Gene data processing method and device, computer device and storage medium

Abstract

The application discloses a gene data processing method and device, computer equipment and a storage medium, and belongs to the technical field of computers. Through a gene function query request of a to-be-tested gene, the application can call a gene association model corresponding to a cell type to which the to-be-tested gene belongs, mine a nonlinear association degree between the to-be-tested gene and a known candidate gene, and use the candidate gene with a higher nonlinear association degree to label function annotation information of the to-be-tested gene. The way of calling the gene association model to extract the nonlinear relationship is completely different from the way of extracting the linear relationship in traditional statistics, can deeply mine the candidate gene with a higher similarity to the to-be-tested gene, and the similarity is not a linear similarity but an implicit nonlinear similarity, so that the accuracy of the gene data processing process is greatly improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a gene data processing method, apparatus, computer equipment, and storage medium. Background Technology

[0002] With the development of computer technology, gene sequencing technology has been widely applied in many fields such as biology, medicine, health, and agriculture. Gene sequencing technology involves determining the function of each gene in a gene expression dataset.

[0003] Currently, it is necessary to calculate the correlation coefficient between the expression levels of each pair of genes, find a set of genes that are significantly associated with the gene to be tested, and then use some statistical methods (such as Fisher's Exact Test) to search the known gene set annotation library. Based on the annotation information of this set of genes, various phenotypic descriptions, such as the biological pathways that the gene to be tested may participate in, are labeled.

[0004] In the above process, when performing functional annotation on the gene to be tested, the key step is to find a set of genes with significant correlation. In bioinformatics, there are many ways to calculate the correlation coefficient between the expression levels of two genes. These methods mainly measure the linear relationship between the two sets of vectors, but the relationship between gene expression levels is more non-linear, which leads to poor accuracy in gene data processing. Summary of the Invention

[0005] This application provides a gene data processing method, apparatus, computer equipment, and storage medium, which can improve the accuracy of the gene data processing process. The technical solution is as follows: On the one hand, a gene data processing method is provided, which includes: Based on the gene function query request of the gene to be tested, the cell type corresponding to the gene to be tested is determined; Obtain the single-cell expression dataset corresponding to the cell type and the gene association model corresponding to the cell type. The gene association model is used to obtain the degree of non-linear association between the gene to be tested and the candidate genes belonging to the cell type in the single-cell expression dataset. The gene association model is called to process the gene to be tested, and the degree of nonlinear association between the gene to be tested and multiple candidate genes belonging to the cell type in the single-cell expression dataset is obtained. Output functional annotation information of candidate genes whose nonlinear association degree meets the first target condition, where the first target condition is used to indicate that the nonlinear association degree is greater than or equal to the association degree threshold.

[0006] On the one hand, a gene data processing device is provided, the device comprising: The first determining module is used to determine the cell type corresponding to the gene to be tested based on the gene function query request of the gene to be tested; The first acquisition module is used to acquire the single-cell expression dataset corresponding to the cell type and the gene association model corresponding to the cell type. The gene association model is used to acquire the degree of non-linear association between the gene to be tested and the candidate genes belonging to the cell type in the single-cell expression dataset. The processing module is used to call the gene association model to process the gene to be tested, and obtain the degree of nonlinear association between the gene to be tested and multiple candidate genes belonging to the cell type in the single cell expression dataset. The output module is used to output functional annotation information of candidate genes whose nonlinear association degree meets the first target condition, wherein the first target condition is used to indicate that the nonlinear association degree is greater than or equal to the association degree threshold.

[0007] In one possible implementation, the processing module is used to: Based on the gene to be tested and the plurality of candidate genes, a plurality of gene pairs are constructed, wherein any gene pair in the plurality of gene pairs includes the gene to be tested and any candidate gene in the plurality of candidate genes; The multiple gene pairs are input into the gene association model, and the multiple gene pairs are convolved by the gene association model to obtain multiple association coefficients. Any one of the multiple association coefficients is used to represent the degree of nonlinear association between the genes contained in any gene pair in the multiple gene pairs.

[0008] In one possible implementation, the device further includes: The second acquisition module is used to acquire multiple sample genes from the single-cell expression dataset, wherein the degree of variation of the multiple sample genes is greater than the variation threshold. The third acquisition module is used to acquire multiple positive gene pairs and multiple negative gene pairs based on the multiple sample genes, wherein the degree of non-linear correlation between the genes contained in any positive gene pair is greater than the correlation degree threshold, and the degree of non-linear correlation between the genes contained in any negative gene pair is less than the correlation degree threshold. The adjustment module is used to adjust the parameters of the initial depth model based on the multiple positive gene pairs and the multiple negative gene pairs to obtain the gene association model.

[0009] In one possible implementation, the third acquisition module is used to: From the multiple sample genes, two sample genes belonging to the same protein complex are identified as a positive gene pair; Any sample gene in the positive gene pair and any sample gene in the plurality of sample genes other than the positive gene pair are identified as a negative gene pair.

[0010] In one possible implementation, the adjustment module is used to: For any one of the plurality of positive gene pairs and the plurality of negative gene pairs, the initial deep model is invoked to predict the association coefficient of any one gene pair; Based on the correlation coefficients of the multiple positive gene pairs and the multiple negative gene pairs, the loss function value of the initial depth model is determined; In response to the loss function value not meeting the stopping condition, the initial deep model is trained iteratively until the loss function value meets the stopping condition, at which point training stops and the gene association model is obtained.

[0011] In one possible implementation, the device further includes: The fourth acquisition module is used to acquire, for any gene pair among the plurality of positive gene pairs and the plurality of negative gene pairs, a plurality of association coefficients predicted by a plurality of gene association models for any gene pair, wherein the plurality of gene association models correspond to a plurality of cell types in at least one single-cell expression dataset. The second determining module is used to determine whether any gene pair meets a second target condition based on the multiple association coefficients of any gene pair. The second target condition is used to indicate that the prediction difficulty of the multiple gene association models for any gene pair is lower than a first target threshold. The adjustment module is also used to adjust the parameters of the gene association model corresponding to the cell type based on gene pairs that meet the second target conditions.

[0012] In one possible implementation, the second determining module is used to: Based on the plurality of correlation coefficients, a first number of coefficients and a second number of coefficients are determined. The first number of coefficients is the number of coefficients among the plurality of correlation coefficients that are greater than or equal to a second target threshold, and the second number of coefficients is the number of coefficients among the plurality of correlation coefficients that are less than the second target threshold. In response to the first coefficient number being greater than the second coefficient number and the gene pair being a positive gene pair, or the first coefficient number being less than the second coefficient number and the gene pair being a negative gene pair, it is determined that the gene pair meets the second target condition.

[0013] On one hand, a computer device is provided, the computer device including one or more processors and one or more memories, the one or more memories storing at least one computer program, the at least one computer program being loaded and executed by the one or more processors to implement the gene data processing method as described in any of the above possible implementations.

[0014] On the one hand, a storage medium is provided that stores at least one computer program, which is loaded and executed by a processor to implement a gene data processing method as described in any of the possible implementations above.

[0015] On one hand, a computer program product or computer program is provided, the computer program product or computer program comprising one or more lines of program code stored in a computer-readable storage medium. One or more processors of a computer device are capable of reading the one or more lines of program code from the computer-readable storage medium, and the one or more processors execute the one or more lines of program code, enabling the computer device to perform the gene data processing method of any of the above possible embodiments.

[0016] The beneficial effects of the technical solutions provided in this application include at least the following: By requesting the gene function of the gene to be tested, a gene association model corresponding to the cell type to which the gene to be tested belongs can be invoked to discover the degree of non-linear association between the gene to be tested and known candidate genes. The candidate genes with a high degree of non-linear association are then used to annotate the functional annotation information of the gene to be tested. This method of extracting non-linear relationships by invoking gene association models is completely different from the method of extracting linear relationships in traditional statistics. It can deeply mine candidate genes with a higher degree of similarity to the gene to be tested. This similarity is not linear but implicit, thus greatly improving the accuracy of the gene data processing process. Attached Figure Description

[0017] 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.

[0018] Figure 1 This is a schematic diagram illustrating the principle of a cloud application scenario provided in an embodiment of this application; Figure 2 This is a schematic diagram of the implementation environment of a gene data processing method provided in an embodiment of this application; Figure 3This is a flowchart of a gene data processing method provided in an embodiment of this application; Figure 4 This is a flowchart illustrating the training process of a gene association model provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of a gene data processing device provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0019] 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.

[0020] In this application, the terms "first," "second," etc., are used to distinguish identical or similar items with essentially the same function. It should be understood that there is no logical or temporal dependency between "first," "second," and "nth," nor are there any restrictions on quantity or execution order.

[0021] In this application, the term "at least one" means one or more, and "multiple" means two or more, for example, multiple first positions means two or more first positions.

[0022] Bioinformatics is a discipline that studies the acquisition, processing, storage, dissemination, analysis, and interpretation of biological information. It is a new discipline that emerged from the rapid development of life sciences and computer science, combining the two fields. It reveals the biological mysteries inherent in vast and complex biological data by comprehensively utilizing biology, computer science, and information technology.

[0023] Gene sequencing technology, also known as DNA (Deoxyribonucleic acid) sequencing technology, is the technique for determining the sequence of DNA. In molecular biology research, DNA sequence analysis is the foundation for further research and modification of target genes. With the widespread application of Next Generation Sequence (NGS) technology, the cost of gene sequencing has gradually decreased, allowing gene sequencing technology to be applied in a wider range of fields such as biology, medicine, health, and agriculture. Compared with traditional methods based on Sanger sequencing and capillary electrophoresis, NGS refers to sequencing technologies with increased throughput, such as the ability to generate hundreds of thousands of relatively small sequence reads at once.

[0024] Single-cell sequencing technology refers to the technique used in DNA research to sequence the relatively simple genomes of single-celled microorganisms. In the life sciences, single-cell sequencing is revolutionary, capable of detecting heterogeneous information that cannot be obtained through sequencing mixed samples, thus ushering the entire field of genetics into a new era.

[0025] Genome annotation refers to the high-throughput annotation of the biological functions of all genes in a genome using bioinformatics methods and tools, and is currently a hot topic in functional genomics research. Annotation typically includes four main aspects: gene structure annotation, gene function annotation, repetitive sequence analysis, and non-coding RNA (ribonucleic acid) annotation.

[0026] With the development of single-cell sequencing technology, the number of publicly available single-cell expression datasets is growing explosively. Single-cell sequencing will become a routine module in R&D for research institutions, universities, hospitals, and life science companies alike. However, data analysis using single-cell sequencing demands high-performance computing equipment. Furthermore, with the number of publicly available single-cell expression datasets increasing exponentially (inevitably reaching tens of thousands), integrating and analyzing all publicly available datasets would be computationally and technically impossible for a single user. Therefore, cloud products providing single-cell sequencing capabilities have emerged. These cloud products enable "run once, use many times," and by offering free limited services or paid customized services, they can increase user stickiness and expand the cloud market share in the life sciences field.

[0027] The following explains some basic concepts involved in the field of cloud technology: Cloud technology refers to a hosting technology that unifies hardware, software, and network resources within a wide area network (WAN) or local area network (LAN) to achieve data computation, storage, processing, and sharing. It encompasses network technologies, information technologies, integration technologies, management platform technologies, and application technologies based on cloud computing business models. These technologies can form resource pools, allowing for on-demand, flexible, and convenient use. Cloud computing technology will become a crucial support in the cloud technology field. Backend services of technical network systems require substantial computing and storage resources, such as those for video websites, image websites, and various portal websites. With the rapid development and application of the internet industry, every item may have its own identification mark in the future, requiring transmission to backend systems for logical processing. Data at different levels will be processed separately, and various industry data will require robust system support, all of which can be achieved through cloud computing.

[0028] Private Cloud: A private cloud is created within a firewall, housing cloud infrastructure and hardware / software resources for various departments within an organization or enterprise to share resources within a data center. In addition to hardware resources, creating a private cloud typically involves cloud equipment (IaaS, Infrastructure as a Service) software.

[0029] Private cloud computing comprises three layers: cloud hardware, cloud platform, and cloud services. The key difference is that cloud hardware consists of the user's own personal computer or server, rather than the cloud computing provider's data center. Cloud computing providers build data centers to provide public cloud services to millions of users, thus requiring hundreds of thousands or even millions of servers. Private cloud computing, for individuals, serves only family and friends; for businesses, it serves only their own employees, customers, and suppliers. Therefore, an individual's or business's own personal computer or server is sufficient to provide cloud services.

[0030] Public cloud: Public cloud typically refers to cloud services provided by third-party providers to users. Public clouds are generally accessible via the Internet and may be free or inexpensive. The core attribute of public clouds is shared resource services. There are many instances of public clouds that provide services across today's open public networks.

[0031] Hybrid cloud: This service model integrates public and private clouds and has become a major model and development direction of cloud computing in recent years. Private clouds are primarily geared towards enterprise users. For security reasons, enterprises prefer to store data in private clouds, but at the same time, they also want access to the computing resources of public clouds. Hybrid clouds are increasingly being adopted in this context. They combine and match public and private clouds to achieve optimal results. This personalized solution achieves both cost-effectiveness and security.

[0032] Artificial Intelligence Cloud Services: Also commonly known as AIaaS (AI as a Service), this is a mainstream service model for artificial intelligence platforms. Specifically, AIaaS platforms break down several common AI (Artificial Intelligence) services and provide them as independent or packaged services in the cloud. This service model is similar to opening an AI-themed marketplace: all developers can access and use one or more AI services provided by the platform through API interfaces. Some experienced developers can also use the platform's AI frameworks and infrastructure to deploy and maintain their own dedicated cloud AI services.

[0033] Since the embodiments of this application involve artificial intelligence cloud services that combine cloud technology and artificial intelligence technology, some basic concepts in the field of artificial intelligence will be explained below: 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 that 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.

[0034] 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 audio processing, computer vision, natural language processing, and machine learning / deep learning.

[0035] In the field of artificial intelligence, machine learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory, among others. 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.

[0036] In related technologies, determining the function of a gene to be tested based on a set of gene expression datasets requires two steps: 1) Calculating the correlation coefficients between the gene to be tested and all genes in the dataset in pairs to find a set of genes that are significantly associated with the gene to be tested; 2) Based on the gene set, using some statistical methods (such as Fisher's Exact Test) to search known gene set annotation libraries, and according to the annotation information of the gene set, labeling various phenotypic descriptions such as biological pathways that the gene to be tested may participate in.

[0037] It can be seen that a key step in functional annotation of genes to be tested is to find a set of genes with significant associations. In bioinformatics, there are various ways to calculate the correlation coefficient between the expression levels of any two genes. These methods mainly measure the linear relationship between the two vectors, but the relationship between gene expression levels is more non-linear, leading to poor accuracy in gene data processing. In addition, the statistical results of statistical methods are greatly affected by sample values ​​and outliers. Therefore, for each gene to be tested, it is necessary to manually determine which statistical method to use to determine significant associations and manually set thresholds for some statistical parameters. Since this is for single-cell expression datasets, many traditional correlation statistical methods are not applicable. Even when using methods specifically developed for single-cell expression datasets, it is still necessary to manually determine which statistical method to use to determine significant associations and manually set thresholds for some statistical parameters. This prevents the automation of the gene functional annotation process, resulting in high human costs in gene data processing.

[0038] In view of this, embodiments of this application provide a cloud-based product that enables cloud computing vendors to provide gene data processing services to various users. The cloud server can discover nonlinear relationships between gene expression levels based on deep learning technology, and provides an automated method to automatically construct a deep learning model (i.e., a gene association model) for any publicly available single-cell expression dataset. This model is used to determine whether there is a significant nonlinear relationship between two input genes (i.e., a gene pair). This method can automatically, universally, and with high accuracy establish gene association models, and can automatically complete the functional annotation of the genes to be tested based on the gene association model.

[0039] Figure 1This is a schematic diagram illustrating the principle of a cloud application scenario provided in this application embodiment, as shown in Figure 100. In this cloud application scenario, both private cloud services and public cloud services can be provided to users. In the data layer, single-cell expression datasets can be deployed within a firewall based on private cloud services, or they can be deployed on cloud servers based on public cloud services. The computing power and algorithms for the aforementioned private or public cloud services are provided based on cloud computing. Further, in the model library layer, each single-cell expression dataset can correspond to one or more gene association models used to determine the degree of nonlinear association between the gene to be tested and candidate genes in the single-cell expression database. These gene association models can be deployed within a firewall based on private cloud services, or they can be deployed on cloud servers based on public cloud services. Further, in the application layer, two interface services are provided externally: 1) Limited free service, where users can select a limited number of publicly available single-cell expression datasets to query the pre-calculated functional annotation results of the gene to be tested. Optionally, the above free service is opened to users who have passed authentication, and the authentication method is to verify whether the user is a registered customer of the cloud product. 2) Customized paid services, such as allowing users to select any gene from all processed, publicly available single-cell expression datasets and view its functional annotation results; or allowing users to select a specific publicly available single-cell expression dataset, input genes of interest, and have the functional annotation results calculated individually for that input gene; or allowing users to upload a new single-cell expression dataset and a gene list, with the functional annotation results calculated individually for each gene in that list. Optionally, these paid services can be made available to users who have passed authentication, with authentication verifying whether the user is a registered and paying customer of the cloud product.

[0040] Figure 2 This is a schematic diagram illustrating the implementation environment of a gene data processing method provided in an embodiment of this application. See also... Figure 2 The implementation environment includes a terminal 201 and a server 202, wherein the terminal 201 and the server 202 are exemplary embodiments of computer devices.

[0041] Terminal 201 has an application that supports gene function annotation installed and running. Optionally, the user logs into the application on terminal 201, inputs the gene to be tested into the application, and terminal 201 sends a gene function query request to server 202, and receives and displays the function annotation information returned by server 202.

[0042] Terminal 201 can be directly or indirectly connected to server 202 via wired or wireless communication. This embodiment of the application does not limit the connection method.

[0043] Server 202 may include at least one of a single server, multiple servers, a cloud computing platform, or a virtualization center. Server 202 is used to provide background services for the aforementioned applications. Optionally, server 202 undertakes the primary computing work, and terminal 201 undertakes the secondary computing work; or, server 202 undertakes the secondary computing work, and terminal 201 undertakes the primary computing work; or, terminal 201 and server 202 collaborate on computing using a distributed computing architecture.

[0044] Optionally, server 202 is an independent physical server, or a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0045] Optionally, the terminal 201 may be a smartphone, tablet computer, laptop computer, desktop computer, smart speaker, smartwatch, MP3 (Moving Picture Experts Group Audio Layer III) player, MP4 (Moving Picture Experts Group Audio Layer IV) player, e-book reader, etc., but is not limited to these.

[0046] Those skilled in the art will understand that the number of terminals 201 described above can be more or less. For example, there may be only one terminal 201, or there may be dozens or hundreds of terminals 201, or even more. This application does not limit the number of terminals or the type of device.

[0047] Figure 3 This is a flowchart of a gene data processing method provided in an embodiment of this application. See also... Figure 3 This embodiment is applied to computer devices. The following description uses a server as an example of a computer device. This embodiment includes the following steps: 301. The server determines the cell type corresponding to the gene to be tested based on the gene function query request of the gene to be tested.

[0048] In some embodiments, the server receives a gene function query request sent by the terminal. Optionally, the gene function query request carries at least a user account identifier and the expression level of the gene to be tested. The server parses the gene function query request to obtain the user account identifier and the expression level of the gene to be tested, and then performs authentication verification on the user account identifier. In response to successful authentication of the user account identifier, the server analyzes the expression level of the gene to be tested to obtain the cell type corresponding to the gene to be tested, and retrieves the single-cell expression dataset corresponding to the cell type from the gene database through step 302 below, and retrieves the gene association model corresponding to the cell type from the model database.

[0049] 302. The server obtains the single-cell expression dataset corresponding to the cell type and the gene association model corresponding to the cell type. The gene association model is used to obtain the degree of non-linear association between the gene to be tested and the candidate genes belonging to the cell type in the single-cell expression dataset.

[0050] In some embodiments, the server stores each single-cell expression dataset in a gene database along with its corresponding cell type. Typically, a single-cell expression dataset includes gene data for multiple cell types. The server uses the cell type as an index to query the gene database for the corresponding index content. When the index matches any index content, it reads that index content to obtain the single-cell expression dataset corresponding to that cell type.

[0051] In some embodiments, the server stores each cell type and its corresponding gene association model in the model database. Using the cell type as an index, the server queries the model database for the corresponding index content. When the index matches any index content, it reads that index content to obtain the gene association model corresponding to that cell type.

[0052] In some embodiments, since a single-cell expression dataset may contain gene data for multiple cell types, after the server trains a gene association model for the gene data of each cell type, it can also store the single-cell expression dataset, cell types, and gene association models accordingly. In this case, when the gene function query request carries the dataset identifier of the single-cell expression dataset, the server can first use the dataset identifier of the single-cell expression dataset as an index to query the index content corresponding to that index from the model database. When the index matches any index content, it reads that index content to obtain the gene association models corresponding to each cell type contained in the single-cell expression dataset, and then queries the gene association model corresponding to the cell type specified in the gene function query request. This improves the efficiency of searching for gene association models.

[0053] In some embodiments, the gene association model can be any deep learning model, such as a CNN (Convolutional Neural Network), a DNN (Deep Neural Networks), or a VGG (Visual Geometry Group). This application does not specifically limit the model structure of the gene association model.

[0054] 303. The server calls the gene association model to process the gene to be tested, and obtains the degree of nonlinear association between the gene to be tested and multiple candidate genes belonging to the cell type in the single cell expression dataset.

[0055] In some embodiments, the server constructs multiple gene pairs based on the gene to be tested and the multiple candidate genes, wherein any gene pair in the multiple gene pairs includes the gene to be tested and any candidate gene in the multiple candidate genes; the multiple gene pairs are input into the gene association model, and the multiple gene pairs are convolved by the gene association model to obtain multiple association coefficients, wherein any one of the multiple association coefficients is used to represent the degree of nonlinear association between the genes contained in any gene pair in the multiple gene pairs.

[0056] Optionally, the server pairs the gene to be tested with each candidate gene in the single-cell expression dataset, iterates through the single-cell expression dataset, and finally obtains the multiple gene pairs, thus enabling comprehensive detection of the entire single-cell expression dataset.

[0057] Optionally, the server can pair the gene to be tested with a subset of candidate genes in the single-cell expression dataset to form multiple gene pairs, thereby saving server computation. For example, the subset of candidate genes can be candidate genes with a query frequency higher than a frequency threshold, or the subset of candidate genes can be candidate genes selected according to a target interval, or the subset of candidate genes can be randomly selected candidate genes. This application does not specifically limit these limitations.

[0058] In some embodiments, taking a CNN model as an example, the gene association model includes an input layer, at least one convolutional layer, and an output layer. For any gene pair among the multiple gene pairs, the expression levels of the gene to be tested and the candidate gene in the gene pair are input into the at least one convolutional layer. The at least one convolutional layer performs convolution processing on the expression levels of the gene to be tested and the candidate gene, and the feature vector output by the last convolutional layer is reduced in dimensionality. The dimensionality-reduced feature vector is input into the output layer. The output layer (i.e., the fully connected layer) performs weighted processing on the dimensionality-reduced feature vector, and the weighted feature vector is normalized by Sigmoid to obtain the association coefficient of the gene pair. The association coefficient is used to represent the degree of nonlinear association between the gene to be tested and the candidate gene in the gene pair.

[0059] Optionally, the correlation coefficient is positively correlated with the degree of nonlinear correlation. That is, the larger the correlation coefficient, the more significant the nonlinear correlation between the two genes in the gene pair, and the smaller the correlation coefficient, the less significant the nonlinear correlation between the two genes in the gene pair.

[0060] In some embodiments, in the input layer of the CNN model, the server determines a 32×32-dimensional feature vector based on the expression levels of the gene to be tested and the candidate genes, and inputs the feature vector into the at least one convolutional layer for convolution processing.

[0061] In some embodiments, based on the above CNN model, an activation layer can be added between any two adjacent convolutional layers. This activation layer is sandwiched between consecutive convolutional layers and is used to perform non-linear mapping on the feature vector output by the previous convolutional layer. The activation layer may include ReLU activation function, Sigmoid activation function, Tanh activation function, etc. This application embodiment does not specifically limit the type of activation function in the activation layer. In one example, the activation function in the activation layer is ReLU.

[0062] Optionally, an activation layer is added between all adjacent convolutional layers in the CNN model, that is, an activation layer is added every other convolutional layer, or an activation layer is added every two convolutional layers in the CNN model, or one or more activation layers are added in the CNN model at other equal or unequal intervals. This application embodiment does not specifically limit this.

[0063] In some embodiments, based on the CNN model described above, a pooling layer can be added between any two adjacent convolutional layers. This pooling layer, sandwiched between consecutive convolutional layers, is used to compress the amount of data and parameters output by the previous convolutional layer, thereby reducing overfitting. Optionally, this pooling layer can be a max pooling layer or a mean pooling layer. This application does not specifically limit the type of pooling layer. In one example, all pooling layers in the CNN model are max pooling layers.

[0064] Optionally, a pooling layer may be added between all adjacent convolutional layers in the CNN model, that is, a pooling layer may be added every other convolutional layer, or a pooling layer may be added every two convolutional layers in the CNN model, or one or more pooling layers may be added in the CNN model at other equal or unequal intervals. This application embodiment does not specifically limit this.

[0065] In some embodiments, the CNN model reduces the dimensionality of the feature vector output by the last convolutional layer through the Flatten layer. Figuratively speaking, the Flatten layer is used to "flatten" the feature vector output by the last convolutional layer, that is, to reduce the multidimensional feature vector to one dimension. The Flatten layer can be used for the transition from the convolutional layer to the fully connected layer (output layer).

[0066] In some embodiments, the output layer includes two fully connected layers, and an activation layer may be embedded between the two fully connected layers. The two fully connected layers can fully extract the association features between the two genes, and the Sigmoid function is used to normalize the feature vector processed by the latter fully connected layer.

[0067] 304. The server outputs functional annotation information of candidate genes whose nonlinear association degree meets the first target condition, where the first target condition is used to indicate that the nonlinear association degree is greater than or equal to the association degree threshold.

[0068] In some embodiments, the server acquires candidate genes with association coefficients greater than an association degree threshold and outputs functional annotation information for these candidate genes. The association degree threshold can be any value greater than or equal to 0 and less than or equal to 1; for example, the association degree threshold is 0.9.

[0069] In some embodiments, the server determines the candidate gene with the highest association coefficient and outputs the functional annotation information of the candidate gene with the highest association coefficient.

[0070] In some embodiments, the server sorts each candidate gene in the single-cell expression dataset in descending order of non-linear correlation and outputs the functional annotation information of the candidate gene ranked in the first target position. The first target position can be any number of digits greater than or equal to 1. For example, the server outputs the functional annotation information of the candidate gene ranked in the first position, or the server outputs the functional annotation information of the candidate gene ranked in the first three positions. This application embodiment does not specifically limit this.

[0071] All of the above-mentioned optional technical solutions can be combined in any way to form optional embodiments of this disclosure, and will not be described in detail here.

[0072] The method provided in this application, through a gene function query request for the gene to be tested, can call a gene association model corresponding to the cell type to which the gene to be tested belongs, to mine the degree of nonlinear association between the gene to be tested and known candidate genes, and use candidate genes with a high degree of nonlinear association to annotate the functional annotation information of the gene to be tested. This method of calling a gene association model to extract nonlinear relationships is completely different from the method of extracting linear relationships in traditional statistics. It can deeply mine candidate genes with higher similarity to the gene to be tested. This similarity is not a linear similarity but an implicit nonlinear similarity, thus greatly improving the accuracy of the gene data processing process.

[0073] Figure 4 This is a flowchart illustrating the training process of a gene association model provided in an embodiment of this application. Please refer to it. Figure 4 The training process of a single gene association model will be used as an example for illustration. This embodiment includes the following steps: 401. The server obtains multiple sample genes from the single-cell expression dataset, and the degree of variation of these multiple sample genes is greater than the variation threshold.

[0074] In some embodiments, the server uses the FindVariableFeatures function in the Seurat package to screen from all genes in the single-cell expression dataset to obtain multiple sample genes (commonly known as highly variable genes) whose degree of variation is greater than the variation threshold.

[0075] In some embodiments, the server uses a local weighted regression method to fit a straight line to the logarithm of the variance of gene expression and the logarithm of the mean of gene expression. Then, it uses the observed mean and expected variance to standardize the gene expression. Finally, it calculates the variance of each window based on the gene expression that retains the largest standardized value, and obtains the vst index value.

[0076] In some embodiments, the server sets the selection mode parameter of the FindVariableFeatures function to selection.method = "vst". The vst selection mode is to use a sliding window approach to calculate the gene expression level of each window on the whole genome of a single cell according to the target step size, and to calculate the vst index value of each window using the window position and gene expression level of each window. Optionally, each gene is set as a window.

[0077] In the above process, VST is an indicator that measures the degree of gene expression differentiation between different populations by calculating the variance of gene expression levels. It can be used to identify highly differentiated regions, that is, to identify highly variable genes. In one example, a Python script is used to calculate the gene expression level in each window, and an R script is used to calculate the VST of each window. Of course, other scripting languages ​​can also be used to screen for highly variable genes. This application does not specifically limit the type of scripting language.

[0078] In some embodiments, after calculating the VST index value for each window, since each window represents a gene, the VST index value for each gene is also obtained. Optionally, the server directly selects genes with VST index values ​​greater than the mutation threshold as the plurality of sample genes. Optionally, the server sorts the genes in descending order of VST index values ​​and selects the gene located at the second target position in the sorting as the plurality of sample genes. The second target position can be any number of bits greater than or equal to 1; for example, the first 2000 genes can be selected as the plurality of sample genes.

[0079] In some embodiments, the server can also set the filtering mode parameter of the FindVariableFeatures function to mean.var.plot or dispersion. This application embodiment does not specifically limit the filtering mode parameter. In mean.var.plot mode, the average expression level and dispersion of each gene are calculated separately. Based on the average expression level, each gene is distributed into a certain number (default 20) sub-bins, and the degree of variation in each bin is calculated. Sample genes with a degree of variation greater than a variation threshold are selected. In dispersion mode, the sample gene with the highest deviation value is directly selected.

[0080] 402. Based on the multiple sample genes, the server obtains multiple positive gene pairs and multiple negative gene pairs, wherein the degree of non-linear association between the genes contained in any positive gene pair is greater than the association degree threshold, and the degree of non-linear association between the genes contained in any negative gene pair is less than the association degree threshold.

[0081] In some embodiments, the server identifies two sample genes belonging to the same protein complex from the plurality of sample genes as a positive gene pair; and identifies any sample gene in the positive gene pair and any sample gene in the plurality of sample genes other than the positive gene pair as a negative gene pair.

[0082] In the above process, two sample genes known to be in the same protein complex are selected as sample gene pairs. This is because the epitope / hypertope relationship between the constituent genes of a protein complex in a cell must be significantly correlated, and the association between constituent genes is often non-linear. Therefore, it can be determined that sample genes belonging to the same protein complex have a significant non-linear association (i.e., the degree of non-linear association is greater than the association threshold). Conversely, based on existing positive gene pairs, one sample gene in the positive gene pair is fixed, and another sample gene that does not belong to the above positive gene pair is randomly selected to construct a negative gene pair. Since the two sample genes in the negative gene pair must not belong to the same protein complex, the degree of non-linear association between the two sample genes can be considered to be poor, i.e., the degree of non-linear association is less than the association threshold.

[0083] In an exemplary embodiment, assuming a total of 2000 sample genes are selected, all positive gene pairs belonging to the same protein complex are first obtained from these 2000 sample genes. It is necessary to ensure that the two sample genes in each positive gene pair are located in the same protein complex. All positive gene pairs can constitute a positive sample set. Then, the same number of negative gene pairs are randomly generated from the positive sample set. It is necessary to ensure that the two sample genes in each negative gene pair are not in the same protein complex.

[0084] For example, the 20S methyltransferase complex comprises three genes: CLNS1A, PRMT5, and WDR77. If any two of these three genes are present in 2000 sample genes, then these two genes constitute a positive gene pair. The above example uses only the methyltransferase complex and should not be construed as limiting the scope of protein complexes. For example, protein complexes also include the CORUM protein complex and the NURSA protein complex, etc., and this application does not limit the scope of these examples.

[0085] In some embodiments, the server may also randomly select a target proportion of gene pairs (containing both positive and negative gene pairs) from either the positive or negative sample set as the training sample set, while the remaining gene pairs serve as the test sample set. This simplifies the process of obtaining the test sample set and saves server processing resources. The target proportion is any value greater than or equal to 0 and less than or equal to 1, for example, 80%.

[0086] 403. Based on the multiple positive gene pairs and the multiple negative gene pairs, the server adjusts the parameters of the initial depth model to obtain the gene association model corresponding to the single-cell expression data.

[0087] In some embodiments, for any one of the plurality of positive gene pairs and the plurality of negative gene pairs, the server calls the initial deep model to predict the association coefficient of any one gene pair; based on the association coefficient of the plurality of positive gene pairs and the plurality of negative gene pairs, the loss function value of the initial deep model is determined; in response to the loss function value not meeting the stopping condition, the initial deep model is iteratively trained until the loss function value meets the stopping condition, at which point training stops, and the gene association model is obtained.

[0088] Optionally, the stopping condition is that the loss function value is less than or equal to the loss threshold, or the stopping condition is that the number of iterations is greater than the number threshold. This application embodiment does not specifically limit the content of the stopping condition.

[0089] In some embodiments, the server inputs any gene pair into the initial depth model, processes the gene pair through the initial depth model, and outputs the correlation coefficient between the two sample genes in the gene pair. Iterates through all positive gene pairs and all negative gene pairs to obtain the correlation coefficient of each gene pair. At this point, the reference coefficient of all positive gene pairs can be regarded as 1, and the reference coefficient of all negative gene pairs can be regarded as 0. Based on the correlation coefficient and reference coefficient of each gene pair, the loss function value of this iteration process can be obtained.

[0090] In some embodiments, the initial deep model can be any deep learning model, such as a CNN, a DNN, or a VGG. This application does not specifically limit the model structure of the initial deep model.

[0091] In some embodiments, taking a CNN model as an example of an initial deep model, the CNN model includes an input layer, at least one convolutional layer, and an output layer. The expression levels of the two sample genes in any gene pair in the input layer are input into the at least one convolutional layer. The expression levels of the two sample genes are convolved by the at least one convolutional layer, and the feature vector output by the last convolutional layer is reduced in dimensionality. The dimensionality-reduced feature vector is input into the output layer, and the dimensionality-reduced feature vector is weighted by the output layer (i.e., the fully connected layer). The weighted feature vector is then normalized by Sigmoid to obtain the correlation coefficient of any gene pair. This correlation coefficient is used to represent the degree of nonlinear correlation between the two sample genes in any gene pair.

[0092] Optionally, the correlation coefficient is positively correlated with the degree of nonlinear correlation. That is, the larger the correlation coefficient, the more significant the nonlinear correlation between the two genes in the gene pair, and the smaller the correlation coefficient, the less significant the nonlinear correlation between the two genes in the gene pair.

[0093] In some embodiments, in the input layer of the CNN model, the server determines a 32×32-dimensional feature vector based on the expression levels of the gene to be tested and the candidate genes, and inputs the feature vector into the at least one convolutional layer for convolution processing.

[0094] In some embodiments, based on the above CNN model, an activation layer can be added between any two adjacent convolutional layers. This activation layer is sandwiched between consecutive convolutional layers and is used to perform non-linear mapping on the feature vector output by the previous convolutional layer. The activation layer may include ReLU activation function, Sigmoid activation function, Tanh activation function, etc. This application embodiment does not specifically limit the type of activation function in the activation layer. In one example, the activation function in the activation layer is ReLU.

[0095] Optionally, an activation layer is added between all adjacent convolutional layers in the CNN model, that is, an activation layer is added every other convolutional layer, or an activation layer is added every two convolutional layers in the CNN model, or one or more activation layers are added in the CNN model at other equal or unequal intervals. This application embodiment does not specifically limit this.

[0096] In some embodiments, based on the CNN model described above, a pooling layer can be added between any two adjacent convolutional layers. This pooling layer, sandwiched between consecutive convolutional layers, is used to compress the amount of data and parameters output by the previous convolutional layer, thereby reducing overfitting. Optionally, this pooling layer can be a max pooling layer or a mean pooling layer. This application does not specifically limit the type of pooling layer. In one example, all pooling layers in the CNN model are max pooling layers.

[0097] Optionally, a pooling layer may be added between all adjacent convolutional layers in the CNN model, that is, a pooling layer may be added every other convolutional layer, or a pooling layer may be added every two convolutional layers in the CNN model, or one or more pooling layers may be added in the CNN model at other equal or unequal intervals. This application embodiment does not specifically limit this.

[0098] In some embodiments, based on the CNN model described above, a random deactivation (Dropout) layer can be added between any two adjacent convolutional layers. This Dropout layer is sandwiched between consecutive convolutional layers and is used to cause a neuron in the previous convolutional layer to stop working with a certain probability p, thereby reducing the overfitting phenomenon of the model and improving the generalization ability of the model.

[0099] Optionally, a Dropout layer is added between all adjacent convolutional layers in the CNN model, that is, a Dropout layer is added every other convolutional layer, or a Dropout layer is added every two convolutional layers in the CNN model, or one or more Dropout layers are added in the CNN model at other equal or unequal intervals. This application embodiment does not specifically limit this.

[0100] In some embodiments, the CNN model reduces the dimensionality of the feature vector output by the last convolutional layer through the Flatten layer. Figuratively speaking, the Flatten layer is used to "flatten" the feature vector output by the last convolutional layer, that is, to reduce the multidimensional feature vector to one dimension. The Flatten layer can be used for the transition from the convolutional layer to the fully connected layer (output layer).

[0101] In some embodiments, the output layer includes two fully connected layers, with an activation layer embedded between them. The two fully connected layers effectively extract the association features between the two genes, and the sigmoid function is used to normalize the feature vector processed by the latter fully connected layer. Optionally, a dropout layer can be embedded after the activation layer between the two fully connected layers to further improve the model's generalization ability.

[0102] In an exemplary embodiment, a 23-layer CNN model is used as an example for illustration. Each layer will be described below.

[0103] The first layer is a 2D convolutional layer, calling the Conv2D function from the Keras package. The feature vector output by the first layer has a dimension of 32, and the width and height of the 2D convolutional window are (3,3). Other hyperparameters include padding=same and input representation dimension input_shape=(32,32,1). Optionally, the server implements the first layer based on the following code: model.add(Conv2D(32, (3, 3), padding='same', input_shape=input_shape)) The second layer: the activation layer, which calls the `Activation` function from the `keras` package, setting the activation parameter to `relu`. Optionally, the server can implement the second layer based on the following code: model.add(Activation('relu')) The third layer is a 2D convolutional layer that uses the Conv2D function from the Keras package. The feature vector output by the third layer has a dimension of 32, and the width and height of the 2D convolutional window are (3,3). Optionally, the server can implement the third layer based on the following code: model.add(Conv2D(32, (3, 3))) The fourth layer: the activation layer, which calls the `Activation` function in the `keras` package, setting the activation parameter to `relu`. Optionally, the server can implement the fourth layer based on the following code: model.add(Activation('relu')) Fifth layer: Pooling layer, calling the MaxPooling2D function from the Keras package, with the maximum pooling window size set to (2,2). Optionally, the server implements the fifth layer based on the following code: model.add(MaxPooling2D(pool_size=(2, 2))) The sixth layer: the Dropout layer, which calls the Dropout function from the Keras package, with the Dropout parameter set to 0.1. Optionally, the server can be implemented using the following code: model.add(Dropout(0.1)) Layer 7: A 2D convolutional layer, calling the Conv2D function from the Keras package. The feature vector output by layer 7 has a dimension of 64, and the width and height of the 2D convolutional window are (3,3). Other hyperparameters include padding=same. Optionally, the server implements layer 7 based on the following code: model.add(Conv2D(64, (3, 3), padding='same')) Layer 8: Activation layer, which calls the Activation function in the Keras package, setting the activation parameter to ReLU. Optionally, the server can implement layer 8 based on the following code: model.add(Activation('relu')) Layer 9: A 2D convolutional layer that uses the Conv2D function from the Keras package. The feature vector output by layer 9 has a dimension of 64, and the width and height of the 2D convolutional window are (3,3). Optionally, the server can implement layer 9 based on the following code: model.add(Conv2D(64, (3, 3))) Layer 10: Activation layer, which calls the Activation function in the Keras package, setting the activation parameter to ReLU. Optionally, the server can implement layer 10 based on the following code: model.add(Activation('relu')) The eleventh layer: the pooling layer, which calls the MaxPooling2D function from the Keras package, with the maximum pooling window size set to (2,2). Optionally, the server can implement the eleventh layer based on the following code: model.add(MaxPooling2D(pool_size=(2, 2))) The twelfth layer: the Dropout layer, which calls the Dropout function from the Keras package, with the Dropout parameter set to 0.1. Optionally, the server can be implemented based on the following code: model.add(Dropout(0.1)) The thirteenth layer is a 2D convolutional layer that uses the Conv2D function from the Keras package. The feature vector output by the thirteenth layer has a dimension of 128, and the width and height of the 2D convolutional window are (3,3). Other hyperparameters include padding=same. Optionally, the server can implement the thirteenth layer based on the following code: model.add(Conv2D(128, (3, 3), padding='same')) The fourteenth layer: the activation layer, which calls the `Activation` function in the `keras` package, setting the activation parameter to `relu`. Optionally, the server can implement the fourteenth layer based on the following code: model.add(Activation('relu')) The fifteenth layer is a 2D convolutional layer that uses the Conv2D function from the Keras package. The feature vector output by the fifteenth layer has a dimension of 128, and the width and height of the 2D convolutional window are (3,3). Optionally, the server can implement the fifteenth layer based on the following code: model.add(Conv2D(128, (3, 3))) The sixteenth layer: the activation layer, which calls the `Activation` function in the `keras` package, setting the activation parameter to `relu`. Optionally, the server can implement the sixteenth layer based on the following code: model.add(Activation('relu')) The seventeenth layer: the pooling layer, which calls the MaxPooling2D function from the Keras package, with the maximum pooling window size set to (2,2). Optionally, the server implements the seventeenth layer based on the following code: model.add(MaxPooling2D(pool_size=(2, 2))) The eighteenth layer: the Dropout layer, which calls the Dropout function from the Keras package, with the Dropout parameter set to 0.1. Optionally, the server implements the eighteenth layer based on the following code: model.add(Dropout(0.1)) Layer 19: The Flatten layer, which calls the Flatten function from the keras package. Optionally, the server implements layer 19 based on the following code: model.add(Flatten()) Layer 20: The fully connected layer, which calls the `dense` function in the Keras package and sets the parameter `units:=512`. Optionally, the server can implement Layer 20 based on the following code: model.add(Dense(512)) Layer 21: Activation function, which calls the Activation function in the Keras package, setting the activation parameter to ReLU. Optionally, the server implements Layer 21 based on the following code: model.add(Activation('relu')) Layer 22: The Dropout layer, which calls the Dropout function from the Keras package, with the Dropout parameter set to 0.1. Optionally, the server can implement Layer 22 based on the following code: model.add(Dropout(0.1)) Layer 23: A fully connected layer that calls the `dense` function from the `keras` package, setting the parameter `units=1` and the normalization parameter `activation=sigmoid`. Optionally, the server can implement Layer 23 based on the following code: model.add(Dense(1, activation='sigmoid')) In summary, the 23-layer CNN model described above can be implemented using the following code: def make_model(input_shape): model=Sequential() # model.add(Conv2D(32, (3, 3), padding='same', input_shape=x_test.shape[1:])) model.add(Conv2D(32, (3, 3), padding='same', input_shape=input_shape)) model.add(Activation('relu')) model.add(Conv2D(32, (3, 3))) model.add(Activation('relu')) model.add(MaxPooling2D(pool_size=(2, 2))) model.add(Dropout(0.1)) model.add(Conv2D(64, (3, 3), padding='same')) model.add(Activation('relu')) model.add(Conv2D(64, (3, 3))) model.add(Activation('relu')) model.add(MaxPooling2D(pool_size=(2, 2))) model.add(Dropout(0.1)) model.add(Conv2D(128, (3, 3), padding='same')) model.add(Activation('relu')) model.add(Conv2D(128, (3, 3))) model.add(Activation('relu')) model.add(MaxPooling2D(pool_size=(2, 2))) model.add(Dropout(0.1)) model.add(Flatten()) model.add(Dense(512)) model.add(Activation('relu')) model.add(Dropout(0.1)) model.add(Dense(1, activation='sigmoid')) return model In some embodiments, when adjusting the parameters of the initial depth model, a gradient descent optimization algorithm can be used, or a stochastic gradient descent (SGD) optimization algorithm can be used, or a Nadam optimization algorithm can be used. This application embodiment does not specifically limit the optimization algorithm of the initial depth model.

[0104] In an exemplary embodiment, the optimization algorithm can be set to the Nadam algorithm by using optimizer=Nadam(clipnorm=1.). The Nadam algorithm can find the optimal optimization path faster and improves the probability of successful model training compared to the SGD method. In addition, apart from modifying optimizerName, developers do not need to manually debug any other parameters.

[0105] In another exemplary embodiment, the server can use a combination of various optimization algorithms to optimize the model. For example, the optimization algorithm for the initial depth model can be set using the following code: ## def load_optimizer(optimizerName): if (optimizerName == ''sgd''): optimizer=SGD(lr=0.1, decay=1e-6, momentum=0.9, nesterov=True) if (optimizerName == ''sgd_clipnorm''): optimizer=SGD(lr=0.1, decay=1e-6, momentum=0.9, nesterov=True,clipnorm=1) if (optimizerName == ''nadam''): optimizer=Nadam(clipnorm=1.) return optimizer The first SGD optimizer is implemented using the following code: optimizer=SGD(lr=0.1, decay=1e-6, momentum=0.9, nesterov=True) The second SGD optimizer has an additional parameter, clipnorm=1, to control gradient clipping, and is implemented using the following code: optimizer=SGD(lr=0.1, decay=1e-6, momentum=0.9, nesterov=True, clipnorm=1) The third Nadam optimizer has an additional parameter, clipnorm=1, that controls gradient clipping. It is implemented using the following code: optimizer=Nadam(clipnorm=1.) In some embodiments, for any gene pair among the plurality of positive gene pairs and the plurality of negative gene pairs, the server can obtain multiple association coefficients predicted by multiple gene association models for any gene pair, wherein the multiple gene association models correspond to multiple cell types in at least one single-cell expression dataset; based on the multiple association coefficients of any gene pair, it is determined whether the gene pair meets a second target condition, wherein the second target condition is used to indicate that the prediction difficulty of the multiple gene association models for any gene pair is lower than a first target threshold; based on gene pairs that meet the second target condition, the parameters of the gene association model corresponding to the cell type are adjusted.

[0106] In an exemplary scenario, for the following five single-cell expression datasets: GSE108989, GSE140228, GSE146771, GSE98638, and GSE99254, after the first round of debugging, 17 gene association models can be successfully debugged. This is due to the inherent characteristics of the five single-cell expression datasets. Each dataset contains multiple cell types, and the server needs to model different gene association models for different cell types. For example, cell types include, but are not limited to, CD8 T cells, DC cells, lymphoid cells, macrophage cells, NK cells, mono cells, ILC cells, firoblast cells, macrophage cells, B cells, and CD4 T cells.

[0107] Based on the 17 gene association models obtained after the first round of debugging, all positive gene pairs and all negative gene pairs can be filtered again to remove some gene pairs that are difficult for the gene association models to predict. Based on the remaining gene pairs after filtering, the parameters of the 17 gene association models can be fine-tuned again to obtain the final 17 gene association models, which can then be stored in the model database.

[0108] In some embodiments, in addition to fine-tuning the parameters of the above 17 gene association models, since there are unfiltered gene pairs with high training difficulty during the first round of debugging, the initial deep model corresponding to a certain cell type will eventually fail to converge, i.e., training will fail. By using the filtered positive gene pairs and negative gene pairs, the initial deep model can be trained again to obtain gene association models corresponding to more cell types.

[0109] In some embodiments, the server can also use the above 17 gene association models to filter gene pairs in other single-cell expression datasets besides the above 5 single-cell expression datasets. Similarly, it can train targeted gene association models for other single-cell expression datasets, and this process is fully automated and does not require manual parameter tuning.

[0110] The above-described method of filtering gene pairs not only improves the accuracy of gene association models on the test sample set, but also automates this fine-tuning process, requiring no manual parameter tuning by developers. Table 1 below shows the accuracy of the 17 gene association models obtained after the first round of debugging on the test sample set, all of which are greater than 0.8.

[0111]

[0112] In some embodiments, the server determines whether any gene pair meets the second target condition based on the following method: based on the plurality of correlation coefficients, determine the first coefficient quantity and the second coefficient quantity, where the first coefficient quantity is the number of coefficients greater than or equal to the second target threshold among the plurality of correlation coefficients, and the second coefficient quantity is the number of coefficients less than the second target threshold among the plurality of correlation coefficients; in response to the first coefficient quantity being greater than the second coefficient quantity and the any gene pair being a positive gene pair, or, the first coefficient quantity being less than the second coefficient quantity and the any gene pair being a negative gene pair, determine that the any gene pair meets the second target condition.

[0113] That is to say, for any gene pair (whether it is a positive gene pair or a negative gene pair) by the server, use multiple gene association models to predict the plurality of correlation coefficients of the two genes in the any gene pair. Assuming that 0.5 is used as the second target threshold, determine that the first coefficient quantity is the number m of correlation coefficients greater than or equal to 0.5, and determine that the second coefficient quantity is the number n of correlation coefficients less than 0.5. For the positive sample set, determine that the positive gene pairs with m > n meet the second target condition; for the negative sample set, determine that the negative gene pairs with m < n meet the second target condition. This can eliminate some sample gene pairs with unclear negative or positive characteristics and high prediction difficulty, thereby facilitating the improvement of the accuracy of the gene association model.

[0114] In some embodiments, it can be set that the number of gene pairs in the negative sample set before filtering is 3 times the number of gene pairs in the positive sample set. After the above filtering process, it is equivalent to downsampling the negative sample set according to the number of gene pairs in the filtered positive sample set.

[0115] All the above optional technical solutions can be combined arbitrarily to form the optional embodiments of the present disclosure, which will not be elaborated here one by one.

[0116] In the embodiments of the present application, an automated model training process is provided. Without additional manual adjustment of the remaining parameters except for changing the optimization algorithm, the accuracy of most gene association models on the test sample set can be quickly greater than 0.8.

[0117] For example, for the 19 gene association models shown in Table 2 below, without the filtering step, a high-accuracy model cannot be trained or the training fails. However, after the filtering step, the accuracy of all models on the test set is greater than 0.8.

[0118]

[0119] Figure 5 It is a schematic structural diagram of a gene data processing device provided by the embodiments of the present application. Please refer to Figure 5 The device includes: The first determining module 501 is used to determine the cell type corresponding to the gene to be tested based on the gene function query request of the gene to be tested. The first acquisition module 502 is used to acquire the single-cell expression dataset corresponding to the cell type and the gene association model corresponding to the cell type. The gene association model is used to acquire the degree of non-linear association between the gene to be tested and the candidate genes belonging to the cell type in the single-cell expression dataset. The processing module 503 is used to call the gene association model to process the gene to be tested and obtain the degree of nonlinear association between the gene to be tested and multiple candidate genes belonging to the cell type in the single cell expression dataset. Output module 504 is used to output functional annotation information of candidate genes whose nonlinear association degree meets the first target condition, wherein the first target condition is used to indicate that the nonlinear association degree is greater than or equal to the association degree threshold.

[0120] The device provided in this application embodiment can call a gene association model corresponding to the cell type to which the gene to be tested belongs by requesting the gene function of the gene to be tested. It can then mine the degree of nonlinear association between the gene to be tested and known candidate genes, and use candidate genes with a high degree of nonlinear association to annotate the functional annotation information of the gene to be tested. This method of calling the gene association model to extract nonlinear relationships is completely different from the method of extracting linear relationships in traditional statistics. It can deeply mine candidate genes with higher similarity to the gene to be tested. This similarity is not a linear similarity but an implicit nonlinear similarity, thus greatly improving the accuracy of the gene data processing process.

[0121] In one possible implementation, the processing module 503 is used to: Based on the gene to be tested and the multiple candidate genes, multiple gene pairs are constructed, and any one of the multiple gene pairs includes the gene to be tested and any one of the multiple candidate genes. The multiple gene pairs are input into the gene association model, and the gene association model is used to perform convolution processing on the multiple gene pairs to obtain multiple association coefficients. Any one of the multiple association coefficients is used to represent the degree of nonlinear association between the genes contained in any gene pair.

[0122] In one possible implementation, based on Figure 5 The device comprises: The second acquisition module is used to acquire multiple sample genes from the single-cell expression dataset, wherein the degree of variation of the multiple sample genes is greater than the variation threshold. The third acquisition module is used to acquire multiple positive gene pairs and multiple negative gene pairs based on the multiple sample genes, wherein the degree of non-linear correlation between the genes contained in any positive gene pair is greater than the correlation threshold, and the degree of non-linear correlation between the genes contained in any negative gene pair is less than the correlation threshold. The adjustment module is used to adjust the parameters of the initial deep model based on the multiple positive gene pairs and the multiple negative gene pairs to obtain the gene association model.

[0123] In one possible implementation, the third acquisition module is used to: From these multiple sample genes, two sample genes belonging to the same protein complex were identified as a positive gene pair; Any sample gene in the positive gene pair and any sample gene in the plurality of sample genes other than the positive gene pair are identified as a negative gene pair.

[0124] In one possible implementation, the adjustment module is used to: For any one of the multiple positive gene pairs and the multiple negative gene pairs, call the initial deep model to predict the association coefficient of any one gene pair; Based on the correlation coefficients between the multiple positive gene pairs and the multiple negative gene pairs, the loss function value of the initial deep model is determined; In response to the loss function value not meeting the stopping condition, the initial deep model is trained iteratively until the loss function value meets the stopping condition, at which point training stops and the gene association model is obtained.

[0125] In one possible implementation, based on Figure 5 The device comprises: The fourth acquisition module is used to acquire multiple association coefficients predicted by multiple gene association models for any gene pair among the multiple positive gene pairs and the multiple negative gene pairs. The multiple gene association models correspond to multiple cell types in at least one single-cell expression dataset. The second determining module is used to determine whether any gene pair meets the second target condition based on the multiple association coefficients of any gene pair. The second target condition is used to indicate that the prediction difficulty of the multiple gene association model for any gene pair is lower than the first target threshold. The adjustment module is also used to adjust the parameters of the gene association model corresponding to the cell type based on gene pairs that meet the second target condition.

[0126] In one possible implementation, the second determining module is used to: Based on the multiple correlation coefficients, the number of first coefficients and the number of second coefficients are determined. The number of first coefficients is the number of coefficients among the multiple correlation coefficients that are greater than or equal to the second target threshold, and the number of second coefficients is the number of coefficients among the multiple correlation coefficients that are less than the second target threshold. In response to the first coefficient being greater than the second coefficient and the gene pair being a positive gene pair, or the first coefficient being less than the second coefficient and the gene pair being a negative gene pair, it is determined that the gene pair meets the second target condition.

[0127] All of the above-mentioned optional technical solutions can be combined in any way to form optional embodiments of this disclosure, and will not be described in detail here.

[0128] It should be noted that the gene data processing device provided in the above embodiments is only illustrated by the division of the above functional modules when processing gene data. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. In addition, the gene data processing device and the gene data processing method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process can be found in the gene data processing method embodiments, which will not be repeated here.

[0129] Figure 6 This is a schematic diagram of a computer device 600 provided in an embodiment of this application. The computer device 600 can vary significantly due to differences in configuration or performance. The computer device 600 includes one or more Central Processing Units (CPUs) 601 and one or more memories 602. The memories 602 store at least one computer program, which is loaded and executed by the one or more processors 601 to implement the gene data processing methods provided in the various embodiments described above. Optionally, the computer device 600 also includes wired or wireless network interfaces, a keyboard, and input / output interfaces for input and output. The computer device 600 also includes other components for implementing device functions, which will not be elaborated here.

[0130] Figure 7This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Optionally, the computer device is described using terminal 700 as an example. The device type of terminal 700 includes: smartphone, tablet computer, MP3 player (Moving Picture Experts Group Audio Layer III), MP4 player (Moving Picture Experts Group Audio Layer IV), laptop computer, or desktop computer. Terminal 700 may also be referred to as user equipment, portable terminal, laptop terminal, desktop terminal, or other names.

[0131] Typically, terminal 700 includes a processor 701 and a memory 702.

[0132] Optionally, the processor 701 includes one or more processing cores, such as a quad-core processor, an octa-core processor, etc. Optionally, the processor 701 is implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). In some embodiments, the processor 701 includes 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, the processor 701 integrates a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content required to be displayed on the screen. In some embodiments, the processor 701 also includes an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0133] In some embodiments, memory 702 includes one or more computer-readable storage media, optionally non-transitory. Optionally, memory 702 also includes 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 memory 702 is used to store at least one program code, which is executed by processor 701 to implement the gene data processing methods provided in the various embodiments of this application.

[0134] In some embodiments, the terminal 700 may also optionally include a peripheral device interface 703 and at least one peripheral device. The processor 701, memory 702, and peripheral device interface 703 can be connected via a bus or signal line. Each peripheral device can be connected to the peripheral device interface 703 via a bus, signal line, or circuit board. Specifically, the peripheral device includes at least one of the following: a radio frequency circuit 704, a display screen 705, a camera assembly 706, an audio circuit 707, a positioning assembly 708, and a power supply 709.

[0135] Peripheral device interface 703 can be used to connect at least one I / O (Input / Output) related peripheral device to processor 701 and memory 702. In some embodiments, processor 701, memory 702 and peripheral device interface 703 are integrated on the same chip or circuit board; in some other embodiments, any one or two of processor 701, memory 702 and peripheral device interface 703 are implemented on separate chips or circuit boards, which is not limited in this embodiment.

[0136] The radio frequency (RF) circuit 704 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The RF circuit 704 communicates with communication networks and other communication devices via electromagnetic signals. The RF circuit 704 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals back into electrical signals. Optionally, the RF circuit 704 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. Optionally, the RF circuit 704 communicates with other terminals via 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 704 also includes circuitry related to NFC (Near Field Communication), which is not limited in this application.

[0137] Display screen 705 is used to display a UI (User Interface). Optionally, the UI includes graphics, text, icons, videos, and any combination thereof. When display screen 705 is a touch display screen, display screen 705 also has the ability to collect touch signals on or above the surface of display screen 705. The touch signals can be input to processor 701 for processing as control signals. Optionally, display screen 705 is also used to provide virtual buttons and / or virtual keyboards, also known as soft buttons and / or soft keyboards. In some embodiments, there is one display screen 705, which is set on the front panel of terminal 700; in other embodiments, there are at least two display screens 705, which are respectively set on different surfaces of terminal 700 or have a folded design; in still other embodiments, display screen 705 is a flexible display screen, which is set on the curved surface or folded surface of terminal 700. Furthermore, optionally, display screen 705 is set as a non-rectangular irregular shape, that is, an irregularly shaped screen. Alternatively, the display screen 705 may be made of materials such as LCD (Liquid Crystal Display) or OLED (Organic Light-Emitting Diode).

[0138] The camera assembly 706 is used to acquire images or videos. Optionally, the camera assembly 706 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 706 also includes a flash. Optionally, the flash is 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, used for light compensation at different color temperatures.

[0139] In some embodiments, the audio circuit 707 includes a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, and convert the sound waves into electrical signals that are input to the processor 701 for processing, or input to the radio frequency circuit 704 to realize voice communication. For stereo acquisition or noise reduction purposes, multiple microphones are used, respectively located at different parts of the terminal 700. Optionally, the microphone is an array microphone or an omnidirectional acquisition microphone. The speaker is used to convert the electrical signals from the processor 701 or the radio frequency circuit 704 into sound waves. Optionally, the speaker is a conventional film speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can not only convert electrical signals into sound waves that are audible to humans, but also convert electrical signals into sound waves that are inaudible to humans for purposes such as distance measurement. In some embodiments, the audio circuit 707 also includes a headphone jack.

[0140] The positioning component 708 is used to determine the current geographic location of the terminal 700 for navigation or LBS (Location Based Service). Optionally, the positioning component 708 is a positioning component based on the US GPS (Global Positioning System), China's BeiDou system, Russia's Granas system, or the EU's Galileo system.

[0141] The power supply 709 is used to power the various components in the terminal 700. Optionally, the power supply 709 is AC power, DC power, a disposable battery, or a rechargeable battery. When the power supply 709 includes a rechargeable battery, the rechargeable battery supports wired or wireless charging. The rechargeable battery also supports fast charging technology.

[0142] In some embodiments, the terminal 700 further includes one or more sensors 710. The one or more sensors 710 include, but are not limited to: an accelerometer 711, a gyroscope 712, a pressure sensor 713, a fingerprint sensor 714, an optical sensor 715, and a proximity sensor 716.

[0143] In some embodiments, the accelerometer 711 detects the magnitude of acceleration along the three coordinate axes of a coordinate system established with the terminal 700. For example, the accelerometer 711 is used to detect the components of gravitational acceleration along the three coordinate axes. Optionally, the processor 701 controls the display screen 705 to display the user interface in either a landscape or portrait view based on the gravitational acceleration signal acquired by the accelerometer 711. The accelerometer 711 is also used for acquiring motion data from games or the user.

[0144] In some embodiments, the gyroscope sensor 712 detects the orientation and rotation angle of the terminal 700. The gyroscope sensor 712 and the accelerometer sensor 711 work together to acquire the user's 3D movements on the terminal 700. Based on the data acquired by the gyroscope sensor 712, the processor 701 implements the following functions: motion sensing (e.g., changing the UI based on the user's tilt operation), image stabilization during shooting, game control, and inertial navigation.

[0145] Optionally, the pressure sensor 713 is disposed on the side bezel of the terminal 700 and / or on the lower layer of the display screen 705. When the pressure sensor 713 is disposed on the side bezel of the terminal 700, it can detect the user's grip signal on the terminal 700, and the processor 701 performs left / right hand recognition or quick operation based on the grip signal collected by the pressure sensor 713. When the pressure sensor 713 is disposed on the lower layer of the display screen 705, the processor 701 controls the operable controls on the UI interface based on the user's pressure operation on the display screen 705. The operable controls include at least one of button controls, scroll bar controls, icon controls, and menu controls.

[0146] The fingerprint sensor 714 is used to collect a user's fingerprint. The processor 701 identifies the user based on the fingerprint collected by the fingerprint sensor 714, or vice versa. When the user's identity is identified as trusted, the processor 701 authorizes the user to perform relevant sensitive operations, including unlocking the screen, viewing encrypted information, downloading software, making payments, and changing settings. Optionally, the fingerprint sensor 714 can be located on the front, back, or side of the terminal 700. When the terminal 700 has a physical button or manufacturer logo, the fingerprint sensor 714 can be integrated with the physical button or manufacturer logo.

[0147] An optical sensor 715 is used to collect ambient light intensity. In one embodiment, the processor 701 controls the display brightness of the display screen 705 based on the ambient light intensity collected by the optical sensor 715. Specifically, when the ambient light intensity is high, the display brightness of the display screen 705 is increased; when the ambient light intensity is low, the display brightness of the display screen 705 is decreased. In another embodiment, the processor 701 also dynamically adjusts the shooting parameters of the camera assembly 706 based on the ambient light intensity collected by the optical sensor 715.

[0148] The proximity sensor 716, also known as a distance sensor, is typically located on the front panel of the terminal 700. The proximity sensor 716 is used to detect the distance between the user and the front of the terminal 700. In one embodiment, when the proximity sensor 716 detects that the distance between the user and the front of the terminal 700 is gradually decreasing, the processor 701 controls the display screen 705 to switch from a screen-on state to a screen-off state; when the proximity sensor 716 detects that the distance between the user and the front of the terminal 700 is gradually increasing, the processor 701 controls the display screen 705 to switch from a screen-off state to a screen-on state.

[0149] Those skilled in the art will understand that Figure 7 The structure shown does not constitute a limitation on terminal 700, which may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0150] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including at least one computer program, which can be executed by a processor in a terminal to perform the gene data processing methods in the various embodiments described above. For example, the computer-readable storage medium includes ROM (Read-Only Memory), RAM (Random-Access Memory), CD-ROM (Compact Disc Read-Only Memory), magnetic tape, floppy disk, and optical data storage devices, etc.

[0151] In an exemplary embodiment, a computer program product or computer program is also provided, including one or more lines of program code stored in a computer-readable storage medium. One or more processors of a computer device are capable of reading the one or more lines of program code from the computer-readable storage medium, and the one or more processors execute the one or more lines of program code, enabling the computer device to perform the gene data processing method described above.

[0152] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. Optionally, the program is stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0153] The above description is merely an optional 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 gene data processing method, characterized in that, The method includes: Based on the gene function query request of the gene to be tested, the cell type corresponding to the gene to be tested is determined; Obtain the single-cell expression dataset corresponding to the cell type and the gene association model corresponding to the cell type. The gene association model is used to obtain the degree of non-linear association between the gene to be tested and the candidate genes belonging to the cell type in the single-cell expression dataset. The gene association model is called to process the gene to be tested, and the degree of nonlinear association between the gene to be tested and multiple candidate genes belonging to the cell type in the single-cell expression dataset is obtained. Output functional annotation information of candidate genes whose nonlinear association degree meets the first target condition, where the first target condition is used to indicate that the nonlinear association degree is greater than or equal to the association degree threshold. The method further includes: each single-cell expression dataset contains gene data of multiple cell types, and different gene association models are trained based on different cell types in at least one single-cell expression dataset to obtain multiple gene association models, wherein the multiple gene association models correspond to multiple cell types in at least one single-cell expression dataset. The training process for each gene association model includes: From the single-cell expression dataset, multiple sample genes are obtained, and the degree of variation of the multiple sample genes is greater than the variation threshold. Based on the multiple sample genes, multiple positive gene pairs and multiple negative gene pairs are obtained, wherein the degree of non-linear association between the genes contained in any positive gene pair is greater than the association degree threshold, and the degree of non-linear association between the genes contained in any negative gene pair is less than the association degree threshold. Based on the multiple positive gene pairs and the multiple negative gene pairs, the parameters of the initial deep model are adjusted to obtain the gene association model. The method further includes: For any gene pair among the plurality of positive gene pairs and the plurality of negative gene pairs, obtain multiple association coefficients predicted by the plurality of gene association models for the any gene pair; based on the multiple association coefficients of the any gene pair, determine whether the any gene pair meets a second target condition, the second target condition being used to indicate that the prediction difficulty of the plurality of gene association models for the any gene pair is lower than a first target threshold; based on gene pairs that meet the second target condition, adjust the parameters of the gene association model corresponding to the cell type.

2. The method according to claim 1, characterized in that, The step of processing the gene to be tested using the gene association model to obtain the degree of nonlinear association between the gene to be tested and multiple candidate genes belonging to the cell type in the single-cell expression dataset includes: Based on the gene to be tested and the plurality of candidate genes, a plurality of gene pairs are constructed, wherein any gene pair in the plurality of gene pairs includes the gene to be tested and any candidate gene in the plurality of candidate genes; The multiple gene pairs are input into the gene association model, and the multiple gene pairs are convolved by the gene association model to obtain multiple association coefficients. Any one of the multiple association coefficients is used to represent the degree of nonlinear association between the genes contained in any gene pair in the multiple gene pairs.

3. The method according to claim 1, characterized in that, The process of obtaining multiple positive gene pairs and multiple negative gene pairs based on the multiple sample genes includes: From the multiple sample genes, two sample genes belonging to the same protein complex are identified as a positive gene pair; Any sample gene in the positive gene pair and any sample gene in the plurality of sample genes other than the positive gene pair are identified as a negative gene pair.

4. The method according to claim 1, characterized in that, The process of adjusting the parameters of the initial deep model based on the multiple positive gene pairs and the multiple negative gene pairs to obtain the gene association model includes: For any one of the multiple positive gene pairs and the multiple negative gene pairs, the initial deep model is invoked to predict the association coefficient of any one gene pair; Based on the correlation coefficients of the multiple positive gene pairs and the multiple negative gene pairs, the loss function value of the initial depth model is determined; In response to the loss function value not meeting the stopping condition, the initial deep model is trained iteratively until the loss function value meets the stopping condition, at which point training stops and the gene association model is obtained.

5. The method according to claim 1, characterized in that, Determining whether any gene pair meets the second target condition based on the multiple association coefficients of any gene pair includes: Based on the plurality of correlation coefficients, a first number of coefficients and a second number of coefficients are determined. The first number of coefficients is the number of coefficients among the plurality of correlation coefficients that are greater than or equal to a second target threshold, and the second number of coefficients is the number of coefficients among the plurality of correlation coefficients that are less than the second target threshold. In response to the first coefficient number being greater than the second coefficient number and the gene pair being a positive gene pair, or the first coefficient number being less than the second coefficient number and the gene pair being a negative gene pair, it is determined that the gene pair meets the second target condition.

6. A gene data processing device, characterized in that, The device includes: The first determining module is used to determine the cell type corresponding to the gene to be tested based on the gene function query request of the gene to be tested; The first acquisition module is used to acquire the single-cell expression dataset corresponding to the cell type and the gene association model corresponding to the cell type. The gene association model is used to acquire the degree of non-linear association between the gene to be tested and the candidate genes belonging to the cell type in the single-cell expression dataset. The processing module is used to call the gene association model to process the gene to be tested, and obtain the degree of nonlinear association between the gene to be tested and multiple candidate genes belonging to the cell type in the single cell expression dataset. The output module is used to output functional annotation information of candidate genes whose nonlinear association degree meets the first target condition, wherein the first target condition is used to indicate that the nonlinear association degree is greater than or equal to the association degree threshold. Each single-cell expression dataset contains gene data of multiple cell types. The device is also used to train different gene association models based on different cell types in at least one single-cell expression dataset to obtain multiple gene association models, which correspond to multiple cell types in at least one single-cell expression dataset. The device further includes: The second acquisition module is used to acquire multiple sample genes from the single-cell expression dataset, wherein the degree of variation of the multiple sample genes is greater than the variation threshold. The third acquisition module is used to acquire multiple positive gene pairs and multiple negative gene pairs based on the multiple sample genes, wherein the degree of non-linear correlation between the genes contained in any positive gene pair is greater than the correlation degree threshold, and the degree of non-linear correlation between the genes contained in any negative gene pair is less than the correlation degree threshold. An adjustment module is used to adjust the parameters of the initial depth model based on the multiple positive gene pairs and the multiple negative gene pairs to obtain the gene association model; The device further includes: The fourth acquisition module is used to acquire, for any gene pair among the plurality of positive gene pairs and the plurality of negative gene pairs, multiple association coefficients predicted by the plurality of gene association models for any gene pair. The second determining module is used to determine whether any gene pair meets a second target condition based on the multiple association coefficients of any gene pair. The second target condition is used to indicate that the prediction difficulty of the multiple gene association models for any gene pair is lower than a first target threshold. The adjustment module is also used to adjust the parameters of the gene association model corresponding to the cell type based on gene pairs that meet the second target conditions.

7. A computer device, characterized in that, The computer device includes one or more processors and one or more memories, wherein at least one computer program is stored in the one or more memories, and the at least one computer program is loaded and executed by the one or more processors to implement the gene data processing method as described in any one of claims 1 to 5.

8. A storage medium, characterized in that, The storage medium stores at least one computer program, which is loaded and executed by a processor to implement the gene data processing method as described in any one of claims 1 to 5.

Images