustered to obtain a clustering label sequence.

[0089] The clustering label sequence is input into the triple recognition network, base feature extraction is performed through the first recognition sub-network, sequence pattern matching is performed through the second recognition sub-network, and quality scoring is performed through the third recognition sub-network to obtain a base candidate sequence set;

[0090] A multiple sequence alignment calculation is performed on the base candidate sequence set to obtain an alignment sequence matrix, and a sequence consistency analysis is performed based on the alignment sequence matrix to obtain a sequence credibility matrix. The sequence is optimized and reconstructed according to the sequence credibility matrix to obtain a DNA base feature sequence.

[0091] In this example, the initial classification probability matrix ,in Representation sample Belongs to category The probability of feature dimensionality reduction is performed. The main classification feature set is extracted through linear discriminant analysis. The goal of linear discriminant analysis is to maximize the scatter matrix between categories. and the within-category scatter matrix The characteristic ratio of . Define the scatter matrix between categories and the within-category scatter matrix They are:

[0092] ;

[0093] in is the number of categories, Yes Category The number of samples, Yes Category The mean vector of is the global mean vector, Yes Category By solving the generalized eigenvalue problem , select the first features to generate the main classification feature set . Extract local feature sets through local preserving projection. Local preserving projection constructs an adjacency graph To maintain the local geometric structure between samples, define for:

[0094] ;

[0095] in is the width parameter of the Gaussian kernel. By optimizing the objective function , that is, retain the distance relationship between adjacent points and obtain the local feature set .Will and Perform feature fusion and obtain the fused feature matrix through weighted linear combination :

[0096] ;

[0097] in is the fusion weight, which controls the contribution ratio of the two features. Based on the fusion feature matrix , construct a distance metric function To calculate the similarity between samples. Let the Euclidean distance be And the Mahalanobis distance is , define the mixed distance function as:

[0098] ;

[0099] in is a weight parameter used to balance the contribution of the two distances. The mixed distance matrix is ​​generated by calculating the mixed distance for all samples. . For the mixed distance matrix Perform iterative clustering and use density clustering algorithms (such as DBSCAN) to automatically identify the clustering structure of samples. The output cluster label sequence ,in Representation sample The clustering category. Input triple recognition network. The first recognition subnetwork is used for base feature extraction, and the feature pattern in the clustering result is extracted through the convolutional neural network; the second recognition subnetwork uses the long short-term memory network for sequence pattern matching to identify the temporal characteristics of the base; the third recognition subnetwork calculates the quality score of each base sequence through the fully connected network and outputs the base candidate sequence set. Perform multiple sequence alignment calculations on the candidate base sequence set and use a dynamic programming algorithm (such as the Smith-Waterman algorithm) to generate an alignment sequence matrix The alignment matrix generates a sequence credibility matrix through sequence consistency analysis. ,in Indicates the first Position belongs to The confidence of the bases. Based on the sequence confidence matrix Optimize and reconstruct the sequence, select the base position with the highest confidence to generate the final DNA base feature sequence For example, in the alignment results, if the confidence of the candidate bases A, T, and G at position 1 is 0.8, 0.1, and 0.1, the optimization result selects base A. The final DNA base feature sequence is optimized through the above process to ensure the accuracy and reliability of the sequence, providing high-quality input for subsequent analysis.

[0100] In one example, based on the DNA base feature sequence and the signal classification probability matrix, the short-term parameters of the sequencing cycle are dynamically adjusted by a convex optimization algorithm, and the long-term sequencing performance is optimized by an integer programming algorithm to obtain a sequencing parameter optimization set and a sequencing cycle allocation strategy, including:

[0101] Perform sequence alignment on DNA base feature sequences to obtain a quality score matrix, and build a short-term optimization model based on the quality score matrix;

[0102] Based on the short-term optimization model and the signal classification probability matrix, quadratic programming constraints are constructed, and the short-term parameter adjustment is obtained by solving the problem through the convex optimization algorithm. The short-term parameter adjustment is then subjected to constraint checking to obtain the short-term sequencing parameter set.

[0103] Perform time series prediction based on the short-term sequencing parameter set to obtain performance prediction data, and build a long-term optimization model based on the performance prediction data;

[0104] Based on the long-term optimization model, integer programming constraints are constructed, and the branch-and-bound algorithm is used to solve and calculate the sequencing cycle allocation plan. The sequencing cycle allocation plan is then verified by resource constraints to obtain the target cycle plan.

[0105] Perform parameter sensitivity analysis on the target cycle plan to obtain a parameter impact matrix, and rank the sequencing parameters by importance based on the parameter impact matrix to obtain a parameter priority sequence;

[0106] Perform multi-objective optimization on the parameter priority sequence and the short-term sequencing parameter set to obtain a comprehensive parameter solution, and perform performance evaluation on the comprehensive parameter solution to obtain a sequencing parameter optimization set;

[0107] The target cycle plan is dynamically adjusted based on the sequencing parameter optimization set to obtain the sequencing cycle allocation strategy.

[0108] In this example, the DNA base feature sequences are aligned, the matching of each position with the reference sequence is calculated, and a quality score matrix is ​​generated. ,in Indicates Positions in base type Assuming that the base matching score is proportional to the alignment score, it is expressed as ,in It is a base In Location The weight of is the alignment score. By calculating the scores of all positions, a complete quality score matrix is ​​formed. Based on the quality score matrix, a short-term optimization model is constructed. The optimization goal is to maximize the accuracy of base recognition in the current sequencing cycle. The objective function of the short-term optimization model is expressed as:

[0109] ;

[0110] in is a decision variable, indicating whether to choose a base In Location The constraint condition is that only one base can be selected at each position, that is:

[0111] ;

[0112] Combined signal classification probability matrix , the optimization conditions are introduced through the quadratic programming model, and the following quadratic programming constraints are constructed:

[0113] ;

[0114] The goal is to minimize the deviation between the optimization result and the classification probability. Use a convex optimization algorithm (such as gradient descent or interior point method) to solve the objective function and obtain the short-term parameter adjustment amount. ,in represents the adjustment vector of the optimization parameters. Perform constraint checks to ensure that parameter adjustments meet the physical limitations of the sequencing equipment, for example:

[0115] ;

[0116] After verification, the short-term sequencing parameter set is obtained Based on the short-term sequencing parameter set, time series prediction is performed to estimate future performance. The performance prediction model uses the autoregressive integrated moving average model (ARIMA) or recurrent neural network (RNN), inputs the sequencing parameters of the current cycle and outputs performance indicators , predicting the future Performance data for each cycle The performance prediction data is used to build a long-term optimization model, the optimization goal is to maximize the future Overall performance for the cycle:

[0117] ;

[0118] in For performance indicators The long-term optimization model is solved by integer programming method to construct allocation constraints, such as the total amount of resources per cycle. Cannot exceed limit:

[0119] ;

[0120] in It is a cycle Assign resources to tasks The integer programming problem is solved by the branch-and-bound algorithm to obtain the sequencing cycle allocation scheme ,in Indicates The resource allocation strategy of the cycle. Further verify the resource constraints of the cycle allocation plan to ensure that the hardware and time constraints are met during actual execution, and generate the target cycle plan . Then, the parameter sensitivity analysis of the target cycle scheme is carried out to construct the parameter impact matrix ,in Indicates sequencing parameters Performance indicators The sensitivity analysis is done by local perturbation method, for example:

[0121] ;

[0122] Based on the parameter impact matrix, the sequencing parameters are ranked in importance and a parameter priority sequence is generated. , high-priority parameters are allocated more resources and optimized calculations first. The parameter priority sequence is combined with the short-term sequencing parameter set to perform multi-objective optimization. The objective function of the multi-objective optimization model is to balance sequencing accuracy, throughput, and resource consumption, and the weighted sum method is used to construct the optimization goal:

[0123] ;

[0124] in It is The objective function, is the corresponding weight. Use evolutionary algorithms (such as NSGA-II) to solve multi-objective optimization problems and generate comprehensive parameter solutions The comprehensive parameter scheme is evaluated for performance, including sequencing data quality, operational efficiency, and resource utilization, and finally an optimized set of sequencing parameters is generated. .based on , for the target cycle plan Perform dynamic adjustments to optimize resource allocation strategies and parameter update plans to obtain the final sequencing cycle allocation strategy This strategy can adapt to changing sequencing conditions in real time while maximizing system performance and data quality.

[0125] In one example, the sequencing parameter optimization set and the sequencing cycle allocation strategy are input into the two-layer nested control model, and the upper-layer resource allocation network and the lower-layer parameter optimization network are jointly solved to obtain a global optimization control parameter set, including:

[0126] The sequencing parameter optimization set is hierarchically decoupled to obtain a computing resource control parameter subset and a sequencing performance control parameter subset, and the computing resource control parameter subset is input into an upper-layer resource allocation network, which includes a resource scheduling layer and a task allocation layer;

[0127] Based on the sequencing cycle allocation strategy, a resource scheduling layer optimization function is constructed. The resource scheduling layer includes four scheduling units, each of which contains a computing resource allocation module and a storage resource allocation module to obtain an initial resource allocation plan.

[0128] The initial resource allocation plan is input into the task allocation layer, which includes a task decomposition unit and a priority allocation unit. The task decomposition unit divides the sequencing task into subtasks, and the priority allocation unit determines the task execution order to obtain a task allocation sequence.

[0129] Input the sequencing performance control parameter subset into the lower parameter optimization network, which includes a parameter update layer and a performance evaluation layer, and perform multi-objective optimization on the parameter update layer. The parameter update layer includes 6 optimization units, each of which includes a parameter generation module and a constraint verification module to obtain a parameter update plan;

[0130] The parameter update scheme is input into the performance evaluation layer, which includes a performance index calculation unit and an error compensation unit, to obtain an optimized parameter scheme;

[0131] The task allocation sequence and optimization parameter scheme are input into the two-layer collaborative optimizer, and the global optimization control parameter set is obtained through interactive iterative operations of the upper and lower layers.

[0132] In this example, the sequencing parameter optimization set is decoupled in layers to clarify the control objectives and tasks at different levels. , by classifying parameter attributes, it is divided into computing resource control parameter subsets and a subset of sequencing performance control parameters .in Includes parameters related to computing resources and storage allocation, such as processor load, storage space allocation, etc.; Focus on key performance parameters that directly affect sequencing quality, such as signal strength, sampling frequency, etc. Input the upper resource allocation network, which consists of the resource scheduling layer and the task allocation layer. The resource scheduling layer contains four scheduling units Each scheduling unit includes a computing resource allocation module and a storage resource allocation module. Based on the sequencing cycle allocation strategy , construct the optimization function of the resource scheduling layer, the goal is to maximize resource utilization and minimize resource conflicts The optimization function is expressed as:

[0133] ;

[0134] in is a weight factor that balances resource utilization and conflict optimization goals. The computing resource allocation module calculates the requirements of the task and available computing resources relationship, ensuring that the allocation satisfies:

[0135] ;

[0136] The storage resource allocation module is based on the task data requirements and available storage resources , ensuring that the allocation satisfies:

[0137] ;

[0138] By solving the optimization function, we get the initial resource allocation plan ,in Represents the scheduling unit The initial resource allocation plan Enter the task allocation layer, which consists of a task decomposition unit and a priority allocation unit. The task decomposition unit will sequence the tasks. Divide into subtasks according to task dependencies and data size , the task decomposition goal is to minimize the execution time of subtasks and communication overhead The priority allocation unit determines the order of task execution based on task urgency and resource requirements. The task assignment layer generates a task assignment sequence , including the execution order of all subtasks. At the same time, the sequencing performance control parameter subset Input the lower parameter optimization network, which consists of a parameter update layer and a performance evaluation layer. In the parameter update layer, multiple performance indicators, such as sequencing accuracy, are optimized simultaneously through a multi-objective optimization method. , Flux and power consumption The multi-objective optimization objective function is expressed as:

[0139] ;

[0140] in is a weight factor that balances the contribution of each optimization objective. The parameter update layer contains six optimization units Each unit includes a parameter generation module and a constraint checking module. The parameter generation module generates the optimized parameter values. , while the constraint checking module ensures that these parameters meet physical constraints and system requirements, such as:

[0141] ;

[0142] Verified parameter update scheme It is input into the performance evaluation layer, where the performance indicator calculation unit calculates the sequencing performance contribution of each optimized parameter, for example:

[0143] ;

[0144] The error compensation unit dynamically adjusts the deviation during the optimization process to ensure the output of the optimized parameter solution Reliability. Assign tasks to a sequence and optimization parameter scheme Input is a two-layer collaborative optimizer, which adjusts resource allocation strategies and performance optimization parameters through interactive iterative operations between the upper and lower layers. The upper-layer resource allocation network provides real-time resource status feedback to the lower-layer parameter optimization network, and the results of the lower-layer parameter optimization network in turn affect the resource allocation decisions of the upper layer. The iteration process continues until the upper and lower layer results converge, and the global optimization control parameter set is output. For example, in a sequencing cycle, if the computing resource requirements is 100 units, and the available resources The resource scheduling layer prioritizes task requirements and allocates remaining resources to low-priority tasks. At the same time, the parameter optimization network adjusts the sampling frequency and gain coefficient to optimize signal quality and energy consumption. The iterative results of the two-layer optimizer ensure the global coordination of computing resources and performance parameters, providing an efficient and dynamic control strategy for the sequencing process.

[0145] Reference Figure 2 This embodiment provides a parallel processing synchronization optimization system for high-throughput gene sequencing, including:

[0146] Initialization module 1 is used to perform channel allocation and signal acquisition initialization for multiple DNA samples to obtain a signal acquisition parameter matrix;

[0147] Analysis module 2, used for performing feature extraction and cluster analysis on the fluorescence signals collected by each channel according to the signal collection parameter matrix, and obtaining the DNA base feature sequence and signal classification probability matrix;

[0148] Optimization module 3 is used to dynamically adjust the short-term parameters of the sequencing cycle through a convex optimization algorithm based on the DNA base feature sequence and the signal classification probability matrix, and optimize the long-term sequencing performance through an integer programming algorithm to obtain the sequencing parameter optimization set and sequencing cycle allocation strategy;

[0149] Solving module 4 is used to input the sequencing parameter optimization set and sequencing cycle allocation strategy into the double-layer nested control model, and jointly solve through the upper-layer resource allocation network and the lower-layer parameter optimization network to obtain the global optimization control parameter set.

[0150] In this embodiment, for the specific implementation of each unit in the above system embodiment, please refer to the above method embodiment, which will not be repeated here.

[0151] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided. The computer device may be a server, and its internal structure may be as follows: Figure 3As shown. The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the above method is implemented.

[0152] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0153] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the above method is implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0154] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided by the present invention and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM.

[0155] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0156] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A parallel processing synchronization optimization method for high-throughput gene sequencing, characterized in that: The following steps are involved: Perform channel allocation and signal acquisition initialization for multiple DNA samples to obtain a signal acquisition parameter matrix; Performing feature extraction and cluster analysis on the fluorescence signals collected by each channel according to the signal collection parameter matrix to obtain a DNA base feature sequence and a signal classification probability matrix; Based on the DNA base feature sequence and the signal classification probability matrix, the short-term parameters of the sequencing cycle are dynamically adjusted by a convex optimization algorithm, and the long-term sequencing performance is optimized by an integer programming algorithm to obtain a sequencing parameter optimization set and a sequencing cycle allocation strategy; The sequencing parameter optimization set and the sequencing cycle allocation strategy are input into a two-layer nested control model, and are jointly solved through an upper-layer resource allocation network and a lower-layer parameter optimization network to obtain a global optimization control parameter set; specifically comprising: hierarchical decoupling of the sequencing parameter optimization set to obtain a computing resource control parameter subset and a sequencing performance control parameter subset, and inputting the computing resource control parameter subset into an upper-layer resource allocation network, wherein the upper-layer resource allocation network includes a resource scheduling layer and a task allocation layer; constructing a resource scheduling layer optimization function based on the sequencing cycle allocation strategy, wherein the resource scheduling layer includes four scheduling units, each scheduling unit includes a computing resource allocation module and a storage resource allocation module, and obtaining an initial resource allocation plan; inputting the initial resource allocation plan into a task allocation layer, wherein the task allocation layer includes a task decomposition layer, and the task allocation layer includes a task decomposition layer, wherein the task decomposition ... The method comprises the steps of: inputting the sequencing performance control parameter subset into a lower parameter optimization network, the lower parameter optimization network comprises a parameter updating layer and a performance evaluation layer, and performing multi-objective optimization on the parameter updating layer, the parameter updating layer comprises 6 optimization units, each of which comprises a parameter generation module and a constraint checking module, and obtaining a parameter updating scheme; inputting the parameter updating scheme into a performance evaluation layer, the performance evaluation layer comprises a performance index calculation unit and an error compensation unit, and obtaining an optimized parameter scheme; inputting the task allocation sequence and the optimized parameter scheme into a two-layer collaborative optimizer, and obtaining a global optimized control parameter set through interactive iterative operations of the upper and lower layers.

2. The method for parallel processing and synchronization optimization of high-throughput gene sequencing according to claim 1, characterized in that: The channel allocation and signal acquisition initialization of multiple DNA samples to obtain a signal acquisition parameter matrix includes: Calculating the channel spatial distribution of multiple DNA samples, establishing a channel allocation table based on the matching relationship between the number of samples and the channel capacity, and determining the concentration gradient of each DNA sample based on the channel allocation table to obtain a sample concentration distribution matrix; Performing signal interference analysis on the sample concentration distribution matrix to obtain channel signal baseline data, and performing quantitative calculation on the excitation light intensity according to the channel signal baseline data to obtain a light intensity control matrix; Performing signal acquisition window optimization on the light intensity control matrix to obtain a time window parameter set, and performing signal gain coefficient calculation on the time window parameter set to obtain a gain coefficient matrix; A state probability calculation is performed on the gain coefficient matrix to obtain a state transition probability matrix, and a multi-parameter coupling analysis is performed based on the state transition probability matrix to obtain a signal acquisition parameter matrix.

3. The parallel processing synchronization optimization method for high-throughput gene sequencing according to claim 2, characterized in that: The method of performing feature extraction and cluster analysis on the fluorescence signals collected by each channel according to the signal collection parameter matrix to obtain a DNA base feature sequence and a signal classification probability matrix includes: Eliminating background noise of the fluorescence signal according to the signal acquisition parameter matrix to obtain pure fluorescence signal data, and performing multi-dimensional feature space mapping on the pure fluorescence signal data to obtain a multi-dimensional feature data set; Normalizing the multidimensional feature data set to obtain a standardized feature matrix, and inputting the standardized feature matrix into a maximum likelihood detection model, calculating the probability of the sample category to which each signal belongs through a probability density function, and obtaining an initial classification probability matrix; Iteratively clustering the initial classification probability matrix to obtain a clustering label sequence, and performing base recognition and sequence matching calculations on the clustering label sequence to obtain a DNA base feature sequence; The DNA base characteristic sequence is subjected to sequence correction to obtain a corrected characteristic sequence, and the classification probability of the corrected characteristic sequence is optimized to obtain a signal classification probability matrix.

4. The method for parallel processing and synchronization optimization of high-throughput gene sequencing according to claim 3, characterized in that: The multidimensional feature data set is normalized to obtain a standardized feature matrix, and the standardized feature matrix is ​​input into a maximum likelihood detection model, and the probability of the sample category to which each signal belongs is calculated by a probability density function to obtain an initial classification probability matrix, including: Performing numerical range calculations on the fluorescence intensity feature, wavelength feature and time series feature in the multidimensional feature data set to obtain a feature range matrix, and performing linear transformation on each feature dimension based on the feature range matrix to obtain preprocessed feature data; Based on the preprocessed feature data, mean calculation and centering processing are performed on the features of each dimension to obtain a zero-mean feature matrix, and variance calculation is performed on the zero-mean feature matrix to obtain a feature variance vector; Scaling the zero-mean feature matrix according to the feature variance vector to obtain a standardized feature matrix, and performing statistical distribution analysis on the standardized feature matrix to obtain a feature distribution function; Inputting the standardized feature matrix and the feature distribution function into a maximum likelihood detection model, obtaining a feature probability density function by Gaussian distribution fitting calculation, and constructing a joint probability distribution according to the feature probability density function; A log-likelihood function is constructed based on the joint probability distribution, the log-likelihood function is maximized by a gradient descent algorithm to obtain an optimal parameter set, and the log-likelihood value of the sample category of each channel is calculated according to the optimal parameter set to obtain a likelihood value matrix; A probability normalization operation is performed on the likelihood value matrix to obtain a category probability vector, and confidence threshold filtering is performed based on the category probability vector to obtain high confidence probability data, and the high confidence probability data is rearranged according to the channel identifier to obtain an initial classification probability matrix.

5. The method for parallel processing and synchronization optimization of high-throughput gene sequencing according to claim 3, characterized in that: The iterative clustering of the initial classification probability matrix to obtain a clustering label sequence, and performing base recognition and sequence matching calculation on the clustering label sequence to obtain a DNA base feature sequence, includes: Performing first feature dimensionality reduction and second feature dimensionality reduction on the initial classification probability matrix, wherein the first feature dimensionality reduction obtains a main classification feature set by linear discriminant analysis, and the second feature dimensionality reduction obtains a local feature set by local preserving projection, and performing feature fusion on the main classification feature set and the local feature set to obtain a fused feature matrix; Constructing a distance metric function based on the fusion feature matrix, and performing weighted combination to obtain a mixed distance matrix, iteratively clustering the mixed distance matrix to obtain a clustering label sequence; Inputting the clustering tag sequence into the triple recognition network, performing base feature extraction through the first recognition subnetwork, performing sequence pattern matching through the second recognition subnetwork, and performing quality scoring through the third recognition subnetwork to obtain a base candidate sequence set; A multiple sequence alignment calculation is performed on the base candidate sequence set to obtain an alignment sequence matrix, and a sequence consistency analysis is performed based on the alignment sequence matrix to obtain a sequence credibility matrix. The sequence is optimized and reconstructed according to the sequence credibility matrix to obtain a DNA base feature sequence.

6. The method for parallel processing and synchronization optimization of high-throughput gene sequencing according to claim 5, characterized in that: The method dynamically adjusts the short-term parameters of the sequencing cycle based on the DNA base feature sequence and the signal classification probability matrix through a convex optimization algorithm, optimizes the long-term sequencing performance through an integer programming algorithm, and obtains a sequencing parameter optimization set and a sequencing cycle allocation strategy, including: Performing sequence alignment on the DNA base feature sequences to obtain a quality score matrix, and constructing a short-term optimization model based on the quality score matrix; Constructing quadratic programming constraints based on the short-term optimization model and the signal classification probability matrix, performing a solving operation through a convex optimization algorithm to obtain a short-term parameter adjustment amount, and performing a constraint check on the short-term parameter adjustment amount to obtain a short-term sequencing parameter set; Performing time series prediction based on the short-term sequencing parameter set to obtain performance prediction data, and constructing a long-term optimization model based on the performance prediction data; Constructing integer programming constraints based on the long-term optimization model, solving and calculating through a branch and bound algorithm to obtain a sequencing cycle allocation plan, and performing resource constraint verification on the sequencing cycle allocation plan to obtain a target cycle plan; Performing parameter sensitivity analysis on the target cycle scheme to obtain a parameter impact matrix, and ranking the sequencing parameters by importance based on the parameter impact matrix to obtain a parameter priority sequence; Performing multi-objective optimization on the parameter priority sequence and the short-term sequencing parameter set to obtain a comprehensive parameter solution, and performing performance evaluation on the comprehensive parameter solution to obtain a sequencing parameter optimization set; The target cycle scheme is dynamically adjusted based on the sequencing parameter optimization set to obtain a sequencing cycle allocation strategy.

7. A parallel processing synchronization optimization system for high-throughput gene sequencing, characterized in that: For implementing the steps of the method according to any one of claims 1 to 6, the system comprises: An initialization module is used to perform channel allocation and signal acquisition initialization for multiple DNA samples to obtain a signal acquisition parameter matrix; An analysis module, used to perform feature extraction and cluster analysis on the fluorescence signals collected by each channel according to the signal collection parameter matrix, to obtain a DNA base feature sequence and a signal classification probability matrix; An optimization module, which is used to dynamically adjust the short-term parameters of the sequencing cycle through a convex optimization algorithm based on the DNA base feature sequence and the signal classification probability matrix, and optimize the long-term sequencing performance through an integer programming algorithm to obtain a sequencing parameter optimization set and a sequencing cycle allocation strategy; A solution module is used to input the sequencing parameter optimization set and the sequencing cycle allocation strategy into a double-layer nested control model, and jointly solve through an upper-layer resource allocation network and a lower-layer parameter optimization network to obtain a global optimization control parameter set; specifically comprising: hierarchical decoupling of the sequencing parameter optimization set to obtain a computing resource control parameter subset and a sequencing performance control parameter subset, and inputting the computing resource control parameter subset into an upper-layer resource allocation network, wherein the upper-layer resource allocation network includes a resource scheduling layer and a task allocation layer; constructing a resource scheduling layer optimization function based on the sequencing cycle allocation strategy, wherein the resource scheduling layer includes four scheduling units, each scheduling unit includes a computing resource allocation module and a storage resource allocation module, and obtaining an initial resource allocation plan; inputting the initial resource allocation plan into a task allocation layer, wherein the task allocation layer includes A task decomposition unit and a priority allocation unit, wherein the task decomposition unit divides the sequencing task into subtasks, and the priority allocation unit determines the task execution order to obtain a task allocation sequence; the sequencing performance control parameter subset is input into a lower parameter optimization network, wherein the lower parameter optimization network includes a parameter update layer and a performance evaluation layer, and the parameter update layer is subjected to multi-objective optimization and solution, wherein the parameter update layer includes 6 optimization units, each of which includes a parameter generation module and a constraint verification module to obtain a parameter update scheme; the parameter update scheme is input into a performance evaluation layer, wherein the performance evaluation layer includes a performance indicator calculation unit and an error compensation unit to obtain an optimized parameter scheme; the task allocation sequence and the optimized parameter scheme are input into a two-layer collaborative optimizer, and a global optimized control parameter set is obtained through interactive iterative operations of the upper and lower layers.

8. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images