Model training method, drug reaction prediction method and device
Patent Information
- Application Number
- CN202510293314.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-12
- Publication Date
- 2026-09-15
Smart Images
Figure CN122761997A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of artificial intelligence (AI) technology, specifically to model training methods, drug response prediction methods, and devices. Background Technology
[0002] With the development of precision medicine, the formulation of personalized drug treatment strategies has become an important direction in clinical research. To predict patient responses to specific drugs, drug response prediction models are widely used in drug screening and efficacy evaluation.
[0003] Existing drug response prediction models typically rely on two-dimensional (2D) cell experimental data, which cannot fully simulate the complex three-dimensional microenvironment in vivo, resulting in poor performance in predicting actual drug responses in patients. Summary of the Invention
[0004] This application proposes a model training method, a drug response prediction method, and an apparatus to obtain a high-precision drug response prediction model, thereby improving the efficiency and accuracy of drug response prediction.
[0005] Specifically, this application provides the following technical solution:
[0006] In a first aspect of this application, a model training method is proposed, the method comprising: acquiring training data, the training data including: cell gene expression information, three-dimensional model gene expression information, and drug molecule structure information; pre-training a deep learning model using the cell gene expression information and the drug molecule structure information to obtain a pre-trained drug response prediction model; and fine-tuning the pre-trained drug response prediction model using the three-dimensional model gene expression information and the drug molecule structure information to obtain a drug response prediction model.
[0007] In some possible implementations, the step of pre-training the deep learning model using the cell gene expression information and the drug molecule structure information to obtain a pre-trained drug response prediction model includes: pre-training the deep learning model using the cell gene expression information and the drug molecule structure information after a first preprocessing to obtain a pre-trained drug response prediction model.
[0008] In some possible implementations, the first preprocessing includes: representing the cell gene expression information in a transcription millions format; performing a logarithmic transformation on the transcription millions format of the cell gene expression information to obtain the cell gene expression information input of the deep learning model; converting the drug molecule structure into a standard expression paradigm; and performing byte-pair encoding on the standard expression paradigm to obtain the drug molecule structure information input of the deep learning model.
[0009] In some implementations, the step of pre-training a deep learning model using the pre-processed cell gene expression information and drug molecule structure information to obtain a pre-trained drug response prediction model includes: extracting features from the pre-processed cell gene expression information using a two-layer fully connected network to obtain a cell gene feature vector; extracting features from the pre-processed drug molecule structure information using a single-layer fully connected network to obtain a drug feature vector; and pre-training the deep learning model based on the cell gene feature vector and the drug feature vector to obtain a pre-trained drug response prediction model.
[0010] In some implementations, the step of pre-training the deep learning model based on the cell gene feature vector and the drug feature vector to obtain a pre-trained drug response prediction model includes: concatenating the cell gene feature vector and the drug feature vector to obtain a cell-drug concatenated feature vector; performing dimensionality upscaling on the cell-drug concatenated feature vector; and using the dimensionality upscaled cell-drug concatenated feature vector to pre-train the deep learning model to obtain a pre-trained drug response prediction model.
[0011] In some implementations, the step of pre-training the deep learning model using the upgraded cell-drug splicing feature vector to obtain a pre-trained drug response prediction model includes: calculating the loss of the deep learning model based on the difference between the output value of the pre-trained drug response prediction model and a first true value; and pre-training the deep learning model based on the loss to obtain a pre-trained drug response prediction model.
[0012] In some possible implementations, the first true value is determined by the AUC value of the drug concentration on the cellular effect.
[0013] In some possible implementations, the step of fine-tuning the pre-trained drug response prediction model using the gene expression information of the three-dimensional model and the drug molecule structure information to obtain a drug response prediction model includes: fine-tuning the pre-trained drug response prediction model using the gene expression information of the three-dimensional model and the drug molecule structure information after a second preprocessing to obtain a drug response prediction model.
[0014] In some implementations, the second preprocessing includes: representing the three-dimensional model gene expression information in a transcription millions format; performing a logarithmic transformation on the transcription millions format to obtain the three-dimensional model gene expression information input of the pre-trained drug response prediction model; converting the drug molecule structure into a standard expression paradigm; and performing byte-pair encoding on the standard expression paradigm to obtain the drug molecule structure information input of the pre-trained drug response prediction model.
[0015] In some implementations, the step of fine-tuning the pre-trained drug response prediction model using the gene expression information of the three-dimensional model after the second preprocessing and the drug molecule structure information to obtain the drug response prediction model includes: extracting features from the gene expression information of the three-dimensional model after the second preprocessing using a two-layer fully connected network to obtain a three-dimensional model gene feature vector; extracting features from the drug molecule structure information after the second preprocessing using a single-layer fully connected network to obtain a drug feature vector; and fine-tuning the pre-trained drug response prediction model based on the three-dimensional model gene feature vector and the drug feature vector to obtain the drug response prediction model.
[0016] In some implementation methods, the step of fine-tuning the pre-trained drug response prediction model based on the three-dimensional model gene feature vector and the drug feature vector to obtain the drug response prediction model includes: concatenating the three-dimensional model gene feature vector and the drug feature vector to obtain a three-dimensional model-drug concatenated feature vector; performing dimensionality upgrading on the three-dimensional model-drug concatenated feature vector; and using the dimensionality upgrading three-dimensional model-drug concatenated feature vector to fine-tune the pre-trained drug response prediction model to obtain the drug response prediction model.
[0017] In some possible implementations, the step of fine-tuning the pre-trained drug response prediction model using the upgraded 3D model-drug splicing feature vector to obtain the drug response prediction model includes: calculating the loss of the drug response prediction model based on the difference between the output value of the drug response prediction model and the second true value; and fine-tuning the drug response prediction model based on the loss to obtain the drug response prediction model.
[0018] In some implementations, the second true value is determined by the AUC value of the effect of drug concentration on the three-dimensional model.
[0019] In some possible implementations, the drug response prediction model is fine-tuned using at least one of L2 regularization or early stop strategies.
[0020] In a second aspect of this application, a method for predicting drug response is proposed, the method comprising: acquiring cell gene expression information and drug molecular structure information; inputting the cell gene expression information and drug molecular structure information into a trained drug response prediction model to obtain drug response prediction information; wherein the trained drug response prediction model is obtained by training using the method described in the first aspect.
[0021] In some possible implementations, the drug is selected from oncology drugs.
[0022] In a third aspect, this application proposes a model training apparatus. According to an embodiment of this application, the apparatus includes: a training data acquisition module for acquiring training data, the training data including: cell gene expression information, three-dimensional model gene expression information, and drug molecule structure information; a pre-training module for pre-training a deep learning model using the cell gene expression information and the drug molecule structure information to obtain a pre-trained drug response prediction model; and a fine-tuning module for fine-tuning the pre-trained drug response prediction model using the three-dimensional model gene expression information and the drug molecule structure information to obtain a drug response prediction model.
[0023] In a fourth aspect of this application, a drug response prediction device is proposed. The system includes: an information acquisition module for acquiring cell gene expression information and drug molecule structure information; and a prediction module for inputting the cell gene expression information and drug molecule structure information into a trained drug response prediction model to obtain drug response prediction information; wherein the trained drug response prediction model is obtained by training using the method described in the first aspect.
[0024] In a fifth aspect of this application, a computing device is provided, the computing device comprising: a processor and a memory; the memory for storing a computer program; and the processor for executing the computer program to implement the model training method as described in the first aspect or the drug response prediction method as described in the second aspect.
[0025] In a sixth aspect, this application proposes a computer-readable storage medium storing computer instructions or programs that, when executed on a computer, cause the model training method as described in the first aspect or the drug response prediction method as described in the second aspect to be performed.
[0026] In a seventh aspect of this application, a computer program product is provided, the computer program product comprising: computer instructions; which, when some or all of the computer instructions are executed on a computer, cause the model training method as described in the first aspect or the drug response prediction method as described in the second aspect to be executed.
[0027] The technical solution of this application trains a drug response prediction model using two-dimensional cell line drug response data and three-dimensional model drug response data to obtain a drug response prediction model with higher prediction accuracy. This model can better simulate the drug response of real patients, improving the efficiency and accuracy of drug response prediction.
[0028] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description
[0029] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0030] Figure 1 These are schematic diagrams of system architecture provided in some examples of this application;
[0031] Figure 2 These are schematic diagrams illustrating the model training methods provided in some examples of this application;
[0032] Figure 3 These are schematic diagrams of the model architecture provided in some examples of this application;
[0033] Figure 4 This is a schematic diagram of a drug response prediction method provided in some examples of this application;
[0034] Figure 5 These are schematic diagrams of model training devices provided in some examples of this application;
[0035] Figure 6 These are schematic diagrams of drug response prediction devices provided in some examples of this application;
[0036] Figure 7 These are schematic diagrams of electronic devices provided in this application;
[0037] Figure 8 These are some examples of illustrations provided in this application to predict drug response assessment results in a real-world clinical setting. Detailed Implementation
[0038] 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 embodiments of the present invention, and not all embodiments. 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.
[0039] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.
[0040] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.
[0041] Before introducing the technical solution of this application, the relevant knowledge of this application will be introduced below:
[0042] In this embodiment, deep learning models are a type of artificial intelligence technology based on neural networks. Due to their powerful nonlinear modeling capabilities and high-dimensional data feature extraction capabilities, they are widely used in fields such as image recognition, natural language processing, and bioinformatics. Deep learning models extract complex features from input data and achieve pattern recognition and prediction functions through the nonlinear mapping of multi-layer neural networks. These models include feedforward neural networks (FNN), convolutional neural networks (CNN), recurrent neural networks (RNN), and attention-based models (such as Transformer), each with different applicability to different types of data.
[0043] The technical problems to be solved, the inventive concept and the system architecture of the embodiments of this application will be described below:
[0044] As mentioned earlier, AI technology has been introduced into the field of drug response prediction. Most existing drug response prediction models are based on 2D cell line experimental models. While these cell line models are widely used in drug screening, they fail to fully simulate the complex three-dimensional microenvironment in vivo, resulting in poor performance in predicting actual patient drug responses. These shortcomings are mainly reflected in the following aspects:
[0045] The simplification of the cellular environment: Traditional two-dimensional cell lines lack the real three-dimensional cellular tissue structure in patients, making it difficult to fully simulate the mechanism of drug action in vivo.
[0046] Insufficient data integration: Existing drug response prediction methods often rely on a single data source (such as cell lines) and lack the ability to effectively integrate multi-dimensional biological data.
[0047] Algorithm limitations: Current common algorithm architectures (such as simple regression models or basic neural networks) have limited performance when dealing with large-scale, heterogeneous biological data.
[0048] These shortcomings result in poor accuracy of drug response prediction models, especially in complex clinical settings, limiting their effectiveness to preclinical models such as cell lines. Therefore, improving the accuracy of drug response prediction models in actual drug responses is a key technical problem that this application aims to solve.
[0049] To address the aforementioned technical issues, this application proposes a model training method to obtain a drug response prediction model that can effectively distinguish patients' clinical responses to drugs and improve the accuracy of drug response prediction.
[0050] The drug response prediction model in this application is a deep learning model, such as LSTM (Long Short-Term Memory Network), GRU (Gated Recurrent Unit), Transformer, or a variant thereof.
[0051] The model training method in this application includes a pre-training phase and a fine-tuning phase.
[0052] The pre-training phase is a crucial step in model training, involving training a model on a large-scale dataset to enable it to learn general feature representations. In this application, the pre-training approach utilizes a two-dimensional cell line dataset, allowing the model to learn the complex interactions between gene expression and drug molecule features. This provides a powerful initial feature representation capability for subsequent smaller datasets (such as three-dimensional model data).
[0053] The fine-tuning phase involves further adjusting the model parameters using task-specific data based on the pre-trained model. The goal of this approach is to transfer the model from the pre-training task to the specific task while retaining the general features learned during pre-training. In this application, the fine-tuning phase utilizes 3D model data from specific organs (such as the colon, bladder, and hepatocellular carcinoma). By fine-tuning to adapt to these highly physiologically relevant samples, the model further optimizes its ability to predict drug responses to specific diseases or tissues.
[0054] In some embodiments of this application, the system architecture is as follows: Figure 1 As shown.
[0055] Figure 1 This is a schematic diagram of a system architecture according to an embodiment of this application. The system architecture includes a user device 101, a data acquisition device 102, a training device 103, an execution module 104, a database 105, and a content library 106.
[0056] The data acquisition device 102 is used to read training samples from the content library 106 and store the read training samples in the database 105. The training samples involved in this embodiment are obtained through public databases or self-test experimental data.
[0057] The training device 103 pre-trains and fine-tunes the deep learning model based on the training samples maintained in the database 105, so that the pre-trained and fine-tuned deep learning model (drug response prediction model) can effectively predict drug responses.
[0058] Additionally, refer to Figure 1 The execution module 104 is equipped with an I / O interface 107 for data interaction with external devices. For example, it receives input commands sent by the user device 101 via the I / O interface; these input commands include cell gene expression information and drug molecule structure information. The calculation module 108 in the execution module 104 processes the input commands using a trained deep learning model to obtain output commands; these output commands include drug response prediction information, and are sent to the user device 101 via the I / O interface.
[0059] User equipment 101 may include mobile phones, tablets, laptops, handheld computers, mobile internet devices (MIDs), desktop computers, or other terminal devices with browser installation capabilities.
[0060] Execution module 104 can be a server.
[0061] For example, the server can be a rack server, blade server, tower server, or cabinet server, etc. The server can be a standalone test server or a test server cluster composed of multiple test servers.
[0062] In this embodiment, the execution module 104 is connected to the user equipment 101 via a network. The network can be an intranet, the Internet, Global System for Mobile Communications (GSM), Wideband Code Division Multiple Access (WCDMA), 4G, 5G, Bluetooth, Wi-Fi, or a voice communication network, or other wireless or wired networks.
[0063] It should be noted that, Figure 1 This is merely a schematic diagram of a system architecture provided in this application embodiment, and the positional relationships between the devices, components, modules, etc. shown in the figure do not constitute any limitation. In some embodiments, the data acquisition device 102, user device 101, training device 103, and execution module 104 may be the same device. The database 105 may be distributed across one server or multiple servers, and the content library 106 may be distributed across one server or multiple servers.
[0064] The technical solution of this application will be described in detail below:
[0065] Figure 2 This application provides a flowchart of a drug response prediction model training method, which can be executed by a training device, for example, by... Figure 1 The training device 103 shown is used for execution, but is not limited to it, such as Figure 2 As shown, the method may include:
[0066] S210: Acquire training data, which includes: cell gene expression information, 3D model gene expression information, and drug molecule structure information;
[0067] In some examples of this application, the aforementioned training data may come from public databases or self-tested cell line experimental data, such as the GDSC2 cell line dataset.
[0068] In some examples of this application, the aforementioned training data may contain labels, such as drug sensitivity measurements (AUC values) for each pair (e.g., cell line-drug pair or 3D model-drug pair).
[0069] It is understandable that the gene expression information of the aforementioned three-dimensional model can be selected from organoids, spheroids, or organ-on-a-chip, etc.
[0070] S220 uses cell gene expression information and drug molecule structure information to pre-train a deep learning model to obtain a pre-trained drug response prediction model.
[0071] The deep learning model, pre-trained drug response prediction model, and drug response prediction model involved in this application are selected from Transformer models or their variants. Transformer is a deep learning model architecture based on a self-attention mechanism, which captures global dependencies by processing all positions in a sequence simultaneously, significantly improving computational efficiency and modeling capabilities. Its core consists of an encoder and a decoder, and it is widely used in natural language processing, computer vision, and speech processing, supporting parallel computing and long sequence modeling, thus overcoming the limitations of traditional recurrent networks in sequence tasks.
[0072] In some examples of this application, a deep learning model is pre-trained using cell gene expression information and drug molecule structure information to obtain a pre-trained drug response prediction model, including: using cell gene expression information and drug molecule structure information that have undergone a first preprocessing to pre-train the deep learning model to obtain a pre-trained drug response prediction model.
[0073] In some examples of this application, the first preprocessing includes: representing cellular gene expression information in Transcripts Per Million (TPM) format; performing a logarithmic transformation on the TPM format of the cellular gene expression information to obtain the cellular gene expression information input for the deep learning model; converting the drug molecule structure into a standard expression paradigm; and performing Byte Pair Encoding (BPE) on the standard expression paradigm to obtain the drug molecule structure information input for the deep learning model. The logarithmic transformation is performed to stabilize the variance and normalize the distribution.
[0074] In some examples of this application, logarithmic transformations are feasible in ways such as log2(TPM+1).
[0075] In some examples of this application, the standard expression paradigm can be the Simplified Molecular Input Line Entry System (SMILES).
[0076] In some examples of this application, each SMILES string is encoded as a fixed-length vector, which is truncated if the sequence exceeds 128 tokens, and padded with zeros if it is not full.
[0077] In some examples of this application, references Figure 3 The deep learning model is pre-trained using pre-processed cellular gene expression information and drug molecule structure information to obtain a pre-trained drug response prediction model. This includes: extracting features from the pre-processed cellular gene expression information using a two-layer fully connected network to obtain cellular gene feature vectors; extracting features from the pre-processed drug molecule structure information using a single-layer fully connected network to obtain drug feature vectors; and pre-training the deep learning model based on the cellular gene feature vectors and drug feature vectors to obtain the pre-trained drug response prediction model. By replacing the traditional Transformer embedding layer with a single / bilinear layer, the model's complexity is simplified while maintaining data validity.
[0078] For example, a bilinear layer (fully connected layer) is used for cell gene feature extraction, with the input being the cell gene expression vector. The output is a cell gene feature vector (same as "embedding vector"), specifically represented as follows:
[0079] First-level linear transformation:
[0080]
[0081] in,
[0082] Second-level linear transformation:
[0083]
[0084] in,
[0085] For example, a single linear layer (fully connected layer) is used for drug feature extraction, with the input being a vector generated by processing SMILES encoding. The output is a drug feature vector (same as "embedding vector"), specifically represented as follows:
[0086] h drug =ReLU(W drug x drug +bdrug )
[0087] in,
[0088] In some examples of this application, a deep learning model is pre-trained based on cell gene feature vectors and drug feature vectors to obtain a pre-trained drug response prediction model. This includes: concatenating the cell gene feature vectors and drug feature vectors to obtain a cell-drug concatenated feature vector; reshaping the cell-drug concatenated feature vector; and using the reshaped cell-drug concatenated feature vector to pre-train the deep learning model to obtain the pre-trained drug response prediction model. By concatenating the cell gene feature vectors and drug feature vectors to obtain richer information representation, the drug response prediction model can more comprehensively consider drug response information during training, enhancing its generalization ability. By reshaping the concatenated feature vectors into a matrix, the model can better understand the relationships between features and effectively capture the dependencies and interactions between different features through a self-attention mechanism.
[0089] For example, the cell-drug splicing feature vector can be represented as:
[0090]
[0091] in,
[0092] For example, the cell-drug splicing feature vector after dimensionality increase is:
[0093] In some examples of this application, a pre-trained deep learning model is obtained by using an upgraded cell-drug splicing feature vector to obtain a pre-trained drug response prediction model. This includes: calculating the loss of the deep learning model based on the difference between the output value of the pre-trained drug response prediction model and the first true value; pre-training the deep learning model based on the loss to obtain the pre-trained drug response prediction model; and optimizing the drug response prediction model based on the loss.
[0094] In some examples of this application, the aforementioned loss is determined based on the difference between the predicted result and the first true value.
[0095] In some examples of this application, the aforementioned loss is selected from mean squared error (MSE).
[0096] In some examples of this application, the first true value is determined by the AUC (area under the curve) value of the drug concentration versus the cellular effect. The AUC value represents the sensitivity of a two-dimensional cell line to the drug. A higher AUC value means that the cells are more resistant to the drug, while a lower AUC value indicates that the cells are more sensitive to the drug.
[0097] In some examples of this application, an X-fold (e.g., 50%, 60%, 70%, 80%) cross-validation strategy is used to ensure the robustness of the model.
[0098] S230 uses three-dimensional model gene expression information and drug molecular structure information to fine-tune the pre-trained drug response prediction model to obtain the drug response prediction model.
[0099] In some examples of this application, the pre-trained drug response prediction model is fine-tuned using three-dimensional model gene expression information and drug molecule structure information to obtain a drug response prediction model, including: fine-tuning the pre-trained drug response prediction model using three-dimensional model gene expression information and drug molecule structure information after a second preprocessing to obtain a drug response prediction model.
[0100] In some examples of this application, the second preprocessing includes: representing the gene expression information of the three-dimensional model in a transcription millions format; performing a logarithmic transformation on the transcription millions format to obtain the three-dimensional model gene expression information input for the pre-trained drug response prediction model; converting the drug molecular structure into a standard expression paradigm; and performing byte-pair encoding on the standard expression paradigm to obtain the drug molecular structure information input for the pre-trained drug response prediction model. The logarithmic transformation is performed to stabilize the variance and normalize the distribution.
[0101] In some examples of this application, logarithmic transformations are feasible in ways such as log2(TPM+1).
[0102] In some examples of this application, the standard expression paradigm can be the Simplified Molecular Input Line Entry System (SMILES).
[0103] In some examples of this application, each SMILES string is encoded as a fixed-length vector, which is truncated if the sequence exceeds 128 tokens, and padded with zeros if it is not full.
[0104] In some examples of this application, references Figure 3The pre-trained drug response prediction model is fine-tuned using the gene expression information and drug molecular structure information of the three-dimensional model after the second preprocessing to obtain the drug response prediction model. This includes: extracting features from the gene expression information of the three-dimensional model after the second preprocessing through a two-layer fully connected network to obtain the gene feature vector of the three-dimensional model; extracting features from the drug molecular structure information after the second preprocessing through a single-layer fully connected network to obtain the drug feature vector; and fine-tuning the pre-trained drug response prediction model based on the gene feature vector and drug feature vector of the three-dimensional model to obtain the drug response prediction model.
[0105] The feature extraction process in this section is the same as the extraction process for gene features and drug features in two-dimensional cell lines, and will not be described again here.
[0106] In some examples of this application, a pre-trained drug response prediction model is fine-tuned based on the gene feature vector and drug feature vector of a three-dimensional model to obtain a drug response prediction model. This includes: concatenating the gene feature vector and drug feature vector of the three-dimensional model to obtain a three-dimensional model-drug concatenated feature vector; and performing dimensionality upscaling on the three-dimensional model-drug concatenated feature vector, using the upscaled three-dimensional model-drug concatenated feature vector to fine-tune the pre-trained drug response prediction model to obtain the drug response prediction model. By concatenating the gene feature vector and drug feature vector of the three-dimensional model, a richer information representation is obtained, enabling the drug response prediction model to more comprehensively consider drug response information during fine-tuning, thus enhancing the prediction accuracy of the drug response prediction model. By reshaping the concatenated feature vector into a matrix, the model can better understand the relationships between features and effectively capture the dependencies and interactions between different features through a self-attention mechanism.
[0107] The process of feature splicing and dimensionality enhancement (or reshaping) in this section is the same as the process of splicing and dimensionality enhancement of cell gene features and drug features, and will not be described again here.
[0108] In some examples of this application, the pre-trained drug response prediction model is fine-tuned using the upgraded 3D model-drug splicing feature vector to obtain the drug response prediction model, including: calculating the loss of the drug response prediction model based on the difference between the output value of the drug response prediction model and the second true value; and fine-tuning the drug response prediction model based on the loss to obtain the drug response prediction model.
[0109] In some examples of this application, the aforementioned loss is selected from mean squared error (MSE).
[0110] In some examples of this application, the second true value is determined by the AUC value of the effect of drug concentration on the three-dimensional model. The AUC value represents the sensitivity of the cells in the three-dimensional model to the drug. A higher AUC value means that the cells are more resistant to the drug, while a lower AUC value indicates that the cells are more sensitive to the drug.
[0111] In some examples of this application, in addition to using mean squared error (MSE) as the loss function, at least one of L2 regularization or early stop strategy can be used to fine-tune the drug response prediction model. By introducing early stop mechanism and regularization strategy, the robustness and generalization ability of the model on three-dimensional model drug experimental data can be improved, and noise or overfitting caused by limited samples can be avoided.
[0112] It should be noted that this application does not limit the type of drug. For example, the aforementioned drugs may be selected from oncology drugs, cardiovascular disease drugs, etc.
[0113] In some examples of this application, clinical drug response assessments were performed on a drug response prediction model (PharmaFormer) trained based on the above methods using first-line clinical drug data for colorectal cancer, bladder cancer, and liver cancer (including pentafluorouracil, oxaliplatin, gemcitabine, carboplatin, and sorafenib).
[0114] To further evaluate PharmaFormer's ability to predict drug response in a real-world clinical setting, we fine-tuned the pre-trained model using data from colorectal cancer organ tissues from 29 patients. We then used both the pre-trained and fine-tuned models to predict drug response in TCGA bulk RNA-seq data from colorectal cancer patients. Since 5-fluorouracil and oxaliplatin are the most commonly used compounds in colorectal cancer, these two molecules were prioritized for further study. Patients were divided into drug-sensitive and drug-resistant groups based on their predicted response scores. Figure 8 As shown in A, the organoid fine-tuning model outperformed the pre-trained model in predicting colorectal cancer patients. Specifically, the hazard ratios of 5-fluorouracil and oxaliplatin improved from 2.5039 (95% CI: 1.1204–5.5956) and 1.9541 (95% CI: 0.8247–4.6301) to 3.9072 (95% CI: 1.5429–9.3941) and 4.4936 (95% CI: 1.7594–11.4765), respectively.
[0115] Similar improvements in predictive accuracy were observed in bladder cancer patients treated with gemcitabine and cisplatin. Figure 8(B) In gemcitabine, the pre-training hazard ratio was 1.7245 (95% CI: 0.8522–3.4895), while the fine-tuned hazard ratio increased to 4.9120 (95% CI: 1.1775–20.4892). In cisplatin, the pre-training hazard ratio was 1.8004 (95% CI: 0.86861–4.7239), and the fine-tuned model's hazard ratio was 6.0137 (95% CI: 1.4329–25.2391). Furthermore, in hepatocellular carcinoma patients treated with sorafenib, the fine-tuned model significantly improved the discrimination of drug response (pre-training hazard ratio: 1.3434, 95% CI: 0.3643–4.9531; fine-tuned hazard ratio: 5.6677, 95% CI: 1.4877–21.5923).
[0116] To verify the efficiency of the proposed transfer learning strategy, we compared four training strategies: cell line only, organoid only, cell line-organoid combination dataset without transfer learning, and cell line pre-training followed by organoid fine-tuning. The results show that the organoid fine-tuning method achieves the best prediction performance among all tested drugs. Figure 8 These findings indicate that transfer learning can effectively integrate two-dimensional cell line data and three-dimensional organoid models, improving the accuracy of clinical drug response prediction.
[0117] In summary, the model obtained based on the model training method of this application has the following beneficial effects:
[0118] Improved prediction accuracy: By combining drug response data from a three-dimensional model, the accuracy of the drug response prediction model is significantly improved. Compared with the traditional two-dimensional cell line data model, it can better simulate the drug response of real patients.
[0119] Improved data utilization: By effectively combining large-scale cell line data with 3D model data using transfer learning methods, this invention can make fuller use of existing multi-source biological data and improve data utilization efficiency.
[0120] Algorithm efficiency optimization: By introducing a feature extraction method based on linear layers, redundant calculations in traditional embedding layers are avoided. At the same time, the transformer architecture is used to process complex data, thus optimizing the efficiency and performance of the algorithm.
[0121] Superior predictive performance compared to classical algorithms: Through evaluation of clinical drug response data for colon cancer and bladder cancer, the algorithm of this invention significantly outperforms classical algorithms such as SVM, MLP and RandomForest in predicting patient drug response accuracy, especially in cross-domain prediction from cell line to three-dimensional model data.
[0122] Advancing Precision Medicine: This invention effectively helps differentiate drug response levels among patients, particularly in the use of major clinical drugs for colorectal, bladder, and liver cancer (such as 5-fluorouracil, oxaliplatin, gemcitabine, carboplatin, and sorafenib), thus contributing to the development of precision medicine. Compared to traditional algorithms, this invention provides more accurate individualized drug response prediction, enabling physicians to develop more precise treatment plans.
[0123] The training method for the drug response prediction model has been provided above. Below, we will provide a method for predicting drug response based on the trained drug response prediction model:
[0124] Figure 4 This is a schematic diagram of a drug response prediction method provided in an embodiment of this application, as shown below. Figure 4 As shown, the execution module can acquire input instructions; the input instructions include: cell gene expression information and drug molecule structure information; the input instructions are input into the drug response prediction model trained by the above training method to obtain output instructions; the output instructions include: drug response prediction information, such as drug sensitivity scores, etc.
[0125] It should be noted that this application does not limit the type of drug. For example, the aforementioned drugs may be selected from oncology drugs, cardiovascular disease drugs, etc.
[0126] Compared to existing machine learning algorithms (such as SVM, MLP, and RandomForest), the drug response model based on this application exhibits better performance and higher prediction accuracy in predicting drug responses, and is better able to distinguish patients' clinical responses to drugs. This model can effectively contribute to precision medicine, helping doctors provide personalized treatment plans for patients.
[0127] The preferred embodiments of this application have been described in detail above with reference to the accompanying drawings. However, this application is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this application, various simple modifications can be made to the technical solutions of this application, and these simple modifications all fall within the protection scope of this application. For example, the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, this application will not describe the various possible combinations separately. Furthermore, various different embodiments of this application can also be arbitrarily combined, as long as they do not violate the spirit of this application, they should also be considered as the content disclosed in this application.
[0128] It should also be understood that, in the various method embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.
[0129] The methods provided in the embodiments of this application have been described above. The model training device and drug response prediction device provided in the embodiments of this application will be described below.
[0130] Figure 5 This is a schematic diagram of a model training device 500 provided in an embodiment of this application, as shown below. Figure 5 As shown, the model training device 500 includes: a training data acquisition module 510, a pre-training module 520, and a fine-tuning module 530. The training data acquisition module 510 is used to acquire training data, which includes: cell gene expression information, three-dimensional model gene expression information, and drug molecule structure information. The pre-training module 520 is used to pre-train the deep learning model using the cell gene expression information and drug molecule structure information to obtain a pre-trained drug response prediction model. The fine-tuning module 530 is used to fine-tune the pre-trained drug response prediction model using the three-dimensional model gene expression information and drug molecule structure information to obtain a drug response prediction model.
[0131] In some examples of this application, the pre-training module 520 is specifically applied to: pre-training a deep learning model using cell gene expression information and drug molecule structure information that have undergone the first preprocessing, to obtain a pre-trained drug response prediction model. The first preprocessing includes: representing cell gene expression information in a transcription millions format; performing a logarithmic transformation on the transcription millions format of cell gene expression information to obtain the cell gene expression information input of the deep learning model; converting the drug molecule structure into a standard expression paradigm; and performing byte-pair encoding on the standard expression paradigm to obtain the drug molecule structure information input of the deep learning model.
[0132] In some examples of this application, the pre-training module 520 is specifically applied to: extracting features from the cell gene expression information after the first preprocessing through a two-layer fully connected network to obtain cell gene feature vectors; extracting features from the drug molecule structure information after the first preprocessing through a single-layer fully connected network to obtain drug feature vectors; and pre-training a deep learning model based on the cell gene feature vectors and drug feature vectors to obtain a pre-trained drug response prediction model.
[0133] In some examples of this application, the pre-training module 520 is specifically applied to: concatenating cell gene feature vectors and drug feature vectors to obtain cell-drug concatenated feature vectors; performing dimensionality upscaling on the cell-drug concatenated feature vectors; and using the dimensionality upscaled cell-drug concatenated feature vectors to pre-train a deep learning model to obtain a pre-trained drug response prediction model.
[0134] In some examples of this application, the pre-training module 520 is specifically applied to: calculating the loss of the deep learning model based on the difference between the output value of the pre-trained drug response prediction model and the first true value; and pre-training the deep learning model based on the loss to obtain the pre-trained drug response prediction model.
[0135] In some examples of this application, the pre-training module 520 is specifically used to: determine a first true value based on the AUC value of the effect of drug concentration on cell effect.
[0136] In some examples of this application, the fine-tuning module 530 is specifically used to: fine-tune the pre-trained drug response prediction model using the gene expression information and drug molecule structure information of the three-dimensional model after the second preprocessing, so as to obtain the drug response prediction model.
[0137] In some examples of this application, the fine-tuning module 530 is specifically applied to: representing the gene expression information of the three-dimensional model in a transcription millions format; performing a logarithmic transformation on the transcription millions format to obtain the three-dimensional model gene expression information input of the pre-trained drug response prediction model; converting the drug molecular structure into a standard expression paradigm; and performing byte-pair encoding on the standard expression paradigm to obtain the drug molecular structure information input of the pre-trained drug response prediction model.
[0138] In some examples of this application, the fine-tuning module 530 is specifically applied to: extracting features from the gene expression information of the three-dimensional model after the second preprocessing through a two-layer fully connected network to obtain the gene feature vector of the three-dimensional model; extracting features from the drug molecular structure information after the second preprocessing through a single-layer fully connected network to obtain the drug feature vector; and fine-tuning the pre-trained drug response prediction model based on the gene feature vector of the three-dimensional model and the drug feature vector to obtain the drug response prediction model.
[0139] In some examples of this application, the fine-tuning module 530 is specifically applied to: splicing the gene feature vector and drug feature vector of the three-dimensional model to obtain the three-dimensional model-drug spliced feature vector; performing dimensionality-upgrading processing on the three-dimensional model-drug spliced feature vector, and using the dimensionality-upgraded three-dimensional model-drug spliced feature vector to fine-tune the pre-trained drug response prediction model to obtain the drug response prediction model.
[0140] In some examples of this application, the fine-tuning module 530 is specifically applied to: calculating the loss of the drug response prediction model based on the difference between the output value of the drug response prediction model and the second true value; and fine-tuning the drug response prediction model based on the loss to obtain the drug response prediction model.
[0141] In some examples of this application, the fine-tuning module 530 is specifically used to determine a second true value by means of the AUC value of the effect of drug concentration on the three-dimensional model.
[0142] In some examples of this application, the fine-tuning module 530 is specifically used to fine-tune the drug response prediction model using at least one of L2 regularization or early stop strategy.
[0143] It should be understood that the device embodiments and method embodiments can correspond to each other, and similar descriptions can be referred to the method embodiments. To avoid repetition, further details will not be provided here. Specifically, Figure 5 The model training device 500 shown can perform Figure 2 The corresponding method embodiments, and the foregoing and other operations and / or functions of each module in the model training device 500 are respectively for implementing Figure 2 For the sake of brevity, the corresponding processes in each method are not described in detail here.
[0144] The model training apparatus 500 of this application embodiment has been described above from the perspective of functional modules in conjunction with the accompanying drawings. It should be understood that this functional module can be implemented in hardware, in software instructions, or in a combination of hardware and software modules. Specifically, the steps of the method embodiments in this application can be completed by the integrated logic circuits in the processor's hardware and / or by software instructions. The steps of the method disclosed in this application embodiment can be directly manifested as execution by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. Optionally, the software module can be located in a mature storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps in the above method embodiments.
[0145] Figure 6 This is a schematic diagram of a drug response prediction device 600 provided in an embodiment of this application, as shown below. Figure 6 As shown, the drug response prediction device 600 includes an information acquisition module 610 and a prediction module 620. The information acquisition module 610 is used to acquire input instructions, which include cell gene expression information and drug molecule structure information. The prediction module 620 is used to input the input instructions into the drug response prediction model trained by the training method provided in this application to obtain output instructions, which include drug response prediction information, such as drug sensitivity scores.
[0146] It should be understood that the device embodiments and method embodiments can correspond to each other, and similar descriptions can be referred to the method embodiments. To avoid repetition, further details will not be provided here. Specifically, Figure 6 The drug response prediction device 600 shown can perform... Figure 3The corresponding method embodiments, and the foregoing and other operations and / or functions of each module in the drug response prediction device 600, are respectively for implementing Figure 3 For the sake of brevity, the corresponding processes in each method are not described in detail here.
[0147] The drug response prediction device 600 of this application embodiment has been described above from the perspective of functional modules in conjunction with the accompanying drawings. It should be understood that this functional module can be implemented in hardware, in software instructions, or in a combination of hardware and software modules. Specifically, the steps of the method embodiments in this application can be completed by the integrated logic circuits in the processor's hardware and / or by software instructions. The steps of the method disclosed in this application embodiment can be directly manifested as execution by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. Optionally, the software module can be located in a mature storage medium in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps in the above method embodiments.
[0148] Figure 7 This is a schematic block diagram of the electronic device 700 provided in an embodiment of this application. The electronic device 700 may be the aforementioned training device or execution module, but is not limited thereto. Figure 7 As shown, the electronic device 700 may include:
[0149] The system includes a memory 710 and a processor 720. The memory 710 stores a computer program 730 and transfers the computer program 730 to the processor 720. In other words, the processor 720 can retrieve and run the computer program 730 from the memory 710 to implement the methods described in the embodiments of this application.
[0150] For example, the processor 720 can be used to execute the steps in the above method according to the instructions in the computer program 730.
[0151] In some embodiments of this application, the processor 720 may include, but is not limited to:
[0152] General-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
[0153] In some embodiments of this application, the memory 710 includes, but is not limited to:
[0154] Volatile memory and / or non-volatile memory. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), and Direct Rambus RAM (DR RAM).
[0155] In some embodiments of this application, the computer program 730 may be divided into one or more modules, which are stored in the memory 710 and executed by the processor 720 to perform the method provided in this application. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program 730 in the electronic device.
[0156] like Figure 7 As shown, the electronic device 700 may further include:
[0157] Transceiver 740, which can be connected to processor 720 or memory 710.
[0158] The processor 720 can control the transceiver 740 to communicate with other devices; specifically, it can send information or data to other devices or receive information or data sent by other devices. The transceiver 740 may include a transmitter and a receiver. The transceiver 740 may further include antennas, and the number of antennas may be one or more.
[0159] It should be understood that the various components in the electronic device 700 are connected through a bus system, which includes a data bus, a power bus, a control bus, and a status signal bus.
[0160] According to one aspect of this application, a computer storage medium is provided that stores a computer program thereon, which, when executed by a computer, enables the computer to perform the methods of the above-described method embodiments. Alternatively, embodiments of this application also provide a computer program product containing instructions that, when executed by a computer, cause the computer to perform the methods of the above-described method embodiments.
[0161] According to another aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the method described in the above-described method embodiments.
[0162] In other words, when implemented using software, it can be implemented entirely or partially as a computer program product. This computer program product includes one or more computer instructions. When these computer program instructions are loaded and executed on a computer, all or part of the processes or functions according to the embodiments of this application are performed. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., digital video disc (DVD)), or a semiconductor medium (e.g., solid-state disk (SSD)).
[0163] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0164] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.
[0165] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to implement the solution of this embodiment according to actual needs. For example, the functional modules in the various embodiments of this application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.
[0166] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.
[0167] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.
[0168] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application without departing from the principles and spirit of this application.
Claims
1. A model training method, characterized in that, include: Acquire training data, which includes: cell gene expression information, three-dimensional model gene expression information, and drug molecule structure information; The cell gene expression information and the drug molecule structure information are used to pre-train the deep learning model to obtain a pre-trained drug response prediction model. The pre-trained drug response prediction model is fine-tuned using the gene expression information of the three-dimensional model and the drug molecular structure information to obtain the drug response prediction model.
2. The method according to claim 1, characterized in that, The step of pre-training a deep learning model using the cell gene expression information and the drug molecule structure information to obtain a pre-trained drug response prediction model includes: The deep learning model is pre-trained using the cell gene expression information and drug molecule structure information after the first preprocessing to obtain a pre-trained drug response prediction model.
3. The method according to claim 2, characterized in that, The first preprocessing includes: The cellular gene expression information is represented in a transcription millions format; Logarithmic transformation is performed on the transcription of the cell gene expression information in millions of formats to obtain the cell gene expression information input for the deep learning model; The drug molecular structure was converted into a standard expression paradigm; The standard expression paradigm is byte-pair encoded to obtain the drug molecule structure information input to the deep learning model.
4. The method according to claim 2 or 3, characterized in that, The process of pre-training a deep learning model using the cell gene expression information and drug molecule structure information after the first preprocessing to obtain a pre-trained drug response prediction model includes: The cell gene expression information after the first preprocessing is extracted using a two-layer fully connected network to obtain cell gene feature vectors. The drug molecule structure information after the first preprocessing is extracted using a single-layer fully connected network to obtain a drug feature vector. The deep learning model is pre-trained based on the cell gene feature vector and the drug feature vector to obtain a pre-trained drug response prediction model.
5. The method according to claim 4, characterized in that, The step of pre-training the deep learning model based on the cell gene feature vector and the drug feature vector to obtain a pre-trained drug response prediction model includes: The cell gene feature vector and the drug feature vector are concatenated to obtain a cell-drug concatenated feature vector. The cell-drug splicing feature vector is subjected to dimensionality increase processing, and the deep learning model is pre-trained using the dimensionality-increased cell-drug splicing feature vector to obtain a pre-trained drug response prediction model.
6. The method according to claim 5, characterized in that, The step of pre-training the deep learning model using the upgraded cell-drug spliced feature vector to obtain a pre-trained drug response prediction model includes: The loss of the deep learning model is calculated based on the difference between the output value of the pre-trained drug response prediction model and the first true value. Based on the loss, the deep learning model is pre-trained to obtain a pre-trained drug response prediction model.
7. The method according to claim 6, characterized in that, The first true value was determined by the AUC value of the effect of drug concentration on cell effect.
8. The method according to claim 1, characterized in that, The step of fine-tuning the pre-trained drug response prediction model using the gene expression information of the three-dimensional model and the drug molecule structure information to obtain the drug response prediction model includes: The pre-trained drug response prediction model is fine-tuned using the gene expression information of the three-dimensional model after the second preprocessing and the drug molecular structure information to obtain the drug response prediction model.
9. The method according to claim 8, characterized in that, The second preprocessing includes: The gene expression information of the three-dimensional model is represented in a transcription millions format; Logarithmic transformation is performed on the millions of transcriptions to obtain the three-dimensional model gene expression information input of the pre-trained drug response prediction model; The drug molecular structure was converted into a standard expression paradigm; The standard expression paradigm is byte-pair encoded to obtain the drug molecular structure information input of the pre-trained drug response prediction model.
10. The method according to claim 8 or 9, characterized in that, The process of fine-tuning the pre-trained drug response prediction model using the gene expression information of the three-dimensional model after the second preprocessing and the drug molecule structure information to obtain the drug response prediction model includes: The gene expression information of the three-dimensional model after the second preprocessing is extracted by a two-layer fully connected network to obtain the gene feature vector of the three-dimensional model. The drug molecule structure information after the second preprocessing is extracted using a single-layer fully connected network to obtain a drug feature vector. The pre-trained drug response prediction model is fine-tuned based on the gene feature vector of the three-dimensional model and the drug feature vector to obtain the drug response prediction model.
11. The method according to claim 10, characterized in that, The step of fine-tuning the pre-trained drug response prediction model based on the gene feature vector of the three-dimensional model and the drug feature vector to obtain the drug response prediction model includes: The gene feature vector of the three-dimensional model and the drug feature vector are spliced together to obtain the three-dimensional model-drug spliced feature vector. The dimensionality of the three-dimensional model-drug splicing feature vector is increased, and the pre-trained drug response prediction model is fine-tuned using the increased dimensionality three-dimensional model-drug splicing feature vector to obtain the drug response prediction model.
12. The method according to claim 11, characterized in that, The process of fine-tuning the pre-trained drug response prediction model using the upgraded 3D model and drug-assembled feature vectors to obtain the drug response prediction model includes: The loss of the drug response prediction model is calculated based on the difference between the output value of the drug response prediction model and the second true value. Based on the aforementioned loss, the drug response prediction model is fine-tuned to obtain the drug response prediction model.
13. The method according to claim 12, characterized in that, The second true value was determined by the AUC value of the effect of drug concentration on the three-dimensional model.
14. The method according to claim 12 or 13, characterized in that, The drug response prediction model is fine-tuned using at least one of L2 regularization or early discontinuation strategies.
15. A method for predicting drug response, characterized in that, include: To obtain information on cellular gene expression and drug molecular structure; The cell gene expression information and drug molecular structure information are input into a trained drug response prediction model to obtain drug response prediction information; The trained drug response prediction model is obtained by training using the method described in any one of claims 1-14.
16. A model training device, characterized in that, include: The training data acquisition module is used to acquire training data, which includes: cell gene expression information, three-dimensional model gene expression information, and drug molecule structure information. The pre-training module is used to pre-train the deep learning model using the cell gene expression information and the drug molecule structure information to obtain a pre-trained drug response prediction model. The fine-tuning module is used to fine-tune the pre-trained drug response prediction model using the gene expression information of the three-dimensional model and the drug molecule structure information to obtain the drug response prediction model.
17. A drug response prediction device, characterized in that, include: The information acquisition module is used to acquire cell gene expression information and drug molecule structure information; The prediction module is used to input the cell gene expression information and drug molecule structure information into the trained drug response prediction model to obtain drug response prediction information. The trained drug response prediction model is obtained by training using the method described in any one of claims 1-14.
18. A computing device, characterized in that, include: Processor and memory; The memory is used to store computer programs; The processor is configured to execute the computer program to implement the model training method as described in any one of claims 1-14 or the drug response prediction method as described in claim 15.
19. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions or programs that, when executed on a computer, cause the model training method as described in any one of claims 1-14 or the drug response prediction method as described in claim 15 to be performed.
20. A computer program product, characterized in that, include: Computer instructions; When some or all of the computer instructions are executed on a computer, the model training method as described in any one of claims 1-14 or the drug response prediction method as described in claim 15 is performed.