Data processing, training method and device of data processing model, equipment and medium

Abstract

The application discloses a data processing method and device, a data processing model training method and device, equipment and a medium, and belongs to the technical field of computers. The method comprises the following steps: extracting target expression values corresponding to each candidate gene under a target cell from target gene expression data of the target cell; obtaining target features corresponding to each candidate gene based on the target expression values corresponding to each candidate gene; extracting interaction information of each candidate gene under the target cell based on the target features corresponding to each candidate gene; and obtaining a prediction result corresponding to the target cell based on the interaction information of each candidate gene under the target cell. In this way, the prediction result corresponding to the target cell is obtained based on the interaction information of each candidate gene under the target cell, the interaction information of each candidate gene under the target cell can reflect the functional characteristics of the target cell, and the accuracy of the obtained prediction result is higher.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a data processing method, apparatus, device, and medium for training data processing models. Background Technology

[0002] With the development of computer technology, cell research has become increasingly widespread. For example, the transcriptome of a single cell is sequenced to obtain the cell's gene expression data, and then the corresponding prediction results (such as classification results, regression results, etc.) are obtained based on the cell's gene expression data.

[0003] In related technologies, researchers determine specific genes that match various prediction results based on experience. In the process of determining the prediction result corresponding to a cell, the expression value of a specific gene in that cell is extracted from the gene expression data of that cell. If the expression value meets the high expression condition, the prediction result that matches the specific gene is taken as the prediction result corresponding to that cell.

[0004] This data processing relies on the researcher's prior knowledge and contains many unstable factors. In addition, the data processing depends on individual specific genes, and the absence or noise of these individual specific genes has a significant impact on the accuracy of the prediction results, making it difficult to obtain relatively accurate prediction results. Summary of the Invention

[0005] This application provides a data processing method, apparatus, device, and medium for training a data processing model, which can be used to improve the stability of data processing and the accuracy of the obtained prediction results. The technical solution is as follows:

[0006] On one hand, embodiments of this application provide a data processing method, the method comprising:

[0007] Obtain target gene expression data of target cells, and extract the target expression values ​​of each candidate gene in the target cells from the target gene expression data;

[0008] Based on the target expression values ​​corresponding to each candidate gene, the target features corresponding to each candidate gene are obtained.

[0009] Based on the target features corresponding to each candidate gene, the interaction information of each candidate gene under the target cell is extracted, and the interaction information of each candidate gene under the target cell is used to characterize the gene interaction features of the target cell.

[0010] Based on the interaction information of each candidate gene in the target cell, the prediction result corresponding to the target cell is obtained.

[0011] A method for training a data processing model is also provided, the method comprising:

[0012] Obtain the gene expression data of the sample cells and the standard results corresponding to the sample cells, and extract the sample expression values ​​of each candidate gene under the sample cells from the gene expression data.

[0013] Based on the sample expression values ​​corresponding to each candidate gene, obtain the sample features corresponding to each candidate gene;

[0014] The first data processing model is invoked to extract the interaction information of each candidate gene in the sample cell based on the sample features corresponding to each candidate gene; based on the interaction information of each candidate gene in the sample cell, the prediction result corresponding to the sample cell is obtained.

[0015] Based on the prediction results and standard results corresponding to the sample cells, a result loss function is obtained; the first data processing model is trained using the result loss function to obtain the target data processing model.

[0016] On the other hand, a data processing apparatus is provided, the apparatus comprising:

[0017] The first acquisition unit is used to acquire target gene expression data of target cells and extract the target expression values ​​corresponding to each candidate gene under the target cells from the target gene expression data.

[0018] The second acquisition unit is used to acquire the target features corresponding to each candidate gene based on the target expression value corresponding to each candidate gene;

[0019] An extraction unit is used to extract the interaction information of each candidate gene in the target cell based on the target features corresponding to each candidate gene. The interaction information of each candidate gene in the target cell is used to characterize the gene interaction features of the target cell.

[0020] The third acquisition unit is used to acquire the prediction result corresponding to the target cell based on the interaction information of each candidate gene in the target cell.

[0021] In one possible implementation, the second acquisition unit is used to convert the target expression value corresponding to the first candidate gene into the expression value feature corresponding to the first candidate gene, wherein the first candidate gene is any one of the candidate genes; and to fuse the expression value feature corresponding to the first candidate gene and the characterization feature corresponding to the first candidate gene to obtain the target feature corresponding to the first candidate gene.

[0022] In one possible implementation, the second acquisition unit is configured to normalize the target expression value corresponding to the first candidate gene to obtain the normalized expression value corresponding to the first candidate gene; determine the target discretized expression value corresponding to the normalized expression value; and use the embedding feature corresponding to the target discretized expression value as the expression value feature corresponding to the first candidate gene.

[0023] In one possible implementation, the target discretized expression value is one of a number of candidate discretized expression values. The second acquisition unit is further configured to perform vectorization transformation on the number of candidate discretized expression values ​​to obtain the embedding features corresponding to the number of candidate discretized expression values ​​respectively.

[0024] In one possible implementation, the extraction unit is used to invoke a target data processing model to extract the interaction information of each candidate gene in the target cell based on the target features corresponding to each candidate gene;

[0025] The third acquisition unit is used to call the target data processing model to obtain the prediction result corresponding to the target cell based on the interaction information of each candidate gene in the target cell.

[0026] In one possible implementation, the prediction result corresponding to the target cell indicates that the category of the target cell is the target category, and the device further includes:

[0027] The determining unit is used to determine, based on the interaction information of each candidate gene in the target cell, the gene that meets the selection criteria and belongs to the cell of the target category.

[0028] A training apparatus for a data processing model is also provided, the apparatus comprising:

[0029] The first acquisition unit is used to acquire sample gene expression data of sample cells and standard results corresponding to the sample cells, and extract sample expression values ​​corresponding to each candidate gene under the sample cells from the sample gene expression data.

[0030] The second acquisition unit is used to acquire the sample features corresponding to each candidate gene based on the sample expression value corresponding to each candidate gene.

[0031] The extraction unit is used to call the first data processing model to extract the interaction information of each candidate gene in the sample cell based on the sample features corresponding to each candidate gene.

[0032] The third acquisition unit is used to acquire the prediction result corresponding to the sample cell based on the interaction information of each candidate gene under the sample cell;

[0033] The training unit is used to obtain a result loss function based on the prediction results and standard results corresponding to the sample cells; and to train the first data processing model using the result loss function to obtain the target data processing model.

[0034] In one possible implementation, the first data processing model includes a first extraction sub-model and a first prediction sub-model; the extraction unit is used to call the first extraction sub-model to extract information from the sample features corresponding to each candidate gene, and obtain the interaction information of each candidate gene under the sample cell.

[0035] The third acquisition unit is used to call the first prediction sub-model to process the interaction information of each candidate gene under the sample cell, and obtain the prediction result corresponding to the sample cell.

[0036] In one possible implementation, the first acquisition unit is further configured to acquire training gene expression data of training cells and extract the training expression values ​​corresponding to each candidate gene under the training cells from the training gene expression data.

[0037] The second acquisition unit is further configured to acquire the training features corresponding to each candidate gene based on the training expression values ​​corresponding to each candidate gene;

[0038] The device further includes:

[0039] The replacement unit is used to replace the training features corresponding to the candidate genes that meet the replacement conditions in each candidate gene with the reference features;

[0040] The extraction unit is further configured to call the initial extraction sub-model to extract information from the reference features corresponding to the candidate genes that meet the replacement conditions and the training features corresponding to the candidate genes that do not meet the replacement conditions, so as to obtain the interaction information of each candidate gene under the training cells; and to obtain the predicted features corresponding to the candidate genes that meet the replacement conditions based on the interaction information of each candidate gene under the training cells.

[0041] The training unit is further configured to obtain a feature loss function based on the predicted features and training features corresponding to the candidate genes that meet the replacement conditions, and to train the initial extraction sub-model using the feature loss function to obtain the first extraction sub-model.

[0042] In one possible implementation, the number of sample cells is at least one, and the sample gene expression data of one sample cell includes the sample expression values ​​corresponding to each measured gene under that sample cell. The device further includes:

[0043] The determining unit is used to count the number of sample cells that each measurement gene hits based on the sample gene expression data of each sample cell. A measurement gene hitting a sample cell is used to indicate that the sample expression value of the measurement gene under the sample cell is not less than a first threshold. Measurement genes that have hit a number of sample cells that are not less than the number threshold are selected as candidate genes.

[0044] In one possible implementation, the first extraction sub-model is a language model obtained by sequentially connecting at least one attention-based encoder.

[0045] On the other hand, a computer device is provided, the computer device including a processor and a memory, the memory storing at least one computer program, the at least one computer program being loaded and executed by the processor to enable the computer device to implement any of the data processing methods or data processing model training methods described above.

[0046] On the other hand, a computer-readable storage medium is also provided, wherein at least one computer program is stored in the computer-readable storage medium, the at least one computer program being loaded and executed by a processor to enable a computer to implement any of the data processing methods or data processing model training methods described above.

[0047] On the other hand, a computer program product is also provided, which includes a computer program or computer instructions, which are loaded and executed by a processor to enable a computer to implement any of the data processing methods or data processing model training methods described above.

[0048] The technical solution provided in this application has at least the following beneficial effects:

[0049] The technical solution provided in this application automatically obtains the prediction results corresponding to the target cells based on the target gene expression data of the target cells, without relying on the researcher's prior knowledge, and the data processing stability is high. Furthermore, the prediction results corresponding to the target cells are obtained based on the interaction information of each candidate gene under the target cells. This interaction information can characterize the gene interaction features of the target cells. Since cells function in organisms through the interaction between genes, the gene interaction features can reflect the functional characteristics of the target cells. Therefore, the prediction results obtained by focusing on the functional characteristics of the target cells have high accuracy. Attached Figure Description

[0050] 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.

[0051] Figure 1 This is a schematic diagram of an implementation environment provided in an embodiment of this application;

[0052] Figure 2 This is a flowchart of a data processing method provided in an embodiment of this application;

[0053] Figure 3 This is a flowchart of a data processing method provided in an embodiment of this application;

[0054] Figure 4 This is a flowchart of a data processing model training method provided in an embodiment of this application;

[0055] Figure 5 This is a schematic diagram illustrating the process of training an initial extraction sub-model according to an embodiment of this application;

[0056] Figure 6 This is a schematic diagram illustrating the training process of a data processing model provided in an embodiment of this application;

[0057] Figure 7 This is a schematic diagram of a data processing apparatus provided in an embodiment of this application;

[0058] Figure 8 This is a schematic diagram of a training device for a data processing model provided in an embodiment of this application;

[0059] Figure 9 This is a schematic diagram of the structure of a server provided in an embodiment of this application;

[0060] Figure 10 This is a schematic diagram of the structure of a terminal provided in an embodiment of this application. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0062] In exemplary embodiments, the data processing method and data processing model training method provided in this application can be applied to various scenarios, including but not limited to cloud technology, artificial intelligence, smart transportation, and assisted driving.

[0063] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science. AI attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to possess the functions of perception, reasoning, and decision-making.

[0064] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies primarily include computer vision, speech processing, natural language processing, as well as machine learning / deep learning, autonomous driving, and intelligent transportation.

[0065] Machine learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and instruction-based learning.

[0066] With the research and advancement of artificial intelligence (AI) technology, AI is being studied and applied in various fields, such as smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, autonomous driving, drones, robots, smart healthcare, smart customer service, vehicle networking, and intelligent transportation. It is believed that with the development of technology, AI will be applied in more fields and play an increasingly important role.

[0067] Figure 1 A schematic diagram of an implementation environment provided in an embodiment of this application is shown. This implementation environment includes: a terminal 11 and a server 12.

[0068] The data processing method provided in this application embodiment can be executed by terminal 11, server 12, or jointly by terminal 11 and server 12; this application embodiment does not limit this. In the case where the data processing method provided in this application embodiment is jointly executed by terminal 11 and server 12, server 12 undertakes the main computational work, and terminal 11 undertakes the secondary computational work; or, server 12 undertakes the secondary computational work, and terminal 11 undertakes the main computational work; or, server 12 and terminal 11 use a distributed computing architecture for collaborative computation.

[0069] The training method for the data processing model provided in this application embodiment can be executed by terminal 11, server 12, or jointly by terminal 11 and server 12; this application embodiment does not limit this. In the case where the training method for the data processing model provided in this application embodiment is jointly executed by terminal 11 and server 12, server 12 undertakes the main computational work, and terminal 11 undertakes the secondary computational work; or, server 12 undertakes the secondary computational work, and terminal 11 undertakes the main computational work; or, server 12 and terminal 11 use a distributed computing architecture for collaborative computation.

[0070] The execution device of the data processing method and the execution device of the data processing model training method can be the same or different, and the embodiments of this application do not limit this.

[0071] In one possible implementation, terminal 11 can be any electronic product capable of human-computer interaction with a user through one or more methods such as a keyboard, touchpad, touchscreen, remote control, voice interaction, or handwriting device. Examples include PCs (Personal Computers), mobile phones, smartphones, PDAs (Personal Digital Assistants), wearable devices, PPCs (Pocket PCs), tablets, smart car systems, smart TVs, smart speakers, smart voice interaction devices, smart home appliances, and in-vehicle terminals. Server 12 can be a single server, a server cluster consisting of multiple servers, or a cloud computing service center. Terminal 11 and server 12 establish a communication connection via wired or wireless network.

[0072] Those skilled in the art should understand that the above-described terminal 11 and server 12 are merely examples. Other existing or future terminals or servers that are applicable to this application should also be included within the scope of protection of this application, and are hereby incorporated by reference.

[0073] Based on the above Figure 1 The implementation environment shown in this application embodiment provides a data processing method, which is executed by a computer device. This computer device can be a terminal 11 or a server 12; this application embodiment does not limit the specific type of computer device. Figure 2 As shown, the data processing method provided in this application embodiment includes the following steps 201 to 203.

[0074] In step 201, target gene expression data of target cells are obtained, and target expression values ​​corresponding to each candidate gene under the target cells are extracted from the target gene expression data.

[0075] The target cell refers to the cell from which the prediction result is to be obtained, and the target gene expression data of the target cell is used to indicate the gene expression status of the target cell. In an exemplary embodiment, the target gene expression data of the target cell is obtained by sequencing the transcriptome of the target cell. The transcriptome of the target cell refers to the total expression level of all mRNAs (messenger ribonucleic acid) in the target cell at a certain time. There is a one-to-one correspondence between mRNAs and genes. By sequencing the transcriptome of the target cell, the target gene expression data of the target cell can be determined. The sequencing technology used to sequence the transcriptome of the target cell can be flexibly adjusted according to the actual application scenario, and this embodiment of the application does not limit it.

[0076] The target gene expression data of the target cell includes the target expression values ​​corresponding to each measurement gene in the target cell. The target expression value of any measurement gene is used to indicate the number of mRNAs corresponding to that measurement gene in the transcriptome of the target cell. Measurement genes refer to genes of interest during sequencing. This application does not limit the type and number of measurement genes. When different sequencing technologies are used to sequence the transcriptome of the target cell, the type and number of measurement genes may be the same or different.

[0077] In an exemplary embodiment, the target gene expression data of the target cell can be pre-acquired and stored. In this way, when the prediction result corresponding to the target cell is needed, the target gene expression data of the target cell is directly extracted. In an exemplary embodiment, the computer device pre-stores the transcriptome of the target cell. In this case, when the prediction result corresponding to the target cell is needed, the target gene expression data of the target cell is obtained by sequencing the transcriptome of the target cell. In an exemplary embodiment, if the computer device does not store the transcriptome and target gene expression data of the target cell, the target gene expression data of the target cell is obtained by: transcribing the target cell to obtain the transcriptome of the target cell; and sequencing the transcriptome of the target cell to obtain the target gene expression data of the target cell.

[0078] After obtaining the target gene expression data of the target cells, the target expression values ​​corresponding to each candidate gene in the target cells are extracted from the target gene expression data. Candidate genes refer to genes of interest during the process of obtaining the prediction results for the target cells; for example, candidate genes are some or all of the sequenced genes. Since the target gene expression data includes the target expression values ​​corresponding to each sequenced gene in the target cells, the target expression values ​​corresponding to each candidate gene in the target cells can be extracted from the target gene expression data after acquisition. The target expression value corresponding to any candidate gene is used to indicate the number of mRNAs corresponding to that candidate gene in the transcriptome of the target cell.

[0079] For example, candidate genes are set based on experience or flexibly adjusted according to the application scenario, and this application embodiment does not limit this. For example, candidate genes are highly reliable genes determined by analyzing gene expression data from a large number of cells.

[0080] In step 202, the target features corresponding to each candidate gene are obtained based on the target expression value of each candidate gene.

[0081] The target features corresponding to each candidate gene are the features used to obtain the prediction results for the target cells. The target features corresponding to each candidate gene are determined based on the target expression value of each candidate gene. This application does not limit the form of the target features corresponding to candidate genes. For example, the target features corresponding to candidate genes can be in the form of a multi-dimensional vector. The dimensions of this multi-dimensional vector can be set empirically or flexibly adjusted according to the application scenario.

[0082] The target feature corresponding to a candidate gene is obtained based on the target expression value of that candidate gene. Taking the first candidate gene as an example, the process of obtaining the target feature corresponding to the first candidate gene based on its target expression value is described. Here, the first candidate gene is any one of the candidate genes. In one possible implementation, the process of obtaining the target feature corresponding to the first candidate gene based on its target expression value includes steps 2021 and 2022.

[0083] In step 2021, the target expression value corresponding to the first candidate gene is converted into the expression value feature corresponding to the first candidate gene.

[0084] The expression value features corresponding to the first candidate gene are used to characterize the target expression value corresponding to the first candidate gene. In one possible implementation, the process of converting the target expression value corresponding to the first candidate gene into the expression value features corresponding to the first candidate gene includes the following steps 1 and 2.

[0085] Step 1: Normalize the target expression value corresponding to the first candidate gene to obtain the normalized expression value corresponding to the first candidate gene.

[0086] Normalizing the target expression value corresponding to the first candidate gene helps to improve the standardization of the expression value corresponding to the first candidate gene. The value obtained after normalizing the target expression value corresponding to the first candidate gene is called the normalized expression value corresponding to the first candidate gene.

[0087] In one possible implementation, the process of normalizing the target expression value corresponding to the first candidate gene to obtain the normalized expression value of the first candidate gene includes: calculating the product of the target expression value corresponding to the first candidate gene and a first multiple to obtain the first expression value; and obtaining the normalized expression value corresponding to the first candidate gene based on the first expression value.

[0088] The first multiple is the ratio of the reference total to the first total, where the reference total is the sum of the expression values ​​of all candidate genes in a desired cell, and the first total is the sum of the expression values ​​of all candidate genes in the target cell. The reference total is set based on experience or can be flexibly adjusted according to the application scenario; this embodiment does not limit this. The first total is obtained by adding the target expression values ​​of all candidate genes in the target cell.

[0089] It should be noted that the first factor is the same for the target expression values ​​of each candidate gene. For example, assuming the reference total is 10,000 and the first total is 2,000, then the first factor is 5. By scaling up the target expression values ​​of each candidate gene by 5 times, the sum of the expression values ​​of each candidate gene can be made to reach the reference total through proportional scaling.

[0090] After obtaining the first expression value by multiplying the target expression value of the first candidate gene by a first fold, the normalized expression value of the first candidate gene is obtained based on the first expression value. In an exemplary embodiment, the first expression value is directly used as the normalized expression value of the first candidate gene. In an exemplary embodiment, the logarithm of the first expression value is used as the normalized expression value of the first candidate gene. By taking the logarithm of the first expression value, the long-tail distribution of the first expression value can be adjusted, further improving the normalization of the obtained normalized expression value.

[0091] Step 2: Determine the target discretized expression value corresponding to the normalized expression value, and use the embedding feature corresponding to the target discretized expression value as the expression value feature corresponding to the first candidate gene.

[0092] After determining the normalized expression value corresponding to the first candidate gene, the target discrete expression value corresponding to that normalized expression value is further determined. In an exemplary embodiment, the computer device stores a correspondence between normalized expression values ​​and candidate discrete expression values, and the target discrete expression value can be directly determined based on this correspondence.

[0093] In an exemplary embodiment, the computer device stores a correspondence between candidate discretized expression values ​​and normalized expression value ranges. In this case, the target normalized expression value range to which the normalized expression value belongs is first determined, and then the candidate discretized expression value corresponding to the target normalized expression value range is used as the target discretized expression value.

[0094] In an exemplary embodiment, the method for obtaining the correspondence between candidate discretized expression values ​​and normalized expression value ranges includes: dividing the candidate normalized expression values ​​into a reference number of statistical bins according to their numerical order from largest to smallest or smallest to largest; setting a candidate discretized expression value for each statistical bin; and using the normalized expression value range formed by the normalized expression values ​​of each candidate in that statistical bin as the normalized expression value range corresponding to the candidate discretized expression value set for that statistical bin. In this way, a reference number of candidate discretized expression values ​​can be set, and each candidate discretized expression value corresponds to a normalized expression value range.

[0095] Candidate normalized expression values ​​refer to the normalized expression values ​​that a gene may correspond to. For example, candidate normalized expression values ​​refer to continuous variables. The number and specific values ​​of candidate normalized expression values ​​are determined based on experience or actual circumstances, and this application embodiment does not limit this. In an exemplary embodiment, dividing candidate normalized expression values ​​into a reference number of statistical bins according to their numerical order (from largest to smallest or smallest to largest) can mean uniformly dividing the candidate normalized expression values ​​into the reference number of statistical bins according to their numerical order (from largest to smallest or smallest to largest), or it can mean non-uniformly dividing the candidate normalized expression values ​​into the reference number of statistical bins according to their numerical order (from largest to smallest or smallest to largest). This application embodiment does not limit this.

[0096] The reference number is used to constrain the number of candidate discretized expression values. The reference number is set based on experience or can be flexibly adjusted according to the application scenario. This application embodiment does not limit this.

[0097] For example, if the reference number is 50, then the number of candidate discretized expression values ​​is 50. Assuming there are 5000 candidate normalized expression values, these 5000 values ​​are evenly distributed into 50 statistical buckets in ascending order of value, with 1000 candidate normalized expression values ​​per bucket. Then, if any candidate normalized expression value is assigned to the x-th bucket (x is a positive integer not greater than 50), and the candidate discretized expression value corresponding to the x-th bucket is set to Bx, then the candidate discretized expression value corresponding to each candidate normalized expression value assigned to that x-th bucket is also Bx. For example, if a normalized expression value is zero, then the corresponding discretized expression value is denoted as Zero.

[0098] After determining the target discretized expression value, the embedding feature corresponding to the target discretized expression value is used as the expression value feature of the first candidate gene. The target discretized expression value is a numerical value, and the embedding feature corresponding to the target discretized expression value refers to a multi-dimensional feature corresponding to that target discretized expression value.

[0099] In one possible implementation, the target discretized expression value is one of a number of candidate discretized expression values ​​that are referenced. The computer device stores the embedding features corresponding to each candidate discretized expression value. By querying, it can be determined which embedding feature corresponds to the target discretized expression value, thereby obtaining the expression value feature corresponding to the first candidate gene.

[0100] For example, the method for obtaining the embedding features corresponding to the reference number of candidate discretized expression values ​​is as follows: the reference number of candidate discretized expression values ​​are vectorized to obtain the embedding features corresponding to the reference number of candidate discretized expression values.

[0101] A reference number of candidate discretized expression values ​​are used to discretize the normalized expression value. These reference number of candidate discretized expression values ​​can be set empirically or flexibly adjusted according to the application scenario; this embodiment does not limit this. By performing vectorization transformation on the reference number of candidate discretized expression values, the embedded features corresponding to each candidate discretized expression value can be obtained. For example, the embedded features corresponding to each candidate discretized expression value are vectors of the same dimension, such as 200-dimensional vectors.

[0102] For example, word vectorization is used to vectorize a reference number of candidate discretized expression values, transforming each candidate discretized expression value into a multi-dimensional vector. This multi-dimensional vector serves as the embedded feature corresponding to the candidate discretized expression value. Word vectorization is a technique for generating multiple embedded features. For example, word vectorization can be implemented using the embedding function in PyTorch (an open-source machine learning library). For example, the principle of vectorization is: the reference number of candidate discretized expression values ​​are transformed into a reference number of embedded features, and the distance between any two embedded features is the same as the distance between any two other embedded features. Through vectorization, each candidate discretized expression value Bx is transformed into a vector form similar to [0.14, -0.33, ..., 0.75, 0.28]. The dimension of this vector can be flexibly set; for example, the dimension of this vector is 200.

[0103] It should be noted that the process of obtaining the expression value features corresponding to the first candidate gene based on the target expression value of the first candidate gene described above is merely an exemplary example, and the embodiments of this application are not limited thereto. In an exemplary embodiment, the discrete expression value corresponding to the target expression value of the first candidate gene can also be directly determined, and the embedding feature corresponding to the discrete expression value can be used as the expression value feature of the first candidate gene. For example, different discrete expression values ​​correspond to different expression value ranges. The method for determining the discrete expression value corresponding to the target expression value of the first candidate gene is as follows: the discrete expression value corresponding to the expression value range to which the target expression value of the first candidate gene belongs is used as the discrete expression value corresponding to the target expression value of the first candidate gene.

[0104] In step 2022, the target features corresponding to the first candidate gene are obtained based on the expression value features corresponding to the first candidate gene.

[0105] After obtaining the expression value features of the first candidate gene, the target features of the first candidate gene are further obtained based on the expression value features of the first candidate gene.

[0106] In one possible implementation, step 2022 is implemented by directly using the expression value feature corresponding to the first candidate gene as the target feature corresponding to the first candidate gene.

[0107] In another possible implementation, step 2022 includes fusing the expression value feature and the characterization feature corresponding to the first candidate gene to obtain the target feature corresponding to the first candidate gene.

[0108] The representational features corresponding to the first candidate gene are used to represent the first candidate gene itself and are independent of the cell. Exemplarily, the representational features corresponding to the first candidate gene are determined based on the semantic information of the first candidate gene. This application does not limit the method of obtaining the representational features corresponding to the first candidate gene. Exemplarily, the representational features corresponding to the first candidate gene are features extracted based on the Gene2Vec method (a gene representation method). The higher the similarity between the representational features corresponding to any two candidate genes extracted based on the Gene2Vec method, the stronger the interaction between the two candidate genes. The dimension of the features extracted based on the Gene2Vec method can be flexibly adjusted according to the application scenario. For example, for the gene EGFR, the features extracted based on the Gene2Vec method are a 200-dimensional vector [0.76, 0.23, ..., -0.49, 0.15]. Of course, in the exemplary embodiment, the representational features corresponding to the first candidate gene can also be extracted based on other methods, as long as they can represent the first candidate gene itself.

[0109] By comprehensively considering the expression value characteristics and the characterization characteristics of the first candidate gene, the target features corresponding to the first candidate gene can be obtained. This can provide richer information for the subsequent process of obtaining the prediction results of the target cells, which is conducive to improving the reliability of the obtained prediction results.

[0110] In an exemplary embodiment, the expression value feature corresponding to the first candidate gene and the characterization feature corresponding to the first candidate gene have the same representation form. The target feature corresponding to the first candidate gene obtained by fusing the expression value feature corresponding to the first candidate gene and the characterization feature corresponding to the first candidate gene has the same representation form as the expression value feature corresponding to the first candidate gene and the characterization feature corresponding to the first candidate gene.

[0111] For example, both the expression value feature and the characterization feature corresponding to the first candidate gene are vectors of a specified dimension. The process of fusing the expression value feature and the characterization feature corresponding to the first candidate gene is achieved by adding or averaging the elements at the same positions in the two vectors. The target feature corresponding to the first candidate gene obtained by fusion is also a vector of a specified dimension. The specified dimension is set empirically or flexibly adjusted according to the application scenario, and this embodiment of the application does not limit it. For example, the specified dimension is 200 dimensions.

[0112] In step 203, based on the target features corresponding to each candidate gene, the interaction information of each candidate gene in the target cell is extracted. The interaction information of each candidate gene in the target cell is used to characterize the gene interaction features of the target cell. Based on the interaction information of each candidate gene in the target cell, the prediction result corresponding to the target cell is obtained.

[0113] After obtaining the target features corresponding to each candidate gene, interaction information of each candidate gene in the target cell is extracted based on these features. This interaction information focuses not only on the expression of each candidate gene itself in the target cell but also on the differences in expression between each candidate gene and other candidate genes. The expression of each candidate gene itself and the differences in expression between them can characterize the gene-interaction features of the target cell. In other words, the interaction information of each candidate gene in the target cell is used to characterize the gene-interaction features of the target cell. Since the function of the target cell in the organism is achieved through gene-gene interactions, the gene-interaction features of the target cell are used to describe its function in the organism; these features can be considered as the essential characteristics of the target cell. Based on the interaction information used to characterize the gene-interaction features of the target cell, more accurate and reliable prediction results can be obtained.

[0114] In an exemplary embodiment, the interaction information of each candidate gene in the target cell can indicate the interaction information between each candidate gene and other candidate genes in the target cell. In an exemplary embodiment, the interaction information of each candidate gene in the target cell is in the form of a weight matrix, the dimension of which is the number of candidate genes. The number of candidate genes is used to determine the pairwise interaction information of candidate genes in the target cell through this matrix. For example, the interaction information of two candidate genes in the target cell is represented by a weight value in the weight matrix. The larger the weight value, the stronger the interaction between the two candidate genes in the target cell; the smaller the weight value, the weaker the interaction between the two candidate genes in the target cell.

[0115] In an exemplary embodiment, the process of extracting the interaction information of each candidate gene in the target cell based on the target features corresponding to each candidate gene is as follows: The target features corresponding to each candidate gene are processed using an attention mechanism to obtain the interaction information of each candidate gene in the target cell. In this approach, the interaction information of each candidate gene in the target cell can also be referred to as an attention weight matrix.

[0116] In exemplary embodiments, the type of prediction result corresponding to the target cell is related to the actual application scenario, and this application embodiment does not limit this. For example, in an application scenario where target cells are classified based on their gene expression data, the prediction result corresponding to the target cell is the classification result corresponding to the target cell, which indicates the category of the target cell. For example, in an application scenario where target cells are regressed based on their gene expression data, the prediction result corresponding to the target cell is the regression result corresponding to the target cell, which indicates the regression value of the target cell. The meaning of the regression value is related to the research purpose of regressing the target cell. For example, if the research purpose of regressing the target cell is to study the development time of the target cell, then the regression value is used to indicate the development time of the target cell, thereby providing data support for downstream developmental trajectory analysis tasks. For example, the application scenario can also be a scenario where target cells are clustered based on their gene expression data. In this scenario, the prediction result corresponding to the target cell is the clustering result corresponding to the target cell.

[0117] In an exemplary embodiment, step 203 can be implemented by running a pre-edited computer program or by calling a target data processing model; this application embodiment does not limit the implementation in this way. This application embodiment uses the example of step 203 being implemented by calling a target data processing model for illustration. That is, the implementation of step 203 includes: calling the target data processing model to extract the interaction information of each candidate gene in the target cell based on the target features corresponding to each candidate gene; and calling the target data processing model to obtain the prediction result corresponding to the target cell based on the interaction information of each candidate gene in the target cell.

[0118] The target data processing model is a trained model capable of outputting relatively accurate predictions of target cells based on the target features corresponding to each candidate gene. The process of training the target data processing model is described in [link to training process]. Figure 4 The embodiments shown are not described in detail here.

[0119] The process of calling the target data processing model to obtain the prediction results corresponding to the target cells is an internal processing process of the target data processing model, which is related to the structure of the target data model. The embodiments of this application do not limit the model structure of the target data processing model.

[0120] In an exemplary embodiment, the target data processing model is a holistic model. In this case, by inputting the target features corresponding to each candidate gene into the target data processing model, the model automatically extracts the interaction information of each candidate gene in the target cell, and then directly obtains and outputs the prediction result corresponding to the target cell based on the interaction information of the candidate genes in the target cell. In an exemplary embodiment, inputting the target features corresponding to each candidate gene into the target data processing model means arranging the target features corresponding to each candidate gene in a specified order according to the candidate genes, forming a feature sequence, and then inputting the feature sequence into the target data processing model. The specified arrangement is set based on experience or flexibly adjusted according to the experience scenario; this embodiment does not limit this.

[0121] In an exemplary embodiment, the target data processing model includes a target extraction sub-model and a target prediction sub-model. See also: Figure 3 The implementation process of step 203 includes: 301, calling the target extraction sub-model to extract information from the target features corresponding to each candidate gene, and obtaining the interaction information of each candidate gene in the target cell; 302, calling the target prediction sub-model to process the interaction information of each candidate gene in the target cell, and obtaining the prediction result corresponding to the target cell.

[0122] In other words, the target features corresponding to each candidate gene are input into the target extraction sub-model, which extracts the interaction information of each candidate gene in the target cell. Then, the interaction information of each candidate gene in the target cell is input into the target prediction model, which obtains and outputs the prediction result corresponding to the target cell.

[0123] The process of calling the target extraction sub-model to extract information from the target features corresponding to each candidate gene and obtain the interaction information of each candidate gene in the target cell is the internal processing of the target extraction sub-model, which is related to the model structure of the target extraction sub-model.

[0124] For example, the target extraction sub-model is a language model obtained by sequentially connecting at least one attention-based encoder, such as BERT (Bidirectional Encoder Representations from Transformers), or a deep bidirectional language model obtained by improving the encoder in BERT. That is, the target extraction sub-model includes at least one attention-based encoder connected in sequence. For example, each attention-based encoder has the function of extracting interaction information from the input features and outputting features with the same feature dimension as the input based on the extracted interaction information. For example, each attention-based encoder includes an attention extraction layer and a feature output layer, wherein the attention extraction layer is used to extract interaction information from the input features, and the feature output layer is used to output features with the same feature dimension as the input based on the extracted interaction information.

[0125] In an exemplary embodiment, where the target extraction sub-model includes at least one attention-based encoder connected in sequence, the interaction information of each candidate gene under the target cell is the interaction information extracted by the last attention-based encoder.

[0126] For example, an attention-based encoder can refer to an encoder based on a single-head or multi-head self-attention mechanism. This application does not limit the type of attention-based encoder; for example, an attention-based encoder can refer to Performer, Transformer, Reformer, Linformer, etc. For example, Performer is an encoder based on a generalized attention mechanism, which can significantly reduce computational complexity and achieve efficient processing of ultra-long sequence data and learning of long-distance feature relationships.

[0127] The embodiments of this application do not limit the number of attention-based encoders included in the target extraction sub-model. The number can be set based on experience or flexibly adjusted according to actual application needs. For example, the number of attention-based encoders included in the target extraction sub-model is 8, or the number of attention-based encoders included in the target extraction sub-model is 5.

[0128] The process of calling the target prediction sub-model to process the interaction information of each candidate gene in the target cell and obtain the prediction result corresponding to the target cell is the internal processing of the target prediction sub-model, which is related to the structure of the target prediction sub-model. In the exemplary embodiment, the structure of the target prediction sub-model is flexibly set according to the type of prediction result corresponding to the target cell to be obtained, and this application embodiment does not limit it in this way.

[0129] For example, when the predicted result for the target cell is a classification result, the target prediction sub-model includes at least one convolutional layer and at least one fully connected layer connected in sequence. The number of convolutional layers and fully connected layers can be set empirically or flexibly adjusted according to the actual application scenario; this embodiment does not limit this. After inputting the interaction information of each candidate gene under the target cell into the target prediction sub-model, it is processed by at least one convolutional layer and at least one fully connected layer in sequence, resulting in the classification result for the target cell output by the last fully connected layer.

[0130] The last fully connected layer can be considered a classifier. The number of classifier heads corresponds to the number of candidate categories, outputting a classification result that includes the probability of each candidate category. Based on this classification result, the candidate category with the highest probability is determined and selected as the category of the target cell. The number of candidate categories can be flexibly adjusted according to the actual classification scenario; as the number of candidate categories changes, the number of classifier heads also changes accordingly.

[0131] For example, a fully connected layer can also be called a feedforward neural network layer. After the convolution operation of the convolutional layer, key information can be effectively extracted, and a more accurate classification result can be obtained after passing through the feedforward neural network layer.

[0132] It should be noted that the above embodiments only illustrate the model structure of the target prediction sub-model by taking the need to obtain the classification results corresponding to the target cells as an example. As the type of prediction results corresponding to the target cells to be obtained changes and as the actual application scenario changes, the structure of the target prediction sub-model can be flexibly adjusted, and the embodiments of this application do not limit it in this regard.

[0133] In the exemplary embodiment, after obtaining the prediction results corresponding to the target cell, further analysis of the cell can be achieved based on the interaction information of each candidate gene under the target cell. In the exemplary embodiment, since the interaction information of each candidate gene under the target cell is extracted during the process of obtaining the prediction results corresponding to the target cell, the interaction information of each candidate gene under the target cell can be directly obtained after obtaining the prediction results corresponding to the target cell.

[0134] In one possible implementation, the prediction result corresponding to the target cell is the classification result corresponding to the target cell, which indicates that the target cell belongs to the target category. In this case, the process of further analyzing the cell based on the interaction information of each candidate gene in the target cell includes: based on the interaction information of each candidate gene in the target cell, determining the gene that meets the selection criteria and belongs to the target category among the candidate genes. For example, the interaction information of each candidate gene in the target cell is in the form of a weight matrix, where each row or column of the weight matrix indicates the interaction information between a candidate gene and other candidate genes in the target cell.

[0135] In an exemplary embodiment, the process of determining the genes that meet the selection criteria for cells belonging to the target category based on the interaction information of each candidate gene in the target cell includes: summing the weight matrix corresponding to the interaction information of each candidate gene in the target cell column by column (or row by row), and selecting the genes corresponding to the columns (or rows) with the largest sums (K, where K is an integer not less than 1) as the genes that meet the selection criteria for cells belonging to the target category. For example, the columns (or rows) with the largest sums (K, where K is an integer not less than 1) can be determined directly, or the columns (or rows) can be sorted according to the size of the sums before determining the columns (or rows) with the largest sums (K, where K is an integer not less than 1).

[0136] The genes corresponding to the top K (K is an integer not less than 1) largest columns (or rows) in the corresponding sum are the genes that have a greater impact on the target cells. These genes that have a greater impact on the target cells can be directly used as the genes that meet the selection criteria for cells belonging to the target category. For example, genes that meet the selection criteria can also be called key genes.

[0137] In an exemplary embodiment, the process of determining the gene that meets the selection criteria and belongs to the target category of cells from among the candidate genes, based on the interaction information of each candidate gene in the target cell, includes: determining the interaction information of each candidate gene in each reference cell based on gene expression data of a known reference cell belonging to the target category; fusing the interaction information of each candidate gene in each reference cell with the interaction information of each candidate gene in the target cell to obtain fused interaction information; and determining the gene that meets the selection criteria and belongs to the target category of cells from among the candidate genes based on the fused interaction information. The reference cell is a cell known to belong to the target category. The reference cell can be determined by the researcher based on experience or by calling the prediction results of a data processing model; this embodiment does not limit this.

[0138] In an exemplary embodiment, the interaction information of each candidate gene in any cell is in the form of a weight matrix. The method for fusing the interaction information of each candidate gene in each reference cell and the interaction information of each candidate gene in the target cell can be to average the elements at corresponding positions in each weight matrix. The form of the fused interaction information is the same as the form of the interaction information of each candidate gene in any cell; for example, the fused interaction information can also be called the fused weight matrix. Exemplarily, this fusion method can be called an element-wise fusion method.

[0139] The principle of identifying the cells corresponding to the target category that meet the selection criteria based on fusion interaction information is the same as the principle of identifying the cells corresponding to the target category that meet the selection criteria based on the interaction information of each candidate gene in the target cell. It will not be repeated here.

[0140] Based on the method provided in the embodiments of this application, for the input data of each cell, a weight matrix can be extracted to indicate the interaction information of each candidate gene in that cell. The dimension of the weight matrix is ​​the number of candidate genes. The number of candidate genes is used, and the weight matrix represents the pairwise interactions between genes. By calculating the element-wise mean of the weight matrices corresponding to each cell category, the fusion weight matrix for each cell category can be obtained. Based on the fusion weight matrix for each cell category, the differences in interaction information between candidate genes in different cell categories can be compared. By summing the fusion weight matrix for a certain cell category column-wise (or row-wise), and then sorting by value and taking the largest subset of values, genes with a greater impact on different cell categories can be identified. These genes with a greater impact on a particular cell category are identified as key genes for that cell category. After obtaining the key genes corresponding to each cell category, functional pathways related to cell categories can be obtained through Gene Ontology and KEGG pathway enrichment analysis.

[0141] In an exemplary embodiment, after obtaining the prediction results corresponding to the target cells, the target cells can be labeled using the prediction results. The data processing method provided in this application embodiment can provide data support for the analysis of single-cell transcriptomes. Transcriptome analysis reveals the molecular mechanisms of complex biological pathways and trait regulatory networks by measuring the abundance and type of gene expression in specific tissue samples, thereby reflecting the differences in the clinical physiological state of the human body. Traditional transcriptome sequencing can only obtain gene expression information at the overall tissue level, but for certain specific tissues, especially tumor tissues, the cell composition is complex and different cell types may play specific physiological functions, so understanding the cell composition and heterogeneity of tissue samples is crucial.

[0142] In recent years, various single-cell transcriptome sequencing methods have become common research tools for obtaining specific gene expression at the individual cell level in tissue samples by achieving single-cell isolation. Single-cell transcriptome sequencing can construct transcriptome maps of different disease stages, reveal changes in gene expression patterns during disease progression, and further provide a molecular basis for the early diagnosis of diseases such as gastric cancer. It can also analyze immune differences related to different clinical features of specific diseases, leading to a deeper understanding of pathogenesis and the development of more targeted treatment strategies.

[0143] In single-cell transcriptome analysis, analyzing the differences in gene expression patterns across different clinical stages and characteristics helps to understand the dynamic changes and directions of tissue development and disease progression, thereby clarifying related biological mechanisms. Whether for early and accurate disease diagnosis or the development of targeted treatment strategies for complex diseases, single-cell transcriptome analysis can greatly facilitate these advancements. Accurate identification of predicted cell types (e.g., cell category) is crucial to realizing the value of single-cell transcriptome analysis; otherwise, erroneous conclusions may be drawn, affecting disease diagnosis and treatment. This application's embodiments rely on the learning of gene-interaction information to achieve a highly accurate and interpretable method for obtaining cell prediction results. Applying the method provided in this application to label cells can promote single-cell transcriptome research and clinical applications, contributing to the discovery of unknown mechanisms and the treatment of complex diseases.

[0144] The data processing method provided in this application automatically obtains the prediction results corresponding to the target cells based on the target gene expression data of the target cells, without relying on the researcher's prior knowledge, and the data processing has high stability. Furthermore, the prediction results corresponding to the target cells are obtained based on the interaction information of each candidate gene in the target cells. This interaction information can characterize the gene interaction features of the target cells. Since cells function in organisms through the interaction between genes, the gene interaction features can reflect the functional characteristics of the target cells. Therefore, the prediction results obtained by focusing on the functional characteristics of the target cells have high accuracy.

[0145] Based on the above Figure 1 The implementation environment shown in this application embodiment provides a method for training a data processing model. This method is executed by a computer device, which can be a terminal 11 or a server 12; this application embodiment does not limit the specific computer device used. Figure 4 As shown, the training method for the data processing model provided in this application embodiment includes the following steps 401 to 404.

[0146] In step 401, sample gene expression data of sample cells and standard results corresponding to sample cells are obtained, and sample expression values ​​corresponding to each candidate gene under sample cells are extracted from the sample gene expression data.

[0147] The sample cells are cells with standard results. These standard results can be set by the researcher based on experience or determined using other cell research methods; this embodiment does not limit this. The type of the standard results corresponding to the sample cells is the same as the type of prediction results to be output by the data processing model, so as to provide supervised information for the training process of the data processing model using the standard results corresponding to the sample cells. Since the sample cells correspond to standard results, the process of training the first data processing model using sample cells is a supervised training process.

[0148] For example, the sample gene expression data of the sample cells and the corresponding standard results are stored in a database, and the sample gene expression data of the sample cells and the corresponding standard results are extracted from the database. For example, the sample gene expression data of the sample cells and the corresponding standard results are stored in PanglaoDB (a single-cell database). For example, the sample cells can refer to labeled cells of a specific tissue.

[0149] In an exemplary embodiment, the sample cell gene expression data refers to the gene expression data of cells that meet quality control conditions, thereby ensuring the reliability of the sample cell gene expression data. Exemplarily, whether a cell meets the quality control conditions is determined based on the cell's gene expression data, which includes the expression values ​​corresponding to each measured gene under that cell's gene expression data. If the ratio of the number of measured genes with non-zero expression values ​​to the total number of measured genes in that cell is not less than a ratio threshold, then the cell is determined to meet the quality control conditions. The ratio threshold is set empirically or flexibly adjusted according to the application scenario; this embodiment does not limit this.

[0150] In an exemplary embodiment, the number of sample cells is at least one. The sample gene expression data of a sample cell includes the sample expression values ​​corresponding to each measured gene under that sample cell. Each candidate gene is determined from each measured gene based on the sample gene expression data of each sample cell. Before extracting the sample expression values ​​corresponding to each candidate gene under the sample cell from the sample gene expression data, the candidate genes need to be determined first. Exemplarily, the process of determining candidate genes includes: based on the sample gene expression data of each sample cell, counting the number of sample cells that each measured gene hits, with one measured gene hitting one sample cell indicating that the sample expression value corresponding to one measured gene under that sample cell is not less than a first threshold; and selecting the measured genes whose number of hit sample cells is not less than the number threshold as candidate genes.

[0151] The first threshold can be set based on experience or adjusted flexibly according to the application scenario; this application embodiment does not limit this. Measurement genes whose number of hit sample cells is not less than the threshold can be considered highly reliable measurement genes. Using highly reliable measurement genes as candidate genes is beneficial to improving the training effect of the data processing model.

[0152] The sample cell is the cell used for training the first data processing model once. The number of sample cells can be one or more; this embodiment does not limit this. This embodiment uses one sample cell as an example for explanation. The process of extracting the sample expression values ​​corresponding to each candidate gene under the sample cell from the sample gene expression data is described in [link to documentation]. Figure 2 The relevant processes in step 201 of the illustrated embodiment will not be repeated here.

[0153] In step 402, based on the sample expression values ​​corresponding to each candidate gene, the sample features corresponding to each candidate gene are obtained.

[0154] The implementation process of step 402 can be found in [link to relevant documentation]. Figure 2 Step 202 in the illustrated embodiment will not be repeated here.

[0155] In step 403, the first data processing model is invoked to extract the interaction information of each candidate gene in the sample cell based on the sample features corresponding to each candidate gene; based on the interaction information of each candidate gene in the sample cell, the prediction result corresponding to the sample cell is obtained.

[0156] The first data processing model refers to the model that is to be trained using the gene expression data of the sample cells and the standard results of the sample cells.

[0157] In one possible implementation, for Figure 2 In the illustrated embodiment, the target data processing model includes a target extraction sub-model and a target prediction sub-model, and the first data processing model includes a first extraction sub-model and a first prediction sub-model. In this case, the implementation of step 403 includes: calling the first extraction sub-model to extract information from the sample features corresponding to each candidate gene, obtaining the interaction information of each candidate gene under the sample cell; and calling the first prediction sub-model to process the interaction information of each candidate gene under the sample cell, obtaining the prediction result corresponding to the sample cell. See [link to implementation details]. Figure 2 The relevant processes in step 203 of the illustrated embodiment will not be repeated here. Exemplarily, the structure of the first extraction sub-model is the same as that of the target extraction sub-model, and the first extraction sub-model is a language model obtained by sequentially connecting at least one encoder based on an attention mechanism.

[0158] In an exemplary embodiment, the first extraction sub-model in the first data processing model can refer to an initialized model or a model obtained after pre-training; this embodiment does not limit this. Similarly, the first prediction sub-model in the first data processing model can refer to an initialized model or a model obtained after pre-training; this embodiment does not limit this.

[0159] This application embodiment uses an example where the first extraction sub-model is a pre-trained model and the first prediction sub-model is an initialized model. In this case, before calling the first extraction sub-model to extract information from the sample features corresponding to each candidate gene and obtain the interaction information of each candidate gene in the sample cells, the first extraction sub-model needs to be trained first.

[0160] In one possible implementation, the process of training the first extraction sub-model includes steps a through d.

[0161] Step a: Obtain the training gene expression data of the training cells, and extract the training expression values ​​corresponding to each candidate gene under the training cells from the training gene expression data.

[0162] Training cells refer to cells without standardized results. In practical applications, there is a large amount of gene expression data from cells without standardized results. Training the initial extraction sub-model using this large-scale gene expression data enables it to learn gene interaction information from gene expression data with different distributions, thus improving its generalization ability. In an exemplary embodiment, gene expression data without standardized results is extracted from a database (e.g., PanglaoDB) as training gene expression data for training cells. The implementation process for extracting the training expression values ​​of each candidate gene from the training gene expression data for training cells is described in [link to documentation]. Figure 2 The relevant processes in step 201 of the illustrated embodiment will not be repeated here.

[0163] Step b: Based on the training expression values ​​of each candidate gene, obtain the training features of each candidate gene, and replace the training features of the candidate genes that meet the replacement conditions with the reference features.

[0164] The process of obtaining the training features corresponding to each candidate gene based on the training expression values ​​of each candidate gene is described in [link to documentation]. Figure 2 Step 202 in the illustrated embodiment will not be repeated here.

[0165] After obtaining the training features corresponding to each candidate gene, candidate genes that meet the replacement conditions are determined from among the candidate genes. In an exemplary embodiment, candidate genes that meet the replacement conditions refer to candidate genes randomly selected from among the candidate genes at a reference proportion. The reference proportion is set based on experience or flexibly adjusted according to the experience scenario; for example, the reference proportion is 15%.

[0166] In an exemplary embodiment, due to the highly sparse nature of the training gene expression data, a candidate gene that meets the replacement condition refers to a candidate gene randomly selected from a reference proportion of candidate genes whose corresponding training expression value is not zero. For example, assuming the reference proportion is 10%, if the training expression value Aij corresponding to candidate gene j in training cell i is not zero, then candidate gene j has a 15% probability of being identified as a candidate gene that meets the replacement condition.

[0167] After identifying candidate genes that meet the replacement criteria, the training features corresponding to these candidate genes are replaced with reference features. The reference features are representation features identified by a special symbol; for example, the special symbol is [Mask]. The representation features identified by special symbols can be set empirically or flexibly adjusted according to the scenario, as long as they can be distinguished from normal training features. It should be noted that the reference features have the same form as the training features; for example, both are vectors of a specified dimension. For example, the process of replacing the training features corresponding to candidate genes that meet the replacement criteria with reference features can be considered as a random replacement process.

[0168] Step c: Call the initial extraction sub-model to extract information from the reference features corresponding to candidate genes that meet the replacement conditions and the training features corresponding to candidate genes that do not meet the replacement conditions, to obtain the interaction information of each candidate gene in the training cells; based on the interaction information of each candidate gene in the training cells, obtain the predicted features corresponding to the candidate genes that meet the replacement conditions.

[0169] After replacing the training features corresponding to candidate genes that meet the replacement conditions with reference features, the reference features corresponding to candidate genes that meet the replacement conditions and the training features corresponding to candidate genes that do not meet the replacement conditions are input into the initial extraction sub-model. The initial extraction sub-model extracts information from the reference features corresponding to candidate genes that meet the replacement conditions and the training features corresponding to candidate genes that do not meet the replacement conditions to obtain the interaction information of each candidate gene in the training cells. For example, inputting the reference features corresponding to candidate genes that meet the replacement conditions and the training features corresponding to candidate genes that do not meet the replacement conditions into the initial extraction sub-model means arranging the reference features corresponding to candidate genes that meet the replacement conditions and the training features corresponding to candidate genes that do not meet the replacement conditions in a specified order according to each candidate gene, forming a feature sequence, and then inputting the feature sequence into the initial extraction sub-model.

[0170] The process of training the initial extraction sub-model using training data from training cells can be regarded as a self-supervised pre-training process. After obtaining the interaction information of each candidate gene in the training cells, the initial extraction sub-model is also called to obtain the predicted features corresponding to the candidate genes that meet the replacement conditions based on the interaction information of each candidate gene in the training cells. This allows the predicted features corresponding to the candidate genes that meet the replacement conditions to be used to construct the loss function used in the self-supervised pre-training process.

[0171] In an exemplary embodiment, the process of obtaining the predicted features corresponding to candidate genes that meet the replacement conditions based on the interaction information of each candidate gene in the training cells is as follows: The initial extraction sub-model is invoked to obtain the predicted features corresponding to each candidate gene based on the interaction information of each candidate gene in the training cells; and the predicted features corresponding to candidate genes that meet the replacement conditions are extracted from the predicted features corresponding to each candidate gene. The process of invoking the initial extraction sub-model to obtain the predicted features corresponding to each candidate gene based on the interaction information of each candidate gene in the training cells is an internal processing procedure of the initial extraction sub-model and is related to the model structure of the initial extraction sub-model.

[0172] For example, the initial extraction sub-model includes at least one attention-based encoder connected in sequence. Each attention-based encoder includes an attention extraction layer and a feature output layer. The attention extraction layer is used to extract interaction information based on the input features, and the feature output layer is used to output features with the same dimension as the input features based on the extracted interaction information. The interaction information of each candidate gene in the training cells is the interaction information extracted by the last attention-based encoder. In this case, by inputting the interaction information of each candidate gene in the training cells into the feature output layer of the last attention-based encoder, the predicted features corresponding to each candidate gene output by the feature output layer can be obtained.

[0173] Step d: Based on the predicted features and training features corresponding to the candidate genes that meet the replacement conditions, obtain the feature loss function, and use the feature loss function to train the initial extraction sub-model to obtain the first extraction sub-model.

[0174] After obtaining the predicted features corresponding to candidate genes that meet the replacement conditions, a feature loss function is obtained based on the predicted features and training features of the candidate genes that meet the replacement conditions to train the initial extraction sub-model. In an exemplary embodiment, the feature loss function is obtained by using the cross-entropy loss function between the predicted features and training features of the candidate genes that meet the replacement conditions as the feature loss function. Of course, the feature loss function can also be obtained in other ways, such as using the mean squared error loss function between the predicted features and training features of the candidate genes that meet the replacement conditions as the feature loss function.

[0175] After obtaining the feature loss function, the initial extraction sub-model is trained using the feature loss function. In an exemplary embodiment, the process of training the initial extraction sub-model using the feature loss function refers to the process of updating the model parameters of the initial extraction sub-model using the feature loss function.

[0176] After training the initial extraction sub-model using the feature loss function, a trained extraction sub-model is obtained. It is then determined whether this trained extraction sub-model meets the first training termination condition. If it does, it is used as the first extraction sub-model. If it does not meet the first training termination condition, the process continues, referring to steps a to d, until an extraction sub-model that meets the first training termination condition is obtained. This sub-model is then used as the first extraction sub-model.

[0177] The first training termination condition can be set based on experience or flexibly adjusted according to the application scenario, and this application embodiment does not limit it in this way. For example, the first training termination condition for the extracted sub-model obtained after training includes, but is not limited to, any one of the following: the number of training iterations performed when obtaining the extracted sub-model has reached a threshold, the feature loss function when obtaining the extracted sub-model is less than a loss function threshold, or the feature loss function when obtaining the extracted sub-model has converged.

[0178] For example, the goal of training the extraction sub-model is to enable the extraction sub-model to predict the training features that are replaced by the reference features more accurately. By being able to predict the training features that are replaced by the reference features accurately, the extraction sub-model can be guaranteed to have a more reliable ability to extract interaction information.

[0179] In an exemplary embodiment, the process of training the initial extraction sub-model is as follows: Figure 5 As shown. The process involves obtaining training features for each candidate gene; replacing the training features of candidate genes that meet the replacement conditions with reference features. The reference features of candidate genes that meet the replacement conditions and the training features of candidate genes that do not meet the replacement conditions are input into the initial extraction sub-model to obtain the predicted features of candidate genes that meet the replacement conditions output by the initial extraction sub-model. Based on the predicted features and training features of candidate genes that meet the replacement conditions, a feature loss function is obtained, and the initial extraction sub-model is trained using the feature loss function. Figure 5 In the initial extraction sub-model, eight Performer encoders are connected sequentially. Figure 5 In this model, the training features and prediction features corresponding to candidate genes are represented by the discretized expression values ​​of the candidate genes (e.g., B2, B15, B7, Zero), and the reference features corresponding to candidate genes are represented by the feature symbol [Mask].

[0180] In step 404, based on the prediction results and standard results corresponding to the sample cells, the result loss function is obtained; the result loss function is used to train the first data processing model to obtain the target data processing model.

[0181] After obtaining the prediction results corresponding to the sample cells, a result loss function is obtained based on the prediction results and the standard results for the sample cells. The result loss function is used to indicate the difference between the prediction results and the standard results for the sample cells. This application does not limit the method of obtaining the result loss function based on the prediction results and the standard results for the sample cells. For example, the cross-entropy loss function or the mean squared error loss function between the prediction results and the standard results for the sample cells can be used as the result loss function.

[0182] After obtaining the result loss function, the first data processing model is trained using the result loss function. Training the first data processing model using the result loss function means updating the parameters of the first data processing model using the result loss function. Updating the parameters of the first data processing model using the result loss function can mean updating all the parameters of the first data processing model using the result loss function, or it can mean updating some of the parameters of the first data processing model using the result loss function. This application embodiment does not limit this.

[0183] In an exemplary embodiment, when the first extraction sub-model in the first data processing model is a model pre-trained using training gene expression data from training cells, and the first prediction sub-model is an initialized model, during the process of updating the parameters of the first data processing model using the result loss function, some parameters of the first extraction sub-model can be kept unchanged, while other parameters of the first extraction sub-model and all parameters of the first prediction sub-model are updated. Which parameters of the first extraction sub-model are kept unchanged can be flexibly set based on experience, and this embodiment does not limit this.

[0184] After training the first data processing model using the resulting loss function, a trained data processing model is obtained. It is then determined whether this trained data processing model meets the second training termination condition. If it does, the trained data processing model is used as the target extraction sub-model. If it does not meet the second training termination condition, the trained data processing model is trained again, following steps 401 to 404, and so on, until a data processing model that meets the second training termination condition is obtained. This data processing model that meets the second training termination condition is then used as the target data processing model.

[0185] The second training termination condition can be set based on experience or flexibly adjusted according to the application scenario, and this application embodiment does not limit it in this way. For example, the data processing model obtained after training may satisfy any one of the following conditions: the number of training iterations performed when obtaining the data processing model has reached a threshold, the loss function of the obtained data processing model is less than a loss function threshold, or the loss function of the obtained data processing model has converged.

[0186] In an exemplary embodiment, the training process of the data processing model is as follows: Figure 6 As shown. Training gene expression data of training cells without standard results is obtained. Based on the training gene expression data, through discretization, random replacement, and vectorization transformation, reference features corresponding to candidate genes that meet the replacement conditions and training features corresponding to candidate genes that do not meet the replacement conditions are obtained. The initial extraction sub-model is then called to process the reference features corresponding to candidate genes that meet the replacement conditions and the training features corresponding to candidate genes that do not meet the replacement conditions, obtaining predicted features corresponding to each candidate gene. Based on the predicted features and training features corresponding to candidate genes that meet the replacement conditions, a feature loss function is obtained. The initial extraction sub-model is then trained using the feature loss function to obtain the first extraction sub-model. The process of training the initial extraction sub-model based on the training gene expression data of training cells can be called a self-supervised pre-training process.

[0187] The process involves acquiring gene expression data from sample cells with standardized results. Based on this data, discretization and vectorization are performed to obtain sample features corresponding to each candidate gene. A first extraction sub-model is then used to process these features, outputting the interaction information of each candidate gene within the sample cells. A convolutional layer in the first prediction sub-model is then used to convolve the interaction information of each candidate gene within the sample cells. The resulting convolutional features are input into a fully connected layer in the first prediction sub-model to obtain the prediction results for the sample cells output by the fully connected layer. Based on the prediction results and the standardized results, a loss function is obtained. This loss function is then used to train the first data processing model, resulting in the target data processing model. The process of training the first data processing model based on the gene expression data from the sample cells can be termed a supervised fine-tuning process.

[0188] In an exemplary embodiment, after obtaining the target data processing model, the target data processing model can be tested to assess its performance. For example, the process of testing the target data processing model includes: pre-calculating the gene expression data of the test cells to obtain test features corresponding to each candidate gene; and calling the target data processing model to obtain prediction results for the test cells based on the test features corresponding to each candidate gene. The performance of the target data processing model can be calculated based on the prediction results and standard results for the test cells.

[0189] This application proposes a data processing model based on an attention mechanism. Compared to traditional methods that rely on manual acquisition of cell prediction results, the data processing model provided in this application can automate the process of acquiring cell prediction results, significantly reducing the cost of cell labeling and avoiding experimental errors caused by human factors. Compared to other methods for automatically acquiring cell prediction results, this application uses large-scale unlabeled data and combines it with Gene2Vec gene encoding, which can effectively learn the complex interactions between genes in different data distributions and has stronger generalization performance. Through the convolutional operation in the supervised training part, key information can be effectively extracted, and more accurate cell prediction results can be obtained through the feedforward neural network. This is conducive to improving the reliability of disease mechanism research and promoting the early diagnosis and personalized treatment of diseases in precision medicine.

[0190] This application embodiment uses large-scale unlabeled data combined with gene correlation relationships for self-supervised pre-training, and then uses a labeled reference dataset for fine-tuning supervised training of the resulting model, ultimately enabling accurate prediction of results for unlabeled cells. This application embodiment fully utilizes large-scale data for gene-interrelationship learning, while incorporating existing gene relationship embeddings, significantly enhancing the model's ability to acquire key information. Currently, the amount of publicly available large-scale unlabeled single-cell transcriptome data is growing exponentially, and there are single-cell atlases corresponding to various tissues, providing important prerequisites for the practical application of this application embodiment. Compared to related technologies, this application embodiment uses labeled reference datasets, eliminating the need for additional cell type-specific genes; large-scale self-supervised pre-training can acquire data distributions from different batches, giving the model sufficient generalization performance; through attention mechanisms, the model can learn biologically meaningful gene-interrelationship information. This method combines data and knowledge, possessing very strong information extraction capabilities, and is of great significance for single-cell transcriptome analysis.

[0191] The data processing model training method provided in this application obtains the prediction results corresponding to the sample cells based on the interaction information of each candidate gene under the sample cells, and then uses the loss function between the prediction results and the standard results to train the first data processing model. This training method is beneficial to improve the ability of the data processing model to extract interaction information, so as to output prediction results that are closer to the standard results based on the extracted interaction information, thereby ensuring the reliability of the interaction information extracted by the trained target data processing model, so that the target data processing model can output more accurate prediction results based on the extracted reliable interaction information.

[0192] See Figure 7 This application provides a data processing apparatus, which includes:

[0193] The first acquisition unit 701 is used to acquire the target gene expression data of the target cell and extract the target expression value corresponding to each candidate gene under the target cell from the target gene expression data.

[0194] The second acquisition unit 702 is used to acquire the target features corresponding to each candidate gene based on the target expression value corresponding to each candidate gene.

[0195] Extraction unit 703 is used to extract the interaction information of each candidate gene in the target cell based on the target features corresponding to each candidate gene. The interaction information of each candidate gene in the target cell is used to characterize the gene interaction features of the target cell.

[0196] The third acquisition unit 704 is used to acquire the prediction results corresponding to the target cell based on the interaction information of each candidate gene in the target cell.

[0197] In one possible implementation, the second acquisition unit 702 is used to convert the target expression value corresponding to the first candidate gene into the expression value feature corresponding to the first candidate gene, wherein the first candidate gene is any candidate gene among the candidate genes; and to fuse the expression value feature corresponding to the first candidate gene and the characterization feature corresponding to the first candidate gene to obtain the target feature corresponding to the first candidate gene.

[0198] In one possible implementation, the second acquisition unit 702 is used to normalize the target expression value corresponding to the first candidate gene to obtain the normalized expression value corresponding to the first candidate gene; determine the target discrete expression value corresponding to the normalized expression value, and use the embedding feature corresponding to the target discrete expression value as the expression value feature corresponding to the first candidate gene.

[0199] In one possible implementation, the target discretized expression value is one of the reference number of candidate discretized expression values. The second acquisition unit 702 is also used to perform vectorization transformation on the reference number of candidate discretized expression values ​​to obtain the embedded features corresponding to the reference number of candidate discretized expression values ​​respectively.

[0200] In one possible implementation, the extraction unit 703 is used to call the target data processing model to extract the interaction information of each candidate gene in the target cell based on the target features corresponding to each candidate gene.

[0201] The third acquisition unit 704 is used to call the target data processing model to obtain the prediction results corresponding to the target cell based on the interaction information of each candidate gene in the target cell.

[0202] In one possible implementation, the prediction result corresponding to the target cell indicates that the target cell belongs to the target category, and the device further includes:

[0203] The determination unit is used to identify the gene that meets the selection criteria and belongs to the cell of the target category, based on the interaction information of each candidate gene in the target cell.

[0204] The data processing apparatus provided in this application automatically obtains the prediction results corresponding to the target cells based on the target gene expression data of the target cells, without relying on the researcher's prior knowledge, and the data processing stability is high. Furthermore, the prediction results corresponding to the target cells are obtained based on the interaction information of each candidate gene under the target cells. This interaction information can characterize the gene interaction features of the target cells. Since cells function in organisms through the interaction between genes, the gene interaction features can reflect the functional characteristics of the target cells. Therefore, the prediction results obtained by focusing on the functional characteristics of the target cells have high accuracy.

[0205] See Figure 8 This application provides a training apparatus for a data processing model, the apparatus comprising:

[0206] The first acquisition unit 801 is used to acquire the sample gene expression data of the sample cells and the standard results corresponding to the sample cells, and to extract the sample expression values ​​corresponding to each candidate gene under the sample cells from the sample gene expression data.

[0207] The second acquisition unit 802 is used to acquire the sample features corresponding to each candidate gene based on the sample expression value corresponding to each candidate gene.

[0208] The extraction unit 803 is used to call the first data processing model to extract the interaction information of each candidate gene in the sample cell based on the sample features corresponding to each candidate gene.

[0209] The third acquisition unit 804 is used to acquire the prediction results corresponding to the sample cells based on the interaction information of each candidate gene in the sample cells;

[0210] Training unit 805 is used to obtain the result loss function based on the prediction results and standard results corresponding to the sample cells; and to train the first data processing model using the result loss function to obtain the target data processing model.

[0211] In one possible implementation, the first data processing model includes a first extraction sub-model and a first prediction sub-model; the extraction unit 803 is used to call the first extraction sub-model to extract information from the sample features corresponding to each candidate gene, and obtain the interaction information of each candidate gene in the sample cell.

[0212] The third acquisition unit 804 is used to call the first prediction sub-model to process the interaction information of each candidate gene under the sample cell and obtain the prediction result corresponding to the sample cell.

[0213] In one possible implementation, the first acquisition unit 801 is further configured to acquire training gene expression data of training cells and extract the training expression values ​​corresponding to each candidate gene under the training cells from the training gene expression data.

[0214] The second acquisition unit 802 is also used to acquire the training features corresponding to each candidate gene based on the training expression value corresponding to each candidate gene.

[0215] The device also includes:

[0216] The replacement unit is used to replace the training features corresponding to the candidate genes that meet the replacement conditions with the reference features.

[0217] The extraction unit 803 is also used to call the initial extraction sub-model to extract information from the reference features corresponding to the candidate genes that meet the replacement conditions and the training features corresponding to the candidate genes that do not meet the replacement conditions, so as to obtain the interaction information of each candidate gene under the training cells; based on the interaction information of each candidate gene under the training cells, the predicted features corresponding to the candidate genes that meet the replacement conditions are obtained.

[0218] The training unit 805 is also used to obtain a feature loss function based on the predicted features and training features corresponding to candidate genes that meet the replacement conditions, and to train the initial extraction sub-model using the feature loss function to obtain the first extraction sub-model.

[0219] In one possible implementation, the number of sample cells is at least one, and the sample gene expression data of one sample cell includes the sample expression values ​​corresponding to each measured gene under one sample cell. The device further includes:

[0220] The determination unit is used to count the number of sample cells that each measurement gene hits based on the sample gene expression data of each sample cell. One measurement gene hitting one sample cell is used to indicate that the sample expression value of a measurement gene under a sample cell is not less than a first threshold. Measurement genes that have hit a number of sample cells that are not less than the number threshold are selected as candidate genes.

[0221] In one possible implementation, the first extraction sub-model is a language model obtained by sequentially connecting at least one attention-based encoder.

[0222] The data processing model training device provided in this application obtains the prediction results corresponding to the sample cells based on the interaction information of each candidate gene under the sample cells, and then uses the loss function between the prediction results and the standard results to train the first data processing model. This training method is beneficial to improve the ability of the data processing model to extract interaction information, so as to output prediction results that are closer to the standard results based on the extracted interaction information, thereby ensuring the reliability of the interaction information extracted by the trained target data processing model, so that the target data processing model can output more accurate prediction results based on the extracted reliable interaction information.

[0223] It should be noted that the apparatus provided in the above embodiments is only illustrated by the division of the above functional units. In practical applications, the above functions can be assigned to different functional units as needed, that is, the internal structure of the device can be divided into different functional units to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process can be found in the method embodiments, which will not be repeated here.

[0224] In an exemplary embodiment, a computer device is also provided, comprising a processor and a memory, wherein at least one computer program is stored in the memory. The at least one computer program is loaded and executed by one or more processors to enable the computer device to implement any of the aforementioned data processing methods or data processing model training methods. The computer device can be a server or a terminal, and this embodiment does not limit this. The structures of the server and the terminal will be described separately below.

[0225] Figure 9This is a schematic diagram of a server structure provided in an embodiment of this application. The server can vary significantly due to differences in configuration or performance. It may include one or more Central Processing Units (CPUs) 901 and one or more memories 902. The one or more memories 902 store at least one computer program, which is loaded and executed by the one or more processors 901 to enable the server to implement the data processing methods or data processing model training methods provided in the various method embodiments described above. Of course, the server may also have wired or wireless network interfaces, a keyboard, and input / output interfaces for input and output. The server may also include other components for implementing device functions, which will not be elaborated upon here.

[0226] Figure 10 This is a schematic diagram of the structure of a terminal provided in an embodiment of this application. The terminal may be, for example, a PC, mobile phone, smartphone, PDA, wearable device, PPC, tablet computer, smart car infotainment system, smart TV, smart speaker, smart voice interaction device, smart home appliance, or in-vehicle terminal. The terminal may also be referred to as user equipment, portable terminal, laptop terminal, desktop terminal, or other names.

[0227] Typically, a terminal includes a processor 1001 and a memory 1002.

[0228] Processor 1001 may include one or more processing cores, such as a quad-core processor, an octa-core processor, etc. Processor 1001 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). Processor 1001 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, processor 1001 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content required to be displayed on the screen. In some embodiments, processor 1001 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0229] The memory 1002 may include one or more computer-readable storage media, which may be non-transitory. The memory 1002 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer-readable storage media in the memory 1002 are used to store at least one instruction, which is executed by the processor 1001 to enable the terminal to implement the data processing method or data processing model training method provided in the method embodiments of this application.

[0230] In some embodiments, the terminal may also optionally include: a peripheral device interface 1003 and at least one peripheral device. The processor 1001, memory 1002, and peripheral device interface 1003 can be connected via a bus or signal line. Each peripheral device can be connected to the peripheral device interface 1003 via a bus, signal line, or circuit board. Specifically, the peripheral device includes at least one of: a radio frequency circuit 1004, a display screen 1005, a camera assembly 1006, an audio circuit 1007, and a power supply 1009.

[0231] Peripheral device interface 1003 can be used to connect at least one I / O (Input / Output) related peripheral device to processor 1001 and memory 1002. In some embodiments, processor 1001, memory 1002 and peripheral device interface 1003 are integrated on the same chip or circuit board; in some other embodiments, any one or two of processor 1001, memory 1002 and peripheral device interface 1003 can be implemented on separate chips or circuit boards, which is not limited in this embodiment.

[0232] The radio frequency (RF) circuit 1004 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The RF circuit 1004 communicates with communication networks and other communication devices via electromagnetic signals. The RF circuit 1004 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals back into electrical signals. Optionally, the RF circuit 1004 includes: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, etc. The RF circuit 1004 can communicate with other terminals through at least one wireless communication protocol. This wireless communication protocol includes, but is not limited to: metropolitan area networks (MANs), various generations of mobile communication networks (2G, 3G, 4G, and 5G), wireless local area networks (WLANs), and / or WiFi (Wireless Fidelity) networks. In some embodiments, the RF circuit 1004 may also include circuitry related to NFC (Near Field Communication), which is not limited in this application.

[0233] Display screen 1005 is used to display a UI (User Interface). This UI may include graphics, text, icons, videos, and any combination thereof. When display screen 1005 is a touch display screen, it also has the ability to collect touch signals on or above its surface. These touch signals can be input as control signals to processor 1001 for processing. In this case, display screen 1005 can also be used to provide virtual buttons and / or a virtual keyboard, also known as soft buttons and / or a soft keyboard. In some embodiments, display screen 1005 may be a single screen, disposed on the front panel of the terminal; in other embodiments, display screen 1005 may be at least two screens, disposed on different surfaces of the terminal or in a folded design; in other embodiments, display screen 1005 may be a flexible display screen, disposed on a curved or folded surface of the terminal. Furthermore, display screen 1005 may be configured as a non-rectangular, irregular shape, i.e., a non-rectangular screen. Display screen 1005 may be made of materials such as LCD (Liquid Crystal Display) or OLED (Organic Light-Emitting Diode).

[0234] The camera assembly 1006 is used to acquire images or videos. Optionally, the camera assembly 1006 includes a front-facing camera and a rear-facing camera. Typically, the front-facing camera is located on the front panel of the terminal, and the rear-facing camera is located on the back of the terminal. In some embodiments, there are at least two rear-facing cameras, which are any one of a main camera, a depth-sensing camera, a wide-angle camera, and a telephoto camera, to achieve background blurring by fusion of the main camera and the depth-sensing camera, panoramic shooting by fusion of the main camera and the wide-angle camera, VR (Virtual Reality) shooting, or other fusion shooting functions. In some embodiments, the camera assembly 1006 may also include a flash. The flash can be a single-color temperature flash or a dual-color temperature flash. A dual-color temperature flash refers to a combination of a warm-light flash and a cool-light flash, which can be used for light compensation at different color temperatures.

[0235] The audio circuit 1007 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, converting the sound waves into electrical signals that are input to the processor 1001 for processing, or input to the radio frequency circuit 1004 for voice communication. For stereo sound acquisition or noise reduction purposes, multiple microphones may be used, each positioned at a different location on the terminal. The microphone may also be an array microphone or an omnidirectional microphone. The speaker is used to convert electrical signals from the processor 1001 or the radio frequency circuit 1004 into sound waves. The speaker may be a conventional diaphragm speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can convert electrical signals not only into audible sound waves but also into inaudible sound waves for purposes such as distance measurement. In some embodiments, the audio circuit 1007 may also include a headphone jack.

[0236] The power supply 1009 is used to power the various components in the terminal. The power supply 1009 can be AC ​​power, DC power, a disposable battery, or a rechargeable battery. When the power supply 1009 includes a rechargeable battery, the rechargeable battery can support wired or wireless charging. The rechargeable battery can also be used to support fast charging technology.

[0237] In some embodiments, the terminal further includes one or more sensors 1010. The one or more sensors 1010 include, but are not limited to: an acceleration sensor 1011, a gyroscope sensor 1012, a pressure sensor 1013, an optical sensor 1015, and a proximity sensor 1016.

[0238] Accelerometer 1011 can detect the magnitude of acceleration along the three coordinate axes of a coordinate system established by the terminal. For example, accelerometer 1011 can be used to detect the components of gravitational acceleration along the three coordinate axes. Processor 1001 can control display screen 1005 to display the user interface in either a landscape or portrait view based on the gravitational acceleration signal acquired by accelerometer 1011. Accelerometer 1011 can also be used for games or for acquiring user motion data.

[0239] The gyroscope sensor 1012 can detect the terminal's orientation and rotation angle. The gyroscope sensor 1012 can work in conjunction with the accelerometer sensor 1011 to collect the user's 3D movements on the terminal. Based on the data collected by the gyroscope sensor 1012, the processor 1001 can perform the following functions: motion sensing (e.g., changing the UI based on the user's tilt), image stabilization during shooting, game control, and inertial navigation.

[0240] The pressure sensor 1013 can be disposed on the side bezel of the terminal and / or the lower layer of the display screen 1005. When the pressure sensor 1013 is disposed on the side bezel of the terminal, it can detect the user's grip signal on the terminal, and the processor 1001 can perform left / right hand recognition or quick operation based on the grip signal collected by the pressure sensor 1013. When the pressure sensor 1013 is disposed on the lower layer of the display screen 1005, the processor 1001 can control the operable controls on the UI interface based on the user's pressure operation on the display screen 1005. The operable controls include at least one of button controls, scroll bar controls, icon controls, and menu controls.

[0241] An optical sensor 1015 is used to collect ambient light intensity. In one embodiment, the processor 1001 can control the display brightness of the display screen 1005 based on the ambient light intensity collected by the optical sensor 1015. Specifically, when the ambient light intensity is high, the display brightness of the display screen 1005 is increased; when the ambient light intensity is low, the display brightness of the display screen 1005 is decreased. In another embodiment, the processor 1001 can also dynamically adjust the shooting parameters of the camera assembly 1006 based on the ambient light intensity collected by the optical sensor 1015.

[0242] The proximity sensor 1016, also known as a distance sensor, is typically installed on the front panel of the terminal. The proximity sensor 1016 is used to detect the distance between the user and the front of the terminal. In one embodiment, when the proximity sensor 1016 detects that the distance between the user and the front of the terminal is gradually decreasing, the processor 1001 controls the display screen 1005 to switch from a screen-on state to a screen-off state; when the proximity sensor 1016 detects that the distance between the user and the front of the terminal is gradually increasing, the processor 1001 controls the display screen 1005 to switch from a screen-off state to a screen-on state.

[0243] Those skilled in the art will understand that Figure 10 The structure shown does not constitute a limitation on the terminal and may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0244] In an exemplary embodiment, a computer-readable storage medium is also provided, which stores at least one computer program, which is loaded and executed by a processor of a computer device to enable the computer to implement any of the above-described data processing methods or data processing model training methods.

[0245] In one possible implementation, the aforementioned computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, and optical data storage device, etc.

[0246] In an exemplary embodiment, a computer program product is also provided, which includes a computer program or computer instructions that are loaded and executed by a processor to enable a computer to implement any of the above-described data processing methods or data processing model training methods.

[0247] It should be noted that the terms "first," "second," etc., used in this application 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 this application described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the above exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0248] It should be understood that "multiple" as used in this article refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0249] The above description is merely an exemplary embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A data processing method, characterized in that, The method includes: Obtain target gene expression data of target cells, and extract the target expression values ​​of each candidate gene in the target cells from the target gene expression data; The target expression value corresponding to the first candidate gene is converted into the expression value feature corresponding to the first candidate gene, wherein the first candidate gene is any one of the candidate genes; The expression value feature and the characterization feature corresponding to the first candidate gene are fused to obtain the target feature corresponding to the first candidate gene; Based on the target features corresponding to each candidate gene, the interaction information of each candidate gene under the target cell is extracted, and the interaction information of each candidate gene under the target cell is used to characterize the gene interaction features of the target cell. Based on the interaction information of each candidate gene in the target cell, the prediction result corresponding to the target cell is obtained.

2. The method according to claim 1, characterized in that, The step of converting the target expression value corresponding to the first candidate gene into the expression value feature corresponding to the first candidate gene includes: The target expression value corresponding to the first candidate gene is normalized to obtain the normalized expression value corresponding to the first candidate gene. Determine the target discretized expression value corresponding to the normalized expression value, and use the embedding feature corresponding to the target discretized expression value as the expression value feature corresponding to the first candidate gene.

3. The method according to claim 2, characterized in that, The target discretized expression value is one of a number of candidate discretized expression values ​​(reference number). Before using the embedding feature corresponding to the target discretized expression value as the expression value feature corresponding to the first candidate gene, the method further includes: The reference number of candidate discretized expression values ​​are vectorized to obtain the embedding features corresponding to the reference number of candidate discretized expression values.

4. The method according to any one of claims 1-3, characterized in that, The step of extracting the interaction information of each candidate gene in the target cell based on the target features corresponding to each candidate gene includes: The target data processing model is invoked to extract the interaction information of each candidate gene in the target cell based on the target features corresponding to each candidate gene; The step of obtaining the prediction result corresponding to the target cell based on the interaction information of each candidate gene in the target cell includes: The target data processing model is invoked to obtain the prediction results corresponding to the target cell based on the interaction information of each candidate gene in the target cell.

5. The method according to any one of claims 1-3, characterized in that, The prediction result corresponding to the target cell indicates that the category of the target cell is the target category, and the method further includes: Based on the interaction information of each candidate gene in the target cell, the gene that meets the selection criteria and belongs to the target category is determined from among the candidate genes.

6. A method for training a data processing model, characterized in that, The method includes: Obtain the gene expression data of the sample cells and the standard results corresponding to the sample cells, and extract the sample expression values ​​of each candidate gene under the sample cells from the gene expression data. The sample expression value corresponding to the first candidate gene is converted into the sample expression value feature corresponding to the first candidate gene, wherein the first candidate gene is any one of the candidate genes; The expression value features of the sample corresponding to the first candidate gene and the characterization features of the first candidate gene are fused together to obtain the sample features corresponding to the first candidate gene. The first data processing model is invoked to extract the interaction information of each candidate gene in the sample cell based on the sample features corresponding to each candidate gene; based on the interaction information of each candidate gene in the sample cell, the prediction result corresponding to the sample cell is obtained. Based on the prediction results and standard results corresponding to the sample cells, a result loss function is obtained; the first data processing model is trained using the result loss function to obtain the target data processing model.

7. The method according to claim 6, characterized in that, The first data processing model includes a first extraction sub-model and a first prediction sub-model; the step of calling the first data processing model to extract the interaction information of each candidate gene in the sample cells based on the sample features corresponding to each candidate gene includes: The first extraction sub-model is invoked to extract information from the sample features corresponding to each candidate gene, thereby obtaining the interaction information of each candidate gene in the sample cells. The step of obtaining the prediction result corresponding to the sample cell based on the interaction information of each candidate gene in the sample cell includes: The first prediction sub-model is invoked to process the interaction information of each candidate gene in the sample cell, and the prediction result corresponding to the sample cell is obtained.

8. The method according to claim 7, characterized in that, Before calling the first extraction sub-model to extract information from the sample features corresponding to each candidate gene to obtain the interaction information of each candidate gene under the sample cells, the method further includes: Acquire training gene expression data of training cells, and extract the training expression values ​​corresponding to each candidate gene under the training cells from the training gene expression data; Based on the training expression values ​​corresponding to each candidate gene, the training features corresponding to each candidate gene are obtained, and the training features corresponding to the candidate genes that meet the replacement conditions are replaced with reference features. The initial extraction sub-model is invoked to extract information from the reference features corresponding to the candidate genes that meet the replacement conditions and the training features corresponding to the candidate genes that do not meet the replacement conditions, thereby obtaining the interaction information of each candidate gene in the training cells; based on the interaction information of each candidate gene in the training cells, the predicted features corresponding to the candidate genes that meet the replacement conditions are obtained. Based on the predicted features and training features corresponding to the candidate genes that meet the replacement conditions, a feature loss function is obtained, and the initial extraction sub-model is trained using the feature loss function to obtain the first extraction sub-model.

9. The method according to any one of claims 6-8, characterized in that, The number of sample cells is at least one, and the sample gene expression data of one sample cell includes the sample expression values ​​corresponding to each measured gene under that sample cell. Before extracting the sample expression values ​​corresponding to each candidate gene under the sample cell from the sample gene expression data, the method further includes: Based on the gene expression data of each sample cell, the number of sample cells that each measurement gene hits is counted. One measurement gene hitting one sample cell is used to indicate that the sample expression value of the measurement gene in one sample cell is not less than a first threshold. Genes that are measured and whose number of sample cells that are hit is not less than the number threshold are selected as candidate genes.

10. The method according to claim 7, characterized in that, The first extraction sub-model is a language model obtained by sequentially connecting at least one attention-based encoder.

11. A data processing apparatus, characterized in that, The device includes: The first acquisition unit is used to acquire target gene expression data of target cells and extract the target expression values ​​corresponding to each candidate gene under the target cells from the target gene expression data. The second acquisition unit is used to convert the target expression value corresponding to the first candidate gene into the expression value feature corresponding to the first candidate gene, wherein the first candidate gene is any one of the candidate genes; and to fuse the expression value feature corresponding to the first candidate gene and the characterization feature corresponding to the first candidate gene to obtain the target feature corresponding to the first candidate gene. An extraction unit is used to extract the interaction information of each candidate gene in the target cell based on the target features corresponding to each candidate gene. The interaction information of each candidate gene in the target cell is used to characterize the gene interaction features of the target cell. The third acquisition unit is used to acquire the prediction result corresponding to the target cell based on the interaction information of each candidate gene in the target cell.

12. The apparatus according to claim 11, characterized in that, The second acquisition unit is used to normalize the target expression value corresponding to the first candidate gene to obtain the normalized expression value corresponding to the first candidate gene; determine the target discretized expression value corresponding to the normalized expression value, and use the embedding feature corresponding to the target discretized expression value as the expression value feature corresponding to the first candidate gene.

13. The apparatus according to claim 12, characterized in that, The target discretized expression value is one of the reference number of candidate discretized expression values. The second acquisition unit is further used to perform vectorization transformation on the reference number of candidate discretized expression values ​​to obtain the embedding features corresponding to the reference number of candidate discretized expression values ​​respectively.

14. The apparatus according to any one of claims 11-13, characterized in that, The extraction unit is used to call the target data processing model to extract the interaction information of each candidate gene in the target cell based on the target features corresponding to each candidate gene. The third acquisition unit is used to call the target data processing model to obtain the prediction result corresponding to the target cell based on the interaction information of each candidate gene in the target cell.

15. The apparatus according to any one of claims 11-13, characterized in that, The prediction result corresponding to the target cell indicates that the category of the target cell is the target category, and the device further includes: The determining unit is used to determine, based on the interaction information of each candidate gene in the target cell, the gene that meets the selection criteria and belongs to the cell of the target category.

16. A training device for a data processing model, characterized in that, The device includes: The first acquisition unit is used to acquire sample gene expression data of sample cells and standard results corresponding to the sample cells, and extract sample expression values ​​corresponding to each candidate gene under the sample cells from the sample gene expression data. The second acquisition unit is used to convert the sample expression value corresponding to the first candidate gene into the sample expression value feature corresponding to the first candidate gene, wherein the first candidate gene is any one of the candidate genes; and to fuse the sample expression value feature corresponding to the first candidate gene and the characterization feature corresponding to the first candidate gene to obtain the sample feature corresponding to the first candidate gene. The extraction unit is used to call the first data processing model to extract the interaction information of each candidate gene in the sample cell based on the sample features corresponding to each candidate gene. The third acquisition unit is used to acquire the prediction result corresponding to the sample cell based on the interaction information of each candidate gene under the sample cell; The training unit is used to obtain a result loss function based on the prediction results and standard results corresponding to the sample cells; and to train the first data processing model using the result loss function to obtain the target data processing model.

17. The apparatus according to claim 16, characterized in that, The first data processing model includes a first extraction sub-model and a first prediction sub-model; the extraction unit is used to call the first extraction sub-model to extract information from the sample features corresponding to each candidate gene, and obtain the interaction information of each candidate gene under the sample cell; The third acquisition unit is used to call the first prediction sub-model to process the interaction information of each candidate gene under the sample cell, and obtain the prediction result corresponding to the sample cell.

18. The apparatus according to claim 17, characterized in that, The first acquisition unit is further configured to acquire training gene expression data of training cells and extract the training expression values ​​corresponding to each candidate gene under the training cells from the training gene expression data; The second acquisition unit is further configured to acquire the training features corresponding to each candidate gene based on the training expression values ​​corresponding to each candidate gene; The device further includes: The replacement unit is used to replace the training features corresponding to the candidate genes that meet the replacement conditions in each candidate gene with the reference features; The extraction unit is used to call the initial extraction sub-model to extract information from the reference features corresponding to the candidate genes that meet the replacement conditions and the training features corresponding to the candidate genes that do not meet the replacement conditions, so as to obtain the interaction information of each candidate gene in the training cells; and to obtain the predicted features corresponding to the candidate genes that meet the replacement conditions based on the interaction information of each candidate gene in the training cells. The training unit is further configured to obtain a feature loss function based on the predicted features and training features corresponding to the candidate genes that meet the replacement conditions, and to train the initial extraction sub-model using the feature loss function to obtain the first extraction sub-model.

19. The apparatus according to any one of claims 16-18, characterized in that, The number of sample cells is at least one, and the sample gene expression data of one sample cell includes the sample expression values ​​corresponding to each measured gene under that sample cell. The device further includes: The determining unit is used to count the number of sample cells that each measurement gene hits based on the sample gene expression data of each sample cell. A measurement gene hitting a sample cell is used to indicate that the sample expression value of the measurement gene under the sample cell is not less than a first threshold. Measurement genes that have hit a number of sample cells that are not less than the number threshold are selected as candidate genes.

20. The apparatus according to claim 17, characterized in that, The first extraction sub-model is a language model obtained by sequentially connecting at least one attention-based encoder.

21. A computer device, characterized in that, The computer device includes a processor and a memory, the memory storing at least one computer program, the at least one computer program being loaded and executed by the processor to enable the computer device to implement the data processing method as described in any one of claims 1 to 5, or the training method for the data processing model as described in any one of claims 6 to 10.

22. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores at least one computer program, which is loaded and executed by a processor to enable the computer to implement the data processing method as described in any one of claims 1 to 5, or the training method for the data processing model as described in any one of claims 6 to 10.

23. A computer program product, characterized in that, The computer program product includes a computer program or computer instructions, which are loaded and executed by a processor to enable a computer to implement the data processing method as described in any one of claims 1 to 5, or the training method for the data processing model as described in any one of claims 6 to 10.

Images