Methods, apparatus, electronic devices and media for determining drug combination synergy

Abstract

This application discloses a method, apparatus, electronic device, and medium for determining the synergy of drug combinations. The method includes: acquiring drug information for each drug in the drug combination to be tested and information on the cell lines to be tested. The cell line information includes mRNA expression information and gene mutation information. The expression information includes the mean expression sequence and the standard deviation expression sequence for each cell line. Based on the mean expression sequence and the standard deviation expression sequence, an initial expression feature sequence for each cell line is obtained. The initial expression feature sequence is input into a pre-trained autoencoder for feature extraction to obtain a dimensionality-reduced expression feature sequence. Based on the drug information, gene mutation information, the dimensionality-reduced expression feature sequence, and a pre-trained deep learning model, it is determined whether the drug combination to be tested has synergy with the cell lines to be tested.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, apparatus, electronic device and medium for determining the synergy of drug combinations. Background Technology

[0002] Currently, combination therapy with anticancer drugs is considered a promising treatment strategy and is widely used in the clinical treatment of tumors. Compared with monotherapy, it can significantly reduce drug dosage or toxic side effects while enhancing or maintaining the same efficacy. However, drug combination effects can also be adverse, even shortening progression-free survival in cancer patients. Generally, drug combination interactions are classified into three types: additive interactions (the effect of the drug combination equals the sum of the effects of the individual drugs taken alone), synergistic effects (the effect of the drug combination is better than the sum of the effects of the individual drugs taken alone), and antagonistic interactions (the effect of the drug combination is less than the sum of the effects of the individual drugs). Therefore, determining whether a drug combination exhibits synergistic effects is of great significance.

[0003] With high-throughput screening and virtual screening becoming important components of modern drug research methods, computer simulation experiments are being used more and more widely in biomedical sciences. Using computational models to predict the synergy of drug combinations is an important solution. Summary of the Invention

[0004] This application provides a method, apparatus, electronic device, and medium for determining the synergy of drug combinations.

[0005] In a first aspect, embodiments of this application provide a method for determining the synergy of drug combinations, including:

[0006] Obtain drug information for each drug in the drug combination to be tested and information on the cell line to be tested. The cell line information includes mRNA expression information and gene mutation information. The expression information includes the mean expression sequence and the standard deviation expression sequence for each cell line.

[0007] Based on the expression mean sequence and the expression standard deviation sequence, an initial expression feature sequence for each cell line is obtained. The initial expression feature sequence is then input into a pre-trained autoencoder for feature extraction to obtain a dimension-reduced expression feature sequence.

[0008] Based on the drug information, the gene mutation information, the dimensionality-reduced expression feature sequence, and the pre-trained deep learning model, it is determined whether the drug combination to be tested has synergistic effects on the cell line to be tested.

[0009] Further, the initial expression characteristic sequence for each cell line is obtained based on the expression mean sequence and the expression standard deviation sequence, including:

[0010] For each expression dimension of each cell line, based on a preset random variable and the expression standard deviation of the expression dimension, the perturbation amount of the expression dimension is obtained, and based on the expression mean of the expression dimension and the perturbation amount, the initial expression characteristics of the expression dimension are obtained.

[0011] Based on the initial expression features of each expression dimension of each cell line, an initial expression feature sequence for each cell line is constructed.

[0012] Furthermore, the random variable is a discrete random variable, and the values ​​of the random variable include: 0, 1, and -1, wherein the probability of the random variable taking the value 0 is p, the probability of the random variable taking the value -1 is (1-p) / 2, the probability of the random variable taking the value 1 is (1-p) / 2, and p is a number greater than 0 and less than 1.

[0013] Furthermore, the random variable is a continuous random variable, taking values ​​from -1 to 1, and the random variable values ​​satisfy a distribution with a mean of 0 and a standard deviation of 1.

[0014] Furthermore, the autoencoder is trained based on a preset loss function, and the autoencoder includes an encoding layer, an intermediate layer, and a decoding layer;

[0015] The intermediate layer is used to output the dimension-reduced expression feature sequence, and the value of the preset loss function is determined based on the difference between the decoding result of the decoding layer and the expression mean sequence corresponding to the initial expression feature sequence input to the encoding layer.

[0016] Further, obtaining the expression mean sequence and expression standard deviation sequence includes:

[0017] Obtain the original mean expression sequence and the original standard deviation expression sequence for each cell line;

[0018] For each expression dimension of each cell line, the original expression mean and original expression standard deviation were standardized to obtain the standardized expression mean sequence and expression standard deviation sequence.

[0019] Furthermore, the drug information includes: structural feature information of the drug; obtaining the drug information for each drug in the drug combination to be tested includes:

[0020] Obtain the SMILES structural information of each drug in the drug combination to be tested;

[0021] The SMILES structural information is vectorized to obtain the structural feature information corresponding to each drug.

[0022] Furthermore, the drug information also includes: drug side effect characteristic information; obtaining the drug information for each drug in the drug combination to be tested further includes:

[0023] Obtain the side effect characteristic information of each drug in the drug combination to be tested. The dimensions of the side effect characteristic information are determined based on the number of side effect types contained in the pre-constructed side effect information database. Each dimension of the side effect characteristic information corresponds to one side effect.

[0024] Furthermore, based on the drug information, the gene mutation information, the dimensionality-reduced expression feature sequence, and the pre-trained deep learning model, it is determined whether the drug combination to be tested has synergistic effects on the cell line to be tested, including:

[0025] The structural feature information and side effect feature information of each drug, as well as the gene mutation information and dimensionality-reduced expression feature sequence of each cell line, are horizontally concatenated to obtain the input feature vector;

[0026] The input feature vector is input into a pre-trained deep learning model to obtain a synergy prediction result, which is used to characterize whether the drug combination to be tested has synergy with the cell line to be tested.

[0027] Secondly, embodiments of this application provide a device for determining the synergy of drug combinations, comprising:

[0028] The acquisition module is used to acquire drug information for each drug in the drug combination to be tested and cell line information to be tested. The cell line information includes mRNA expression information and gene mutation information. The expression information includes the mean expression sequence and the standard deviation expression sequence.

[0029] The expression feature extraction module is used to obtain the initial expression feature sequence of each cell line based on the expression mean sequence and the expression standard deviation sequence, and input the initial expression feature sequence into a pre-trained autoencoder for feature extraction to obtain the dimensionality-reduced expression feature sequence.

[0030] The determination module is used to determine whether the drug combination to be tested has synergistic effects on the cell line to be tested, based on the drug information, the gene mutation information, the dimensionality-reduced expression feature sequence, and the pre-trained deep learning model.

[0031] Thirdly, embodiments of this application provide an electronic device, the electronic device comprising:

[0032] One or more processors;

[0033] Memory;

[0034] One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, the one or more applications being configured to: perform the drug combination synergy determination method described in the first aspect above.

[0035] Fourthly, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the drug combination synergy determination method described in the first aspect above.

[0036] The technical solutions provided in this application have at least the following technical effects or advantages:

[0037] The drug combination synergy determination method, apparatus, electronic device, and medium provided in this application embodiment can predict the synergy of the drug combination under test end-to-end through a deep learning model. Furthermore, in the process of feature extraction of mRNA expression information, expression standard deviation data is integrated, which helps to enrich the extracted expression information, improve the accuracy of expression feature extraction, thereby improving the accuracy of synergy prediction results and reducing unnecessary losses in subsequent experimental stages.

[0038] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0039] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0040] Figure 1 This is a flowchart of a method for determining the synergy of drug combinations in an embodiment of this application;

[0041] Figure 2 This is a network architecture diagram of an exemplary autoencoder in an embodiment of this application;

[0042] Figure 3 This is a schematic diagram of a collaborative prediction process in an embodiment of this application;

[0043] Figure 4 This is a schematic diagram of a drug combination synergy determination device according to an embodiment of this application;

[0044] Figure 5This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0045] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art. Throughout this document, the term "a plurality of" includes two or more cases.

[0046] In this article, mRNA (Messenger RNA) is a large class of RNA molecules that transmits genetic information from DNA to ribosomes, where it serves as a template for protein synthesis and determines the amino acid sequence of the peptide chain of the protein product expressed by the gene. SMILES (Simplified Molecular Input Line Entry System) is a linear representation of drug structures, using letters, numbers, and special characters to represent a molecule. For example, "C" represents a carbon atom, and "=" represents a covalent bond between two atoms. Carbon dioxide can be represented as O=C=O, and aspirin can be represented as O=C(C)OC1CCCCC1C(=O)O. The longest SMILES expression for a drug contains 188 positions.

[0047] Figure 1 A flowchart illustrating a method for determining the synergy of drug combinations provided in an embodiment of this application is shown. Figure 1 As shown, the method for determining the synergy of drug combinations may include at least the following steps S101 to S103.

[0048] Step S101: Obtain drug information for each drug in the drug combination to be tested and information on the cell line to be tested. Cell line information includes mRNA expression information and gene mutation information. Expression information includes the mean expression sequence and the standard deviation expression sequence.

[0049] The drug combination to be tested is a combination of drugs whose synergistic effect needs to be assessed, and can include a combination of two or more drugs. For example, the drug combination to be tested could be an anti-tumor drug combination. The cell line to be tested is the collection of cells that the drug combination can act on. For example, in the field of oncology, the cell line can be a collection of multiple tumor cells. From a biological mechanism perspective, the combined use of drugs can interact and produce a synergistic effect on the cell line to kill the tumor cell line.

[0050] In some examples, drug information may include structural feature information corresponding to the drug. In this case, step S101, the process of obtaining drug information for each drug in the drug combination to be tested, may include: obtaining the SMILES structural information for each drug in the drug combination to be tested; and vectorizing the SMILES structural information to obtain the structural feature information corresponding to each drug. Vectorizing the SMILES structural information and using it as one of the input features, directly inputting it into the deep learning model to obtain the synergy prediction result, is beneficial for improving the prediction efficiency of drug combination synergy.

[0051] For example, the SMILES structure information of drugs can be obtained from drug databases such as DrugBank. Embedding training can be performed beforehand on all drugs with SMILES structures in the database to obtain a vector representation of the SMILES structure for each drug. The vector representation dimension can be set according to actual needs. For example, the output dimension of the vector transformation model can be set to 6, then each letter, number, and special character in the SMILES structure can be converted into a 6-dimensional vector, and the vectorized format of the drug SMILES structure is 6 * the original structure length. For example, word vector models such as Word2Vec can be used for vector transformation.

[0052] Taking the SMILES structure information O=C(C)OC1CCCCC1C(=O)O as an example, through embedding training, we can learn the vector representation of each character such as "O", "=", "C", etc., and thus obtain the vector representation of the overall structure.

[0053] Furthermore, in addition to structural feature information, drug information can also include: side effect feature information corresponding to the drug. In this case, the process of obtaining drug information for each drug in the drug combination to be tested can also include: obtaining side effect feature information for each drug in the drug combination to be tested, wherein the dimensions of the side effect feature information are determined based on the number of side effect types contained in a pre-constructed side effect information database, and each dimension of the side effect feature information corresponds to one side effect.

[0054] The side effect information database stores information on various potential side effects of a drug, which can be obtained from existing databases such as SIDER 4.1. In SIDER 4.1, drug side effect information is presented as phrases. To simplify the processing, 0 and 1 can be used to represent the presence or absence of the side effect, respectively. Finally, for each drug in the test drug combination, a vector consisting of 0s and 1s is generated. For example, if the side effect information database contains N types of side effects, the side effect feature information will be an N-dimensional vector.

[0055] In addition to drug structure information, cell line gene mutation information, and expression information, introducing drug side effect information can enrich the input features of deep learning models, enabling them to learn more pharmacological information and thus improve the accuracy of collaborative prediction results.

[0056] Gene mutation information of cell lines belongs to cell line genomic data. For example, gene mutation information can be obtained from the GDSC database (Genomics of Drug Sensitivity in Cancer). Each cell line corresponds to a 310-dimensional gene mutation vector, where each dimension corresponds to a gene. A dimension value of "1" represents the presence of a mutation in the corresponding gene, and a dimension value of "0" represents the absence of a mutation. For example, if there are 100 cell lines to be tested, there will be 100 corresponding gene mutation vectors.

[0057] In some examples, mRNA expression data can be obtained from the Affymetrix Human Genome U219 array. The E-MTAB-3610 dataset includes expression mean and standard deviation data. Individual gene expression levels are influenced by temporal and spatial characteristics and interference, making them inherently dynamic; therefore, standard deviation better reflects actual biological significance. That is, each cell line corresponds to one mRNA expression mean sequence and one mRNA expression standard deviation sequence. For each cell line, the expression mean sequence includes M dimensions of expression mean, and the expression standard deviation sequence includes M dimensions of expression standard deviation, with a one-to-one correspondence between the M dimensions of expression mean and M dimensions of expression standard deviation.

[0058] Typically, the dimensionality of expressed information can reach tens of thousands, meaning the M value can exceed ten thousand, far greater than the dimensionality of gene mutation information and drug information. Therefore, to simplify computation and reduce model training and prediction time, feature compression of the expressed information can be performed by executing the following step S102.

[0059] In some examples, there may be significant differences in the expression information of the same dimension among different cell lines. In order to increase the convergence of the model, the expression mean and expression standard deviation of each cell line can be standardized before compressing the expression information. For example, min-max standardization can be used to map the values ​​to the range [0, 1].

[0060] Therefore, in step S101, the process of obtaining the expression mean sequence and the expression standard deviation sequence may include: obtaining the original expression mean sequence and the original expression standard deviation sequence corresponding to each cell line; and standardizing the original expression mean and the original expression standard deviation for each expression dimension of each cell line to obtain the standardized expression mean sequence and the expression standard deviation sequence.

[0061] Taking min-max normalization as an example, for the normalization of expression mean, we can compare the original expression mean values ​​of each cell line in the same expression dimension of the original expression mean sequence to obtain the maximum expression mean x in that expression dimension. max and minimal expression x min Then, according to formula x norm =(xx) min ) / (x max -x min This yields the standardized mean expression value for each cell line within this expression dimension. Where x represents the original mean expression value, x... norm This represents the standardized mean expression value. By standardizing each expression dimension according to the above standardization process, the standardized mean expression value sequence for each cell line can be obtained.

[0062] For standardization of expression standard deviation: For each expression dimension in the original expression standard deviation sequence corresponding to each cell line, the standard deviation can be calculated using the formula σ. norm =(σ / x)*x norm The standardized standard deviation of the expression is obtained. Where σ norm σ represents the standard deviation of the standardized expression, and σ represents the original standard deviation of the expression.

[0063] Step S102: Based on the expression mean sequence and the expression standard deviation sequence, the initial expression feature sequence of each cell line is obtained. The initial expression feature sequence is then input into a pre-trained autoencoder for feature extraction to obtain the dimensionality-reduced expression feature sequence.

[0064] Autoencoders are unsupervised methods used to train a low-dimensional representation of input data, such as... Figure 2As shown, it mainly consists of three parts: the encoder layer, which is composed of a series of layers with progressively decreasing numbers of neurons; the latent vector layer, which represents the minimum dimension that maximizes the preservation of the input's expressive information; and the decoder layer, which is a mirror image of the encoder architecture, except that the number of neurons in each layer increases progressively. For example, the first layer of the encoder layer has 2000 neurons, the second and third layers have 1000 and 500 neurons respectively; the decoder layer is the opposite, with the first, second, and third layers having 500, 1000, and 2000 neurons respectively.

[0065] In this embodiment of the application, when using an autoencoder to extract expression features of mRNA, the input of the autoencoder is optimized. In addition to the expression mean data, the input data also incorporates the expression standard deviation data, which helps to enrich the extracted expression information and thus improve the accuracy of feature extraction.

[0066] In some examples, the process of obtaining the initial expression feature sequence of each cell line based on the expression mean sequence and the expression standard deviation sequence may include: for each expression dimension of each cell line, based on a preset random variable and the expression standard deviation of that expression dimension, obtaining the perturbation amount of that expression dimension, and obtaining the initial expression feature of that expression dimension based on the expression mean and the perturbation amount; then, the initial expression feature sequence of each cell line can be constructed based on the initial expression features of each expression dimension of each cell line.

[0067] This allows for the addition of small perturbations to the input expression mean data, and cleverly correlates the perturbation amount with the expression standard deviation. This not only improves the generalization ability and robustness of the autoencoder, but also incorporates the expression standard deviation into the input expression information on top of the expression mean, effectively enriching the expression information for feature extraction and improving the accuracy of expression feature extraction. This, in turn, helps to improve the accuracy of collaborative prediction results.

[0068] For example, for each expression dimension of each cell line, the following formula can be used:

[0069]

[0070] Obtain initial expression features Where b is a preset random variable, b*σ norm This represents the disturbance quantity.

[0071] In some examples, the random variable b can be a discrete random variable, taking values ​​of 0, 1, and -1. The probability of b being 0 is p, the probability of b being -1 is (1-p) / 2, and the probability of b being 1 is (1-p) / 2, where p is a number greater than 0 and less than 1. The specific value can be configured according to actual needs and multiple experiments. Since the expression data of a single cell line can be in the tens of thousands, this method can guarantee the expected ± expression standard deviation.

[0072] In some examples, considering the special nature of standard deviation data—that is, reflecting the dispersion of the data and not simply being 1 or -1—the random variable b can be defined as a random function with a mean of 0 and a standard deviation of 1. In this case, the random variable b is a continuous random variable, taking values ​​from -1 to 1. This makes the changes in the perturbation more continuous. More perturbation improves the robustness of the autoencoder, thereby enhancing the reliability and accuracy of feature extraction.

[0073] During the training of the autoencoder, the expression mean data and expression standard deviation data of the training samples can be collected in advance, and preprocessed according to the above process to obtain the initial expression feature sequence of the training samples. The autoencoder can then be trained using the training samples.

[0074] Specifically, each initial expression feature in the initial expression feature sequence In the input autoencoder, after passing through the encoding layer and reaching the intermediate layer, the signal becomes y, which can be represented by the following formula: Here, s is a nonlinear function, such as the sigmoid function, W is the link weight from the input layer to the intermediate layer, and b is the bias of the intermediate layer. y is decoded by the decoding layer and output to the output layer, where the signal becomes z, as shown in the following formula: z = s(W'y + b'). The autoencoder is trained using training samples on W, W', b, and b' until the preset loss function value meets the convergence condition.

[0075] The preset loss function value can be determined based on the difference between the decoding result of the decoding layer and the mean expression sequence corresponding to the initial expression feature sequence of the input encoding layer. For example, the loss function of an autoencoder can be:

[0076] L(x0, z) = ||x0 - z|| 2

[0077] Where x0 is the expression mean without perturbation, i.e., the x mentioned above. norm .

[0078] In practice, the initial expression feature sequence corresponding to the cell line to be tested is input into the trained autoencoder, which then outputs a dimensionality-reduced expression feature sequence from the intermediate layer. This expression feature sequence represents the lowest dimension that maximizes the preservation of information from the initial expression feature sequence, meaning that while compressing high-dimensional expression information, it extracts as accurate feature information as possible.

[0079] Step S103: Based on drug information, gene mutation information, dimensionality-reduced expression feature sequences, and pre-trained deep learning models, determine whether the drug combination to be tested has synergistic effects on the cell lines to be tested.

[0080] Before executing step S103, the deep learning model needs to be trained. The training process may include: constructing an initial deep learning model, training the initial deep learning model using pre-collected sample data, and obtaining a trained deep learning model. The input to the deep learning model is: drug information for each drug in the drug combination to be tested, gene mutation information of the cell line to be tested, and expression characteristic sequences of mRNA in the cell line to be tested. The output is: synergy prediction results, used to characterize whether the drug combination to be tested has synergy with the cell line to be tested.

[0081] Specifically, the sample data includes the input feature data and classification labels. Classification labels include two categories: those indicating synergy and those indicating no synergy. For example, taking the synergy prediction of anti-tumor drug combinations as an example, existing anti-tumor drug combinations can be collected as training samples. Following steps S101 and S102 above, relevant data for drug combination synergy prediction is obtained from these training samples, resulting in the input feature data of the samples. The classification label value is then determined based on the synergy data of these drug combinations. For example, drug synergy is usually measured using reference models, including the Loewe model. According to the definition of the Loewe score, a true score greater than a threshold t is considered a strong synergistic effect; less than a threshold -t is considered a strong antagonistic effect; and a score between the two is called an additive effect. On the DrugComb dataset, the threshold t is usually set to 10. Thus, the synergy classification label value of the drug combinations used as training samples can be obtained, such as 1 indicating synergy and 0 indicating no synergy.

[0082] There are several ways to construct an initial deep learning model. For example, a fully connected network can be used, or other suitable deep learning models can be used; this embodiment does not limit this. Taking a fully connected network as an example, the number of layers and the number of neurons in each layer can be set according to actual needs and multiple experiments. It is necessary to ensure that the number of neurons decreases layer by layer, and the last layer, i.e., the output layer, has 1 neuron.

[0083] For example, in one application scenario, a fully connected network can be divided into four layers, with 500, 200, 100, and 1 neurons respectively. Cross-entropy loss is used as the loss function for training, and the calculation method is as follows:

[0084]

[0085] Where L is the loss function value, n is the number of samples, y′ is the predicted value, and y is the true value, i.e., the classification label value (0 or 1) corresponding to the sample. For example, the relevant training parameters can be configured as follows: the optimizer uses stochastic gradient descent (SGD), batch size = 32, dropout = 0.001, and epochs = 100. The specific training process can be referred to relevant technologies, which will not be detailed in this embodiment.

[0086] After training the deep learning model, it can be used to perform synergistic predictions on the drug combination under test. Specifically, the drug information of each drug in the drug combination, the gene mutation information of the cell line to be tested, and the expression feature sequence of mRNA in the cell line to be tested can be concatenated to obtain the input feature vector.

[0087] For example, for each cell line, the dimension of the expression feature sequence after dimensionality reduction by the autoencoder is 1*500, and the dimension of gene mutation information is 1*310; for each drug in the drug combination to be tested, the dimension of structural feature information is 6*structural length, and the dimension of side effect feature information is 1*N. Considering that the dimensions of the above features are relatively similar, a horizontal concatenation method can be used to process these features. That is, as... Figure 3 As shown, the structural feature information of each drug, the side effect feature information of each drug, the gene mutation information of each cell line, and the expression feature sequence of mRNA in each cell line are horizontally spliced ​​to obtain the above input feature vector.

[0088] Furthermore, by inputting the input feature vector into the deep learning model, the synergy prediction results can be output end-to-end, thereby determining whether the drug combination to be tested is synergistic with the cell line to be tested.

[0089] The drug combination synergy determination method provided in this application can be applied before the experimental stage. By adopting the technical solution of this embodiment, it is possible to detect whether any drug combination has a synergistic effect on certain cell lines. If synergy is found, it can be further verified experimentally. This avoids directly conducting experiments on all combined drugs, effectively reducing experimental losses and improving experimental efficiency. Furthermore, in the process of compressing mRNA expression information, expression standard deviation data is incorporated, which helps to enrich the extracted expression information, improve the accuracy of expression feature extraction, and thus improve the accuracy of synergy prediction results, further reducing experimental losses, saving experimental costs and experimental cycle time.

[0090] Based on the same inventive concept, embodiments of this application also provide a device for determining the synergy of drug combinations, such as... Figure 4 As shown, the drug combination synergy determination device 40 includes:

[0091] The acquisition module 401 is used to acquire drug information for each drug in the drug combination to be tested and cell line information to be tested. The cell line information includes mRNA expression information and gene mutation information. The expression information includes the mean expression sequence and the standard deviation expression sequence.

[0092] The expression feature extraction module 402 is used to obtain an initial expression feature sequence for each cell line based on the expression mean sequence and the expression standard deviation sequence, and input the initial expression feature sequence into a pre-trained autoencoder for feature extraction to obtain a dimension-reduced expression feature sequence.

[0093] The determination module 403 is used to determine whether the drug combination to be tested has synergy with the cell line to be tested, based on the drug information, the gene mutation information, the dimensionality-reduced expression feature sequence, and the pre-trained deep learning model.

[0094] In some examples, the above-mentioned expression feature extraction module 402 can be specifically used for:

[0095] For each expression dimension of each cell line, based on a preset random variable and the expression standard deviation of the expression dimension, the perturbation amount of the expression dimension is obtained, and based on the expression mean of the expression dimension and the perturbation amount, the initial expression characteristics of the expression dimension are obtained.

[0096] Based on the initial expression features of each expression dimension of each cell line, an initial expression feature sequence for each cell line is constructed.

[0097] In some examples, the above random variable is a discrete random variable, and the values ​​of the random variable include: 0, 1, -1, where the probability of the random variable being 0 is p, the probability of the random variable being -1 is (1-p) / 2, the probability of the random variable being 1 is (1-p) / 2, and p is a number greater than 0 and less than 1.

[0098] In some examples, the random variable is a continuous random variable with a value from -1 to 1, and the random variable's value satisfies a distribution with a mean of 0 and a standard deviation of 1.

[0099] In some examples, the autoencoder described above is trained based on a preset loss function. The autoencoder includes an encoding layer, an intermediate layer, and a decoding layer. The intermediate layer outputs a dimensionality-reduced representation feature sequence, and the value of the preset loss function is determined based on the difference between the decoding result of the decoding layer and the mean representation sequence corresponding to the initial representation feature sequence input to the encoding layer.

[0100] In some examples, the aforementioned acquisition module 401 can specifically be used for:

[0101] Obtain the original mean expression sequence and the original standard deviation expression sequence for each cell line;

[0102] For each expression dimension of each cell line, the original expression mean and original expression standard deviation were standardized to obtain the standardized expression mean sequence and expression standard deviation sequence.

[0103] In some examples, drug information includes: structural characteristic information of the drug. The aforementioned acquisition module 401 can specifically be used for:

[0104] Obtain the SMILES structural information of each drug in the drug combination to be tested;

[0105] The SMILES structural information is vectorized to obtain the structural feature information corresponding to each drug.

[0106] In some examples, the drug information also includes: drug side effect characteristic information, and the acquisition module 401 mentioned above can specifically be used for:

[0107] Obtain the side effect characteristic information of each drug in the drug combination to be tested. The dimensions of the side effect characteristic information are determined based on the number of side effect types contained in the pre-constructed side effect information database. Each dimension of the side effect characteristic information corresponds to one side effect.

[0108] In some examples, the aforementioned determining module 403 is specifically used for:

[0109] The structural feature information and side effect feature information of each drug, as well as the gene mutation information and dimensionality-reduced expression feature sequence of each cell line, are horizontally concatenated to obtain the input feature vector;

[0110] The input feature vector is input into a pre-trained deep learning model to obtain a synergy prediction result, which is used to characterize whether the drug combination to be tested has synergy with the cell line to be tested.

[0111] It should be noted that the specific manner in which each module of the drug combination synergy determination device 40 provided in this application performs its operation has been described in detail in the method embodiment provided in the first aspect above. The specific implementation process can be referred to the method embodiment provided in the first aspect above, and will not be described in detail here.

[0112] Based on the same inventive concept, such as Figure 5 As shown in the figure, this application embodiment also provides an electronic device, including one or more processors 502, a memory 504, and one or more application programs (not shown in the figure) stored on the memory 504 and capable of running on the processor 502. These one or more programs are configured to implement the various processes of the drug combination synergy determination method embodiment provided in the first aspect above, and can achieve the same technical effects; therefore, to avoid repetition, they will not be described again here.

[0113] It should be noted that the electronic device can be any device with data processing capabilities, and this embodiment does not limit it. For example, the electronic device can be a user terminal such as a personal computer, mobile phone, tablet computer, laptop computer, or smart wearable device, or it can be a server.

[0114] exist Figure 5 In this document, a bus architecture (represented by bus 500) is used. Bus 500 may include any number of interconnected buses and bridges, linking various circuits including one or more processors represented by processor 502 and memory represented by memory 504. Bus 500 may also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. Bus interface 505 provides an interface between bus 500 and receiver 501 and transmitter 503. Receiver 501 and transmitter 503 may be the same element, i.e., a transceiver, providing a unit for communicating with various other devices over a transmission medium. Processor 502 is responsible for managing bus 500 and general processing, while memory 504 can be used to store data used by processor 502 during operation.

[0115] Understandable, Figure 5The structure shown is for illustrative purposes only. The electronic device provided in this application embodiment may also include components that are more advanced than those shown in the diagram. Figure 5 The more or fewer components shown, or having the same Figure 5 The different configurations shown. Figure 5 The components shown can be implemented using hardware, software, or a combination thereof.

[0116] Based on the same inventive concept, this application also provides a computer-readable storage medium storing computer instructions. When these instructions are executed on a computer, the computer performs the various processes of the drug combination synergy determination method embodiment provided in the first aspect, achieving the same technical effect. To avoid repetition, further details are omitted here. For example, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0117] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0118] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus, and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0119] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0120] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0121] In this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, without necessarily requiring or implying any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0122] Although preferred embodiments have been described in this specification, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this specification.

Claims

1. A method for determining the synergistic effect of drug combinations, characterized in that, The method comprises the following steps: obtain drug information of each drug in the drug combination to be tested and cell line information to be tested, wherein the cell line information comprises mRNA expression information and gene mutation information, and the expression information comprises expression mean value sequence and expression standard deviation sequence of each cell line; based on the expression mean value sequence and the expression standard deviation sequence, obtain an initial expression feature sequence of each cell line, and input the initial expression feature sequence into a pre-trained autoencoder for feature extraction to obtain a reduced expression feature sequence; based on the drug information, the gene mutation information, the reduced expression feature sequence and a pre-trained deep learning model, determine whether the drug combination to be tested has synergy on the cell line to be tested; wherein, based on the expression mean value sequence and the expression standard deviation sequence, the initial expression feature sequence of each cell line is obtained, which comprises: for each expression dimension of each cell line, based on a pre-set random variable and the expression standard deviation of the expression dimension, a perturbation amount of the expression dimension is obtained, and based on the expression mean value of the expression dimension and the perturbation amount, an initial expression feature of the expression dimension is obtained; based on the initial expression feature of each expression dimension of each cell line, the initial expression feature sequence of each cell line is constituted.

2. The method of claim 1, wherein, The random variable is a discrete random variable, and the value of the random variable includes 0, 1 and -1, wherein the probability of the random variable taking 0 is p, the probability of the random variable taking -1 is (1-p) / 2, and the probability of the random variable taking 1 is (1-p) / 2, and p is a number greater than 0 and less than 1.

3. The method of claim 1, wherein, The random variable is a continuous random variable, and the value of the random variable is -1 to 1, and the value of the random variable satisfies the distribution with a mean value of 0 and a standard deviation of 1.

4. The method of claim 1, wherein, The autoencoder is trained based on a pre-set loss function, and the autoencoder comprises an encoding layer, an intermediate layer and a decoding layer; wherein, the intermediate layer is used to output the reduced expression feature sequence, and the value of the pre-set loss function is determined based on the difference between the decoding result of the decoding layer and the expression mean value sequence corresponding to the initial expression feature sequence input into the encoding layer.

5. The method of claim 1, wherein, Obtaining the expression mean value sequence and the expression standard deviation sequence comprises: obtaining the original expression mean value sequence and the original expression standard deviation sequence corresponding to each cell line; for each expression dimension of each cell line, the original expression mean value and the original expression standard deviation are standardized respectively to obtain the standardized expression mean value sequence and the expression standard deviation sequence.

6. The method of claim 1, wherein, The drug information comprises structural feature information of the drug, and the drug information of each drug in the drug combination to be tested comprises: obtaining SMILES structure information of each drug in the drug combination to be tested; vectorizing the SMILES structure information to obtain the structural feature information corresponding to each drug.

7. The method of claim 6, wherein, The drug information also comprises side effect feature information of the drug, and the drug information of each drug in the drug combination to be tested further comprises: obtain side effect feature information of each drug in the drug combination to be tested, a dimension of the side effect feature information being determined based on a number of side effect types contained in a pre-constructed side effect information library, each dimension in the side effect feature information corresponding to a side effect.

8. The method of claim 7, wherein, Based on the drug information, the gene mutation information, the reduced expression feature sequence, and a pre-trained deep learning model, it is determined whether the drug combination to be tested has synergy on the cell line to be tested, including: performing horizontal splicing processing on the structure feature information and the side effect feature information of each drug, and the gene mutation information and the reduced expression feature sequence of each cell line to obtain an input feature vector; inputting the input feature vector into the pre-trained deep learning model to obtain a synergy prediction result, the synergy prediction result being used to represent whether the drug combination to be tested has synergy on the cell line to be tested.

9. A drug combination synergism determining apparatus, characterized by comprising: a drug combination synergism determining program for determining synergism of a drug combination; and a computer for executing the drug combination synergism determining program. including: an obtaining module, configured to obtain drug information of each drug in a drug combination to be tested and cell line information to be tested, the cell line information including mRNA expression information and gene mutation information, the expression information including an expression mean sequence and an expression standard deviation sequence; an expression feature extraction module, configured to obtain an initial expression feature sequence of each cell line based on the expression mean sequence and the expression standard deviation sequence, and input the initial expression feature sequence into a pre-trained autoencoder for feature extraction to obtain a reduced expression feature sequence; a determination module, configured to determine whether the drug combination to be tested has synergy on the cell line to be tested based on the drug information, the gene mutation information, the reduced expression feature sequence, and a pre-trained deep learning model; wherein the initial expression feature sequence of each cell line based on the expression mean sequence and the expression standard deviation sequence includes: for each expression dimension of each cell line, obtaining a perturbation amount of the expression dimension based on a pre-set random variable and an expression standard deviation of the expression dimension, and obtaining an initial expression feature of the expression dimension based on an expression mean value of the expression dimension and the perturbation amount; and constructing the initial expression feature sequence of each cell line based on the initial expression feature of each expression dimension of each cell line.

10. An electronic device, comprising: The electronic device includes: one or more processors; a memory; one or more application programs, wherein the one or more application programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs are configured to perform the drug combination synergy determination method of any one of claims 1 to 8.

11. A computer readable storage medium having stored thereon a computer program, characterized in that The computer program, when executed by a processor, implements the drug combination synergy determination method of any one of claims 1 to 8.

Images