A circRNA drug resistance prediction method based on a hybrid graph neural network

Abstract

The application particularly relates to a circRNA drug resistance prediction method based on a hybrid graph neural network, which comprises the following steps: firstly, constructing a biological similarity information network; then inputting the hybrid graph neural network; fusing features of a graph neural network combined with two layers of GCN and GAT and a CNN layer; adding disease correlation information into circRNA and drug representation; finally, splicing comprehensive representation of circRNA and drugs, inputting a classifier, completing a final prediction task, and obtaining a predicted result. The method can obtain a prediction result of a reaction relationship between circRNA and drugs with higher accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of circRNA and drug resistance prediction, and in particular to a circRNA drug resistance prediction method based on a hybrid graph neural network. Background Art

[0002] Among related technologies, methods for predicting the relationship between circRNAs and drug resistance can be categorized into two main categories: traditional methods and deep learning approaches. Traditional analysis relies on high-cost experiments in biology and medicine to generate a series of pairings, which are then organized into a series of biological databases. However, these approaches generally require extensive observation time and expensive and demanding experimental resources. A range of deep learning approaches have been developed that can shorten development time and have achieved commendable results in predicting circRNA drug resistance, but they still face a number of limitations. For example, the features used by each model are relatively simple; for example, the GATECDA and MNCLCDA models only utilize a limited set of features related to circRNAs and drugs. Furthermore, there are certain deficiencies in feature extraction and utilization, resulting in low prediction accuracy.

[0003] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present invention, and therefore may include information that does not constitute prior art known to ordinary technicians in this field. Summary of the Invention

[0004] The present invention provides a circRNA drug resistance prediction method based on a hybrid graph neural network, a storage medium, a computer program product, and an electronic device, which can overcome the defects in the prior art to a certain extent.

[0005] Other features and advantages of the present invention will become apparent from the following detailed description, or may be learned in part by practice of the present invention.

[0006] According to a first aspect of the present invention, a method for predicting circRNA drug resistance based on a hybrid graph neural network is provided, the method comprising:

[0007] Construct drug similarity matrix, disease similarity matrix, and circRNA similarity matrix;

[0008] Based on the drug similarity matrix, disease similarity matrix, and circRNA similarity matrix, the circRNA-drug correlation matrix, circRNA-disease correlation matrix, and drug-disease correlation matrix were constructed;

[0009] The drug similarity matrix, disease similarity matrix, and circRNA similarity matrix were used to integrate the data, combined with the circRNA-drug correlation matrix, circRNA-disease correlation matrix, and drug-disease correlation matrix to obtain the drug similarity network, disease similarity network, and circRNA similarity network, respectively.

[0010] Using a feature extraction model based on a hybrid graph neural network, we extract features from the drug similarity network, disease similarity network, and circRNA similarity network to obtain drug feature data, disease feature data, and circRNA feature data.

[0011] Performing feature fusion processing on the drug feature data and the circRNA feature data to obtain fused feature data;

[0012] The fusion feature data are classified to obtain prediction result data of the reaction relationship between circRNA and drug.

[0013] In some exemplary embodiments, constructing a drug similarity matrix comprises:

[0014] Convert the molecular structure sequence of each drug in the drug database into corresponding binary molecular fingerprint data;

[0015] The drug similarity between drugs is calculated using binary molecular fingerprint data, and the drug similarity matrix is ​​constructed using the drug similarity.

[0016] In some exemplary embodiments, constructing a disease similarity matrix comprises:

[0017] For the diseases in the disease database, calculate the similarity between any two diseases;

[0018] The disease similarity matrix is ​​constructed by combining the similarity between diseases, the number of genes related to the disease, and the correlation between disease and drugs.

[0019] In some exemplary embodiments, constructing a circRNA similarity matrix comprises:

[0020] Based on the names and genomic locations of circRNAs, data from multiple different disease-related circRNA datasets were integrated to construct a target circRNA dataset. The target circRNA dataset included circRNAs, drug resistance-associated ncRNAs, and drug-targeted ncRNAs.

[0021] The similarity of circRNAs was calculated based on their sequence information, and a circRNA similarity matrix was constructed based on the similarity of circRNAs.

[0022] In some exemplary embodiments, the drug similarity matrix, the disease similarity matrix, and the circRNA similarity matrix are combined with the circRNA-drug correlation matrix, the circRNA-disease correlation matrix, and the drug-disease correlation matrix to perform data integration processing to obtain a drug similarity network, a disease similarity network, and a circRNA similarity network, including:

[0023] Data integration and processing of drug similarity matrix, circRNA-drug correlation matrix, and drug-disease correlation matrix were performed to obtain drug similarity network;

[0024] Data integration and processing of disease similarity matrix, circRNA-disease correlation matrix, and drug-disease correlation matrix were performed to obtain disease similarity network;

[0025] The circRNA similarity matrix, circRNA-drug correlation matrix, and circRNA-disease correlation matrix were integrated to obtain the circRNA similarity network.

[0026] In some exemplary embodiments, the hybrid graph neural network-based feature extraction model includes at least two hybrid graph neural network layers;

[0027] Each hybrid graph neural network layer includes a graph convolutional neural network layer and a graph attention neural network layer arranged in sequence.

[0028] In some exemplary embodiments, when performing feature extraction using a feature extraction model based on a hybrid graph neural network, the method further includes:

[0029] Based on the disease characteristic data, constraints on the drug characteristic data and circRNA characteristic data are configured.

[0030] In some exemplary embodiments, the fusion feature data includes: multiple pairs of fusion descriptors constructed based on circRNA features and drug features.

[0031] According to a second aspect of the present invention, a storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the above-mentioned circRNA drug resistance prediction method based on hybrid graph neural network is implemented.

[0032] According to a third aspect of the present invention, a computer program product is provided, on which a computer program is stored, and when the computer program is executed by a processor, the above-mentioned circRNA drug resistance prediction method based on hybrid graph neural network is implemented.

[0033] According to a fourth aspect of the present invention, there is provided an electronic device, comprising:

[0034] processor; and

[0035] a memory for storing executable instructions of the processor;

[0036] Wherein, the processor is configured to implement the above-mentioned circRNA drug resistance prediction method based on hybrid graph neural network by executing the executable instructions.

[0037] The embodiment of the present invention provides a circRNA drug resistance prediction method based on a hybrid graph neural network. By constructing a drug similarity network, a disease similarity network, and a circRNA similarity network, disease background information is integrated into the benchmark data, thereby enabling the integration of specific disease context into the analysis of circRNA drug interactions. A feature extraction model based on the hybrid graph neural network is then used to extract drug feature data, disease feature data, and circRNA feature data, and the feature data is fused and then classified, thereby obtaining a more accurate prediction result of the reaction relationship between circRNA and drug.

[0038] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings are incorporated into and constitute a part of this specification, illustrate embodiments consistent with the present invention, and together with the description, serve to explain the principles of the present invention. Obviously, the drawings described below are only some embodiments of the present invention, and it is clear that those skilled in the art can derive other drawings based on these drawings without inventive effort.

[0040] Figure 1 A schematic diagram schematically illustrates a method for predicting circRNA drug resistance based on a hybrid graph neural network according to an exemplary embodiment of the present invention;

[0041] Figure 2 A schematic diagram schematically illustrates a model framework of an exemplary embodiment of the present invention;

[0042] Figure 3 A schematic diagram schematically illustrates a comparative prediction probability heat map of a circular RNA-drug interaction according to an exemplary embodiment of the present invention;

[0043] Figure 4 A schematic diagram schematically illustrates the results of predictive analysis of the top 30 circular RNAs for drug resistance in multiple models according to an exemplary embodiment of the present invention;

[0044] Figure 5The figure schematically shows the composition of an electronic device in an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0045] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be embodied in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0046] In addition, the accompanying drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the blocks shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0047] In related technologies, recent scientific research has demonstrated that some circRNAs can bind to proteins, affecting their function or stability, thereby regulating intracellular signaling and metabolic processes. Aberrant circRNA expression has been implicated in the development and progression of numerous diseases. In the field of cancer, in particular, the persistently high mortality rate caused by cancer, exacerbated by genomic mutations that confer drug resistance, highlights the challenges of cancer treatment and clinical trial implementation. However, traditional approaches to identifying circRNAs and drug resistance consume significant time, resources, and expertise. Deep learning approaches have significantly accelerated this process. Existing models have achieved commendable results in predicting circRNA drug resistance, but they still face a number of limitations and areas for improvement. First, previous models, such as GATECDA and MNCLCDA, utilize only a limited set of features related to circRNAs and drugs. Currently, existing models fail to incorporate disease context into their predictions. This comprehensive disease information could enhance understanding of how circRNAs contribute to drug resistance in specific disease states. While methods such as MKGCN and GraphCDD utilize graph neural network-based approaches, there is room for improvement in how these models extract and utilize features from data.

[0048] In view of the shortcomings and deficiencies of the existing technology, this example embodiment provides a circRNA drug resistance prediction method based on a hybrid graph neural network. Figure 1As shown, the circRNA drug resistance prediction method based on hybrid graph neural network can specifically include the following steps:

[0049] Step S11, constructing a drug similarity matrix, a disease similarity matrix, and a circRNA similarity matrix;

[0050] Step S12, constructing a circRNA-drug correlation matrix, a circRNA-disease correlation matrix, and a drug-disease correlation matrix based on the drug similarity matrix, the disease similarity matrix, and the circRNA similarity matrix;

[0051] Step S13, using the drug similarity matrix, disease similarity matrix, and circRNA similarity matrix, combined with the circRNA-drug correlation matrix, circRNA-disease correlation matrix, and drug-disease correlation matrix to perform data integration processing to obtain a drug similarity network, a disease similarity network, and a circRNA similarity network, respectively;

[0052] Step S14, using a feature extraction model based on a hybrid graph neural network, extracting features from the drug similarity network, the disease similarity network, and the circRNA similarity network, respectively, to obtain drug feature data, disease feature data, and circRNA feature data;

[0053] Step S15, performing feature fusion processing on the drug feature data and the circRNA feature data to obtain fused feature data;

[0054] Step S16: classify the fusion feature data to obtain prediction result data of the reaction relationship between circRNA and drug.

[0055] Below, the various steps of the circRNA drug resistance prediction method based on a hybrid graph neural network in this exemplary embodiment will be described in more detail with reference to the accompanying drawings and examples.

[0056] In step S11, a drug similarity matrix, a disease similarity matrix, and a circRNA similarity matrix are constructed.

[0057] Exemplarily, the constructing of the drug similarity matrix includes: converting the molecular structure sequence of each drug in the drug database into corresponding binary molecular fingerprint data; and

[0058] The drug similarity between drugs is calculated using binary molecular fingerprint data, and the drug similarity matrix is ​​constructed using the drug similarity.

[0059] Specifically, the drug database may be the DrugBank database, from which molecular structure information (SMILES) and pharmacological data may be extracted to calculate similarity information between drugs.

[0060] Specifically, a drug similarity calculation method based on chemical structure can be used. For example, the rdkit package in Python is used to process the SMILES sequence and convert it into a binary molecular fingerprint (MACC fingerprint). Subsequently, the Tanimoto similarity method is used to calculate drug similarity and compare these binary MACC molecular fingerprints. The calculation formula is as follows:

[0061] Among them, the drug similarity matrix The calculation formula is:

[0062]

[0063] Among them, Dr. i and Dr. j are the binary MACC molecular fingerprints of the two drugs i and j being compared, respectively. i,k and Dr. j,k is the kth position in the fingerprint of each molecule.

[0064] Exemplarily, the constructing of the disease similarity matrix includes: calculating the similarity between any two diseases in the disease database; and

[0065] The disease similarity matrix is ​​constructed by combining the similarity between diseases, the number of genes related to the disease, and the correlation between disease and drugs.

[0066] Specifically, disease information data from the Disease Ontology database can be compiled as a disease database. To assess the similarity between various diseases associated with circRNAs (circRNAs) and drugs, a network can be constructed with similarity as the connection weight. For example, the FNSemSim method can be used to calculate the similarity between diseases.

[0067] For a given disease d a and disease b ; Among them, d a ∈D,d b ∈D, D is the set of disease terms; consider D is the set of disease terms related to circRNAs and drugs, then disease d a and disease b The similarity between them can be calculated by FNSemSim, expressed as FNSim(d a , d b ). The formula includes:

[0068]

[0069] Among them, |Ga| and |Gb| represent the disease d a d b Number of related genes; g ai ∈G a ,g bj ∈G b ; R G (g) represents the connection weight in the fusion function association network; |G MICA |Represents the disease a and disease b The number of genes associated with a common ancestor.

[0070] After calculating the pairwise disease similarities, the min-max normalization method can be applied to appropriately scale these values. On this basis, the similarity calculation results are used to construct the disease similarity matrix DFS, which can be used to represent the association between diseases and drugs associated with circRNAs.

[0071] Exemplarily, constructing a circRNA similarity matrix includes: integrating data from multiple different disease-related circRNA datasets based on the names and genomic locations of the circRNAs to construct a target circRNA dataset; wherein the target circRNA dataset includes: circRNAs, drug resistance-related ncRNAs, and drug-targeted ncRNAs; and

[0072] The similarity of circRNAs was calculated based on their sequence information, and a circRNA similarity matrix was constructed based on the similarity of circRNAs.

[0073] Specifically, four disease-related circRNA datasets can be used, such as the circRNAdisease dataset, the circR2Disease dataset, the circR2Disease dataset, and the MNDR dataset. Among them, ncRNADrug brings together manually curated and computationally predicted non-coding RNAs (ncRNAs) associated with drug resistance and drug-targeted ncRNAs data. These datasets are crucial for establishing the connection between specific circRNAs and a range of diseases. To ensure consistency, the circRNAs in each dataset are mapped to the circBase database by referring to the circRNA name (alias) and genomic location, and combined with the original circBase database to obtain the target circRNA dataset.

[0074] To accurately assess the similarity between circRNAs, the sequence information of each circRNA was extracted from circBase, and the Levenshtein distance between the circRNA sequence information was calculated to represent the similarity. This metric quantifies the minimum number of single-character edits required to transform one sequence into another. For two circRNA sequences c1 and c2, a specific recursive formula is used to calculate the Levenshtein distance between the first i characters of c1 and the first j characters of c2. The corresponding formula is:

[0075]

[0076] If the characters of the two circRNA sequences are equal, then (c1[i]≠c2[j]) is 0; otherwise, it is 1. The function max(i, j) returns the larger value of i and j, while min(i, j) represents the minimum value among the listed options.

[0077] Based on the similarity calculation results between any two circRNA sequences, the circRNA similarity matrix CSeqSim is constructed to represent the circRNA sequence similarity in the benchmark dataset.

[0078] In step S12, based on the drug similarity matrix, the disease similarity matrix, and the circRNA similarity matrix, a circRNA-drug correlation matrix, a circRNA-disease correlation matrix, and a drug-disease correlation matrix are constructed.

[0079] For example, based on the similarity matrix of each entity mentioned above, the circRNA-drug correlation matrix A corresponding to the circRNA, drug and disease included in the benchmark dataset can be calculated: circ-dis , circRNA-disease correlation matrix A circ-dis , and the drug-disease correlation matrix A drug-dis .

[0080] Specifically, the Gaussian interaction profile (GIP) can be used to measure the similarity of these three entities. For circRNA, two types of GIP similarities can be calculated. One is based on the association between circRNA and disease (CGS_dis), which is represented by the correlation matrix A. circ-dis Calculation; Another circRNA-based drug resistance (CGS_dr) is calculated by the correlation matrix A circ-drug calculate.

[0081] Based on disease and drug resistance, the similarity of circRNA i and j can be expressed as:

[0082]

[0083] Where ρ is the kernel bandwidth, which is used to adjust the smoothness of the similarity function.

[0084] If M and N represent a set of entities in circRNA, disease or drug. According to the above correlation matrix A M-N ,for Represents the interaction profile of an entity m and an entity set N, that is, Where n∈N.

[0085] Similarly, the GIP similarity of diseases and drugs was calculated using the same method.

[0086] Based on diseases and circRNAs, the similarity of drugs can be expressed as:

[0087]

[0088] Similarly, based on drugs and circRNAs, the similarity of diseases can be expressed as:

[0089]

[0090] In step S13, the drug similarity matrix, disease similarity matrix, and circRNA similarity matrix are used to combine the circRNA-drug correlation matrix, the circRNA-disease correlation matrix, and the drug-disease correlation matrix for data integration processing to obtain the drug similarity network, the disease similarity network, and the circRNA similarity network, respectively.

[0091] Exemplarily, the above step S13 may specifically include:

[0092] Data integration and processing of drug similarity matrix, circRNA-drug correlation matrix, and drug-disease correlation matrix were performed to obtain drug similarity network;

[0093] Data integration and processing of disease similarity matrix, circRNA-disease correlation matrix, and drug-disease correlation matrix were performed to obtain disease similarity network;

[0094] The circRNA similarity matrix, circRNA-drug correlation matrix, and circRNA-disease correlation matrix were integrated to obtain the circRNA similarity network.

[0095] Specifically, based on the biological entity similarity matrix and GIP information obtained in the above steps, the similarity network of each biological entity is reintegrated.

[0096] Taking circRNA as an example, the formula of the similarity network is:

[0097]

[0098] Similarly, the similarity networks of diseases and drugs can be expressed as:

[0099]

[0100] In step S14, a feature extraction model based on a hybrid graph neural network is used to extract features from the drug similarity network, the disease similarity network, and the circRNA similarity network, respectively, to obtain drug feature data, disease feature data, and circRNA feature data.

[0101] Exemplarily, the feature extraction model based on the hybrid graph neural network includes at least two layers of hybrid graph neural network layers; each hybrid graph neural network layer includes a graph convolutional neural network layer and a graph attention neural network layer arranged in sequence.

[0102] Specifically, the feature extraction model uses a hybrid graph neural network architecture with at least two layers. Each layer combines the output of a graph convolutional neural network (GCN) and a graph attention neural network (GAT) to produce a unified feature representation, which is then passed as input to subsequent layers. In the following examples, a feature extraction model including two hybrid graph neural network layers is used as an example.

[0103] The GCN layer learns node representations by leveraging the topological structure of the graph. The formula includes:

[0104]

[0105] Among them, H (l) represents the node feature matrix of the lth layer, A is the adjacency matrix of the graph containing self-loops, represents the degree matrix corresponding to A, W (l) is the weight matrix of the lth layer, and σ represents the nonlinear activation function.

[0106] The GAT layer calculates node features by incorporating an attention mechanism, enabling the model to prioritize more important nodes in the neighborhood. The formula for the GAT layer includes:

[0107]

[0108] in, and represents the features of nodes i and j at layer l, is the attention coefficient between nodes i and j, which determines the significance of node j to node i, W (l) is the weight matrix at layer l.

[0109] In order to effectively combine the advantages of GCN and GAT, the outputs of the two layers can be averaged to form a unified feature representation for each node. The formula is:

[0110]

[0111] Then, the eigenvector H of this combination (l+1) It is used as the input to the next layer of the network. After two iterations of this process, the obtained feature representation has rich local and global graph structure information, and the CNN completes the representation integration.

[0112] Exemplarily, when feature extraction is performed using a feature extraction model based on a hybrid graph neural network, the method further includes: configuring constraints on the drug feature data and circRNA feature data based on the disease feature data.

[0113] For example, the focus of the training phase after obtaining the above representations is to integrate these features through disease contextual representation learning. We use the feature matrices F of circRNA, drug and disease respectively. circ 、F drug and F disease To calculate the interaction score matrix. These scores represent the potential associations between these entities, expressed as follows:

[0114]

[0115] These scores range from 0 to 1, indicating the likelihood of association between the pair. The training process uses known associations as constraints to guide the calculation of losses and the optimization of parameters. The training goal is to minimize the loss between the score matrix and the true association matrix. For the loss function, we use binary cross entropy loss. This loss function is crucial in measuring the difference between the predicted scores and the actual labels from our dataset. The formula is as follows:

[0116]

[0117] Among them, y i is the actual label, is the predicted probability from the score matrix. At the same time, this method focuses on drug-disease interactions, which ensures that the model not only learns from circRNA and drug features, but also effectively integrates disease information, enhancing its predictive ability for drug resistance affected by the disease environment.

[0118] For example, in the present invention, we prepare the prediction stage by constructing a balanced dataset that includes verified associations between circRNAs and drugs in positive samples. For negative samples, we randomly select the same number of circRNAs and unconfirmed drug pairs.

[0119] In step S15, feature fusion processing is performed on the drug feature data and the circRNA feature data to obtain fused feature data.

[0120] For example, a classifier may be provided, including a feature fusion processing layer and a classifier. In the feature fusion processing layer, the feature matrix F corresponding to the drug feature data and the circRNA feature data is circ and F drug In , the features of each circRNA and drug are extracted to form a fusion descriptor for each pair, which is defined as:

[0121] F pij =[F circ (i),F drug (j)]

[0122] This formula combines the i-th circRNA representation vector F circ (i) and the characterization vector F of the jth drug drug (j). This descriptor can be used as input to the data classification layer.

[0123] In step S16, the fusion feature data is classified to obtain prediction result data of the reaction relationship between circRNA and drug.

[0124] Specifically, the data classification layer of the classifier can use an MLP binary classifier. In the data classification layer of the classifier, the descriptor input from the data fusion processing layer is received, and the prediction result of the interaction between circRNA and drug resistance is output.

[0125] For example, this method uses eight commonly used evaluation metrics: accuracy, precision, recall, f1 score, Matthews correlation coefficient (MCC), specificity, area under the receiver operating characteristic (ROC) curve (AUROC), and area under precision (AUPR). In the experiment, 5-fold and 10-fold cross-validation were used to verify the performance of the model. The specific steps are as follows:

[0126] (1) First, based on the benchmark dataset ncRNADrug we compiled, we constructed a balanced dataset to prepare for the prediction stage, which includes the associations between verified circular RNAs and drugs in positive samples. For negative samples, we randomly selected the same number of circular RNA and unconfirmed drug pairs.

[0127] (2) Drug and circRNA characterization information extracted through the hybrid graph neural network stage is obtained from the learned feature matrix F circ and F d□ug Extract the features of each circular RNA and drug to form a fusion descriptor for each pair, defined as: F pij =[F circ (i),F drug (j)], combined with the i-th circRNA representation vector F circ (i) and the characterization vector F of the jth drug drug (j).

[0128] (3) The descriptor obtained in (2) is used as input to an MLP binary classifier to complete the prediction and obtain the prediction result.

[0129] (4) Calculate various indicators with the true labels of the samples retained in advance to obtain the results of various indicators.

[0130] (5) The superiority of our invented model is demonstrated by comparing it with the effect indicators of seven existing mainstream models in the same data set. The specific indicators are shown in Table 1.

[0131] Table 1

[0132]

[0133] In addition, to fully demonstrate the model generalization ability of the present invention, we introduced another dataset, NoncoRNA, and conducted a secondary verification with the seven mainstream models found, including:

[0134] (1) Normalize the input of our model on the NoncoRNA dataset.

[0135] (2) The required circRNA and drug characterization information is obtained through the hybrid graph neural network and training process, and the learned feature matrix F circ and F drug The features of each circular RNA and drug are extracted to form a fusion descriptor for each pair.

[0136] (3) Input the descriptor obtained in step (2) into our MLP binary classifier.

[0137] (4) Calculate various indicators based on the results of step (3) and the sample labels to complete the comparison.

[0138] The comparison results show that the model of this method has the best comprehensive evaluation performance among various indicators. The specific indicator results are shown in Table 2.

[0139] Table 2

[0140]

[0141] In addition, we deleted the original successful prediction examples of the model's prediction answers under the NCRNADRUG dataset, and then searched in public resources and papers (especially the NONCORNA database) to obtain multiple verifiable successful predictions. We took the top 20 high-scoring results and listed them in Table 3. The corresponding verification results are also shown in Table 3, which further illustrates the effectiveness of the present invention.

[0142] Table 3

[0143]

[0144] The method provided by the embodiment of the present invention first constructs a biological similarity information network, then inputs a hybrid graph neural network, fuses features with the CNN layer through a two-layer GCN and GAT combined graph neural network, and then comprehensively performs disease contextual representation learning training, adding disease association information to the circRNA and drug representation. Finally, we splice the comprehensive representation of circRNA and drug, input a three-layer MLP classifier, complete the final prediction task, and obtain the prediction result. Figure 3 Figure 2 shows a heatmap of the comparative prediction probabilities of circRNA-drug interactions across multiple models. It provides a comprehensive visualization of the predictive performance of various computational models across circular RNA-drug interactions. Each panel contains a heatmap representing the predicted probabilities of 25 drugs against 30 circular RNAs, with darker shades indicating a higher probability of drug resistance. Each heatmap has a scatterplot showing the predictive performance of each drug against the 30 circular RNAs, with red dots indicating true positive signatures where the circular RNA actually exhibits drug resistance. Figure 4 As shown, it represents the predictive analysis of the top 30 circular RNAs for drug resistance in various models. It shows heat maps and scatter plots, illustrating the predictive performance of various computational models in identifying the top 30 circular RNAs most likely to show resistance to 25 drugs. Each heat map describes the prediction probability, and the darker the shade, the higher the probability. The scatter plot corresponding to the right side of each heat map visualizes the distribution of these predictions, marked with the true resistance label as a red dot. The position of the red dot toward the right side of the scatter plot, especially exceeding the probability threshold of 0.8, indicates a higher prediction accuracy. The above results all show that: a circRNA drug resistance prediction method based on a hybrid graph neural network constructed by the present invention has good experimental results and has the potential for real-world application, and can be used for the prediction of circRNA and cancer drug resistance in the future.

[0145] It should be noted that the above figures are merely illustrative of the processes included in the method according to exemplary embodiments of the present invention and are not intended to be limiting. It is readily understood that the processes illustrated in the above figures do not indicate or limit the temporal order of these processes. Furthermore, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0146] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to an embodiment of the present invention, the features and functions of two or more modules or units described above can be concretized in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units to be concretized.

[0147] Figure 5 A schematic diagram of an electronic device suitable for implementing an embodiment of the present invention is shown.

[0148] It should be noted that Figure 5 The electronic device 1000 shown is only an example and should not limit the functions and scope of use of the embodiments of the present invention.

[0149] like Figure 5 As shown, electronic device 1000 includes a central processing unit (CPU) 1001, which can perform various appropriate actions and processes according to the program stored in read-only memory (ROM) 1002 or the program loaded from storage portion 1008 into random access memory (RAM) 1003. Various programs and data required for system operation are also stored in RAM 1003. CPU 1001, ROM 1002 and RAM 1003 are connected to each other via bus 1004. Input / output (I / O) interface 1005 is also connected to bus 1004.

[0150] The following components are connected to the I / O interface 1005: an input section 1006 including a keyboard, a mouse, and the like; an output section 1007 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and a speaker; a storage section 1008 including a hard disk and the like; and a communication section 1009 including a network interface card such as a LAN (Local Area Network) card or a modem. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to the I / O interface 1005 as needed. Removable media 1011, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 1010 as needed, so that computer programs read therefrom can be installed into the storage section 1008 as needed.

[0151] In particular, according to an embodiment of the present invention, the process described below with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product that includes a computer program carried on a storage medium, the computer program containing program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 1009 and / or installed from a removable medium 1011. When the computer program is executed by the central processing unit (CPU) 1001, the various functions defined in the system of the present application are performed.

[0152] It should be noted that the storage medium shown in the embodiments of the present invention can be a computer-readable signal medium or a computer-readable storage medium or any combination of the above. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, device or device. In the present invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any storage medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code contained on the storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0153] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the above-mentioned module, program segment, or a part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0154] The units involved in the embodiments of the present invention may be implemented in software or hardware, and the units described may also be provided in a processor. In some cases, the names of these units do not limit the units themselves.

[0155] It should be noted that, as another aspect, the present application also provides a storage medium, which can be included in an electronic device; or it can exist independently without being installed in the electronic device. The above storage medium carries one or more programs, and when the above one or more programs are executed by an electronic device, the electronic device implements the method described in the following embodiments. For example, the electronic device can implement the following Figure 1 The individual steps of the method are shown.

[0156] In one embodiment, the present application provides a computer program product, including a computer program, which implements the steps in the above-mentioned method embodiments when executed by a processor.

[0157] Furthermore, the above-described figures are merely illustrative of the processes included in the method according to exemplary embodiments of the present invention and are not intended to be limiting. It is readily understood that the processes illustrated in the above-described figures do not indicate or limit the temporal order of these processes. Furthermore, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0158] Other embodiments of the present invention will readily occur to those skilled in the art after considering the specification and practicing the invention herein. This application is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the invention being indicated by the claims.

[0159] It should be understood that the present invention is not limited to the exact construction described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof, which is limited only by the appended claims.

Claims

1. A circRNA drug resistance prediction method based on a hybrid graph neural network, characterized in that: The method comprises: Construct drug similarity matrix, disease similarity matrix, and circRNA similarity matrix; Based on the drug similarity matrix, disease similarity matrix, and circRNA similarity matrix, the circRNA-drug correlation matrix, circRNA-disease correlation matrix, and drug-disease correlation matrix were constructed; The drug similarity matrix, disease similarity matrix, and circRNA similarity matrix were used to integrate the data, combined with the circRNA-drug correlation matrix, circRNA-disease correlation matrix, and drug-disease correlation matrix to obtain the drug similarity network, disease similarity network, and circRNA similarity network, respectively. Using a feature extraction model based on a hybrid graph neural network, we extract features from the drug similarity network, disease similarity network, and circRNA similarity network to obtain drug feature data, disease feature data, and circRNA feature data. Performing feature fusion processing on the drug feature data and the circRNA feature data to obtain fused feature data; The fusion feature data are classified to obtain prediction result data of the reaction relationship between circRNA and drug.

2. The method according to claim 1, characterized in that The constructing of the drug similarity matrix comprises: Convert the molecular structure sequence of each drug in the drug database into corresponding binary molecular fingerprint data; The drug similarity between drugs is calculated using binary molecular fingerprint data, and the drug similarity matrix is ​​constructed using the drug similarity.

3. The method according to claim 1, characterized in that The constructing of the disease similarity matrix includes: For the diseases in the disease database, calculate the similarity between any two diseases; The disease similarity matrix is ​​constructed by combining the similarity between diseases, the number of genes related to the disease, and the correlation between disease and drugs.

4. The method according to claim 1, wherein The constructing of the circRNA similarity matrix comprises: Based on the names and genomic locations of circRNAs, data from multiple different disease-related circRNA datasets were integrated to construct a target circRNA dataset. The target circRNA dataset included circRNAs, drug resistance-associated ncRNAs, and drug-targeted ncRNAs. The similarity of circRNAs was calculated based on their sequence information, and a circRNA similarity matrix was constructed based on the similarity of circRNAs.

5. The method according to claim 1, wherein The drug similarity matrix, disease similarity matrix, and circRNA similarity matrix are used to integrate data in combination with the circRNA-drug correlation matrix, the circRNA-disease correlation matrix, and the drug-disease correlation matrix to obtain a drug similarity network, a disease similarity network, and a circRNA similarity network, including: Data integration and processing of drug similarity matrix, circRNA-drug correlation matrix, and drug-disease correlation matrix were performed to obtain drug similarity network; Data integration and processing of disease similarity matrix, circRNA-disease correlation matrix, and drug-disease correlation matrix were performed to obtain disease similarity network; The circRNA similarity matrix, circRNA-drug correlation matrix, and circRNA-disease correlation matrix were integrated to obtain the circRNA similarity network.

6. The method according to claim 1, characterized in that The hybrid graph neural network-based feature extraction model includes at least two hybrid graph neural network layers; Each hybrid graph neural network layer includes a graph convolutional neural network layer and a graph attention neural network layer arranged in sequence.

7. The method according to claim 1 or 6, characterized in that When feature extraction is performed using a feature extraction model based on a hybrid graph neural network, the method further includes: Based on the disease characteristic data, constraints on the drug characteristic data and circRNA characteristic data are configured.

8. The method according to claim 1, characterized in that The fusion feature data includes: multiple pairs of fusion descriptors constructed based on circRNA features and drug features.

9. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method for predicting circRNA drug resistance based on a hybrid graph neural network according to any one of claims 1 to 8 is implemented.

10. An electronic device, characterized in that: include: processor; as well as a memory for storing executable instructions of the processor; The processor is configured to execute the circRNA drug resistance prediction method based on a hybrid graph neural network according to any one of claims 1 to 8 by executing the executable instructions.

Images