A method and system for predicting anticancer drug-drug interactions

Abstract

This invention discloses a method and system for predicting anticancer drug-drug interactions, belonging to the field of drug action prediction technology. The method includes: constructing known drug pairs and test drug pairs for a target anticancer drug combination; determining multiple known coded drug feature pairs for each known drug pair and multiple test coded drug feature pairs for each test drug pair; determining a trained pulsed drug prediction model based on each known coded drug feature pair and inputting it into each test coded drug feature pair; generating anticancer pulsed features and test pulsed features for each test coded drug feature pair using a pulsed twin pulsed convolutional network based on a neuron reuse mechanism, concatenating them into a pulsed joint feature, inputting it into an IF pulsed perceptron to output the corresponding perceptual features; using a feature weighter to weight and fuse the perceptual features before inputting them into a pulsed multilayer perceptron for drug-drug interaction prediction, and outputting the target drug action result. This improves the reliability of drug-drug interaction prediction.

Description

Technical Field

[0001] This invention relates to the field of drug action prediction technology, and in particular to a method and system for predicting anticancer drug-drug interactions. Background Technology

[0002] The human body is a highly complex system. In the treatment of diseases, traditional monotherapy typically targets specific molecular pathways or physiological mechanisms, which often fails to adequately address complex diseases such as cancer or chronic illnesses. Combination therapy, through the synergistic effects of multiple targets and mechanisms, generates drug-drug interactions (DDIs), enabling more effective management of complex diseases. Furthermore, it usually uses lower drug doses, thus reducing drug toxicity and side effects. The combination of drugs can also effectively mitigate or overcome drug resistance. However, not all drug combination DDIs are beneficial; some DDIs may lead to antagonistic effects, thereby reducing overall efficacy. Therefore, it is necessary to predict the DDIs of drug combinations.

[0003] Traditional drug-drug interaction prediction methods rely on laboratory validation, which requires numerous and expensive experiments and clinical trials. Furthermore, they must meet stringent regulatory requirements to ensure safety and efficacy, making them unsuitable for practical applications. To address these issues, existing technologies employ a Transformer-like substructure extraction module to obtain a fixed number of representative vectors associated with various substructure patterns of drug molecules. The interaction strength between two drug substructures is then captured by a similarity-based interaction module. However, this method depends on single drug features, which can lead to information loss, reduced model generalization ability, and an inability to fully capture complex drug-drug interactions, thus lowering prediction reliability. Summary of the Invention

[0004] This invention provides a method and system for predicting anticancer drug-drug interactions, which solves the technical problem that existing technologies rely on single drug characteristics when predicting drug-drug interactions, which easily reduces the reliability of prediction.

[0005] The first aspect of this invention provides a method for predicting anticancer drug-drug interactions, comprising:

[0006] The target anticancer drug is combined with known interacting drugs and test interacting drugs, and the corresponding combinations are known drug pairs and test drug pairs;

[0007] Multiple drug features of the target anticancer drug, each of the known interacting drugs, and each of the drugs to be tested are obtained and encoded respectively to construct corresponding coded drug features;

[0008] The encoded drug features of each known drug pair and each drug pair to be tested are classified and aggregated to determine multiple known encoded drug feature pairs of each known drug pair and multiple drug feature pairs to be tested of each drug pair to be tested.

[0009] The trained pulse drug prediction model is determined based on the known coded drug feature pairs, and the coded drug feature pairs to be tested are input; the trained pulse drug prediction model includes a pulse twin pulse convolutional network, an IF pulse perceptron, a feature weighter, and a pulse multilayer perceptron;

[0010] Feature extraction is performed on each of the drug-encoded feature pairs using a pulsed twin pulsed convolutional network based on a neuron reuse mechanism, generating anticancer pulse features and test pulse features for each drug-encoded feature pair.

[0011] After concatenating the anti-cancer pulse features with the corresponding pulse features to be tested into a pulse joint feature, the feature is input into an IF pulse sensor for dimensionality reduction and nonlinear mapping, and the corresponding sensor feature is output.

[0012] The feature weighter is used to perform weighted fusion of each of the sensing features to determine the fused features, and the fused features are input into a pulsed multilayer perceptron for drug-drug interaction prediction, and the target drug effect result is output.

[0013] Optionally, the pulse twin pulse convolutional network includes two sub-pulse convolutional networks, each sub-pulse convolutional network comprising a LIF pulse convolutional block and a flattening layer; the step of extracting features from each of the drug-encoded feature pairs using the pulse twin pulse convolutional network based on a neuron reuse mechanism to generate anticancer pulse features and test pulse features for each drug-encoded feature pair includes:

[0014] Input any of the drug feature pairs to be tested into the pulse twin pulse convolutional network;

[0015] After extracting the encoded drug features of the target anticancer drug in the target encoded drug feature pair and resetting the membrane potential in real time by using the LIF pulse convolution block cascaded in the first sub-pulse convolution network, the anticancer pulse features are flattened by using a flattening layer.

[0016] The LIF neurons in the first sub-pulse convolutional network are reused in real time by cascading LIF pulse convolutional blocks in the second sub-pulse convolutional network. After feature extraction and real-time membrane potential reset of the encoded drug features of the drug to be tested in the drug to be tested feature pair, the features are flattened by a flattening layer to determine the pulse features to be tested.

[0017] The next pair of drug features to be tested is input into the pulse twin pulse convolutional network. The LIF neurons in the second sub-pulse convolutional network are reused in real time through the cascaded LIF pulse convolutional block in the first sub-pulse convolutional network until the anti-cancer pulse feature and the pulse feature to be tested for each pair of drug features to be tested are determined.

[0018] Optionally, the LIF pulsed convolutional block includes a one-dimensional convolutional layer, LIF neurons, and a pooling layer; the processing procedure of the LIF pulsed convolutional block includes:

[0019] The convolutional output features are determined by performing convolution operations on the convolutional input features of the input LIF pulse convolutional block through a one-dimensional convolutional layer.

[0020] After nonlinearly mapping the convolutional output features using LIF neurons, pooling dimensionality reduction is performed based on pooling layers to determine the pooling output features.

[0021] Optionally, the step of using the feature weighter to perform weighted fusion of each of the perceptual features to determine the fused feature includes:

[0022] The various perceptual features are concatenated to generate a combined feature;

[0023] The combined features are linearly transformed using a fully connected layer to determine the weight matrix;

[0024] The weight matrix is ​​nonlinearly transformed using the softmax activation function to determine the values ​​of multiple weight elements.

[0025] After performing a Hadamard product between each weight element value and the corresponding perceptual feature, the fused feature is output by concatenating the results.

[0026] Optionally, the step of determining the trained pulse drug prediction model based on the known coded drug features includes:

[0027] The known coded drug features are input into the pulse drug prediction model to be trained to predict drug-drug interactions, and the predicted drug action results are output.

[0028] The loss function value is calculated based on the predicted drug action results and the corresponding actual drug action results;

[0029] If the loss function value does not converge, the gradient descent method is used to update the model parameters until the loss function value converges, thus determining the trained pulse drug prediction model.

[0030] Optionally, the calculation process of the loss function value includes:

[0031] ;

[0032] In the formula, For loss function, To reflect the actual effects of the drug, To predict drug action outcomes, Category labels for drug action results , This represents the total number of categories.

[0033] A second aspect of the present invention provides an anticancer drug-drug interaction prediction system, comprising:

[0034] The drug data acquisition module is used to acquire target anticancer drugs with associated known interacting drugs and drugs to be tested, and the corresponding combinations are known drug pairs and drug pairs to be tested;

[0035] The data preprocessing module is used to acquire multiple drug features of the target anticancer drug, each of the known interacting drugs, and each of the drug to be tested, and encode them respectively to construct corresponding coded drug features;

[0036] The drug pair combination module is used to classify and aggregate the encoded drug features of each known drug pair and each drug pair to be tested, and determine multiple known encoded drug feature pairs of each known drug pair and multiple drug feature pairs to be tested of each drug pair.

[0037] The model training module is used to train and determine the trained pulse drug prediction model based on each of the known coded drug feature pairs, and to input each of the coded drug feature pairs to be tested; the trained pulse drug prediction model includes a pulse twin pulse convolutional network, an IF pulse perceptron, a feature weighter, and a pulse multilayer perceptron;

[0038] The pulse extraction module is used to extract features from each of the drug feature pairs to be tested using a pulse twin pulse convolutional network based on a neuron reuse mechanism, and to generate anticancer pulse features and test pulse features for each of the drug feature pairs to be tested.

[0039] The dimension reduction extraction module is used to concatenate the anti-cancer pulse features with the corresponding pulse features to be tested into a pulse joint feature, and then input it into the IF pulse sensor for dimension reduction nonlinear mapping, and output the corresponding sensor features.

[0040] The prediction module is used to perform weighted fusion of each of the perceived features using the feature weighter to determine the fused features, and input them into the pulse multilayer perceptron to predict drug-drug interactions, and output the target drug effect results.

[0041] A computer device provided in a third aspect of the present invention includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the anticancer drug-drug interaction prediction method as described in any of the preceding claims.

[0042] The fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed, implements the anticancer drug-drug interaction prediction method as described in any of the preceding claims.

[0043] The fifth aspect of the present invention provides a computer program product comprising a computer program / instruction, wherein the computer program / instruction, when executed by a processor, implements the anticancer drug-drug interaction prediction method as described in any of the preceding claims.

[0044] As can be seen from the above technical solutions, the present invention has the following advantages:

[0045] The above-described scheme of the present invention provides a method for predicting anticancer drug-drug interactions, comprising: using a target anticancer drug and associated known interacting drugs and test interacting drugs, correspondingly combining them into known drug pairs and test drug pairs; acquiring multiple drug features of the target anticancer drug, each known interacting drug, and each test interacting drug and encoding them respectively, constructing corresponding encoded drug features; classifying and aggregating the encoded drug features of each known drug pair and each test drug pair respectively, determining multiple known encoded drug feature pairs of each known drug pair and multiple test encoded drug feature pairs of each test drug pair; training and determining a trained pulse drug prediction model based on each known encoded drug feature pair, and inputting each test drug pair into the target drug pair. The model consists of a pulsed drug feature pair and a trained pulsed drug prediction model. The model includes a pulsed twin pulsed convolutional network, an IF pulsed perceptron, a feature weighter, and a pulsed multilayer perceptron. The pulsed twin pulsed convolutional network extracts features from each drug feature pair based on neuron reuse, generating anticancer pulsed features and target pulsed features for each pair. These anticancer pulsed features are concatenated with their corresponding target pulsed features to form a joint pulsed feature, which is then input into the IF pulsed perceptron for dimensionality reduction and nonlinear mapping, outputting the corresponding perceptual features. A feature weighter is used to weight and fuse these perceptual features to determine the fused features, which are then input into the pulsed multilayer perceptron for drug-drug interaction prediction, outputting the target drug action result. Based on the above scheme, drug features are encoded into binary data using multiple drug features as a foundation. By introducing spiking neurons into the spiking drug prediction model, the sparsity problem related to the encoded drug features is improved. A spiking twin spiking convolutional network is used to enhance the expressive power of the drug features. The model complexity is reduced through a neuron reuse mechanism. Furthermore, based on a feature weighter, different drug pairs are adaptively weighted and fused for fine-grained adjustment, effectively capturing the interactions between drugs and improving the reliability of drug-drug interaction prediction. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 A flowchart illustrating the steps of a method for predicting anticancer drug-drug interactions provided in an embodiment of the present invention;

[0048] Figure 2 This is a schematic diagram of the structure of a pulse twin pulse convolutional network provided in an embodiment of the present invention;

[0049] Figure 3 This is a flowchart illustrating the feature weighting function provided in an embodiment of the present invention.

[0050] Figure 4 This is a schematic diagram of data flow for an anticancer drug-drug interaction prediction method provided in an embodiment of the present invention;

[0051] Figure 5 This is a structural block diagram of an anticancer drug-drug interaction prediction system provided in an embodiment of the present invention. Detailed Implementation

[0052] This invention provides a method and system for predicting anticancer drug-drug interactions, which addresses the technical problem that existing technologies rely on single drug characteristics when predicting drug-drug interactions, easily reducing the reliability of predictions.

[0053] To make the objectives, features, and advantages of this invention more apparent and understandable, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described below are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0054] Please see Figure 1 , Figure 1 The flowchart illustrates the steps of a method for predicting anticancer drug-drug interactions provided in this embodiment of the invention.

[0055] This invention provides a method for predicting anticancer drug-drug interactions, comprising:

[0056] Step 101: The target anticancer drug is combined with the associated known interacting drugs and the drug to be tested, and the corresponding combinations are known drug pairs and drug pairs to be tested.

[0057] It should be noted that after selecting a certain anticancer drug as the target anticancer drug, multiple known interacting drugs with its known drug-drug interaction results and multiple test interacting drugs with its predicted drug-drug interaction results are extracted from the Drugbank database. The target anticancer drug is then paired with any known interacting drug to form a known drug pair, and the target anticancer drug is paired with any test interacting drug to form a test drug pair.

[0058] Understandably, in practical implementation, to facilitate visualization, multiple known drug pairs can be used to form a known drug pair set, and multiple drug pairs to be tested can be used to form a drug pair to be tested set. The sets are saved in the form of a list. Thus, each row in the known drug pair set represents each known drug pair with a known drug-drug interaction result label, including information such as drug name, drug ID, interaction type, and interaction description. Similarly, each row in the drug pair to be tested set represents each drug pair to be tested with a drug-drug interaction result label, including information such as drug name and drug ID.

[0059] Step 102: Obtain multiple drug characteristics of the target anticancer drug, each known interacting drug, and each drug to be tested, and encode them respectively to construct the corresponding coded drug characteristics.

[0060] It should be noted that drug characteristics refer to features that reflect the efficacy and pharmacological properties of a drug. Databases such as Drugbank, Uniport, and KEGG can be used to obtain multiple drug characteristics for each drug among the target anticancer drug, known interacting drugs, and drugs to be tested for interaction. These characteristics include chemical substructure features, target features, pathway features, and enzyme features. These drug characteristics are then encoded and converted into binary data, thus representing the encoded drug characteristics. In one implementation, chemical substructure features are represented by a simplified linear input canonical representation of the molecule and encoded using molecular fingerprints, while target features, pathway features, and enzyme features can be encoded using one-dimensional embedding vectors.

[0061] Understandably, in practical implementation, to better display the different dimensions of the data, the coded drug features of each drug can be stored in the corresponding view matrix according to the type of coded drug features. In the middle, in the formula Number the view. To encode drug characteristics, For the first One drug.

[0062] Step 103: Classify and aggregate the coded drug features of each known drug pair and each drug pair to be tested to determine multiple known coded drug feature pairs of each known drug pair and multiple drug feature pairs to be tested of each drug pair.

[0063] It should be noted that, in this embodiment, each drug pair is considered as a unit, and the multiple coded drug features associated with the two drugs in each known drug pair are classified and aggregated according to type. For example, the chemical substructure features of the target anticancer drug and the chemical substructure features of the known interacting drug in the known drug pair are combined into a known coded drug feature pair, thereby determining multiple known coded drug feature pairs for each known drug pair. Similarly, the multiple coded drug features associated with the two drugs in any drug pair to be tested are classified and combined. For example, the target features of the target anticancer drug and the target features of the interacting drug in the drug pair to be tested are combined into a test coded drug feature pair, thereby obtaining multiple test coded drug feature pairs for each drug pair to be tested.

[0064] Step 104: Based on the known coded drug feature pairs, train and determine the trained pulse drug prediction model, and input the coded drug feature pairs to be tested; the trained pulse drug prediction model includes a pulse twin pulse convolutional network, an IF pulse perceptron, a feature weighter, and a pulse multilayer perceptron.

[0065] In one specific implementation, a well-trained pulse drug prediction model is determined based on the known coded drug features, including:

[0066] The known coded drug features are input into the pulse drug prediction model to be trained to predict drug-drug interactions and output the predicted drug action results.

[0067] The loss function value is calculated based on the predicted drug action results and the corresponding actual drug action results;

[0068] If the loss function value does not converge, the gradient descent method is used to update the model parameters until the loss function value converges, thus determining the trained pulse drug prediction model.

[0069] It should be noted that, since the form of binary data limits the continuous representation of features, there may be a problem that the drug prediction model cannot be optimized using gradient information. In this embodiment, we consider introducing spiking neurons to build a spiking drug prediction model. The introduction of spiking neurons can effectively solve the sparsity problem related to binary encoding and incorporate nonlinearity into the model, so that it can approximate any function through learning.

[0070] In practice, the known coded drug features are used to predict the pulsatile drug prediction model to be trained, so as to generate the predicted drug effect. A loss function is defined to calculate the loss function value based on the predicted drug effect and the corresponding real drug effect. The convergence of the loss function value is judged. If it does not converge, the gradient descent method is used to update the model parameters with the goal of minimizing the loss function value until the loss function value converges, thereby determining the trained pulsatile drug prediction model. Then, the coded drug features to be tested are used to process the trained pulsatile drug prediction model.

[0071] In one implementation, the loss function is defined as the multi-class cross-entropy: , For loss function, To reflect the actual effects of the drug, To predict drug action outcomes, Category labels for drug action results , This represents the total number of categories.

[0072] Step 105: Extract features from each drug feature pair to be tested using a pulse twin pulse convolutional network based on the neuron reuse mechanism, and generate anticancer pulse features and test pulse features for each drug feature pair to be tested.

[0073] In one specific implementation, the pulsed twin pulsed convolutional network includes two sub-pulsed convolutional networks, each sub-pulsed convolutional network including a LIF pulsed convolutional block and a flattening layer; step 105 includes the following sub-steps:

[0074] S11. Input any coded drug feature pair into a pulse twin pulse convolutional network;

[0075] S12. After extracting the coded drug features of the target anticancer drug in the coded drug feature pair of the target anticancer drug through the cascaded LIF pulse convolution block in the first sub-pulse convolution network and resetting the membrane potential in real time, the anticancer pulse features are flattened by using a flattening layer.

[0076] S13. The LIF neurons in the first sub-pulse convolutional network are reused in real time through the LIF pulse convolutional block cascaded in the second sub-pulse convolutional network. After extracting the features of the coded drug features of the interacting drug in the target drug feature pair and resetting the membrane potential in real time, the feature is flattened through the flattening layer to determine the pulse features to be tested.

[0077] S14. Input the next drug feature pair to be tested into the pulse twin pulse convolutional network. Reuse the LIF neurons in the second sub-pulse convolutional network in real time through the cascaded LIF pulse convolutional block in the first sub-pulse convolutional network until the anti-cancer pulse feature and the pulse feature to be tested for each drug feature pair to be tested are determined.

[0078] In a more specific implementation, the LIF pulsating convolutional block includes a one-dimensional convolutional layer, LIF neurons, and a pooling layer; the processing of the LIF pulsating convolutional block includes:

[0079] The convolutional output features are determined by performing convolution operations on the convolutional input features of the input LIF pulse convolutional block through a one-dimensional convolutional layer.

[0080] After using LIF neurons to perform nonlinear mapping on the convolution output features, pooling dimensionality reduction is performed based on pooling layers to determine the pooling output features.

[0081] It should be noted that in the two sub-pulse convolutional networks of the pulse twin convolutional network, the parameters are shared, ensuring that two similar input features are converted into two similar feature representations by the corresponding sub-pulse convolutional networks. The neuron reuse mechanism refers to the LIF neurons in the pulse twin convolutional network being reused alternately in the two sub-pulse convolutional networks, thereby reducing the number of neurons, reducing storage requirements and hardware implementation complexity.

[0082] like Figure 2 As shown, each sub-pulse convolutional network includes LIF pulse convolutional blocks and flattened layers. Each LIF pulse convolutional block includes a one-dimensional convolutional layer, LIF neurons, and pooling layers. When the encoded drug features of the target anticancer drug or the encoded drug features of the interacting drug in any drug pair to be tested are input into the corresponding sub-pulse convolutional network:

[0083] 1) Feature extraction of the encoded drug features of the target anticancer drug, which serves as the input feature of the convolution, is performed through the one-dimensional convolutional layer of the first LIF pulse convolutional block of the first sub-pulse convolutional network: ,in, In the formula, For the first drug pair One drug, For the first drug pair The convolutional output features of a drug For convolution kernel weights, Input features for convolution, For convolution bias;

[0084] 2) After determining the convolution output features, pulses are generated through LIF neurons: In the formula, For time steps , For the first One LIF neuron, For the first LIF neurons time step membrane potential, For the first LIF neurons time step membrane potential, As the attenuation factor, Threshold voltage, For time steps The pulse; when At that time, a pulse is generated. Conversely ;

[0085] 3) The pulses generated by LIF neurons After pooling operation: In the formula, For pooling operations, Pooling output features;

[0086] 4) Clear the membrane potential in the LIF neuron to complete the reset, restoring it to its initial state: In the formula, For the first The membrane potential of a LIF neuron This represents the initial membrane potential.

[0087] 5) On the one hand, the pooling output features from step 3) are input into the next LIF pulse convolution block, and feature processing is performed according to steps 1) to 4). Then, the features are input into the flattening layer to perform the flattening operation, resulting in the final anti-cancer pulse features. Simultaneously, the LIF neurons in the first and second LIF pulse convolutional blocks of the first sub-pulse convolutional network are reused sequentially through the first LIF pulse convolutional block and the second LIF pulse convolutional block of the second sub-pulse convolutional network, and feature processing is performed according to steps 1) to 5) to obtain the pulse features to be tested. ;

[0088] 6) Input the next drug feature pair to be tested into the pulse twin pulse convolutional network. In the first sub-pulse convolutional network, reuse the LIF neurons reset by the second sub-pulse convolutional network in the previous time step. Perform feature processing according to steps 1) to 5) above until the anticancer pulse feature and the pulse feature to be tested for each drug feature pair to be tested are determined.

[0089] Step 106: After concatenating each anti-cancer pulse feature with the corresponding pulse feature to be tested into a pulse joint feature, input it into an IF pulse perceptron for dimensionality reduction nonlinear mapping and output the corresponding perceptron feature.

[0090] It should be noted that the characteristic pairs of each coded drug to be tested... Merging them into a one-dimensional column vector yields the pulse joint feature. Then, the input is fed into half the number of corresponding elements of the IF neurons using a fully connected approach for feature processing, thereby achieving dimensionality reduction. Finally, the membrane potential is updated in the IF output neuron of the IF impulse sensor. In the formula, For the first One IF output neuron, For the first Each IF output neuron time step membrane potential, For time step, For the first The input of each IF output neuron; based on the membrane potential Has the threshold voltage been reached? The IF output neuron triggers the corresponding pulse, and the sensory feature can be obtained by combining all the pulse results. Since simple serial connection only connects feature vectors, it lacks the ability to explore nonlinear relationships between features. However, the pulse perceptron can extract more abstract and complex features in feature fusion by simulating complex nonlinear dynamics. Furthermore, since it transmits information based on discrete spike signals, it can better filter noise and focus on meaningful interactive information.

[0091] Step 107: Use a feature weighter to perform weighted fusion of each sensing feature to determine the fused feature, and input it into a pulsed multilayer perceptron to predict drug-drug interactions, and output the target drug action results.

[0092] It should be noted that the feature weighter employs a self-attention mechanism to automatically assign weights to multiple perceptual features, thereby performing weighted fusion to determine the fused feature. This fused feature is then used as input to a spiking MLP for drug-drug interaction prediction, and the output is the target drug action result as the classification result. , It is a pulse multilayer sensor. For pulse multilayer sensor parameters, To predict the effects of drugs;

[0093] It is understandable that, such as Figure 4 As shown, the drug action results include, but are not limited to, categories such as synergistic, toxicity, and antagonism.

[0094] For example, when the spiking multilayer perceptron is a LIF spiking multilayer perceptron employing LIF neurons, its specific processing procedure includes:

[0095] ;

[0096] In the formula, For the LIF pulse multilayer sensor, the first Layer LIF neurons, For the first LIF spiking multilayer perceptron weights for LIF neurons in layers. For the first Pulses of LIF neurons in the layer For the first LIF spiking multilayer perceptron bias for LIF neurons in layers. For the first Linear characteristics of LIF neurons in layers For the first Layer LIF neuron time steps membrane potential, For the first Layer LIF neuron time steps pulse, For the total time step, The total number of LIF neurons in the LIF spiking multilayer perceptron; at each time step The membrane potential of LIF neurons is updated according to their input pulses and synaptic weights. The pulse firing of each LIF neuron in the last layer is regarded as a vote for a specific class. By counting the number of pulses fired by each LIF neuron in all time steps, the total support of each class can be obtained. After normalization, a probability distribution is obtained, which represents the probability that the input sample belongs to each class.

[0097] In one specific implementation, a feature weighter is used to perform weighted fusion of each perceived feature to determine the fused feature, including:

[0098] The various sensory features are concatenated to generate combined features;

[0099] The weight matrix is ​​determined by performing a linear transformation on the combined features through a fully connected layer.

[0100] The softmax activation function is used to perform a nonlinear transformation on the weight matrix to determine the values ​​of multiple weight elements.

[0101] After performing a Hadamard product between each weight element value and its corresponding perceptual feature, the fused feature is output by concatenating the results.

[0102] It should be noted that the feature weighter uses a self-attention mechanism to automatically assign weights to multiple perceptual features, such as... Figure 3 As shown, the feature weighter includes a fully connected layer and a softmax activation function; firstly, there are multiple perceptual features under each view number, and these perceptual features are concatenated into a single combined feature. , As a combination feature, For view The sensory features are then transformed into a weight matrix using a fully connected layer. , This is the weight matrix. For fully connected weights, For the fully connected bias, the softmax activation function is then used to compute the weight element values. , For output index , For view indexing, For the first Each view output index The weight element value, For input index Furthermore, the feature representation of each view is reweighted by performing a Hadamard product between the weighted element values ​​and the perceptual features. , For view Weighted perception features, Weight element value matrix The The weighted element values ​​are then listed, and finally, the newly weighted feature representations are aggregated to form a fused output. , This is a feature of fusion.

[0103] To verify the effectiveness of this scheme, a performance comparison experiment was conducted with existing predictive classification models. It is understood that there are various evaluation metrics for performance comparison:

[0104] ;

[0105] ;

[0106] ;

[0107] In the formula, For accuracy, The number of samples correctly predicted as positive is True Positive. True Negative means the number of samples correctly predicted as negative. The number of samples that are false positives, i.e., incorrectly predicted as positive. False Negative means the number of samples incorrectly predicted as negative. For accuracy, Recall rate;

[0108] Taking accuracy as the evaluation metric, the performance comparison results are shown in Table 1:

[0109] Table 1 Performance Comparison Results

[0110]

[0111] When predicting drug-drug interactions of the anticancer drugs Erlotinib and Gefitnib, this scheme uses LIF neuron parameters including a membrane time constant of 20 ms, a membrane resistance of 10 megohms, a resting potential of -70 mV, a threshold potential of -50 mV, and a reset potential of -65 mV. In the pulse twin pulse convolutional network, the parameters of the convolutional layer are set to 16 input channels, 32 output channels, 3 kernels, 1 stride, "same" padding, and no bias term. The pooling type is max pooling, the pooling window size is 2, the stride is 2, and the padding is "valid".

[0112] As shown in Table 1, the anticancer drugs Erlotinib and Gefitnib were selected as target anticancer drugs. Among them, 1019 relevant drug interaction data were collected for Erlotinib, including 5 different types of DDIs; and 1458 relevant drug interaction data were collected for Gefitnib, including 8 different types of DDIs. According to the accuracy results in Table 1, the method proposed in this embodiment has a significantly better accuracy in predicting drug-drug interactions than other algorithms.

[0113] In this embodiment of the invention, based on multiple drug features, the drug features are encoded into binary data. The sparsity problem related to the encoded drug features is improved by introducing spiking neurons into the spiking drug prediction model. The spiking twin spiking convolutional network is used to enhance the expressive power of the drug features. The neuron reuse mechanism can also reduce the model complexity. Furthermore, based on the feature weighter, different drug pairs are adaptively weighted and fused for fine-grained adjustment, effectively capturing the interactions between drugs and improving generalization ability and prediction reliability.

[0114] Please see Figure 5 , Figure 5 This is a structural block diagram of an anticancer drug-drug interaction prediction system provided in an embodiment of the present invention.

[0115] This invention provides an anticancer drug-drug interaction prediction system, comprising:

[0116] The drug data acquisition module 501 is used to acquire target anticancer drugs with associated known interacting drugs and drugs to be tested, and the corresponding combinations are known drug pairs and drug pairs to be tested;

[0117] The data preprocessing module 502 is used to acquire multiple drug features of the target anticancer drug, each known interacting drug, and each test interacting drug, and encode them respectively to construct the corresponding coded drug features;

[0118] The drug pair combination module 503 is used to classify and aggregate the coded drug features of each known drug pair and each drug pair to be tested, and to determine multiple known coded drug feature pairs of each known drug pair and multiple coded drug feature pairs to be tested of each drug pair.

[0119] The model training module 504 is used to train and determine the trained pulse drug prediction model based on each known coded drug feature pair, and input each coded drug feature pair to be tested; the trained pulse drug prediction model includes a pulse twin pulse convolutional network, an IF pulse perceptron, a feature weighter, and a pulse multilayer perceptron;

[0120] The pulse extraction module 505 is used to extract features from each pair of drug codes to be tested based on the neuron reuse mechanism through a pulse twin pulse convolutional network, and generate anticancer pulse features and test pulse features for each pair of drug codes to be tested.

[0121] The dimension reduction extraction module 506 is used to concatenate each anti-cancer pulse feature with the corresponding pulse feature to be tested into a pulse joint feature, and then input it into the IF pulse sensor for dimension reduction nonlinear mapping, and output the corresponding sensor feature.

[0122] The prediction module 507 is used to perform weighted fusion of each perceived feature using a feature weighter to determine the fused feature, and input it into the pulse multilayer perceptron to predict drug-drug interactions, and output the target drug effect result.

[0123] Optionally, the pulse twin pulse convolutional network includes two sub-pulse convolutional networks, each sub-pulse convolutional network comprising a LIF pulse convolutional block and a flattening layer; the pulse extraction module 505 is specifically used for:

[0124] Input any pair of coded drug features into a pulse twin pulse convolutional network;

[0125] After extracting the coded drug features of the target anticancer drug from the cascaded LIF pulse convolutional blocks in the first sub-pulse convolutional network and resetting the membrane potential in real time, a flattening layer is used to flatten the anticancer pulse features.

[0126] The LIF neurons in the first sub-pulse convolutional network are reused in real time by cascading LIF pulse convolutional blocks in the second sub-pulse convolutional network. After feature extraction and real-time membrane potential reset of the encoded drug features of the interacting drugs in the target encoded drug feature pair, the features are flattened through a flattening layer to determine the pulse features to be tested.

[0127] The next drug feature pair to be tested is input into the pulse twin pulse convolutional network. The LIF neurons in the second sub-pulse convolutional network are reused in real time through the cascaded LIF pulse convolutional block in the first sub-pulse convolutional network until the anticancer pulse feature and the test pulse feature of each drug feature pair to be tested are determined.

[0128] Optionally, the LIF pulsating convolutional block includes a one-dimensional convolutional layer, LIF neurons, and a pooling layer; the processing procedure of the LIF pulsating convolutional block includes:

[0129] The convolutional output features are determined by performing convolution operations on the convolutional input features of the input LIF pulse convolutional block through a one-dimensional convolutional layer.

[0130] After using LIF neurons to perform nonlinear mapping on the convolution output features, pooling dimensionality reduction is performed based on pooling layers to determine the pooling output features.

[0131] Optionally, a feature weighter is used to perform weighted fusion of the perceptual features to determine the fused features, including:

[0132] The various sensory features are concatenated to generate combined features;

[0133] The weight matrix is ​​determined by performing a linear transformation on the combined features through a fully connected layer.

[0134] The softmax activation function is used to perform a nonlinear transformation on the weight matrix to determine the values ​​of multiple weight elements.

[0135] After performing a Hadamard product between each weight element value and its corresponding perceptual feature, the fused feature is output by concatenating the results.

[0136] Optionally, a well-trained pulse drug prediction model is determined based on the known coded drug features, including:

[0137] The known coded drug features are input into the pulse drug prediction model to be trained to predict drug-drug interactions and output the predicted drug action results.

[0138] The loss function value is calculated based on the predicted drug action results and the corresponding actual drug action results;

[0139] If the loss function value does not converge, the gradient descent method is used to update the model parameters until the loss function value converges, thus determining the trained pulse drug prediction model.

[0140] Optionally, the calculation process of the loss function value includes:

[0141] ;

[0142] In the formula, For loss function, To reflect the actual effects of the drug, To predict drug action outcomes, Category labels for drug action results , This represents the total number of categories.

[0143] This invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program; when the computer program is executed by the processor, the processor performs the steps of the anticancer drug-drug interaction prediction method as described in any of the above embodiments.

[0144] This invention also provides a computer-readable storage medium storing a computer program / instructions thereon, which, when executed by a processor, implements the steps of the anticancer drug-drug interaction prediction method as described in any of the above embodiments.

[0145] This invention also provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the anticancer drug-drug interaction prediction method as described in any of the above embodiments.

[0146] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system and modules described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0147] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0148] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0149] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0150] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0151] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting anticancer drug-drug interactions, characterized in that, include: The target anticancer drug is combined with known interacting drugs and test interacting drugs, and the corresponding combinations are known drug pairs and test drug pairs; Multiple drug features of the target anticancer drug, each of the known interacting drugs, and each of the drugs to be tested are obtained and encoded respectively to construct corresponding coded drug features; The encoded drug features of each known drug pair and each drug pair to be tested are classified and aggregated to determine multiple known encoded drug feature pairs of each known drug pair and multiple drug feature pairs to be tested of each drug pair to be tested. The trained pulse drug prediction model is determined based on the known coded drug feature pairs, and the coded drug feature pairs to be tested are input; the trained pulse drug prediction model includes a pulse twin pulse convolutional network, an IF pulse perceptron, a feature weighter, and a pulse multilayer perceptron; Feature extraction is performed on each of the drug-encoded feature pairs using a pulsed twin pulsed convolutional network based on a neuron reuse mechanism, generating anticancer pulse features and test pulse features for each drug-encoded feature pair. After concatenating the anti-cancer pulse features with the corresponding pulse features to be tested into a pulse joint feature, the feature is input into an IF pulse sensor for dimensionality reduction and nonlinear mapping, and the corresponding sensor feature is output. The feature weighter is used to perform weighted fusion of each of the sensing features to determine the fused features, and the fused features are input into a pulse multilayer sensor to predict drug-drug interactions, and the target drug effect results are output. The pulse twin pulse convolutional network includes two sub-pulse convolutional networks, each sub-pulse convolutional network comprising a LIF pulse convolutional block and a flattening layer; the feature extraction of each of the drug-encoded feature pairs by the pulse twin pulse convolutional network based on a neuron reuse mechanism, generating anticancer pulse features and test pulse features for each drug-encoded feature pair, includes: Input any of the drug feature pairs to be tested into the pulse twin pulse convolutional network; After extracting the encoded drug features of the target anticancer drug in the target encoded drug feature pair and resetting the membrane potential in real time by using the LIF pulse convolution block cascaded in the first sub-pulse convolution network, the anticancer pulse features are flattened by using a flattening layer. The LIF neurons in the first sub-pulse convolutional network are reused in real time by cascading LIF pulse convolutional blocks in the second sub-pulse convolutional network. After feature extraction and real-time membrane potential reset of the encoded drug features of the drug to be tested in the drug to be tested feature pair, the features are flattened by a flattening layer to determine the pulse features to be tested. The next pair of drug features to be tested is input into the pulse twin pulse convolutional network. The LIF neurons in the second sub-pulse convolutional network are reused in real time through the cascaded LIF pulse convolutional block in the first sub-pulse convolutional network until the anti-cancer pulse feature and the pulse feature to be tested for each pair of drug features to be tested are determined. The step of using the feature weighter to perform weighted fusion of each of the perceptual features to determine the fused feature includes: The various perceptual features are concatenated to generate a combined feature; The combined features are linearly transformed using a fully connected layer to determine the weight matrix; The weight matrix is ​​nonlinearly transformed using the softmax activation function to determine the values ​​of multiple weight elements. After performing a Hadamard product between each weight element value and the corresponding perceptual feature, the fused feature is output by concatenating the results.

2. The method for predicting anticancer drug-drug interactions according to claim 1, characterized in that, The LIF pulsed convolutional block comprises a one-dimensional convolutional layer, LIF neurons, and a pooling layer; the processing procedure of the LIF pulsed convolutional block includes: The convolutional output features are determined by performing convolution operations on the convolutional input features of the input LIF pulse convolutional block through a one-dimensional convolutional layer. After nonlinearly mapping the convolutional output features using LIF neurons, pooling dimensionality reduction is performed based on pooling layers to determine the pooling output features.

3. The method for predicting anticancer drug-drug interactions according to claim 1, characterized in that, The training of the pulse drug prediction model based on the known coded drug features includes: The known coded drug features are input into the pulse drug prediction model to be trained to predict drug-drug interactions, and the predicted drug action results are output. The loss function value is calculated based on the predicted drug action results and the corresponding actual drug action results; If the loss function value does not converge, the gradient descent method is used to update the model parameters until the loss function value converges, thus determining the trained pulse drug prediction model.

4. The method for predicting anticancer drug-drug interactions according to claim 3, characterized in that, The calculation process for the loss function value includes: ； In the formula, For loss function, To reflect the actual effects of the drug, To predict drug action outcomes, Category labels for drug action results , This represents the total number of categories.

5. A drug-drug interaction prediction system for cancer, characterized in that, include: The drug data acquisition module is used to acquire target anticancer drugs with associated known interacting drugs and drugs to be tested, and the corresponding combinations are known drug pairs and drug pairs to be tested; The data preprocessing module is used to acquire multiple drug features of the target anticancer drug, each of the known interacting drugs, and each of the test interacting drugs, and encode them respectively to construct corresponding coded drug features; The drug pair combination module is used to classify and aggregate the encoded drug features of each known drug pair and each drug pair to be tested, and determine multiple known encoded drug feature pairs of each known drug pair and multiple drug feature pairs to be tested of each drug pair. The model training module is used to train and determine the trained pulse drug prediction model based on each of the known coded drug feature pairs, and input each of the coded drug feature pairs to be tested; the trained pulse drug prediction model includes a pulse twin pulse convolutional network, an IF pulse perceptron, a feature weighter, and a pulse multilayer perceptron; The pulse extraction module is used to extract features from each of the drug feature pairs to be tested using a pulse twin pulse convolutional network based on a neuron reuse mechanism, and to generate anticancer pulse features and test pulse features for each of the drug feature pairs to be tested. The dimension reduction extraction module is used to concatenate the anti-cancer pulse features with the corresponding pulse features to be tested into a pulse joint feature, and then input it into the IF pulse sensor for dimension reduction nonlinear mapping, and output the corresponding sensor features. The prediction module is used to perform weighted fusion of each of the perceived features using the feature weighter to determine the fused features, and input them into the pulse multilayer sensor to predict drug-drug interactions, and output the target drug effect results; The pulse twin pulse convolutional network includes two sub-pulse convolutional networks, each sub-pulse convolutional network comprising a LIF pulse convolutional block and a flattening layer; the pulse extraction module is specifically used for: Input any of the drug feature pairs to be tested into the pulse twin pulse convolutional network; After extracting the encoded drug features of the target anticancer drug in the target encoded drug feature pair and resetting the membrane potential in real time by using the LIF pulse convolution block cascaded in the first sub-pulse convolution network, the anticancer pulse features are flattened by using a flattening layer. The LIF neurons in the first sub-pulse convolutional network are reused in real time by cascading LIF pulse convolutional blocks in the second sub-pulse convolutional network. After feature extraction and real-time membrane potential reset of the encoded drug features of the drug to be tested in the drug to be tested feature pair, the features are flattened by a flattening layer to determine the pulse features to be tested. The next pair of drug features to be tested is input into the pulse twin pulse convolutional network. The LIF neurons in the second sub-pulse convolutional network are reused in real time through the cascaded LIF pulse convolutional block in the first sub-pulse convolutional network until the anti-cancer pulse feature and the pulse feature to be tested for each pair of drug features to be tested are determined. The step of using the feature weighter to perform weighted fusion of each of the perceptual features to determine the fused feature includes: The various perceptual features are concatenated to generate a combined feature; The combined features are linearly transformed using a fully connected layer to determine the weight matrix; The weight matrix is ​​nonlinearly transformed using the softmax activation function to determine the values ​​of multiple weight elements. After performing a Hadamard product between each weight element value and the corresponding perceptual feature, the fused feature is output by concatenating the results.

6. A computer device, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor causes the processor to perform the steps of the anticancer drug-drug interaction prediction method as described in any one of claims 1-4.

7. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the anticancer drug-drug interaction prediction method as described in any one of claims 1-4.

8. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the anticancer drug-drug interaction prediction method as described in any one of claims 1-4.

Images