A nucleotide drug nanopore feature classification method and related apparatus

Abstract

This invention belongs to the field of bioinformatics and data processing technology, and discloses a method and related device for classifying nucleotide drugs using nanopore features. The method involves acquiring the raw current signal generated when a nucleotide drug passes through a nanopore, extracting multidimensional feature vectors characterizing the physical properties of the drug molecules, constructing a nanopore feature dataset and performing preprocessing. Based on the preprocessed nanopore feature dataset, a moderately complex decision tree model and a quadratic discriminant analysis (QDIA) classification model are constructed and trained. Depending on the classification requirements of the nucleotide drug to be tested, the trained moderately complex decision tree model and QDIA classification model are selected. The nanopore feature data of the drug to be tested are input into these models, and the classification result and confidence level of the drug to be tested are calculated. This invention effectively improves the model's generalization ability and classification accuracy by analyzing the electrical signal data generated by nanopore detection, realizing the automatic classification and identification of nucleotide drugs.

Description

Technical Field

[0001] This invention belongs to the field of bioinformatics and data processing technology, and relates to a method and related device for classifying the nanopore features of nucleotide drugs. Background Technology

[0002] Nanopore single-molecule detection technology has been widely used in drug analysis and gene sequencing due to its advantages of being label-free, high-throughput, and low-cost. When nucleotide drug molecules pass through nanopores, they trigger changes in the ionic current within the pore, generating a characteristic blocking current signal. By analyzing the characteristics of this electrical signal, different drug molecules can be identified.

[0003] However, existing nanopore feature classification techniques for nucleotide drugs still have many shortcomings: The signal characteristics of different nucleotide drugs exhibit a complex nonlinear distribution, and the covariance matrices of each category differ significantly. Traditional linear classification methods such as linear discriminant analysis (LDA) assume that all categories share the same covariance matrix, which cannot adapt to this characteristic and limits the classification accuracy.

[0004] Decision tree models are prone to underfitting or overfitting. Simple coarse decision trees suffer from insufficient discrimination due to underfitting, while fully grown fine decision trees or complex neural networks are prone to overfitting in noisy biological signal data, resulting in a significant decrease in the model's generalization ability.

[0005] Some schemes fail to fully exploit the higher-order statistical characteristics of current signals, such as skewness and kurtosis, resulting in the loss of effective classification information and further limiting classification accuracy.

[0006] Therefore, there is an urgent need for a nucleotide drug nanopore feature classification method that can adapt to nonlinear complex data distributions and effectively control model complexity and prevent overfitting. Summary of the Invention

[0007] The purpose of this invention is to provide a method and related apparatus for classifying the nanopore features of nucleotide drugs, thereby solving the problems of poor model generalization ability, low nonlinear classification accuracy, and insufficient feature utilization in the prior art.

[0008] To achieve the above objectives, the present invention employs the following technical solution: A method for classifying the nanopore features of nucleotide drugs, comprising: The raw current signal generated when the nucleotide drug passes through the nanopore is obtained, the multidimensional feature vector characterizing the physical properties of the drug molecule is extracted, the nanopore feature dataset is constructed and preprocessed to obtain the preprocessed nanopore feature dataset. Based on the preprocessed nanopore feature dataset, a moderately complex decision tree model and a quadratic discriminant analysis classification model were constructed and trained to obtain a moderately complex decision tree model and a quadratic discriminant analysis classification model that are protected against overfitting. Based on the requirements of the nucleotide drug classification scenario, a medium-complexity decision tree model and a secondary discriminant analysis classification model were selected after training. The nanopore feature data of the drug to be tested were input into them to calculate the classification result and confidence level of the drug to be tested.

[0009] Furthermore, the preprocessing of the nanopore feature dataset includes: The nanopore feature dataset is cleaned by removing outlier samples containing null values ​​or non-numerical values, and samples with a dwell time of less than [a certain duration]. The invalid samples are used to obtain the cleaned valid dataset; A stratified sampling strategy was adopted, and the cleaned effective dataset was divided into training set and test set according to the proportion of each drug category in the nanopore feature dataset, to ensure that the sample category distribution in the training set and test set is consistent with the overall distribution. Based on the feature statistics of the training set, the Z-Score standardization method is used to standardize the training set and the test set respectively to eliminate the scale difference between features of different dimensions, and obtain the standardized training set and test set.

[0010] Furthermore, the multidimensional feature vectors characterizing the physical properties of drug molecules include the percentage of blocking current, current standard deviation, residence time, skewness, and kurtosis.

[0011] Furthermore, the construction process of a moderately complex decision tree model is as follows: The Gini coefficient is used as the evaluation criterion for node splitting. The Gini impurity of all possible feature splitting points of the current node is calculated, and the feature with the smallest Gini impurity and the threshold are selected for splitting. The tree's growth structure is monitored in real time, and a maximum number of leaf nodes is set. When the total number of leaf nodes in the tree reaches the maximum number of leaf nodes, the splitting stops, resulting in a decision tree model of medium complexity.

[0012] Furthermore, the formula for calculating Gini impurity is as follows:

[0013] in, This represents the dataset of the current node. This indicates the total number of drug categories. Indicates the first The probability of a sample of a certain type appearing in the current node.

[0014] Furthermore, the construction process of the secondary discriminant analysis classification model is as follows: Assuming that the feature covariance matrices of different nucleotide drug categories are different, the mean vector and covariance matrix of each drug category are calculated separately, and a quadratic discriminant function is constructed based on Bayes' theorem. The drug category with the largest quadratic discriminant function value is determined as the prediction result, forming a quadratic curve or quadratic surface classification boundary in the feature space. This adapts to the nonlinearity and heterogeneity of the distribution of nucleotide drug characteristics, resulting in a quadratic discriminant analysis classification model.

[0015] Furthermore, the quadratic discriminant function is:

[0016] in, Describes the quadratic discriminant function. This represents the input feature vector. Indicates the first The mean vector of the drug class, Indicates the first The covariance matrix of the drug class, Represents the transpose of a vector. Indicates the first The inverse of a covariance matrix, Represents a determinant. Indicates the first Prior probabilities of a class of drugs.

[0017] A nucleotide drug nanopore feature classification system includes: The data module is used to acquire the raw current signal generated when nucleotide drugs pass through nanopores, extract multidimensional feature vectors characterizing the physical properties of drug molecules, construct a nanopore feature dataset and perform preprocessing to obtain a preprocessed nanopore feature dataset. The modeling module is used to construct and train a moderately complex decision tree model and a quadratic discriminant analysis classification model based on the preprocessed nanopore feature dataset, thereby obtaining a trained moderately complex decision tree model and a quadratic discriminant analysis classification model that are protected against overfitting. The calculation module is used to select a trained medium-complexity decision tree model and a secondary discriminant analysis classification model according to the requirements of the nucleotide drug classification scenario. The nanopore feature data of the drug to be tested are input into the module to calculate the classification result and confidence level of the drug to be tested.

[0018] A terminal device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method.

[0019] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method.

[0020] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a method for classifying nucleotide drugs by nanopore features. By acquiring the original current signal generated when the nucleotide drug passes through the nanopore, extracting multidimensional feature vectors that characterize the physical properties of the drug molecule, constructing a nanopore feature dataset and performing preprocessing, constructing a moderately complex decision tree model and a secondary discriminant analysis classification model and training them, the appropriate model can be flexibly selected according to the actual classification scenario to perform feature input analysis on the nucleotide drug to be tested, and accurately outputting the classification results and corresponding confidence scores. To address the problems of existing technologies, such as linear models being unable to adapt to nonlinear distributions, decision trees being prone to overfitting or underfitting, and insufficient utilization of high-order features, this invention constructs a moderately complex decision tree with a limited maximum number of leaf nodes and a quadratic discriminant analysis model based on the independent covariance matrix. Machine learning algorithms are used to analyze the electrical signal data generated by nanopore detection, achieving automated and accurate classification and identification of nucleotide drugs. This eliminates the reliance on traditional manual analysis and effectively improves the efficiency, accuracy, and reliability of nucleotide drug nanopore detection and classification. Furthermore, the model exhibits strong adaptability and good generalization ability, meeting the practical classification application needs of different nucleotide drugs to be tested. It provides an efficient and feasible intelligent technical solution for rapid identification and feature recognition of nucleotide drugs and can be applied to automatic identification scenarios for nucleotide drug nanopore detection.

[0021] Furthermore, by strictly limiting the maximum number of leaf nodes in the decision tree, the overfitting problem of traditional decision tree algorithms on biological signal data is avoided, and the model's predictive stability for new samples is significantly improved while ensuring high accuracy.

[0022] Furthermore, this invention introduces quadratic discriminant analysis, which uses the independent covariance matrices of each category to construct a quadratic classification boundary. This effectively solves the problem that linear models cannot handle the uneven covariance in nucleotide drug data, thereby improving classification accuracy.

[0023] Furthermore, this invention comprehensively utilizes five-dimensional features, including the percentage of blocking current, current standard deviation, residence time, skewness, and kurtosis, which, compared to single feature analysis, can capture more subtle physical fingerprints of drug molecules.

[0024] Furthermore, the nanopore feature dataset undergoes multi-step refined preprocessing. First, null values, non-numerical outliers, and invalid samples with a dwell time of less than 10ms are cleaned and removed, effectively filtering out noise interference and invalid data, thus improving the overall quality of the dataset. Then, a stratified sampling strategy is adopted to divide the training set and test set according to the proportion of each category, ensuring that the sample category distribution is consistent with the population and avoiding sampling bias that could lead to model training distortion. At the same time, Z-Score standardization is applied to the training set and test set based on the training set statistics to uniformly process them, eliminating differences in the dimensions and scales of different features and avoiding the impact of differences in feature numerical magnitudes on the model's classification accuracy. This improves the training effect, classification accuracy, and generalization ability of subsequent decision tree models and secondary discriminant analysis classification models, ensuring the stability and reliability of the nucleotide drug nanopore feature classification results. Attached Figure Description

[0025] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a flowchart illustrating the implementation of the nucleotide drug nanopore feature classification method of the present invention.

[0027] Figure 2 This is a comparison chart of the classification accuracy of the two improved models (Medium Tree and QDA) of this invention with other conventional models.

[0028] Figure 3 This is a schematic diagram of the structure of a preferred embodiment of the nucleotide drug nanopore feature classification system of the present invention.

[0029] Figure 4 This is a schematic diagram of the electronic device structure according to a preferred embodiment of the present invention. Detailed Implementation

[0030] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of this application, including various details to aid understanding. These should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0031] Obviously, the described embodiments are only some, not all, of the embodiments in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0032] It should be noted that the terminals involved in the embodiments of this application may include, but are not limited to, mobile phones, personal digital assistants (PDAs), wireless handheld devices, tablet computers, personal computers (PCs), MP3 players, MP4 players, wearable devices (e.g., smart glasses, smartwatches, smart bracelets), smart home devices, and other smart devices.

[0033] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0034] The present invention will now be described in further detail with reference to the accompanying drawings: See Figure 1 This invention provides a method for classifying the nanopore features of nucleotide drugs, specifically including the following steps: Step 1: Collect the raw current signal generated when the nucleotide drug passes through the nanopore, extract the multidimensional feature vector characterizing the physical properties of the drug molecule, construct the nanopore feature dataset and perform preprocessing, and complete data cleaning, dataset partitioning and standardization preprocessing.

[0035] Feature extraction: Extracting multidimensional feature vectors characterizing the physical properties of drug molecules from the raw nanopore current signal, including at least the following five features: percentage of blocking current ( ): The percentage of the average blocking current as drug molecules pass through nanopores relative to the baseline current; Current standard deviation ( ): The amplitude of the current signal fluctuation during the passage of drug molecules through the nanopore; Duration of stay ( ): The residence time of drug molecules within nanopores; Skewness ( ): The degree of asymmetry in the probability distribution of the current signal; Kubularity ( ): The steepness of the probability distribution of the current signal.

[0036] Specifically, the preprocessing implementation flow for the nanopore feature dataset is as follows: Data cleaning: Traverse the nanopore feature dataset, automatically identify and remove outlier sample rows containing null values ​​(NaN) or infinity (Inf), and also remove rows with dwell time. Invalid samples are removed from the source to eliminate interference from electrical noise and non-specific transient adsorption, ensuring the validity of the input data and obtaining a cleaned valid dataset. Dataset shuffling and partitioning: First, the cleaned effective dataset is randomly shuffled. Then, a stratified sampling strategy is adopted to divide the data into training and test sets in an 8:2 ratio. This ensures that the distribution ratio of each drug category in the training and test sets is completely consistent with the overall dataset, thus solving the problem of drug sample imbalance. Standardized processing: for dwell time ( ) and percentage of blocking current ( To address the issue of significant differences in dimensions and numerical magnitudes, the Z-Score normalization method is employed, calculating the mean of each feature based solely on the training set data. with standard deviation Then, the mean and standard deviation are used to standardize the training and test sets respectively, eliminating scale differences between features of different dimensions, to obtain the standardized training and test sets. The calculation formula is as follows:

[0037] In the formula, These are the original eigenvalues. This is the mean of the feature in the training set. The standard deviation of this feature in the training set. These are the standardized feature values.

[0038] Step 2: Construct a medium-complexity decision tree classification model based on a growth constraint strategy.

[0039] Specifically, the construction process of a moderately complex decision tree is as follows: The Gini impurity is used as the evaluation criterion for node splitting. The Gini impurity of all possible split points of the current node is calculated, and the feature with the lowest Gini impurity and the corresponding threshold are selected for splitting. The formula for calculating Gini impurity is as follows:

[0040] in, This represents the dataset of the current node. This indicates the total number of drug categories. Indicates the first The probability of a sample of a certain type appearing in the current node.

[0041] To prevent overfitting and avoid underfitting, a maximum leaf node limit parameter is introduced during model training. When the total number of leaf nodes in the decision tree reaches the maximum number of leaf nodes, splitting is forcibly stopped. This setting ensures that the model complexity is in a balance between preventing underfitting and overfitting, i.e., at a "moderate" level, focusing on capturing the core characteristics of nucleotide drugs rather than noise.

[0042] In a preferred embodiment of the present invention, the maximum number of leaf nodes is set to 20.

[0043] Step 3: Construct a classification model based on quadratic discriminant analysis (QDA) using independent covariance matrices.

[0044] Specifically, the construction process of the quadratic discriminant analysis model is as follows: Parameter estimation: Unlike linear discriminant analysis, which assumes all categories share the same covariance matrix, this invention assumes that the feature covariance matrices of different nucleotide drug categories are different. Based on the training set, the parameters are respectively... Calculate the mean vector of drugs and independent covariance matrices .

[0045] Discriminant function construction: Based on Bayes' theorem, a quadratic discriminant function is constructed. (QuadraticDiscriminant Function):

[0046] in, Describes the quadratic discriminant function. This represents the input feature vector. Indicates the first The mean vector of the drug class, Indicates the first The covariance matrix of the drug class, Represents the transpose of a vector. Indicates the first The inverse of a covariance matrix, Represents a determinant. Indicates the first Prior probabilities of a class of drugs.

[0047] Nonlinear boundary generation: due to the introduction of quadratic terms The drug category with the largest quadratic discriminant function value is determined as the prediction result, forming a quadratic curve or quadratic surface classification boundary in the feature space, thereby effectively adapting to the nonlinearity and heterogeneity of the nucleotide drug feature distribution.

[0048] Step 4: Model selection, prediction, and result output.

[0049] Based on the requirements of the nucleotide drug classification scenario, select any of the above-mentioned trained medium-complexity decision tree (Medium Tree) model or quadratic discriminant analysis (QDA) model, input the standardized nanopore feature data of the drug to be tested into it, the model calculates and outputs the posterior probability of each drug category, selects the category with the highest posterior probability as the final prediction result, and outputs the corresponding category confidence score.

[0050] The present invention will be further described in detail below through specific embodiments: Example 1: Classification using a medium-complexity decision tree model Step 1: Obtain nanopore feature data Nanopore detection signals for four nucleotide drugs (2'-FL, 3'-FL, 6'-FL, and LNT) were collected, and five-dimensional features were extracted, including Amp SD (SD), toff (residence time), I / I0 (relative current blockade), skewness, and kurtosis, to construct a training dataset. The dataset contains 2235 valid samples, with sufficient sample size for each drug class. The dataset was randomly divided into a training set (1788 samples) and a test set (447 samples) in an 8:2 ratio.

[0051] Step 2: Model Training A moderately complex decision tree model was constructed, and the feature data from the training set was input into the model. During training, the Gini coefficient was used as the evaluation criterion for node splitting, and a maximum leaf node limit was introduced to control the tree's growth scale—splitting was forcibly stopped when the total number of leaf nodes reached 20, keeping the model complexity at a "moderate" level and balancing underfitting and overfitting. A fixed random seed was used to ensure the reproducibility of training results. Ten-fold stratified cross-validation (with randomly shuffled data) was performed on the training set, resulting in a mean cross-validation accuracy of 96.24% and a standard deviation of 1.08%.

[0052] Step 3: Model Prediction and Evaluation The test set data was input into the trained medium-complexity decision tree model for prediction, and the test set accuracy was 95.08%.

[0053] Taking a certain boundary test sample as an example, its true class is 3'-FL, and the feature parameters are: Amp SD=6.861, t off=1.238, I / I0=0.883, skew=0.235, kurt=-0.993. This sample is located in the overlapping region of the 2'-FL and 3'-FL categories in the feature space scatter plot.

[0054] Model output: The predicted category is 3'-FL, with a confidence level of 100%. The sample features clearly fall within the region divided by the decision tree as 3'-FL, therefore the model gives a judgment with full confidence.

[0055] In this embodiment 1, after training is completed, the contribution of each feature output by the model to the classification is extracted. The contribution of the 5-dimensional features is ranked as follows: SD>t off >I / I0>skew>kurt. This ranking indicates that, among the characteristics of nanopore signals, current fluctuation amplitude (SD) and residence time (toff) are the most discriminative features.

[0056] Example 2: Classification using a Quadratic Discriminant Analysis (QDA) model Step 1: Obtain nanopore feature data The dataset and partitioning method were exactly the same as in Example 1, with 4 nucleotide drugs, 5-dimensional nanopore features, 2235 samples, and an 8:2 training / test set partitioning.

[0057] Step 2: Model Training A quadratic discriminant analysis (PCA) classification model was constructed, and the training set feature data was input into the PCA classification model. Unlike linear discriminant analysis, which assumes that all categories share the same covariance matrix, this embodiment calculates independent mean vectors and covariance matrices for the four drug categories: 2'-FL, 3'-FL, 6'-FL, and LNT. During training, an option was set to store the covariance matrices of each category for subsequent difference analysis of the correlation between features of each category. Ten-fold stratified cross-validation (data randomly shuffled) was performed on the training set, with a mean cross-validation accuracy of 94.81% and a standard deviation of 0.90%. Compared with the decision tree model in Example 1 (mean cross-validation 96.24%, standard deviation 1.08%), this embodiment has a smaller standard deviation, and the difference between the mean cross-validation and the test set accuracy (93.74%) is also smaller, indicating that the model performs more consistently under different data partitions and has more stable generalization ability.

[0058] Step 3: Model Prediction and Evaluation The test set data was input into the trained secondary discriminant analysis classification model for prediction, and the test set accuracy was 93.74%.

[0059] To provide a direct comparison with Example 1, the same boundary test sample as in Example 1 was used for prediction. The true class of this sample was 3'-FL, and the feature parameters were: Amp SD=6.861, t off =1.238, I / I0=0.883, skew=0.235, kurt=-0.993.

[0060] Since the quadratic discriminant analysis classification model does not assume that all classes share the same covariance matrix, this sample, when calculating its posterior probability, considers both the Mahalanobis distance to the mean vectors of each class and the penalty term resulting from the size of the determinant of the covariance matrix of each class. Model output: Predicted class is 3'-FL, with a confidence level of 99.97%.

[0061] Compared to the 100% confidence level given by the decision tree model in Example 1, the confidence level of this model is slightly lower than 100%. This is because the sample is located in the overlapping region of 2'-FL and 3'-FL in the feature space, and its covariance structure is partially similar to that of the 2'-FL class, resulting in a probability assignment of approximately 0.03% when calculating the posterior probability of 2'-FL. This is not a defect in the model's uncertainty, but rather reflects the ability of quadratic discriminant analysis to perform more refined probabilistic modeling of class boundaries through independent covariance matrices.

[0062] In Example 2, after training, the quadratic discriminant analysis classification model stores the covariance matrix of each drug class. Comparison reveals that the covariance signs of 2'-FL and 3'-FL on the skew-kurt feature pairs are opposite: skew in 2'-FL is negatively correlated with kurt, while skew in 3'-FL is positively correlated with kurt. This indicates that the distribution directions of the two drug classes in this two-dimensional feature space are completely different. It is precisely because the quadratic discriminant analysis classification model allows each drug class to have an independent covariance matrix that this structural difference can be captured in the discriminant function, generating the quadratic classification boundary.

[0063] Figure 2 This is a bar chart comparing the mean 10-fold cross-validation accuracy of 13 classification models on the dataset of 2235 nanopore features. The horizontal axis represents different algorithm models, the vertical axis represents the mean cross-validation accuracy, and the error bars indicate a range of ±1 standard deviation.

[0064] Comparative analysis revealed that the average cross-validation accuracy of all models ranged from 89% to 97%, indicating that the 5D nanopore feature system constructed in this invention possesses stable and efficient discrimination capabilities for four drug classes: 2'-FL, 3'-FL, 6'-FL, and LNT. Various classification paradigms achieved superior performance. Ensemble methods (Bagged Trees, mean 96.42%) and nonlinear support vector machines (Cubic SVM, mean 95.88%) accounted for the majority of the results. Figure 2 At the highest position, the linear models (Linear SVM, Linear Discriminant) also showed high accuracy, indicating that the overall distribution of each category in the feature space has a good linear separability basis.

[0065] The medium-complexity decision tree in this invention has a mean cross-validation accuracy of 96.24% and a standard deviation of 1.08%, placing it at the top of the histogram. This, along with the 95.08% test set accuracy in Example 1, demonstrates its high classification accuracy. The quadratic discriminant analysis (QDA) classification has a mean cross-validation accuracy of 94.81% and a standard deviation of 0.90%, placing it in the upper-middle range, consistent with the 93.74% test set accuracy in Example 2.

[0066] Further observation of the error bar length reveals that the standard deviation of Medium Tree (1.08%) is slightly greater than that of QDA (0.90%). Furthermore, the discrepancy between the cross-validation mean and the test set accuracy of Medium Tree is 1.16 percentage points, while the discrepancy for QDA is only 1.07 percentage points. This indicates that QDA performs more consistently across multiple data partitions, exhibiting less generalization bias. The covariance structure analysis in Example 2 provides a mechanism explanation for this phenomenon: QDA is modeled based on a globally independent covariance matrix, making it less affected by local data partitions, while a single decision tree is more sensitive to subtle changes in the training data, thus resulting in a slightly higher variance.

[0067] Combine the same boundary sample described in Examples 1 and 2 (true value 3'-FL, feature located in the overlapping region of 2'-FL and 3'-FL). Figure 2 The model characteristics reflected are also confirmed at the single-sample level. Medium Tree classifies the sample as 3'-FL with 100% confidence, exhibiting "hard classification" characteristics; while QDA gives the same prediction with 99.97% confidence, while retaining approximately 0.03% of the 2'-FL probability. This fine-grained difference in probability is consistent with... Figure 2 The lower standard deviation of QDA corresponds to the fact that both global performance and single-sample behavior demonstrate that QDA can provide a smoother and more realistic posterior probability estimate.

[0068] comprehensive Figure 2 The global comparison, the importance analysis of decision tree features in Example 1, and the covariance difference analysis in Example 2 can provide a clear basis for model selection in practical applications: for application scenarios that pursue the highest classification accuracy and allow for some performance fluctuations, priority can be given to selecting [the appropriate model]. Figure 2 Among the leading models, such as Bagged Trees, Cubic SVM, or Medium Tree, QDA is more suitable for applications that require outstanding generalization stability and the ability to output smooth probabilities to support risk-controlled decisions (such as identifying samples with low confidence in drug screening for further validation). This is especially true when the covariance structures of different drug categories differ significantly.

[0069] This invention also provides a nucleotide drug nanopore feature classification system, such as... Figure 3 As shown, the system includes: a data module, a modeling module, and a computing module.

[0070] The data module is used to acquire the raw current signal generated when nucleotide drugs pass through nanopores, extract multidimensional feature vectors characterizing the physical properties of drug molecules, construct a nanopore feature dataset and perform preprocessing to obtain a preprocessed nanopore feature dataset. The modeling module is used to construct and train a moderately complex decision tree model and a quadratic discriminant analysis classification model based on the preprocessed nanopore feature dataset, thereby obtaining a trained moderately complex decision tree model and a quadratic discriminant analysis classification model that are protected against overfitting. The calculation module is used to select a trained medium-complexity decision tree model and a secondary discriminant analysis classification model according to the requirements of the nucleotide drug classification scenario. The nanopore feature data of the drug to be tested are input into the module to calculate the classification result and confidence level of the drug to be tested.

[0071] It is understood that the nucleotide drug nanopore feature classification system provided by the present invention corresponds to the nucleotide drug nanopore feature classification method provided in the foregoing embodiments. The relevant technical features of the nucleotide drug nanopore feature classification system can be referred to the relevant technical features of the nucleotide drug nanopore feature classification method, and will not be repeated here.

[0072] Another object of the present invention is to provide an electronic device, such as... Figure 4 As shown, it includes a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor performing the steps of the nucleotide drug nanopore feature classification method.

[0073] The method for classifying the nanopore features of nucleotide drugs includes the following steps: The raw current signal generated when the nucleotide drug passes through the nanopore is obtained, the multidimensional feature vector characterizing the physical properties of the drug molecule is extracted, the nanopore feature dataset is constructed and preprocessed to obtain the preprocessed nanopore feature dataset. Based on the preprocessed nanopore feature dataset, a moderately complex decision tree model and a quadratic discriminant analysis classification model were constructed and trained to obtain a moderately complex decision tree model and a quadratic discriminant analysis classification model that are protected against overfitting. Based on the requirements of the nucleotide drug classification scenario, a medium-complexity decision tree model and a secondary discriminant analysis classification model were selected after training. The nanopore feature data of the drug to be tested were input into them to calculate the classification result and confidence level of the drug to be tested.

[0074] A fourth objective of this invention is to provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the nucleotide drug nanopore feature classification method.

[0075] The method for classifying the nanopore features of nucleotide drugs includes the following steps: The raw current signal generated when the nucleotide drug passes through the nanopore is obtained, the multidimensional feature vector characterizing the physical properties of the drug molecule is extracted, the nanopore feature dataset is constructed and preprocessed to obtain the preprocessed nanopore feature dataset. Based on the preprocessed nanopore feature dataset, a moderately complex decision tree model and a quadratic discriminant analysis classification model were constructed and trained to obtain a moderately complex decision tree model and a quadratic discriminant analysis classification model that are protected against overfitting. Based on the requirements of the nucleotide drug classification scenario, a medium-complexity decision tree model and a secondary discriminant analysis classification model were selected after training. The nanopore feature data of the drug to be tested were input into them to calculate the classification result and confidence level of the drug to be tested.

[0076] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0077] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0078] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0079] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0080] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for classifying the nanopore characteristics of nucleotide drugs, characterized in that, include: The raw current signal generated when the nucleotide drug passes through the nanopore is obtained, the multidimensional feature vector characterizing the physical properties of the drug molecule is extracted, the nanopore feature dataset is constructed and preprocessed to obtain the preprocessed nanopore feature dataset. Based on the preprocessed nanopore feature dataset, a moderately complex decision tree model and a quadratic discriminant analysis classification model were constructed and trained to obtain a moderately complex decision tree model and a quadratic discriminant analysis classification model that are protected against overfitting. Based on the requirements of the nucleotide drug classification scenario, a medium-complexity decision tree model and a secondary discriminant analysis classification model were selected after training. The nanopore feature data of the drug to be tested were input into them to calculate the classification result and confidence level of the drug to be tested.

2. The method of claim 1, wherein the nucleotide drug nanopore feature classification method is characterized by, The preprocessing of the nanopore feature dataset includes: The nanopore feature dataset is cleaned by removing outlier samples containing null values ​​or non-numerical values, and samples with a dwell time of less than [a certain duration]. The invalid samples are used to obtain the cleaned valid dataset; A stratified sampling strategy was adopted, and the cleaned effective dataset was divided into training set and test set according to the proportion of each drug category in the nanopore feature dataset, to ensure that the sample category distribution in the training set and test set is consistent with the overall distribution. Based on the feature statistics of the training set, the Z-Score standardization method is used to standardize the training set and the test set respectively to eliminate the scale difference between features of different dimensions, and obtain the standardized training set and test set.

3. The method for classifying the nanopore features of nucleotide drugs according to claim 1, characterized in that, Multidimensional feature vectors characterizing the physical properties of drug molecules include percentage of blocking current, standard deviation of current, residence time, skewness, and kurtosis.

4. The method of claim 1, wherein the nucleotide drug nanopore feature classification method is characterized by, The construction process of a moderately complex decision tree model is as follows: The Gini coefficient is used as the evaluation criterion for node splitting. The Gini impurity of all possible feature splitting points of the current node is calculated, and the feature with the smallest Gini impurity and the threshold are selected for splitting. The tree's growth structure is monitored in real time, and a maximum number of leaf nodes is set. When the total number of leaf nodes in the tree reaches the maximum number of leaf nodes, the splitting stops, resulting in a decision tree model of medium complexity.

5. The method of claim 4, wherein the nucleotide drug nanopore feature classification method is characterized by, The formula for calculating Gini impurity is: in, This represents the dataset of the current node. This indicates the total number of drug categories. Indicates the first The probability of a sample of a certain type appearing in the current node.

6. The method of claim 1, wherein the nucleotide drug nanopore feature classification method is characterized by, The construction process of the quadratic discriminant analysis classification model is as follows: Assuming that the feature covariance matrices of different nucleotide drug categories are different, the mean vector and covariance matrix of each drug category are calculated separately, and a quadratic discriminant function is constructed based on Bayes' theorem. The drug category with the largest quadratic discriminant function value is determined as the prediction result, forming a quadratic curve or quadratic surface classification boundary in the feature space. This adapts to the nonlinearity and heterogeneity of the distribution of nucleotide drug characteristics, resulting in a quadratic discriminant analysis classification model.

7. The method of claim 6, wherein the nucleotide drug nanopore feature classification method is characterized by, The quadratic discriminant function is: in, Describes the quadratic discriminant function. This represents the input feature vector. Indicates the first The mean vector of the drug class, Indicates the first The covariance matrix of the drug class, Represents the transpose of a vector. Indicates the first The inverse of a covariance matrix, Represents a determinant. Indicates the first Prior probabilities of a class of drugs.

8. A nucleotide drug nanopore feature classification system characterized by, include: The data module is used to acquire the raw current signal generated when nucleotide drugs pass through nanopores, extract multidimensional feature vectors characterizing the physical properties of drug molecules, construct a nanopore feature dataset and perform preprocessing to obtain a preprocessed nanopore feature dataset. The modeling module is used to construct and train a moderately complex decision tree model and a quadratic discriminant analysis classification model based on the preprocessed nanopore feature dataset, thereby obtaining a trained moderately complex decision tree model and a quadratic discriminant analysis classification model that are protected against overfitting. The calculation module is used to select a trained medium-complexity decision tree model and a secondary discriminant analysis classification model according to the requirements of the nucleotide drug classification scenario. The nanopore feature data of the drug to be tested are input into the module to calculate the classification result and confidence level of the drug to be tested.

9. A terminal device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images