A machine learning-based method for judging ad five-target inhibitors

Abstract

This invention relates to a machine learning-based method for identifying five-target inhibitors in Alzheimer's disease (AD). It involves collecting known inhibitor and non-inhibitor molecular structure data for five key target proteins associated with Alzheimer's disease, forming an initial dataset. Two-dimensional molecular descriptors for each molecular structure in the initial dataset are calculated, and a systematic feature selection operation incorporating multiple attribute screening methods and search strategies is used to select a subset of features for modeling from all molecular descriptors. Various machine learning algorithms are used to construct classification models for the five key targets and non-inhibitors, respectively, and these models are integrated to form a multi-target prediction system. The molecular descriptors of the molecules to be predicted are input into the multi-target prediction system. The multi-target prediction system processes the input descriptors and outputs the inhibitor identification result. Compared with existing technologies, this invention has advantages such as high efficiency, strong generalization ability, and strong robustness.

Description

Technical Field

[0001] This invention relates to the field of bioinformatics, and in particular to a method for identifying five-target inhibitors of Alzheimer's disease (AD) based on machine learning. Background Technology

[0002] Alzheimer's disease (AD) is a neurodegenerative disease characterized by progressive memory loss, cognitive impairment, and personality changes. It is the fourth leading cause of death after heart disease, cancer, and stroke, and is also the most common cause of dementia in the elderly. With the increasing aging of the global population, the incidence of AD continues to rise, placing a heavy burden on society and healthcare systems. The etiology of this disease is complex, involving multiple pathological targets, including acetylcholinesterase (AChE), butyrylcholinesterase (BChE), and monoamine oxidase B (MAO-B). β - Secretase 1 (BACE1) and Tau protein, etc., are typical multi-target diseases, so single-target intervention is often difficult to achieve ideal results.

[0003] In recent years, artificial intelligence and machine learning technologies have been widely applied in bioinformatics and drug development, providing new ideas for disease diagnosis and drug discovery. For example, existing studies have used machine learning methods such as deep learning, support vector machines, and random forests to assist in the analysis and prediction of Alzheimer's disease (AD) from the perspectives of gesture testing, biomarker screening, and inhibitor activity prediction. However, existing machine learning-based drug activity prediction schemes still have significant limitations: First, most methods only model a single target or a single task, lacking the ability to simultaneously discriminate the inhibitory spectrum of the same candidate small molecule across multiple targets, resulting in insufficient decision-making information in complex disease scenarios; second, the structural characterization of small molecule compounds is usually characterized by high dimensionality, high redundancy, and strong collinearity. Existing methods often lack a systematic, multi-strategy feature screening and optimization process, making the model susceptible to noise interference, resulting in unstable generalization ability and affecting its reliability in real drug screening scenarios. Therefore, how to provide an efficient virtual screening method that can simultaneously determine the inhibitory activity of small molecules against multiple key AD targets and accurately identify their targets, while possessing high generalization ability and stability, is a technical problem that needs to be solved in this field. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the existing technology by providing a machine learning-based method for identifying five-target inhibitors in Alzheimer's disease (AD). By combining three attribute screening methods—CfsSubsetEval, ConsistencySubsetEval, and FilteredSubsetEval—with three search strategies—BestFirst, GeneticSearch, and GreedyStepwise—a systematic feature selection process is formed. This process can efficiently screen out key feature subsets with strong discriminative power and complementary information from high-dimensional and redundant molecular descriptors, thereby improving the prediction accuracy, generalization ability, and stability of subsequent multi-target classification models.

[0005] The objective of this invention can be achieved through the following technical solutions: According to one aspect of the present invention, a machine learning-based method for determining five-target inhibitors in Alzheimer's disease (AD) is provided, characterized by the following specific steps: S1. Collect known inhibitor molecular structure data and non-inhibitor molecular structure data of five key target proteins related to Alzheimer's disease to form the original dataset; S2. Calculate the two-dimensional molecular descriptors of each molecular structure in the original dataset, and use a systematic feature selection operation that integrates multiple attribute filtering methods and search strategies to select a feature subset for modeling from all molecular descriptors. S3. Based on the selected feature subset, various machine learning algorithms are used to construct classification models for the five key targets and non-inhibitors respectively, and the classification models are integrated to form a multi-target prediction system; the molecular descriptor of the molecule to be predicted is input into the multi-target prediction system. S4. The multi-target prediction system processes the input descriptor and outputs the judgment result of whether the molecule to be predicted is an inhibitor of the five key target proteins.

[0006] Furthermore, the five key target proteins in S1 include acetylcholinesterase, butyrylcholinesterase, monoamine oxidase B, β-secretase 1, and Tau protein.

[0007] Furthermore, in S2, a total of 45 two-dimensional molecular descriptors are calculated, including electric field parameters, structural parameters, and thermodynamic parameters; the electric field parameters include dipole length, electronic energy, highest occupied orbital energy, and lowest empty orbital energy; the structural parameters include the number of rotatable bonds, polar surface area, and total connectivity; the thermodynamic parameters include Henry's law constant, vapor pressure, and lipid-water partition coefficient.

[0008] Furthermore, the specific steps of the systematic feature selection operation in S2 include: The attribute filtering methods are paired with search strategies one by one to obtain multiple filtering combinations. The attribute filtering methods include the relevance-based feature subset evaluator CfsSubsetEval, the consistency-based feature subset evaluator ConsistencySubsetEval, and the filter-based feature subset evaluator FilteredSubsetEval. The search strategies include BestFirst search, GeneticSearch, and GreedyStepwise search. The 45 two-dimensional molecular descriptors were screened using screening combinations to obtain the screening results of each screening combination, which were used as the initial selection results. Further filtering of the initial selection results yields a feature subset, including: Traverse all descriptor subsets in the initial selection results, retain the subset size (i.e., the number of descriptors included in the subset), and form the first intermediate result set for subsets with a size between 4 and 15, where the subset size is the number of descriptors contained in the subset. Each descriptor subset in the first intermediate result set is sorted, and the names of the molecular descriptors in the descriptor subset are sorted according to a unified, predefined lexicographical order to generate a unique normalized string representation for the corresponding descriptor subset; based on the normalized string representation, all subsets are compared, and those that are completely identical are considered duplicate subsets, and only one of them is retained to form the second intermediate result set; All subsets of the second intermediate result set are output as multiple distinct final feature subsets.

[0009] Furthermore, in S3, the various machine learning algorithms include Naïve Bayes, Artificial Neural Networks (ANNs), Polynomial Kernel Support Vector Machines (Poly-SVM), K-Nearest Neighbors (KNN), Logit Boost, Decision Tree (C4.5), and Random Forest.

[0010] Furthermore, the judgment result output in S4 specifically includes: when the molecule to be predicted is identified as an inhibitor, specifying the name of the specific target protein; otherwise, the judgment result is that it is identified as a non-inhibitor.

[0011] Furthermore, in step S3, the performance of each constructed classification model is evaluated using a ten-fold cross-validation method. The specific steps include: The dataset used for model training is randomly and hierarchically divided into ten non-overlapping subsets; Each subset is used as a validation set, and the remaining nine subsets are used as training sets. A classification model is trained based on the training set data, and predictions are made on the corresponding validation sets. Summarize the prediction results on the ten validation sets and calculate the overall performance index of each classification model.

[0012] Furthermore, a classification model is selected based on the results of the ten-fold cross-validation, and the selection rules include: Based on the comprehensive performance index, the consistency of the prediction performance of the same classification model for different target categories is compared, and the model with balanced consistency among the categories is selected. Among models that meet the preset consistency requirements, the overall prediction accuracy of the models is compared, and the model with the higher overall prediction accuracy is selected first; if the difference in overall prediction accuracy is less than the preset threshold, the model with fewer descriptors in the feature subset on which it is based is selected first.

[0013] According to a second aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the method described thereon.

[0014] According to a third aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method described thereon.

[0015] Compared with the prior art, the present invention has the following beneficial effects: (1) Solved the problem of simultaneous prediction of multiple targets and improved the information value and efficiency of virtual screening: In view of the lack of simultaneous determination of the inhibition spectrum of multiple targets for the same candidate molecule in the existing technology, the present invention uses a variety of machine learning algorithms to construct independent classification models for five key targets, AChE, BChE, MAO-B, BACE1, and Tau protein and non-inhibitors, and integrates them into a unified multi-target prediction system. The system can perform a calculation on the input molecular descriptor once and output the inhibitory activity judgment of the molecule on all five targets and the specific target identification result. This overcomes the limitation of traditional single-target models that require multiple independent predictions and provides more comprehensive and more decision-making activity spectrum information. This method can improve screening efficiency and information output in large-scale virtual screening of AD drug lead compounds, provide direct candidate molecules and action target references for multi-target drug design, and improve the accuracy of drug targeting specific targets.

[0016] (2) The model input was optimized through systematic feature selection, which enhanced the generalization ability and stability of the model: In view of the problem that the model is susceptible to noise interference due to the high dimensionality and high redundancy of small molecule descriptors, this invention sampled a systematic feature selection process, combined three attribute selection methods with different principles, namely CfsSubsetEval, ConsistencySubsetEval and FilteredSubsetEval, with three search strategies, namely BestFirst, GeneticSearch and GreedyStepwise, to form multiple selection combinations for parallel exploration and selection. By controlling the number range and standardizing the deduplication of the initial selection results, multiple concise and distinct feature subsets were obtained. This can efficiently remove redundant and noisy features from a large number of original descriptors and select key molecular descriptor combinations with strong discriminative ability and complementary information, thereby reducing data dimensionality and collinearity, improving the model's generalization ability, reducing the risk of overfitting, and enhancing the overall prediction stability, making it more reliable when dealing with new unknown compounds.

[0017] (3) A rigorous model evaluation and selection mechanism was established to ensure the optimal overall performance of the final prediction system: In order to select the optimal configuration from the many models built based on different feature subsets and algorithms, this invention adopted the ten-fold cross-validation method for comprehensive performance evaluation and formulated a three-level progressive selection rule that considers the prediction consistency of each category, the overall accuracy and the simplicity of the feature subset in turn. The cross-validation provides robust performance estimation and avoids the randomness of the model evaluation results. Furthermore, the multi-level selection rule ensures that the selected model not only has high overall accuracy, but also has a relatively balanced prediction ability for each target. At the same time, the model structure is not too complex, so that the final integrated multi-target prediction system achieves a good balance between accuracy, robustness and practicality, ensuring that the overall performance of the deployed system prediction results is better and more suitable for application scenarios in actual drug development where the stability and interpretability of prediction results are required. Attached Figure Description

[0018] Figure 1 The flowchart shows a machine learning-based method for identifying five-target inhibitors in Alzheimer's disease (AD). Figure 2 This is a comparison chart of the ROC curves of the model prediction performance in this embodiment; Figure 3 This is a comparison chart of the AUC values ​​of each algorithm in this implementation; Figure 4 This is a diagram of the architecture of the AD multi-target inhibitor online prediction system (MVC). Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0020] This embodiment combines multiple attribute screening methods and search strategies to reduce and purify high-dimensional molecular descriptors. Based on this, it integrates various machine learning algorithms to construct a two-dimensional structure-activity relationship model, thereby establishing a high-precision prediction system capable of simultaneously predicting the inhibitory activity of small molecules against five key AD targets. This method can simultaneously distinguish and accurately identify whether a small molecule is an inhibitor of acetylcholinesterase, butyrylcholinesterase, monoamine oxidase B, β-secretase 1, and Tau protein, providing an efficient tool for large-scale, low-cost virtual screening of lead compounds.

[0021] like Figure 1 The image shows a machine learning-based method for identifying five-target inhibitors in Alzheimer's disease (AD). The specific steps include: S1. Collect known inhibitor molecular structure data and non-inhibitor molecular structure data of five key target proteins related to Alzheimer's disease to form the original dataset; S2. Calculate the two-dimensional molecular descriptors of each molecular structure in the original dataset, and use a systematic feature selection operation that integrates multiple attribute filtering methods and search strategies to select a feature subset for modeling from all molecular descriptors. S3. Based on the selected feature subset, various machine learning algorithms are used to construct classification models for five key targets and non-inhibitors respectively. The classification models are integrated to form a multi-target prediction system. The molecular descriptor of the molecule to be predicted is input into the multi-target prediction system. S4. The multi-target prediction system processes the input descriptor and outputs the judgment result of whether the molecule to be predicted is an inhibitor of five key target proteins.

[0022] Specifically, the five key target proteins in S1 include AChE, butyrylcholinesterase (BChE), monoamine oxidase B (MAO-B), Tau protein, and β-secretase 1 (BACE1). Known, experimentally validated inhibitors for each of these targets were collected by searching databases such as Web of Science and relevant literature. For the collected compounds, their corresponding chemical structures were plotted using ChemSketch software and saved as .mol ​​files. 185 non-inhibitor molecules were obtained from the DUD database as negative samples and uniformly converted to .mol ​​format for subsequent model training and evaluation. The number of inhibitor and non-inhibitor samples collected is shown in Table 1, with the training and test sets allocated in a 3:1 ratio.

[0023] Table 1. Collection of inhibitors and non-inhibitors targeting different proteins. In S2, a total of 45 two-dimensional molecular descriptors were calculated, including electric field parameters, structural parameters, and thermodynamic parameters. Electric field parameters included dipole length, electronic energy, highest occupied orbital energy, and lowest unoccupied orbital energy. Structural parameters included the number of rotatable bonds, polar surface area, and total connectivity. Thermodynamic parameters included Henry's law constant, vapor pressure, and lipid-water partition coefficient. Specific two-dimensional molecular descriptors are shown in Table 2.

[0024] Table 2. Distribution of 45 molecular descriptors calculated by ChemOffice The specific steps of the systematic feature selection operation in S2 include: The attribute filtering methods are paired with search strategies one by one to obtain multiple filtering combinations. The attribute filtering methods include the relevance-based feature subset evaluator CfsSubsetEval, the consistency-based feature subset evaluator ConsistencySubsetEval, and the filter-based feature subset evaluator FilteredSubsetEval. The search strategies include BestFirst search, GeneticSearch, and GreedyStepwise search. The 45 two-dimensional molecular descriptors were screened using screening combinations, and the screening results of each screening combination were used as the initial selection results. Further filtering of the initial selection results yields a feature subset, including: Iterate through all descriptor subsets in the initial selection results, retain the size of the subset (i.e., the number of descriptors contained in the subset), and form the first intermediate result set for subsets with a size between 4 and 15. The size of the subset is the number of descriptors contained in the subset. For each descriptor subset in the first intermediate result set, sort the names of the molecular descriptors in the descriptor subset according to a unified, predefined lexicographical order to generate a unique normalized string representation for the corresponding descriptor subset; based on the normalized string representation, compare all subsets, and treat identical subsets as duplicate subsets, retaining only one of them to form the second intermediate result set; All subsets of the second intermediate result set are output as multiple distinct final feature subsets.

[0025] In the implementation of this embodiment, various attribute filtering methods and search strategies are used for pairing, as shown in Table 3. The attribute filtering methods used include: correlation-based feature subset evaluator (CfsSubsetEvaluator), chi-square attribute evaluator, classifier subset evaluator, consistency subset evaluator, filtered attribute evaluator, filtered subset evaluator, and gain ratio attribute evaluator. The evaluation criteria include: Information Gain Attribute Evaluator, Latent Semantic Analysis, OneR Attribute Evaluator, Principal Components Analysis, ReliefF Attribute Evaluator, Polynomial Kernel Support Vector Machine Attribute Evaluator, Symmetrical Uncertainty Attribute Evaluator, and WrapperSubset Evaluator. The search methods employed primarily include: Best First Search, Exhaustive Search, Genetic Search, Greedy Stepwise Search, Linear Forward Selection, Rank Search, Random Search, Scatter Search V1, and Subset Size Forward Selection.

[0026] Table 3. Screening results of molecular descriptors The seven feature subsets obtained in this embodiment and the specific molecular descriptors they contain are shown in Table 4.

[0027] Table 4. Distribution of molecular descriptors for each subset In S3, various machine learning algorithms are used, including Naïve Bayes, Artificial Neural Networks (ANNs), Polynomial Kernel Support Vector Machine (Poly-SVM), K Nearest Neighbors (KNN), Logit Boost, Decision Tree (C4.5), and Random Forest.

[0028] The judgment results output in S4 specifically include: when the molecule to be predicted is identified as an inhibitor, specifying the name of the specific target protein; otherwise, the judgment result is that it is identified as a non-inhibitor.

[0029] In S3, the performance of each constructed classification model is evaluated using the ten-fold cross-validation method. The specific steps include: The dataset used for model training is randomly and hierarchically divided into ten non-overlapping subsets; Each subset is used as the validation set, and the remaining nine subsets are used as the training set. The classification model is trained based on the training set data, and predictions are made on the corresponding validation set. Summarize the prediction results on the ten validation sets and calculate the overall performance index of each classification model.

[0030] The classification model is selected based on the results of 10-fold cross-validation, and the selection criteria include: Based on comprehensive performance indicators, compare the consistency of the prediction performance of the same classification model for different target categories, and select the model with balanced consistency among categories. Among models that meet the preset consistency requirements, the overall prediction accuracy of the models is compared, and the model with the higher overall prediction accuracy is selected first; if the difference in overall prediction accuracy is less than the preset threshold, the model with fewer descriptors in the feature subset on which it is based is selected first.

[0031] In this embodiment, based on the obtained seven feature subsets, various machine learning algorithms were used to construct classification models for five key targets and non-inhibitors. The performance of all models was evaluated using the ten-fold cross-validation method, and the evaluation results of this embodiment are shown in Tables 5 and 6.

[0032] Specifically, Table 5 shows the prediction accuracy obtained by using different feature subsets and different machine learning algorithms for 10-fold cross-validation. On different feature subsets, the random forest algorithm generally shows better or more competitive overall prediction accuracy, and in most cases, its prediction performance across different categories is relatively balanced. In particular, the random forest model built based on feature subset 4 shows good performance in both overall prediction accuracy and prediction of each target category, which provides crucial data support for subsequent model optimization.

[0033] Table 5. Forecast accuracy of 10 sets of cross-validation models using different feature subsets. As shown in Table 6, the random forest model constructed based on feature subset 4 achieved excellent overall prediction accuracy on the independent test set, and also maintained a high level of prediction accuracy for each target category. Its performance is superior to or equivalent to that of other algorithms on the test set. This further confirms that the classification model constructed by combining feature subset 4 with the random forest algorithm not only performs well within the training and verification framework of this invention, but also has strong practical application potential and stability.

[0034] Table 6 Modeling results of molecular descriptors in the test set In this embodiment, to comprehensively evaluate the discriminative performance and stability of the model, receiver operating characteristic (ROC) curves are further used to analyze the modeling results. For example... Figure 2 and Figure 3 As shown, by comparing the ROC curves and corresponding area under the curves of various machine learning algorithms on the multi-target classification task, it can be seen that the random forest algorithm exhibits the best AUC value in all category predictions. Based on the above evaluation results, the random forest algorithm is selected in this embodiment, and the final prediction model is constructed based on feature subset 4.

[0035] Based on the optimal classification model constructed above, this embodiment develops and deploys an online prediction system for AD multi-target inhibitors based on the method of this embodiment, such as... Figure 4 As shown, this system employs J2EE technology and is implemented based on the MVC framework. The system architecture is specifically divided into: a view layer, which provides the user interface, allowing users to upload .csv files containing descriptors of the small molecules to be predicted and submit prediction requests, while also viewing the prediction results returned by the system; a control layer, implemented by a Servlet component, which receives user requests and data from the view layer, performs necessary data transformations, calls the core prediction component, and returns the prediction results to the view layer; and a model layer, the core of the system, which encapsulates a prediction model built based on the random forest algorithm and feature subset 4. The predictor component uses this model to calculate the input information and generate prediction results. This three-layer architecture separates data, logic, and display, improving the system's maintainability and scalability.

[0036] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the described module can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0037] The electronic device of this invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in read-only memory (ROM) or loaded from a storage unit into random access memory (RAM). The RAM may also store various programs and data required for device operation. The CPU, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0038] Multiple components in the device are connected to an I / O interface, including: input units such as a keyboard, mouse, etc.; output units such as various types of displays, speakers, etc.; storage units such as disks, optical disks, etc.; and communication units such as network interface cards, modems, wireless transceivers, etc. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks. The processing unit performs the various methods and processes described above, such as the method of the present invention. For example, in some embodiments, the method of the present invention may be implemented as a computer software program tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via ROM and / or the communication unit. When the computer program is loaded into RAM and executed by the CPU, one or more steps of the method of the present invention described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute the method of the present invention by any other suitable means (e.g., by means of firmware).

[0039] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0040] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0041] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0042] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A machine learning-based method for identifying five-target inhibitors for Alzheimer's disease (AD), characterized in that, The specific steps include: S1. Collect known inhibitor molecular structure data and non-inhibitor molecular structure data of five key target proteins related to Alzheimer's disease to form the original dataset; S2. Calculate the two-dimensional molecular descriptors of each molecular structure in the original dataset, and use a systematic feature selection operation that integrates multiple attribute filtering methods and search strategies to select a feature subset for modeling from all molecular descriptors. S3. Based on the selected feature subset, various machine learning algorithms are used to construct classification models for the five key targets and non-inhibitors respectively, and the classification models are integrated to form a multi-target prediction system; the molecular descriptor of the molecule to be predicted is input into the multi-target prediction system. S4. The multi-target prediction system processes the input descriptor and outputs the judgment result of whether the molecule to be predicted is an inhibitor of the five key target proteins.

2. The method for determining five-target inhibitors of Alzheimer's disease based on machine learning according to claim 1, characterized in that, The five key target proteins in S1 include acetylcholinesterase, butyrylcholinesterase, monoamine oxidase B, β-secretase 1, and Tau protein.

3. The method for determining five-target inhibitors of Alzheimer's disease (AD) based on machine learning according to claim 1, characterized in that, In S2, a total of 45 two-dimensional molecular descriptors are calculated, including electric field parameters, structural parameters, and thermodynamic parameters. The electric field parameters include dipole length, electronic energy, highest occupied orbital energy, and lowest empty orbital energy. The structural parameters include the number of rotatable bonds, polar surface area, and total connectivity. The thermodynamic parameters include Henry's law constant, vapor pressure, and lipid-water partition coefficient.

4. The method for determining five-target inhibitors of Alzheimer's disease (AD) based on machine learning according to claim 1, characterized in that, The specific steps of the systematic feature selection operation in S2 include: The attribute filtering methods are paired with search strategies to obtain multiple filtering combinations; wherein, the attribute filtering methods include the relevance-based feature subset evaluator CfsSubsetEval, the consistency-based feature subset evaluator ConsistencySubsetEval, and the filter-based feature subset evaluator FilteredSubsetEval; the search strategies include BestFirst search, GeneticSearch, and GreedyStepwise search. The 45 two-dimensional molecular descriptors were screened using screening combinations to obtain the screening results of each screening combination, which were used as the initial selection results. Further filtering of the initial selection results yields a feature subset, including: Traverse all descriptor subsets in the initial selection results, retain the subset size (i.e., the number of descriptors included in the subset), and form the first intermediate result set for subsets with a size between 4 and 15, where the subset size is the number of descriptors contained in the subset. Each descriptor subset in the first intermediate result set is sorted, and the names of the molecular descriptors in the descriptor subset are sorted according to a unified, predefined lexicographical order to generate a unique normalized string representation for the corresponding descriptor subset; based on the normalized string representation, all subsets are compared, and those that are completely identical are considered duplicate subsets, and only one of them is retained to form the second intermediate result set; All subsets of the second intermediate result set are output as multiple distinct final feature subsets.

5. The method for determining five-target inhibitors of Alzheimer's disease (AD) based on machine learning according to claim 1, characterized in that, In S3, the various machine learning algorithms include Naïve Bayes, Artificial Neural Networks (ANNs), Polynomial Kernel Support Vector Machine (Poly-SVM), K-Nearest Neighbors (KNN), Logit Boost, Decision Tree (C4.5), and Random Forest.

6. The method for determining five-target inhibitors of Alzheimer's disease (AD) based on machine learning according to claim 1, characterized in that, The judgment result output in S4 specifically includes: when the molecule to be predicted is identified as an inhibitor, the name of the specific target protein is specified; otherwise, the judgment result is that it is identified as a non-inhibitor.

7. The method for determining five-target inhibitors of Alzheimer's disease (AD) based on machine learning according to claim 1, characterized in that, In step S3, the performance of each constructed classification model is evaluated using a ten-fold cross-validation method. The specific steps include: The dataset used for model training is randomly and hierarchically divided into ten non-overlapping subsets; Each subset is used as a validation set, and the remaining nine subsets are used as training sets. A classification model is trained based on the training set data, and predictions are made on the corresponding validation sets. Summarize the prediction results on the ten validation sets and calculate the overall performance index of each classification model.

8. The method for determining five-target inhibitors of Alzheimer's disease (AD) based on machine learning according to claim 7, characterized in that, The classification model is selected based on the results of the ten-fold cross-validation, and the selection rules include: Based on the comprehensive performance index, the consistency of the prediction performance of the same classification model for different target categories is compared, and the model with balanced consistency among the categories is selected. Among models that meet the preset consistency requirements, the overall prediction accuracy of the models is compared, and the model with the higher overall prediction accuracy is selected first; if the difference in overall prediction accuracy is less than the preset threshold, the model with fewer descriptors in the feature subset on which it is based is selected first.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 8.

Images