Method, apparatus, device, and medium for predicting cell gene expression after drug perturbation

Abstract

The application relates to the technical field of artificial intelligence and medical health, and discloses a method and device for predicting cell gene expression after drug disturbance, equipment and a medium, wherein the method comprises the following steps: inputting initial cell gene expression data into a first encoder to extract a first hidden vector; inputting cell line gene expression data into a second encoder to extract a second hidden vector; inputting compound expression data into a third encoder to extract a third hidden vector; and using a target decoder to decode the first hidden vector, the second hidden vector and the third hidden vector to obtain target cell gene expression data of a target cell after drug disturbance of a target drug. The decoding of cell gene expression is realized based on the potential relationship among cell lines, drugs and genes, and the accuracy of cell gene expression prediction after drug disturbance is improved.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and medical health technology, and in particular to a method, apparatus, device and medium for predicting cell gene expression after drug perturbation. Background Technology

[0002] Developing new drugs for specific diseases is extremely time-consuming and costly, with only about 5% of new drugs ultimately gaining market approval. Accurately predicting gene expression in cells after drug perturbation can effectively shorten the drug development cycle and reduce costs. With the development of machine learning technology, using machine learning methods to predict gene expression in cells after drug perturbation has effectively assisted in new drug development, shortening the development cycle, reducing the failure rate, and lowering costs. However, most current machine learning methods randomly initialize drug and gene representations, leading to low accuracy in predicting gene expression in cells after drug perturbation. Summary of the Invention

[0003] Based on this, it is necessary to address the technical problem that existing machine learning methods randomly initialize the representations of drugs and genes, resulting in low accuracy in predicting cell gene expression after drug perturbation. Therefore, a method, device, equipment, and medium for predicting cell gene expression after drug perturbation are proposed.

[0004] In a first aspect, a method for predicting cellular gene expression after drug perturbation is provided, the method comprising:

[0005] Obtain a pre-trained target model, wherein the target model includes a first encoder, a second encoder, a third encoder, and a target decoder;

[0006] The initial cell gene expression data of the target cells without drug perturbation by the target drug is obtained, and the initial cell gene expression data is input into the first encoder for latent vector extraction to obtain the first latent vector;

[0007] Obtain the cell line gene expression data corresponding to the cell line to which the target cell belongs, and input the cell line gene expression data into the second encoder for latent vector extraction to obtain the second latent vector;

[0008] Obtain the compound expression data corresponding to the target drug, and input the compound expression data into the third encoder for latent vector extraction to obtain the third latent vector;

[0009] Using the target decoder, the first hidden vector, the second hidden vector, and the third hidden vector are used to decode the cellular gene expression of the target cell after the drug perturbation of the target drug, so as to obtain the target cell gene expression data.

[0010] Secondly, a device for predicting cellular gene expression after drug perturbation is provided, the device comprising:

[0011] The data acquisition module is used to acquire a pre-trained target model, wherein the target model includes a first encoder, a second encoder, a third encoder, and a target decoder;

[0012] The first latent vector determination module is used to acquire the initial cell gene expression data of the target cells under drug perturbation without the target drug, and input the initial cell gene expression data into the first encoder for latent vector extraction to obtain the first latent vector;

[0013] The second latent vector determination module is used to obtain the cell line gene expression data corresponding to the cell line to which the target cell belongs, and input the cell line gene expression data into the second encoder for latent vector extraction to obtain the second latent vector.

[0014] The third latent vector determination module is used to acquire the compound expression data corresponding to the target drug, and input the compound expression data into the third encoder for latent vector extraction to obtain the third latent vector.

[0015] The prediction module is used to use the target decoder to decode the first hidden vector, the second hidden vector, and the third hidden vector to obtain the target cell gene expression data after the drug perturbation of the target drug.

[0016] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for predicting cell gene expression after drug perturbation.

[0017] Fourthly, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described method for predicting cell gene expression after drug perturbation.

[0018] The method for predicting cell gene expression after drug perturbation in this application involves inputting the initial cell gene expression data into a first encoder for latent vector extraction to obtain a first latent vector, inputting the cell line gene expression data into a second encoder for latent vector extraction to obtain a second latent vector, and inputting the compound expression data into a third encoder for latent vector extraction to obtain a third latent vector. A target decoder is then used to decode the cell gene expression of the target cell after drug perturbation by the target drug using the first, second, and third latent vectors, thus obtaining the target cell gene expression data. By decoding the cell gene expression of the target cell after drug perturbation by the target drug using the first, second, and third latent vectors, decoding cell gene expression based on the potential relationship between cell lines, drugs, and genes is achieved, improving the accuracy of predicting cell gene expression after drug perturbation. Furthermore, the accuracy of predicting cell gene expression after drug perturbation is further improved by using the second latent vector extracted from the cell line gene expression data input into the second encoder. Attached Figure Description

[0019] 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.

[0020] in:

[0021] Figure 1 This is a diagram illustrating the application environment of a method for predicting cellular gene expression after drug perturbation in one embodiment.

[0022] Figure 2 This is a flowchart of a method for predicting cellular gene expression after drug perturbation in one embodiment;

[0023] Figure 3 This is a structural block diagram of a device for predicting cellular gene expression after drug perturbation in one embodiment;

[0024] Figure 4 This is a structural block diagram of a computer device in one embodiment;

[0025] Figure 5 This is another structural block diagram of a computer device in one embodiment. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] The method for predicting cellular gene expression after drug perturbation provided in this invention can be applied to, for example... Figure 1 In the application environment, client 110 communicates with server 120 via the network. Server 120 can obtain a pre-trained target model through client 110, wherein the target model includes a first encoder, a second encoder, a third encoder, and a target decoder; Server 120 can obtain initial cell gene expression data of target cells without drug perturbation by the target drug through client 110, and then Server 120 inputs the initial cell gene expression data into the first encoder for latent vector extraction to obtain a first latent vector; Server 120 can obtain cell line gene expression data corresponding to the cell line to which the target cell belongs through client 110, and then Server 120 inputs the cell line gene expression data into the second encoder for latent vector extraction to obtain a second latent vector; Server 120 can obtain compound expression data corresponding to the target drug through client 110, and then Server 120 inputs the compound expression data into the third encoder for latent vector extraction to obtain a third latent vector; Server 120 uses the target decoder to decode the first latent vector, the second latent vector, and the third latent vector to obtain the target cell gene expression data. By decoding the first, second, and third hidden vectors to obtain the cellular gene expression of the target cells after drug perturbation by the target drug, decoding of cellular gene expression based on the potential relationship between cell lines, drugs, and genes is achieved, improving the accuracy of cellular gene expression prediction after drug perturbation. Furthermore, by inputting the second hidden vector obtained from the cell line gene expression data into the second encoder, the accuracy of cellular gene expression prediction after drug perturbation is further improved.

[0028] The client 110 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server 120 can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will now be described in detail through specific embodiments.

[0029] Please see Figure 2 As shown, Figure 2A flowchart illustrating a method for predicting cellular gene expression after drug perturbation provided in an embodiment of the present invention includes the following steps:

[0030] S1: Obtain a pre-trained target model, wherein the target model includes a first encoder, a second encoder, a third encoder, and a target decoder;

[0031] The pre-trained target model is a pre-trained predictive model for cell gene expression after drug perturbation.

[0032] The first encoder, second encoder, and third encoder are all encoders. The target decoder uses a decoder.

[0033] The outputs of the first encoder, the second encoder, and the third encoder are all used as inputs to the target decoder.

[0034] Specifically, it can obtain a pre-trained target model from user input, a pre-trained target model from storage space, a pre-trained target model from the client, or a pre-trained target model from a third-party application.

[0035] S2: Obtain the initial cell gene expression data of the target cells under drug perturbation without the target drug, and input the initial cell gene expression data into the first encoder for latent vector extraction to obtain the first latent vector;

[0036] The target cell is the cell whose gene expression is to be disturbed by the drug.

[0037] Initial cellular gene expression data represents the gene expression level of target cells without drug perturbation by the target drug. This initial cellular gene expression data is a multidimensional tensor at the gene level, where the dimension of the tensor represents the number of genes of interest. Gene expression level encompasses both transcriptional and translational levels, including both mRNA transcription and protein expression. mRNA generally refers to messenger ribonucleic acid (MRNA). MRNA is a single-stranded ribonucleic acid transcribed from one strand of DNA as a template, carrying genetic information and guiding protein synthesis.

[0038] A target drug is a drug that is intended to cause drug disturbances in target cells.

[0039] Specifically, it can obtain the initial cell gene expression data of the target cells without drug perturbation by the target drug, input by the user; it can also obtain the initial cell gene expression data of the target cells without drug perturbation by the target drug from the storage space; it can also obtain the initial cell gene expression data of the target cells without drug perturbation by the target drug from the client; and it can also obtain the initial cell gene expression data of the target cells without drug perturbation by the target drug from third-party applications.

[0040] In this process, the initial cell gene expression data is input into the first encoder for latent vector extraction. The first encoder encodes the initial cell gene expression data into a denser latent vector with high-dimensional features, and this latent vector is used as the first latent vector. The dimension of the first latent vector is much lower than that of the initial cell gene expression data. This can accelerate the training speed of the target model and enable the target model to extract more effective and higher-dimensional gene information from the gene expression dimension, reducing redundant information input decoding steps.

[0041] S3: Obtain the cell line gene expression data corresponding to the cell line to which the target cell belongs, and input the cell line gene expression data into the second encoder for latent vector extraction to obtain the second latent vector;

[0042] Cell line gene expression data is the comprehensive expression data of all cell types in the cell line to which the target cell belongs.

[0043] Specifically, the system can obtain the cell line gene expression data corresponding to the cell line to which the target cell belongs, either by obtaining the data from the storage space, the client, or a third-party application.

[0044] In this process, the cell line gene expression data is input into the second encoder for latent vector extraction. The second encoder encodes the cell line gene expression data into a denser latent vector, which is then used as the second latent vector. The dimension of the second latent vector is much lower than that of the cell line gene expression data, achieving dimensionality reduction and reducing redundant information, thereby obtaining higher-dimensional gene expression data.

[0045] S4: Obtain the compound expression data corresponding to the target drug, and input the compound expression data into the third encoder for latent vector extraction to obtain the third latent vector;

[0046] Compound expression data is the expression of the compound corresponding to the target drug in vector space.

[0047] Specifically, it can obtain compound expression data corresponding to the target drug input by the user, or obtain compound expression data corresponding to the target drug from storage space, or obtain compound expression data corresponding to the target drug from the client, or obtain compound expression data corresponding to the target drug from third-party applications.

[0048] Specifically, the compound expression data is input into the third encoder for latent vector extraction. The third encoder encodes the compound expression data into a denser latent vector, which is then used as the third latent vector.

[0049] S5: Using the target decoder, decode the first hidden vector, the second hidden vector, and the third hidden vector to obtain the target cell gene expression data after the drug perturbation of the target drug.

[0050] Specifically, the first hidden vector, the second hidden vector, and the third hidden vector are fused together, and the fused data is input into the target decoder to decode the cell gene expression of the target cell after the drug perturbation of the target drug, and the decoded data is used as the target cell gene expression data.

[0051] This embodiment decodes the cellular gene expression of the target cells after drug perturbation by the target drug using the first hidden vector, the second hidden vector, and the third hidden vector. This enables decoding of cellular gene expression based on the potential relationship between cell lines, drugs, and genes, improving the accuracy of cellular gene expression prediction after drug perturbation. Furthermore, by inputting the second hidden vector obtained from the cell line gene expression data into the second encoder, the accuracy of cellular gene expression prediction after drug perturbation is further improved.

[0052] In one embodiment, prior to the step of obtaining the pre-trained target model, the method further includes:

[0053] S11: Obtain a training sample set and an initial model, wherein the initial model includes: a first initial encoder, a second initial encoder, a third initial encoder, a first classifier, a second classifier, and an initial decoder. The output of the first initial encoder is used as the input of the first classifier and the second classifier. The first classifier is used to perform cell line classification prediction and drug classification prediction.

[0054] The first, second, and third initial encoders are all encoders. The outputs of the first, second, and third initial encoders are all inputs to the initial decoder. The initial decoder is a decoder. The first classifier is a multilayer perceptron or multiple linear layers. The second classifier is a multilayer perceptron or multiple linear layers.

[0055] The training sample set contains multiple training samples. These training samples include: initial cell gene expression sample data, cell line gene expression sample data, compound expression sample data, cell line calibration values, drug calibration values, and cell gene expression calibration data. The initial cell gene expression sample data, cell line gene expression sample data, and compound expression sample data all correspond to the same cell, which is used as the cell to be analyzed. The cell line calibration value is the accurate label of the cell line to which the cell to be analyzed belongs. The drug calibration value is the accurate label of the drug added to the cell to be analyzed for drug perturbation. The cell gene expression calibration data represents the cell gene expression of the cell to be analyzed after drug perturbation with the drug corresponding to the compound expression sample data. The initial cell gene expression sample data represents the gene expression level of the cell to be analyzed without drug perturbation with the drug corresponding to the compound expression sample data. The cell line gene expression sample data is the comprehensive expression data of all cell types within the cell line to which the cell to be analyzed belongs. The compound expression sample data represents the expression of the compound corresponding to the drug used for drug perturbation in the vector space.

[0056] Specifically, it can obtain the training sample set and initial model input by the user, or obtain the training sample set and initial model from the storage space, or obtain the training sample set and initial model from the client, or obtain the training sample set and initial model from a third-party application.

[0057] S12: Using the training sample set, perform adversarial training on the initial model to improve cellular gene expression after drug perturbation;

[0058] Specifically, the training sample set is used to perform adversarial training on the initial model to assess cell gene expression after drug perturbation. Adversarial loss is introduced through the first classifier and the second classifier to ensure that the data output by the first initial encoder does not contain any specific cell line or specific drug (other than the drug to be perturbed), thereby improving the accuracy of the initial model after training.

[0059] S13: The first initial encoder, the second initial encoder, the third initial encoder, and the initial decoder of the initial model after training are used as the pre-trained target model.

[0060] This embodiment introduces adversarial loss through a first classifier and a second classifier to learn the latent vector information of cell gene expression. This reduces gene information in gene expression that is specific to a certain cell line or a certain drug, thereby reducing the coupling between genes, cell lines, and drugs. This ensures that the data output by the first initial encoder contains only the effective information of this measurement as much as possible, and avoids including any specific cell line or specific drug, thus improving the accuracy of the initial model after training. By introducing cell line gene expression sample data corresponding to the cell line to which the cell belongs, the accuracy of model prediction is improved, and the generalization ability of the model in application scenarios is enhanced, so that the model can also predict new cell lines during the prediction process.

[0061] In one embodiment, the step of using the training sample set to perform adversarial training on the initial model for drug-perturbed cellular gene expression includes:

[0062] S121: Using a preset batch sample size, the training sample set is divided to obtain multiple training sample subsets;

[0063] Specifically, the training sample set is divided using a preset batch size, and each resulting subset is considered a training sample subset. The number of training samples contained in each training sample subset is equal to the preset batch size.

[0064] Optionally, the preset batch sample size is 128. It is understood that the preset batch sample size can also be set to other values, which are not limited here.

[0065] S122: Get the counter and initialize the counter value to 1;

[0066] S123: Obtain the training sample subsets with the same order as the counter from each of the training sample subsets, and use them as the training subsets;

[0067] Specifically, a training sample subset with the same order as the counter is obtained from each of the training sample subsets, and the obtained training sample subset is used as the training subset.

[0068] S124: If the value of the counter is not even, then based on the batch training method, the initial model is subjected to adversarial training on cell gene expression after drug perturbation using the subset to be trained. During the adversarial training, the network parameters of the first initial encoder, the first classifier, the second classifier and the initial decoder are updated.

[0069] Specifically, if the value of the counter is not even, then the value of the counter is odd. Based on the batch training method, the initial model is subjected to adversarial training on cell gene expression after drug perturbation using the subset to be trained. During the adversarial training, the network parameters of the first initial encoder, the first classifier, the second classifier, and the initial decoder are updated. It can be understood that although the subset to be trained contains multiple training samples, the adversarial training using the subset to be trained only updates the network parameters once.

[0070] If the value of the counter is not even, the network parameters of the first initial encoder, the first classifier, the second classifier, and the initial decoder are updated, but the network parameters of the second initial encoder and the third initial encoder are not updated.

[0071] In this step, the step of using the batch training method to perform adversarial training on the initial model for drug-perturbed cell gene expression using the training subset specifically includes: inputting the initial cell gene expression sample data, cell line gene expression sample data, and compound expression sample data of each training sample in the training subset into the initial model; calculating the single-sample loss value based on the data output by the initial model and the cell line calibration value, drug calibration value, and cell gene expression calibration data corresponding to the training sample; and updating the network parameters of the first initial encoder, the first classifier, the second classifier, and the initial decoder based on all the single-sample loss values ​​corresponding to the training subset.

[0072] S125: If the value of the counter is even, then based on the batch training method, the initial model is subjected to adversarial training on cell gene expression after drug perturbation using the subset to be trained. During the adversarial training, the network parameters of the second initial encoder and the third initial encoder are updated.

[0073] Specifically, when the counter value is even, the network parameters of the second and third initial encoders are updated, but the network parameters of the first initial encoder, the first classifier, the second classifier, and the initial decoder are not updated. It is understood that although the training subset contains multiple training samples, the adversarial training using this training subset only updates the network parameters once.

[0074] In this step, the step of using the batch training method to perform adversarial training on the initial model for drug-perturbed cell gene expression using the subset to be trained specifically includes: inputting the initial cell gene expression sample data, cell line gene expression sample data, and compound expression sample data of each training sample in the subset to be trained into the initial model; calculating the single-sample loss value based on the data output by the initial model and the cell line calibration value, drug calibration value, and cell gene expression calibration data corresponding to the training sample; and updating the network parameters of the second initial encoder and the third initial encoder based on all the single-sample loss values ​​corresponding to the subset to be trained.

[0075] S126: Increment the value of the counter by 1, and jump to the step of re-executing the step of obtaining the training sample subset with the same order as the counter from each of the training sample subsets as the training subset, until the preset training end condition is reached.

[0076] The preset training termination condition is that the value of the counter is greater than a first preset value or the average loss value of the initial model converges to a second preset value.

[0077] The average loss value of the initial model is the average of the loss values ​​of all single samples corresponding to the subset to be trained.

[0078] This embodiment further improves the accuracy of model predictions by considering the coupling between drug, gene, and cell line information and updating the network parameters of different units in batches.

[0079] In one embodiment, the step of obtaining initial cellular gene expression data of target cells under drug perturbation conditions without the target drug includes:

[0080] S21: Obtain the cell gene expression data of the target cells under any of the following conditions: no target drug was applied in the experiment, or a drug other than the target drug was applied in the experiment, as the initial cell gene expression data of the target cells under the drug perturbation condition without the target drug.

[0081] Specifically, the system can acquire cell gene expression data of the target cells under any of the following conditions: no target drug applied in the experiment, or drug other than the target drug applied in the experiment, as input by the user; it can also acquire cell gene expression data of the target cells under any of the following conditions: no target drug applied in the experiment, or drug other than the target drug applied in the experiment, as input by the user; it can also acquire cell gene expression data of the target cells under any of the following conditions: no target drug applied in the experiment, or drug other than the target drug applied in the experiment, as input by the user; or it can acquire cell gene expression data of the target cells under any of the following conditions: no target drug applied in the experiment, or drug other than the target drug applied in the experiment, as input by the user; and it can also acquire cell gene expression data of the target cells under any of the following conditions: no target drug applied in the experiment, or drug other than the target drug applied in the experiment, as input by the user; wherein, the acquired cell gene expression data is used as the initial cell gene expression data of the target cells under the condition of not being subjected to drug perturbation by the target drug.

[0082] In one embodiment, before the step of obtaining the cell line gene expression data corresponding to the cell line to which the target cell belongs, the method further includes:

[0083] S31: Obtain the cell gene expression data of each cell in the cell line to which the target cell belongs under natural conditions, and use it as standard cell gene expression data;

[0084] Specifically, the system can obtain the cell gene expression data of each cell in the cell line to which the target cell belongs under natural conditions, input by the user; it can also obtain the cell gene expression data of each cell in the cell line to which the target cell belongs under natural conditions from storage space; it can also obtain the cell gene expression data of each cell in the cell line to which the target cell belongs under natural conditions from the client; or it can obtain the cell gene expression data of each cell in the cell line to which the target cell belongs under natural conditions from a third-party application. The obtained cell gene expression data of each cell in the cell line to which the target cell belongs under natural conditions is used as standard cell gene expression data.

[0085] S32: Calculate the average value of the gene expression data of each of the standard cells to obtain the gene expression data of the cell line to which the target cell belongs.

[0086] Specifically, the average value of gene expression data for each of the standard cells is calculated, and the calculated data is used as the gene expression data of the cell line to which the target cell belongs.

[0087] In this embodiment, the average value of the gene expression data of each standard cell is used as the gene expression data of the cell line to which the target cell belongs, thereby obtaining characteristic data that characterizes a cell line.

[0088] In one embodiment, prior to the step of obtaining the compound expression data corresponding to the target drug, the method further includes:

[0089] S41: Obtain the compound information of the target drug;

[0090] Specifically, the compound information of the target drug can be obtained from user input, from storage space, from the client, or from a third-party application.

[0091] S42: Using the RDKit tool, the compound information is expressed in vector space to obtain the compound expression data corresponding to the target drug.

[0092] RDKit is a toolkit for handling cheminformatics problems. Written in C++, it integrates numerous cheminformatics processing methods and tools. RDKit is an open-source toolkit for cheminformatics that uses machine learning methods to generate compound descriptors based on 2D and 3D molecular manipulation. Therefore, RDKit can be used to generate vector space representations of compounds.

[0093] Specifically, the compound information is input into the RDKit tool, which generates the mathematical expression of the compound corresponding to the compound information, thereby generating a vector space expression of the compound and obtaining the compound expression data corresponding to the target drug.

[0094] This embodiment uses the RDKit tool to express the compound information in vector space, providing a basis for the expression of cellular genes in the target cells after drug perturbation by the target drug.

[0095] In one embodiment, the step of using the target decoder to decode the first hidden vector, the second hidden vector, and the third hidden vector to obtain the target cell gene expression data after drug perturbation by the target drug includes:

[0096] S51: The first hidden vector, the second hidden vector, and the third hidden vector are weighted and summed to obtain the target hidden vector;

[0097] S52: Input the target latent vector into the target decoder to decode the cell gene expression of the target cell after the drug perturbation of the target drug, and obtain the target cell gene expression data.

[0098] Specifically, the target latent vector is input into the target decoder to decode the cellular gene expression of the target cell after the drug perturbation of the target drug, and the decoded data is used as the gene expression data of the target cell.

[0099] In this embodiment, the first hidden vector, the second hidden vector, and the third hidden vector are weighted and summed to obtain the target hidden vector, thereby containing information about cell lines, genes, and drugs. The target hidden vector is then input into the target decoder to decode the cell gene expression of the target cells after drug perturbation by the target drug. This realizes the decoding of cell gene expression based on the potential relationship between cell lines, drugs, and genes, and improves the accuracy of cell gene expression prediction after drug perturbation.

[0100] Please see Figure 3 As shown, in one embodiment, a device for predicting cellular gene expression after drug perturbation is provided, the device comprising:

[0101] The data acquisition module 801 is used to acquire a pre-trained target model, wherein the target model includes a first encoder, a second encoder, a third encoder, and a target decoder;

[0102] The first latent vector determination module 802 is used to acquire the initial cell gene expression data of the target cell under the condition of drug perturbation without the target drug, and input the initial cell gene expression data into the first encoder for latent vector extraction to obtain the first latent vector;

[0103] The second latent vector determination module 803 is used to obtain the cell line gene expression data corresponding to the cell line to which the target cell belongs, and input the cell line gene expression data into the second encoder for latent vector extraction to obtain the second latent vector.

[0104] The third latent vector determination module 804 is used to acquire the compound expression data corresponding to the target drug, and input the compound expression data into the third encoder for latent vector extraction to obtain the third latent vector.

[0105] The prediction module 805 is used to use the target decoder to decode the first hidden vector, the second hidden vector and the third hidden vector to obtain the target cell gene expression data after the drug perturbation of the target drug.

[0106] This embodiment decodes the cellular gene expression of the target cells after drug perturbation by the target drug using the first hidden vector, the second hidden vector, and the third hidden vector. This enables decoding of cellular gene expression based on the potential relationship between cell lines, drugs, and genes, improving the accuracy of cellular gene expression prediction after drug perturbation. Furthermore, by inputting the second hidden vector obtained from the cell line gene expression data into the second encoder, the accuracy of cellular gene expression prediction after drug perturbation is further improved.

[0107] In one embodiment, the apparatus further includes: a model training module, the model training module being used for:

[0108] A training sample set and an initial model are obtained, wherein the initial model includes: a first initial encoder, a second initial encoder, a third initial encoder, a first classifier, a second classifier, and an initial decoder. The output of the first initial encoder is used as the input of the first classifier and the second classifier. The first classifier is used to perform cell line classification prediction and drug classification prediction.

[0109] Using the training sample set, the initial model is subjected to adversarial training on cellular gene expression after drug perturbation;

[0110] The first initial encoder, the second initial encoder, the third initial encoder, and the initial decoder of the initial model after training are used as the pre-trained target model.

[0111] In one embodiment, the step of using the training sample set to perform adversarial training on the initial model regarding cell gene expression after drug perturbation in the model training module includes:

[0112] The training sample set is divided into multiple training sample subsets using a preset batch sample size.

[0113] Get the counter and initialize its value to 1;

[0114] From each of the training sample subsets, obtain the training sample subsets that have the same order as the counter, and use them as the training subsets;

[0115] If the value of the counter is not even, then based on the batch training method, the initial model is subjected to adversarial training on cell gene expression after drug perturbation using the subset to be trained. During the adversarial training, the network parameters of the first initial encoder, the first classifier, the second classifier and the initial decoder are updated.

[0116] If the value of the counter is even, then based on the batch training method, the initial model is subjected to adversarial training on cell gene expression after drug perturbation using the subset to be trained. During the adversarial training, the network parameters of the second initial encoder and the third initial encoder are updated.

[0117] Increment the value of the counter by 1, and jump to the step of re-executing the step of obtaining the training sample subsets with the same order as the counter from each of the training sample subsets as the training subsets, until the preset training termination condition is reached.

[0118] In one embodiment, the first latent vector determination module 802 includes the following steps in acquiring initial cellular gene expression data of the target cells under drug perturbation conditions without the target drug:

[0119] Acquire the cell gene expression data of the target cells under any of the following conditions: no target drug was applied in the experiment, or a drug other than the target drug was applied in the experiment. This data is used as the initial cell gene expression data of the target cells under the drug perturbation condition without the target drug.

[0120] In one embodiment, before the step of obtaining the cell line gene expression data corresponding to the cell line to which the target cell belongs, the second latent vector determination module 803 further includes:

[0121] Obtain the gene expression data of each cell in the cell line to which the target cell belongs under natural conditions, and use it as standard cell gene expression data;

[0122] The average value of the gene expression data of each of the standard cells is calculated to obtain the gene expression data of the cell line to which the target cell belongs.

[0123] In one embodiment, the third latent vector determination module 804 further includes, before the step of obtaining the compound expression data corresponding to the target drug:

[0124] Obtain the compound information of the target drug;

[0125] The RDKit tool is used to represent the compound information in vector space to obtain the compound expression data corresponding to the target drug.

[0126] In one embodiment, the step of the prediction module 805 using the target decoder to decode the first hidden vector, the second hidden vector, and the third hidden vector to obtain the target cell gene expression data after drug perturbation by the target drug includes:

[0127] The target hidden vector is obtained by weighted summation of the first hidden vector, the second hidden vector, and the third hidden vector.

[0128] The target latent vector is input into the target decoder to decode the cellular gene expression of the target cell after the drug perturbation of the target drug, thereby obtaining the target cell gene expression data.

[0129] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When executed by the processor, the computer program implements the server-side functions or steps of a method for predicting cellular gene expression after drug perturbation.

[0130] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When executed by the processor, the computer program implements client-side functions or steps of a method for predicting cell gene expression after drug perturbation.

[0131] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, performs the following steps:

[0132] Obtain a pre-trained target model, wherein the target model includes a first encoder, a second encoder, a third encoder, and a target decoder;

[0133] The initial cell gene expression data of the target cells without drug perturbation by the target drug is obtained, and the initial cell gene expression data is input into the first encoder for latent vector extraction to obtain the first latent vector;

[0134] Obtain the cell line gene expression data corresponding to the cell line to which the target cell belongs, and input the cell line gene expression data into the second encoder for latent vector extraction to obtain the second latent vector;

[0135] Obtain the compound expression data corresponding to the target drug, and input the compound expression data into the third encoder for latent vector extraction to obtain the third latent vector;

[0136] Using the target decoder, the first hidden vector, the second hidden vector, and the third hidden vector are used to decode the cellular gene expression of the target cell after the drug perturbation of the target drug, so as to obtain the target cell gene expression data.

[0137] This embodiment decodes the cellular gene expression of the target cells after drug perturbation by the target drug using the first hidden vector, the second hidden vector, and the third hidden vector. This enables decoding of cellular gene expression based on the potential relationship between cell lines, drugs, and genes, improving the accuracy of cellular gene expression prediction after drug perturbation. Furthermore, by inputting the second hidden vector obtained from the cell line gene expression data into the second encoder, the accuracy of cellular gene expression prediction after drug perturbation is further improved.

[0138] In one embodiment, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program that, when executed by a processor, performs the following steps:

[0139] Obtain a pre-trained target model, wherein the target model includes a first encoder, a second encoder, a third encoder, and a target decoder;

[0140] The initial cell gene expression data of the target cells without drug perturbation by the target drug is obtained, and the initial cell gene expression data is input into the first encoder for latent vector extraction to obtain the first latent vector;

[0141] Obtain the cell line gene expression data corresponding to the cell line to which the target cell belongs, and input the cell line gene expression data into the second encoder for latent vector extraction to obtain the second latent vector;

[0142] Obtain the compound expression data corresponding to the target drug, and input the compound expression data into the third encoder for latent vector extraction to obtain the third latent vector;

[0143] Using the target decoder, the first hidden vector, the second hidden vector, and the third hidden vector are used to decode the cellular gene expression of the target cell after the drug perturbation of the target drug, so as to obtain the target cell gene expression data.

[0144] This embodiment decodes the cellular gene expression of the target cells after drug perturbation by the target drug using the first hidden vector, the second hidden vector, and the third hidden vector. This enables decoding of cellular gene expression based on the potential relationship between cell lines, drugs, and genes, improving the accuracy of cellular gene expression prediction after drug perturbation. Furthermore, by inputting the second hidden vector obtained from the cell line gene expression data into the second encoder, the accuracy of cellular gene expression prediction after drug perturbation is further improved.

[0145] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0146] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0147] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0148] 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, and should all be included within the protection scope of the present invention.

Claims

1. A method for predicting cellular gene expression after drug perturbation, the method comprising: Obtain a pre-trained target model, wherein the target model includes a first encoder, a second encoder, a third encoder, and a target decoder; The initial cell gene expression data of the target cells under drug perturbation without the target drug is obtained, and the initial cell gene expression data is input into the first encoder for latent vector extraction to obtain the first latent vector; Obtain the cell line gene expression data corresponding to the cell line to which the target cell belongs, and input the cell line gene expression data into the second encoder for latent vector extraction to obtain the second latent vector; Obtain the compound expression data corresponding to the target drug, and input the compound expression data into the third encoder for latent vector extraction to obtain the third latent vector; Using the target decoder, the first hidden vector, the second hidden vector, and the third hidden vector are used to decode the cellular gene expression of the target cell after the drug perturbation of the target drug, so as to obtain the target cell gene expression data. Before the step of obtaining the pre-trained target model, the method further includes: A training sample set and an initial model are obtained. The initial model includes: a first initial encoder, a second initial encoder, a third initial encoder, a first classifier, a second classifier, and an initial decoder. The output of the first initial encoder serves as the input to the first and second classifiers. The first classifier is used for cell line classification prediction, and the second classifier is used for drug classification prediction. The first classifier is a multilayer perceptron or multiple linear layers. The second classifier is a multilayer perceptron or multiple linear layers. The training samples include: initial cell gene expression sample data, cell line gene expression sample data, compound expression sample data, cell line calibration values, drug calibration values, and cell gene expression calibration data. Using the training sample set, the initial model is subjected to adversarial training on cell gene expression after drug perturbation; adversarial loss is introduced through the first classifier and the second classifier to reduce the coupling between genes, cell lines and drugs; The first initial encoder, the second initial encoder, the third initial encoder, and the initial decoder of the initial model after training are used as the pre-trained target model; The step of using the training sample set to perform adversarial training on the initial model for drug-induced perturbation of cellular gene expression includes: The training sample set is divided into multiple training sample subsets using a preset batch sample size. Get the counter and initialize its value to 1; From each of the training sample subsets, obtain the training sample subsets that have the same order as the counter, and use them as the training subsets; If the value of the counter is not even, then based on the batch training method, the initial model is subjected to adversarial training on cell gene expression after drug perturbation using the subset to be trained. During the adversarial training, the network parameters of the first initial encoder, the first classifier, the second classifier and the initial decoder are updated. If the value of the counter is even, then based on the batch training method, the initial model is subjected to adversarial training on cell gene expression after drug perturbation using the subset to be trained. During the adversarial training, the network parameters of the second initial encoder and the third initial encoder are updated. Increment the value of the counter by 1, and jump to the step of re-executing the step of obtaining the training sample subsets with the same order as the counter from each of the training sample subsets as the training subsets, until the preset training termination condition is reached.

2. The method for predicting cellular gene expression after drug perturbation according to claim 1, characterized in that, The step of obtaining initial cellular gene expression data of target cells under drug perturbation conditions without the target drug includes: Acquire the cell gene expression data of the target cells under any of the following conditions: no target drug was applied in the experiment, or a drug other than the target drug was applied in the experiment. This data is used as the initial cell gene expression data of the target cells under the drug perturbation condition without the target drug.

3. The method for predicting cellular gene expression after drug perturbation according to claim 1, characterized in that, Before the step of obtaining the cell line gene expression data corresponding to the cell line to which the target cell belongs, the method further includes: Obtain the gene expression data of each cell in the cell line to which the target cell belongs under natural conditions, and use it as standard cell gene expression data; The average value of the gene expression data of each of the standard cells is calculated to obtain the gene expression data of the cell line to which the target cell belongs.

4. The method for predicting cellular gene expression after drug perturbation according to claim 1, characterized in that, Before the step of obtaining the compound expression data corresponding to the target drug, the method further includes: Obtain the compound information of the target drug; The RDKit tool is used to represent the compound information in vector space to obtain the compound expression data corresponding to the target drug.

5. The method for predicting cellular gene expression after drug perturbation according to claim 1, characterized in that, The step of using the target decoder to decode the first hidden vector, the second hidden vector, and the third hidden vector to obtain the target cell gene expression data after drug perturbation by the target drug, includes: The target hidden vector is obtained by weighted summation of the first hidden vector, the second hidden vector, and the third hidden vector. The target latent vector is input into the target decoder to decode the cellular gene expression of the target cell after the drug perturbation of the target drug, thereby obtaining the target cell gene expression data.

6. A device for predicting cellular gene expression after drug perturbation, characterized in that, The device includes: The data acquisition module is used to acquire a pre-trained target model, wherein the target model includes a first encoder, a second encoder, a third encoder, and a target decoder; The first latent vector determination module is used to acquire the initial cell gene expression data of the target cells under drug perturbation without the target drug, and input the initial cell gene expression data into the first encoder for latent vector extraction to obtain the first latent vector; The second latent vector determination module is used to obtain the cell line gene expression data corresponding to the cell line to which the target cell belongs, and input the cell line gene expression data into the second encoder to extract the latent vector to obtain the second latent vector. The third latent vector determination module is used to acquire the compound expression data corresponding to the target drug, and input the compound expression data into the third encoder for latent vector extraction to obtain the third latent vector. The prediction module is used to use the target decoder to decode the first hidden vector, the second hidden vector and the third hidden vector to obtain the target cell gene expression data after the drug perturbation of the target drug. The model training module, prior to the step of obtaining the pre-trained target model, is used for: A training sample set and an initial model are obtained. The initial model includes: a first initial encoder, a second initial encoder, a third initial encoder, a first classifier, a second classifier, and an initial decoder. The output of the first initial encoder serves as the input to the first and second classifiers. The first classifier is used for cell line classification prediction, and the second classifier is used for drug classification prediction. The first classifier is a multilayer perceptron or multiple linear layers. The second classifier is a multilayer perceptron or multiple linear layers. The training samples include: initial cell gene expression sample data, cell line gene expression sample data, compound expression sample data, cell line calibration values, drug calibration values, and cell gene expression calibration data. Using the training sample set, the initial model is subjected to adversarial training on cell gene expression after drug perturbation; adversarial loss is introduced through the first classifier and the second classifier to reduce the coupling between genes, cell lines and drugs; The first initial encoder, the second initial encoder, the third initial encoder, and the initial decoder of the initial model after training are used as the pre-trained target model; The model training module is used to perform adversarial training on the initial model based on the training sample set, specifically for: The training sample set is divided into multiple training sample subsets using a preset batch sample size. Get the counter and initialize its value to 1; From each of the training sample subsets, obtain the training sample subsets that have the same order as the counter, and use them as the training subsets; If the value of the counter is not even, then based on the batch training method, the initial model is subjected to adversarial training on cell gene expression after drug perturbation using the subset to be trained. During the adversarial training, the network parameters of the first initial encoder, the first classifier, the second classifier and the initial decoder are updated. If the value of the counter is even, then based on the batch training method, the initial model is subjected to adversarial training on cell gene expression after drug perturbation using the subset to be trained. During the adversarial training, the network parameters of the second initial encoder and the third initial encoder are updated. Increment the value of the counter by 1, and jump to the step of re-executing the step of obtaining the training sample subsets with the same order as the counter from each of the training sample subsets as the training subsets, until the preset training termination condition is reached.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method for predicting cell gene expression after drug perturbation as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method for predicting cell gene expression after drug perturbation as described in any one of claims 1 to 5.

Images