Single-cell rna sequencing data dimension reduction representation method and device based on prior constraint

Abstract

The application relates to the technical field of artificial intelligence, and discloses a single-cell RNA sequencing data dimension reduction representation method and device based on prior constraints, which are used for solving the technical problem that generated dimension reduction representation data is difficult to accurately reflect biological information when single-cell RNA sequencing data is subjected to dimension reduction processing in the prior art. The method mainly comprises four parts: a prior variational autoencoder with a prior information hidden layer is constructed based on biological prior constraint information; wherein the biological prior information is used to constrain the connection relationship between different gene features in the encoding process; single-cell RNA sequencing data is acquired and preprocessed to generate a cell gene expression matrix used for encoding input; the trained prior variational autoencoder is called to encode the cell gene expression matrix in combination with biological prior constraints and a cell similarity matrix to generate an initial latent representation vector; and a denoising diffusion probability model is called to denoise the initial latent representation vector to obtain a sequencing data dimension reduction representation vector.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method, apparatus, electronic device, computer storage medium, and computer program product for dimensionality reduction representation of single-cell RNA sequencing data based on prior constraints. Background Technology

[0002] Single-cell RNA sequencing (scRNA-seq) technology can acquire gene expression profiles at single-cell resolution and has been widely used in recent years in fields such as cell type identification, developmental trajectory analysis, and disease mechanism research. Furthermore, because single-cell RNA sequencing data typically exhibits high dimensionality, high sparsity, and significant noise, dimensionality reduction methods are usually required to obtain low-dimensional embedding representations to support subsequent analysis tasks.

[0003] In existing technologies, commonly used dimensionality reduction methods (such as PCA, t-SNE, UMAP, etc.) can extract structural features between cells to a certain extent, but they are usually based on statistical data features for modeling, and rarely consider the hierarchical relationship between gene expression layers and biological functional layers, as well as prior information such as biological signaling pathways. Furthermore, in the latent space learning process, the preservation of cell proximity relationships largely relies on reconstruction errors or manifold embedding mechanisms, lacking a unified modeling framework that combines explicit cell similarity matrices for structural constraints. In addition, after obtaining the latent representation, existing dimensionality reduction methods usually directly use it for downstream analysis, with relatively few mechanisms for structured denoising optimization of the latent representation itself, making it difficult to further improve the stability and structural consistency of the low-dimensional embedding representation.

[0004] Therefore, how to maximize the preservation of biological semantics in dimensionality-reduced single-cell data after dimensionality reduction processing, thereby obtaining a low-dimensional embedding representation that combines biological semantic expression and structural preservation capabilities, has become a technical problem that needs further research in this field. Summary of the Invention

[0005] The main objective of this invention is to solve the technical problem that biological information is easily lost when performing dimensionality reduction representation on single-cell RNA sequencing data in the prior art, resulting in inaccurate dimensionality reduction representation data.

[0006] The first aspect of this invention provides a method for dimensionality reduction representation of single-cell RNA sequencing data based on prior constraints, comprising: A priori variational autoencoder with a priori information hidden layer is constructed based on biological prior constraint information; wherein, the biological prior information is used to constrain the connection relationship between different gene features during the encoding process; Acquire single-cell RNA sequencing data and preprocess the single-cell RNA sequencing data to generate a cell gene expression matrix for encoding input; The trained prior variational autoencoder is invoked, and the cell gene expression matrix is ​​encoded by combining biological prior constraints and cell similarity matrix to generate an initial latent representation vector; The trained denoising diffusion probability model is invoked to perform denoising processing on the initial latent representation vector to obtain a denoised sequencing data dimensionality-reduced representation vector.

[0007] Furthermore, in a first implementation of the first aspect of the present invention, before acquiring single-cell RNA sequencing data, the method further includes: A single-cell sample dataset is constructed, and the prior variational autoencoder is pre-trained; wherein, the single-cell sample dataset contains gene expression matrices of multiple cell samples and is labeled with gene regulation representations; The pre-trained prior variational autoencoder and the denoised diffusion probability model are jointly fine-tuned on a specific tissue or disease dataset.

[0008] Optionally, in a second implementation of the first aspect of the present invention, the joint fine-tuning training of the pre-trained prior variational autoencoder and the denoising diffusion probability model on a specific tissue or disease dataset includes: The similarity index between cells is calculated based on the cell gene expression matrix, a cell similarity matrix is ​​constructed, and a neighborhood preservation loss term is constructed based on the cell similarity matrix. The parameters of the prior variational autoencoder and the denoising diffusion probability model are updated by minimizing the joint loss function, which includes the reconstruction loss term, the Kullback-Leibler divergence loss term, and the neighborhood preservation loss term, until joint fine-tuning training is completed.

[0009] Optionally, in a third implementation of the first aspect of the present invention, the construction of a priori variational autoencoder with a priori information hidden layer based on biological prior constraint information includes: Obtain the associations between genes and signaling pathways, protein complexes, or transcription factor regulatory units from biological databases to obtain biological prior constraints information; Each signaling pathway, protein complex, or transcription factor regulatory unit is set as a priori node; Based on the biological prior constraint information, connections are established between gene feature nodes and corresponding prior nodes, thereby forming a structured sparse connection prior information hiding layer. A connection is established based on the structure of the prior information hiding layer and the standard variational autoencoder to obtain the prior variational autoencoder.

[0010] Optionally, in a fourth implementation of the first aspect of the present invention, the preprocessing of the single-cell RNA sequencing data to generate a cell gene expression matrix for encoding the input includes: Quality control was performed on the raw sequencing data to remove low-quality cell samples and low-expression genes; Normalize the gene expression data; The normalized gene expression data were screened for highly variable genes to generate the cell gene expression matrix.

[0011] Optionally, in a fifth implementation of the first aspect of the present invention, the step of calling the trained prior variational autoencoder to encode the cell gene expression matrix in combination with biological prior constraints and a cell similarity matrix to generate an initial latent representation vector includes: The prior variational autoencoder outputs the sampled mean vector and variance vector of the latent variables through its coding layer after training. The initial latent representation vector is generated by randomly sampling based on the sampled mean vector and the variance vector output by the variational autoencoder. In the encoding layer, the encoder parameters are adjusted according to the cell similarity matrix so that the distance relationship between cells in the latent representation space is consistent with the similarity relationship in the cell similarity matrix.

[0012] Optionally, in a sixth implementation of the first aspect of the present invention, the step of invoking the trained denoising diffusion probability model to perform denoising processing on the initial latent representation vector includes: Random noise is gradually added to the initial latent representation vector at multiple preset time steps to form a forward diffusion process; A reverse denoising model is constructed based on a neural network structure to predict the noise components at each time step. The noise is gradually removed based on the predicted noise components to generate the denoised sequencing data dimensionality reduction representation vector.

[0013] A second aspect of the present invention provides a dimensionality reduction representation device for single-cell RNA sequencing data based on prior constraints, comprising: An encoder construction module is used to construct a prior variational autoencoder with a prior information hidden layer based on biological prior constraint information; wherein, the biological prior information is used to constrain the connection relationship between different gene features during the encoding process; The data acquisition module is used to acquire single-cell RNA sequencing data and preprocess the single-cell RNA sequencing data to generate a cell gene expression matrix for encoding input. The dimensionality reduction encoding module is used to call the trained prior variational autoencoder, combine biological prior constraints and cell similarity matrix to encode the cell gene expression matrix, and generate an initial latent representation vector; The denoising module is used to call the trained denoising diffusion probability model to perform denoising processing on the initial latent representation vector to obtain the denoised sequencing data dimensionality-reduced representation vector.

[0014] A third aspect of the present invention provides a prior-constraint-based single-cell RNA sequencing data dimensionality reduction representation device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the prior-constraint-based single-cell RNA sequencing data dimensionality reduction representation device to perform the steps of the prior-constraint-based single-cell RNA sequencing data dimensionality reduction representation method described above.

[0015] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the steps of the above-described method for dimensionality reduction representation of single-cell RNA sequencing data based on prior constraints.

[0016] A fifth aspect of the present invention provides a computer program product comprising a computer program / instructions that, when executed by a processor, implement the steps of the above-described method for dimensionality reduction representation of single-cell RNA sequencing data based on prior constraints.

[0017] The technical solution provided by this invention constructs a priori variational autoencoder with a priori information hidden layer based on biological prior constraint information. The biological prior information is used to constrain the connection relationships between different gene features during the encoding process. Single-cell RNA sequencing data is acquired and preprocessed to generate a cell gene expression matrix for encoding input. The trained priori variational autoencoder is invoked, and the cell gene expression matrix is ​​encoded by combining biological prior constraints and a cell similarity matrix to generate an initial latent representation vector. A trained denoising diffusion probability model is invoked to denoise the initial latent representation vector to obtain a denoised, reduced-dimensional representation vector of the sequencing data. This method enables the dimensionality-reduced single-cell RNA sequencing data to maintain maximum accurate biological semantics, thereby obtaining a low-dimensional embedding representation that combines biological semantic expression capabilities with structural preservation capabilities. The device, electronic device, computer-readable storage medium, and computer program product provided by this invention also solve the corresponding technical problems. Attached Figure Description

[0018] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart illustrating the first embodiment of the method for dimensionality reduction representation of single-cell RNA sequencing data based on prior constraints in this invention. Figure 2 This is a schematic diagram of the prior variational autoencoder constructed in the first embodiment of the method for dimensionality reduction representation of single-cell RNA sequencing data based on prior constraints in this invention. Figure 3 This is a schematic diagram of the hierarchy of the prior variational autoencoder constructed in the first embodiment of the method for dimensionality reduction representation of single-cell RNA sequencing data based on prior constraints in this invention. Figure 4 This is a schematic diagram of the data dimensionality reduction representation obtained in the first embodiment of the single-cell RNA sequencing data dimensionality reduction representation method based on prior constraints in the present invention, used for cell clustering tasks; Figure 5 This is a schematic diagram illustrating the use of the data dimensionality reduction representation obtained in the first embodiment of the single-cell RNA sequencing data dimensionality reduction representation method based on prior constraints for the cell subpopulation discovery task in this invention. Figure 6 This is a schematic diagram illustrating the use of the data dimensionality reduction representation obtained in the first embodiment of the single-cell RNA sequencing data dimensionality reduction representation method based on prior constraints for pathology-related comparison and interpretation tasks in this invention. Figure 7 This is a schematic diagram of an embodiment of the single-cell RNA sequencing data dimensionality reduction representation device based on prior constraints in this invention. Figure 8 This is a schematic diagram of an embodiment of a single-cell RNA sequencing data dimensionality reduction representation device based on prior constraints in this invention. Figure 9 This is a schematic diagram illustrating the principle of a computer-readable medium according to an embodiment of the present invention. Detailed Implementation

[0019] Exemplary embodiments of the invention will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limiting the invention to the embodiments set forth herein. Rather, these exemplary embodiments are provided to make the invention more comprehensive and complete, and to facilitate a full communication of the inventive concept to those skilled in the art. The same reference numerals in the drawings denote the same or similar elements, components, or parts, and therefore repeated descriptions of them will be omitted.

[0020] Subject to the technical concept of this invention, the features, structures, characteristics or other details described in a particular embodiment may be combined in one or more other embodiments in a suitable manner.

[0021] In the description of specific embodiments, the features, structures, characteristics, or other details described in this invention are intended to enable those skilled in the art to fully understand the embodiments. However, it is not excluded that those skilled in the art can practice the technical solutions of this invention without one or more of the specific features, structures, characteristics, or other details.

[0022] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0023] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0024] The terms “and / or” or “and / or” include all combinations of any one or more of the listed items.

[0025] See Figures 1-6 The first embodiment of the single-cell RNA sequencing data dimensionality reduction representation method based on prior constraints in this invention includes: S101. Construct a prior variational autoencoder with a prior information hidden layer based on biological prior constraint information; It is understood that the executing entity of this invention can be a device for dimensionality reduction representation of single-cell RNA sequencing data based on prior constraints, or it can be a terminal or a server; the specific implementation is not limited here. This embodiment of the invention will be described using a server as an example.

[0026] Biological prior information is used to constrain the connections between different gene features during the encoding process. The construction of a variational autoencoder specifically includes: obtaining the associations between genes and signaling pathways, protein complexes, or transcription factor regulatory units from a biological database to obtain biological prior constraint information; setting each signaling pathway, protein complex, or transcription factor regulatory unit as a prior node; establishing connections between gene feature nodes and corresponding prior nodes based on the biological prior constraint information, thereby forming a structured sparsely connected prior information hiding layer; and establishing connections based on the prior information hiding layer and the structure of a standard variational autoencoder to obtain a prior-guided variational autoencoder (PVAED).

[0027] In existing technologies, the encoder and decoder structures in typical variational autoencoders are usually composed of fully connected neural networks. However, the prior variational autoencoder described in this embodiment redesigns the encoder and decoder architecture by introducing structured sparse connections in order to utilize established biological prior information, such as signaling pathway membership relationships, protein complexes, transcription factor-target relationships, and gene interactions.

[0028] See Figure 2 and Figure 3 Specifically, we added a gene layer and a prior layer as hidden layers after the input gene expression layer. The neuronal nodes in the gene layer represent individual genes, while the neuronal nodes in the prior layer represent signaling pathways, complexes, and transcription factors. In one specific implementation, the nodes in the gene layer are connected to the nodes in the input gene expression layer based on gene relationships stored in a pre-constructed professional transcriptome database (such as the Reactome database). Simultaneously, the nodes in the prior layer are explicitly connected to the nodes in the gene layer based on a curated prior database.

[0029] The input layer takes gene expression data as its input. n is the number of cells, m is the number of genes or neuron nodes, and each neuron in the input layer represents the normalized expression of a gene.

[0030] The second layer is the gene layer, which has the same number of neurons as the input layer. If nodes in this layer are associated with nodes in the input layer in a specialized transcriptome database, a connection will be established between them. The propagation process from the input layer to the gene layer is as follows: Where A is the adjacency matrix between the input layer and the gene layer. If gene node i and gene node j are related, then . It is a trainable weight matrix, where the subscript gene indicates the current gene layer.

[0031] The third layer consists of the prior layer and the signal path nodes. Complex nodes and TF nodes Composition. If a gene node is in the prior item, then the prior layer node is connected to the node in the gene layer. Taking the signaling pathway item as an example, if the following conditions are met... Conditions, then gene nodes Connected to signal path node j. The correlation matrix of this layer is: ,in, The activation function is defined as: in, It is a trainable weight matrix, and the subscript prior indicates the current prior layer.

[0032] Potential layer and Similar to the standard VAE, where, and Represents the probability distribution of latent variables. Let be the mean of the latent variable distribution. Let Variance be the variance, and let them have respectively Neuron, This represents the dimension of the space. Nodes in the latent layer are fully connected to nodes in the prior layer. The sample layer z has... Neurons, from the latent layer and Medium sampling: The remaining layers construct the decoder section, which has the same structure as described above but in the opposite direction.

[0033] After constructing the encoder, the process includes a step of pre-training the prior variational autoencoder. Specifically, this involves constructing a single-cell sample dataset and pre-training the prior variational autoencoder. The single-cell sample dataset contains gene expression matrices of multiple cell samples and is labeled with gene regulation representations. The pre-trained prior variational autoencoder and the denoising diffusion probabilistic models (DDPM) are then jointly fine-tuned on a specific tissue or disease dataset.

[0034] In one specific implementation, a similarity index between cells is calculated based on the cell gene expression matrix to construct a cell similarity matrix, and a neighborhood preservation loss term is constructed based on the cell similarity matrix. The parameters of the prior variational autoencoder and the denoising diffusion probability model are updated by minimizing a joint loss function including a reconstruction loss term, a Kullback-Leibler (KL) divergence loss term, and the neighborhood preservation loss term, until joint fine-tuning training is completed. The prior variational autoencoder model obtained after fine-tuning training can then provide a dimensionality-reduced representation of specific single-cell RNA sequencing data based on new task-related input data.

[0035] In this embodiment, the prior variational autoencoder, compared to the standard variational autoencoder, incorporates a confidence index into the loss function and uses weighting factors. By integrating these elements, the reliability of the variational autoencoder is enhanced, thereby guiding and constraining the preservation of local similarity during the encoding process. in It is the original unit outline With the reconstructed The reconstruction losses between It is the Kourbach-Leibler divergence, used to measure the approximate posterior distribution. ( | ; ) and prior distribution ( The difference between the two, as a regularization component, serves to ensure that the learned implicit distribution always closely approximates the selected prior distribution, typically the standard normal distribution. (0, ). It is the total number of individual cells. Each cell is considered the closest A neighboring superior cell, In the reduced-dimensional space, it belongs to the target cell. of A set of nearest neighbor cells, but a set of cells that do not belong to this set in the original space; ( , ) is a unit in the original space Neighborhood unit Rank. The confidence value is quantified between zero and one, where 1 indicates that the local neighborhood is perfectly preserved and 0 indicates that the neighborhood is completely destroyed.

[0036] After training the prior variational autoencoder, we use the original cell embedding x as the starting point of the denoising diffusion probability model. Then, x undergoes a forward pass to become a noisy embedding x. Finally, the loss function of the prior variational autoencoder described in this embodiment is: in, This represents the loss of the denoising diffusion probability model.

[0037] The denoising diffusion probability model described in this embodiment is a latent variable model, which includes a forward denoising process. ( | This process consists of ) gradually destroying the structure of the original data x; and a reverse denoising process ( ( It learns to recover the original data x from noisy input. The forward noise addition process is modeled using a first-order Markov chain with Gaussian transitions, which remain fixed throughout the training process; while noise scheduling... to It can be set as a fixed value or learned through practice.

[0038] The forward process can be summarized as follows, allowing In any It has a closed form: in, ,and The posterior distribution of the forward process can also be handled, and its expression is as follows: When sample size Large enough and When the variance distribution follows a good pattern, the distribution... It will approximate an isotropic Gaussian distribution. The entire probability system can be trained end-to-end through variational inference. During sampling, it can be obtained from... (Selected as an isotropic Gaussian distribution) to extract a latent variable (whose size is the same as that of the training data points). (The same), and performs the reverse process to generate new samples from the underlying data distribution. The inverse process in the denoising diffusion probability model minimizes the upper bound of the following negative log-likelihood through training: All KL divergences involve Gaussian distributions and can therefore be obtained using closed-form expressions. It is worth noting that the conditional probability of the inverse process... ; can be obtained through the posterior probability of the forward process ; Parameterization allows for parameterization at any time step The second term of the above equation can be simplified to the following objective function: in, This means that the inverse process in the denoising diffusion probability model is trained to predict what is added to the input at any time step t. The noise on the surface. Based on this, the loss in the denoising diffusion probability model is: It should be noted that, in the embodiments, the denoising diffusion probability model is implemented using a U-Net (U-shaped Convolutional Neural Network, image segmentation network) architecture based on time step embedding, including but not limited to.

[0039] S102. Obtain single-cell RNA sequencing data and preprocess the single-cell RNA sequencing data to generate a cell gene expression matrix for encoding input. In single-cell RNA sequencing technology, sequencing platforms (such as those based on droplets or microplates) typically output transcript count data for each cell. This raw data generally exists in the form of sequencing reads and needs to be converted into a gene expression count matrix through bioinformatics processing such as alignment and quantification. This count matrix is ​​usually represented as a two-dimensional matrix of "cell × gene," where rows correspond to different cells, columns correspond to different genes, and matrix elements represent the expression count or expression intensity of the corresponding gene in the corresponding cell.

[0040] In practical analysis, to improve data quality and the stability of subsequent model training, preprocessing of the single-cell RNA sequencing data is usually required. This preprocessing may include, but is not limited to, the following steps: First, the raw sequencing data is subjected to quality control to remove low-quality cells (such as cells with too few detected genes or abnormal mitochondrial gene ratios) and low-expressed genes, thereby reducing the impact of noisy data on subsequent analysis. Secondly, the retained count data are normalized to eliminate the impact of differences in sequencing depth among different cells. For example, scale normalization can be performed based on the total expression level of each cell, and logarithmic transformation can be applied as needed to make the expression value distribution more stable. In addition, in some implementations, highly variable genes can be selected based on the variance characteristics of gene expression to reduce data dimensionality and retain key biological variation information.

[0041] After the above processing, a structured cell gene expression matrix is ​​obtained. This cell gene expression matrix serves as input data for the subsequent encoder network, used to extract low-dimensional representations of cells.

[0042] S103. Call the trained prior variational autoencoder, combine biological prior constraints and cell similarity matrix to encode the cell gene expression matrix, and generate the initial latent representation vector. The prior variational autoencoder outputs the sampled mean vector and variance vector of the latent variables through its encoding layer after training; random sampling is performed based on the sampled mean vector and variance vector output by the variational autoencoder to generate the initial latent representation vector; in the encoding layer, the encoder parameters are adjusted according to the cell similarity matrix so that the distance relationship between cells in the latent representation space is consistent with the similarity relationship in the cell similarity matrix.

[0043] In this embodiment, when the trained prior variational autoencoder is invoked to encode the cellular gene expression matrix in conjunction with biological prior constraints and a cell similarity matrix, a random walk based on the prior variational autoencoder on a hypergraph is also included. Generally, a hypergraph is a generalization of a graph, where an edge, called a hyperedge, can connect any number of vertices (two or more), thus representing a collective relationship or group interaction between a set of nodes. In this work, to explore multipath relationships between gene sets rather than being limited to binary relationships between two genes, we first screen contributing regulatory elements from the prior neurons of the prior variational autoencoder based on the node weights during training, and then construct a hypergraph. In this hypergraph, prior neurons resembling signaling pathways are represented as hyperedges, while genes associated with signaling pathways are represented as nodes. Based on this hypergraph, we can use random walks and spectral clustering methods to not only identify genes of significant value but also predict potential gene sets that may jointly participate in biological processes such as disease progression.

[0044] Specifically, a hypergraph is defined as... ( , ),in, It is a set of hyperedges. It is a set of vertices. Hyperedge ∈ yes A subset of, when node When it belongs to a superedge, that is ∈ Then the superedge is said to be connected to the node. Adjacency. The adjacency relationship between nodes is defined by a hyperedge: if ∈ and ∈ Then it is called a node. With nodes In the super-edge Adjacent to each other.

[0045] After constructing the hypergraph, we will implement a random walk with restart on it. A random walk on a standard graph is a Markov process, where a walker starts from a given vertex and moves to an adjacent vertex after each discrete time step t. It's important to note that the transition probability from one vertex to another is completely independent of time t, meaning the transition probability remains constant throughout the walk.

[0046] We can define a transition matrix for the entire random walk process. The transition matrix P represents the transition probabilities between all states when the surfer moves randomly between vertices. The transition matrix P is a key component of the random walk process and can be computed based on multi-domain knowledge. Next, we introduce random walks in the context of hypergraphs. Essentially, the surfer only moves when two nodes are adjacent to the currently visited hyperedge. In short, the process can be viewed as a two-step operation: first, the surfer randomly selects a hyperedge adjacent to the currently visited node; second, it selects a target adjacent node within the selected hyperedge. The hypergraph random walk process can be defined as a Markov chain: at each time step t, the surfer moves to another target vertex within an adjacent hyperedge. We will then focus on deriving the transition matrix corresponding to the hypergraph random walk. .

[0047] Consider an unweighted hypergraph ( , In this matrix, neither the hyperedges nor the vertices have weights. We define the incidence matrix. as follows: in, ( ) is a hyperedge The degree, ( ) is a vertex degree, | | indicates a superedge The number of nodes within, ( ) is related to the vertex The set of adjacent hyperedges. From this, the diagonal hyperedge degree matrix can be obtained. and the degree matrix of the diagonal vertices .

[0048] Considering the above motivations, we now define a hypergraph. ( , The weighted adjacency matrix of ) : in, Indicates that vertex v is on the hyperedge In our model, the weight of vertex v is equal to the weight of the hyperedge. Number of nodes in the middle. Then, we redefine the hyperedge degree. and hyperboundary matrix as follows: in, It is a diagonal hyperside degree matrix, whose elements are: ( ).

[0049] Now, we can compute from the vertex as follows: To the top Transition probability: It can also be represented in matrix form as follows: in, ∈ It is a diagonal matrix of hyperedge weights. Based on the practical considerations mentioned above, we treat all diseases as equal here, which means that all hyperedges have the same weight, i.e. Naturally, the transition matrix It has randomness, meaning that the sum of the elements in each row is equal to 1.

[0050] S104. Call the trained denoising diffusion probability model to perform denoising processing on the initial latent representation vector to obtain the denoised sequencing data dimensionality-reduced representation vector.

[0051] In multiple preset time steps, random noise is gradually added to the initial latent representation vector to form a forward diffusion process; a reverse denoising model is constructed based on a neural network structure to predict the noise components in each time step; noise is gradually removed according to the predicted noise components to generate the denoised sequencing data dimensionality reduction representation vector.

[0052] The denoising diffusion probability model used in this embodiment is a latent variable model, consisting of a forward denoising process that gradually destroys the structure of the original data, and a reverse denoising process that learns to recover the original data from the noisy input. In this embodiment, the denoising diffusion probability model is used to further denoise the latent embeddings in the latent space of the prior variational autoencoder during the denoising process, thereby generating a more refined embedding representation for downstream analysis tasks.

[0053] Furthermore, after obtaining the dimensionality-reduced representation vector of the denoised sequencing data, the generated low-dimensional embedding representation can also be used for cell clustering tasks. When performing cell clustering tasks, the ARI (Adjusted Rand Index) can be used to evaluate clustering performance.

[0054] Wherein, RI represents the RAND Corporation index: This indicates the number of sample pairs belonging to the same cluster in the two partitions. represents the number of sample pairs belonging to the same true category but different clusters, c represents the number of sample pairs belonging to the same cluster but different true categories, and d represents the number of sample pairs belonging to different clusters and different true categories. [ ] represents the expected RI value under random clustering (hypergeometric distribution).

[0055] MI quantifies the shared information between two clusters, U (true labels) and V (predicted clusters). A higher value indicates better alignment.

[0056] in, Represents the true category Clustering with prediction The number of samples in the middle Is it a real category? Sample size It is predictive clustering Sample size It represents the total number of samples.

[0057] NMI normalizes MI to the [0,1] interval, where 1 represents perfect alignment.

[0058] AMI applies random adjustments to MI, similar to ARI. The value ranges from 0 (random) to 1 (perfect).

[0059] Qualitative measures whether each cluster contains only samples from a single true class.

[0060] Completeness measures whether all samples in the true category are assigned to the same cluster.

[0061] V-measure is the harmonic mean of homogeneity and integrity, and its value ranges from 0 to 1.

[0062] The method provided in this invention integrates biological prior knowledge and diffusion-based denoising through a constructed prior variational autoencoder, achieving superior dimensionality reduction. This ensures that the dimensionality-reduced single-cell RNA sequencing data retains maximized accuracy in biological semantics, thus obtaining a low-dimensional embedding representation that combines biological semantic expression with structural preservation. Further, see [link to related documentation]. Figure 4 as well as Figure 5 Furthermore, it can perform clustering based on the dimensionality reduction representation of the acquired single-cell RNA sequencing data to obtain pre-defined cell cluster subgroups and extract useful information contained in the single-cell RNA sequencing data, providing input data for information processing, prediction, modeling, and other tasks.

[0063] For further details, please refer to [link / reference]. Figure 6 Furthermore, a hypergraph model can be built on the basis of the trained prior encoder to analyze pathology-related comparative tasks, and then used for interpretability tasks, providing an explanatory basis and source information for the prediction and identification of pathology-related tasks.

[0064] The above describes the method for dimensionality reduction representation of single-cell RNA sequencing data based on prior constraints in the embodiments of the present invention. The following describes the device for dimensionality reduction representation of single-cell RNA sequencing data based on prior constraints in the embodiments of the present invention. (See reference...) Figure 7 One embodiment of the single-cell RNA sequencing data dimensionality reduction representation device based on prior constraints in this invention includes: The encoder construction module 701 is used to construct a prior variational autoencoder with a prior information hidden layer based on biological prior constraint information; wherein, the biological prior information is used to constrain the connection relationship between different gene features during the encoding process; The data acquisition module 702 is used to acquire single-cell RNA sequencing data and preprocess the single-cell RNA sequencing data to generate a cell gene expression matrix for encoding input. The dimensionality reduction encoding module 703 is used to call the trained prior variational autoencoder, combine biological prior constraints and cell similarity matrix to encode the cell gene expression matrix, and generate an initial latent representation vector; The denoising module 704 is used to call the trained denoising diffusion probability model to perform denoising processing on the initial latent representation vector to obtain the denoised sequencing data dimensionality-reduced representation vector.

[0065] The apparatus provided in this invention enables single-cell RNA sequencing data to maintain maximum accurate biological semantics after dimensionality reduction, thereby obtaining a low-dimensional embedding representation that combines biological semantic expression and structural preservation capabilities.

[0066] In another embodiment of this application, a training module is further included, the training module being used for: A single-cell sample dataset is constructed, and the prior variational autoencoder is pre-trained; wherein, the single-cell sample dataset contains gene expression matrices of multiple cell samples and is labeled with gene regulation representations; The pre-trained prior variational autoencoder and the denoised diffusion probability model are jointly fine-tuned on a specific tissue or disease dataset.

[0067] In another embodiment of this application, the joint fine-tuning training of the pre-trained prior variational autoencoder and the denoising diffusion probability model on a specific tissue or disease dataset includes: The similarity index between cells is calculated based on the cell gene expression matrix, a cell similarity matrix is ​​constructed, and a neighborhood preservation loss term is constructed based on the cell similarity matrix. The parameters of the prior variational autoencoder and the denoising diffusion probability model are updated by minimizing the joint loss function, which includes the reconstruction loss term, the Kullback-Leibler divergence loss term, and the neighborhood preservation loss term, until joint fine-tuning training is completed.

[0068] In another embodiment of this application, the encoder construction module 701 is specifically used for: Obtain the associations between genes and signaling pathways, protein complexes, or transcription factor regulatory units from biological databases to obtain biological prior constraints information; Each signaling pathway, protein complex, or transcription factor regulatory unit is set as a priori node; Based on the biological prior constraint information, connections are established between gene feature nodes and corresponding prior nodes, thereby forming a structured sparse connection prior information hiding layer. A connection is established based on the structure of the prior information hiding layer and the standard variational autoencoder to obtain the prior variational autoencoder.

[0069] In another embodiment of this application, the dimensionality reduction encoding module 703 is specifically used for: The prior variational autoencoder outputs the sampled mean vector and variance vector of the latent variables through its coding layer after training. The initial latent representation vector is generated by randomly sampling based on the sampled mean vector and the variance vector output by the variational autoencoder. In the encoding layer, the encoder parameters are adjusted according to the cell similarity matrix so that the distance relationship between cells in the latent representation space is consistent with the similarity relationship in the cell similarity matrix.

[0070] In another embodiment of this application, the noise reduction module 704 is specifically used for: Random noise is gradually added to the initial latent representation vector at multiple preset time steps to form a forward diffusion process; A reverse denoising model is constructed based on a neural network structure to predict the noise components at each time step. The noise is gradually removed based on the predicted noise components to generate the denoised sequencing data dimensionality reduction representation vector.

[0071] The apparatus provided in this invention enables single-cell RNA sequencing data to maintain maximum accurate biological semantics after dimensionality reduction, thereby obtaining a low-dimensional embedding representation that combines biological semantic expression and structural preservation capabilities.

[0072] This invention also provides a dimensionality reduction representation system for single-cell RNA sequencing data based on prior constraints. The dimensionality reduction representation system for single-cell RNA sequencing data based on prior constraints can be implemented by combining the hardware of the aforementioned device embodiment with the information processing flow described in the aforementioned method embodiment, and will not be described again here.

[0073] Based on the same inventive concept, this specification also provides an electronic device for dimensionality reduction representation of single-cell RNA sequencing data based on prior constraints. The electronic device for dimensionality reduction representation of single-cell RNA sequencing data based on prior constraints in this embodiment of the invention will be described in detail below from the perspective of hardware processing.

[0074] Figure 8 This is a schematic diagram of an electronic device provided as an embodiment of this specification. Refer to the following... Figure 8To describe the electronic device 800 according to this embodiment of the invention. Figure 8 The electronic device 800 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.

[0075] like Figure 8 As shown, the electronic device 800 is presented in the form of a general-purpose computing device. The components of the electronic device 800 may include, but are not limited to: at least one processing unit 810, at least one storage unit 820, a bus 830 connecting different system components (including storage unit 820 and processing unit 810), a display unit 840, etc.

[0076] The storage unit stores program code that can be executed by the processing unit 810, causing the processing unit 810 to perform the steps described in the processing method section of this specification according to various exemplary embodiments of the present invention. For example, the processing unit 810 can perform, as follows: Figure 1 The steps are shown.

[0077] The storage unit 820 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 8201 and / or a cache storage unit 8202, and may further include a read-only memory unit (ROM) 8203.

[0078] The storage unit 820 may also include a program / utility 8204 having a set (at least one) program module 8205, such program module 8205 including but not limited to: an operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.

[0079] Bus 830 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.

[0080] Electronic device 800 can also communicate with one or more external devices 100 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with the electronic device 800, and / or with any device that enables the electronic device 800 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 850. Furthermore, electronic device 800 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 860. Network adapter 860 can communicate with other modules of electronic device 800 via bus 830. It should be understood that, although... Figure 8 As not shown in the diagram, other hardware and / or software modules may be used in conjunction with the electronic device 800, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0081] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described in this invention can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this invention can be embodied in the form of a software product, which can be stored in a computer-readable storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, or network device, etc.) to execute the method described above according to this invention. When the computer program is executed by a data processing device, it enables the computer-readable medium to implement the method described above, i.e.: as... Figure 1 The method shown.

[0082] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described in this invention can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this invention can be embodied in the form of a software product, which can be stored in a computer-readable storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, or network device, etc.) to execute the method described above according to this invention. When the computer program is executed by a data processing device, it enables the computer-readable medium to implement the method described above, i.e.: as... Figure 1 The method shown.

[0083] Figure 9This is a schematic diagram of a computer-readable medium provided for embodiments of this specification.

[0084] accomplish Figure 1 The computer program of the method shown can be stored on one or more computer-readable media. A computer-readable medium can be a readable signal medium or a readable storage medium. A readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0085] The computer-readable storage medium may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The readable storage medium may also be any readable medium other than a readable storage medium, capable of transmitting, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0086] In addition, the present invention provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the dimensionality reduction representation method for single-cell RNA sequencing data based on prior constraints as described in any of the above embodiments.

[0087] Program code for performing the operations of this invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, and Python, as well as conventional procedural programming languages ​​such as C, Python, or similar languages. The program code can execute entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0088] In summary, the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that in practice, general-purpose data processing devices such as microprocessors or digital signal processors (DSPs) can be used to implement some or all of the functions of some or all of the components according to the embodiments of the present invention. The present invention can also be implemented as a device or apparatus program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such programs implementing the present invention can be stored on a computer-readable medium or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

[0089] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the present invention is not inherently related to any specific computer, virtual device, or electronic device, and various general-purpose devices can also implement the present invention. The above descriptions are merely specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0090] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0091] If the technical solution of this application involves personal information, the product using this technical solution has clearly informed the user of the personal information processing rules and obtained the user's voluntary consent before processing the personal information. If the technical solution of this application involves sensitive personal information, the product using this technical solution has obtained the user's separate consent before processing the sensitive personal information, and also meets the requirement of "express consent". For example, at personal information collection devices such as cameras, clear and prominent signs are set up to inform users that they have entered the scope of personal information collection and that personal information will be collected. If an individual voluntarily enters the collection scope, it is deemed that they have agreed to the collection of their personal information; or on the personal information processing device, with clear signs / information informing users of the personal information processing rules, authorization is obtained from the individual through pop-up information or by asking the individual to upload their personal information; wherein, the personal information processing rules may include information such as the personal information processor, the purpose of personal information processing, the processing method, and the types of personal information processed.

[0092] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for dimensionality reduction representation of single-cell RNA sequencing data based on prior constraints, characterized in that, include: A priori variational autoencoder with a priori information hidden layer is constructed based on biological prior constraint information; wherein, the biological prior information is used to constrain the connection relationship between different gene features during the encoding process; Acquire single-cell RNA sequencing data and preprocess the single-cell RNA sequencing data to generate a cell gene expression matrix for encoding input; The trained prior variational autoencoder is invoked, and the cell gene expression matrix is ​​encoded by combining biological prior constraints and cell similarity matrix to generate an initial latent representation vector; The trained denoising diffusion probability model is invoked to perform denoising processing on the initial latent representation vector to obtain a denoised sequencing data dimensionality-reduced representation vector.

2. The method for dimensionality reduction representation of single-cell RNA sequencing data based on prior constraints according to claim 1, characterized in that, Before obtaining single-cell RNA sequencing data, the following is also included: A single-cell sample dataset is constructed, and the prior variational autoencoder is pre-trained; wherein, the single-cell sample dataset contains gene expression matrices of multiple cell samples and is labeled with gene regulation representations; The pre-trained prior variational autoencoder and the denoised diffusion probability model are jointly fine-tuned on a specific tissue or disease dataset.

3. The method for dimensionality reduction representation of single-cell RNA sequencing data based on prior constraints according to claim 2, characterized in that, The joint fine-tuning training of the pre-trained prior variational autoencoder and the denoised diffusion probability model on a specific tissue or disease dataset includes: The similarity index between cells is calculated based on the cell gene expression matrix, a cell similarity matrix is ​​constructed, and a neighborhood preservation loss term is constructed based on the cell similarity matrix. The parameters of the prior variational autoencoder and the denoising diffusion probability model are updated by minimizing the joint loss function, which includes the reconstruction loss term, the Kullback-Leibler divergence loss term, and the neighborhood preservation loss term, until joint fine-tuning training is completed.

4. The method for dimensionality reduction representation of single-cell RNA sequencing data based on prior constraints according to claim 1, characterized in that, The aforementioned prior variational autoencoder, constructed based on biological prior constraint information and featuring a prior information hidden layer, includes: Obtain the associations between genes and signaling pathways, protein complexes, or transcription factor regulatory units from biological databases to obtain biological prior constraints information; Each signaling pathway, protein complex, or transcription factor regulatory unit is set as a priori node; Based on the biological prior constraint information, connections are established between gene feature nodes and corresponding prior nodes, thereby forming a structured sparse connection prior information hiding layer. A connection is established based on the structure of the prior information hiding layer and the standard variational autoencoder to obtain the prior variational autoencoder.

5. The method for dimensionality reduction representation of single-cell RNA sequencing data based on prior constraints according to claim 1, characterized in that, The method involves calling the trained prior variational autoencoder, which encodes the cell gene expression matrix by combining biological prior constraints and the cell similarity matrix, to generate an initial latent representation vector, including: The prior variational autoencoder outputs the sampled mean vector and variance vector of the latent variables through its coding layer after training. The initial latent representation vector is generated by randomly sampling based on the sampled mean vector and the variance vector output by the variational autoencoder. In the encoding layer, the encoder parameters are adjusted according to the cell similarity matrix so that the distance relationship between cells in the latent representation space is consistent with the similarity relationship in the cell similarity matrix.

6. The method for dimensionality reduction representation of single-cell RNA sequencing data based on prior constraints according to claim 1, characterized in that, The step of calling the trained denoising diffusion probability model to perform denoising processing on the initial latent representation vector includes: Random noise is gradually added to the initial latent representation vector at multiple preset time steps to form a forward diffusion process; A reverse denoising model is constructed based on a neural network structure to predict the noise components at each time step. The noise is gradually removed based on the predicted noise components to generate the denoised sequencing data dimensionality reduction representation vector.

7. A device for dimensionality reduction representation of single-cell RNA sequencing data based on prior constraints, characterized in that, The prior-constraint-based single-cell RNA sequencing data dimensionality reduction representation device includes: An encoder construction module is used to construct a prior variational autoencoder with a prior information hidden layer based on biological prior constraint information; wherein, the biological prior information is used to constrain the connection relationship between different gene features during the encoding process; The data acquisition module is used to acquire single-cell RNA sequencing data and preprocess the single-cell RNA sequencing data to generate a cell gene expression matrix for encoding input. The dimensionality reduction encoding module is used to call the trained prior variational autoencoder, combine biological prior constraints and cell similarity matrix to encode the cell gene expression matrix, and generate an initial latent representation vector; The denoising module is used to call the trained denoising diffusion probability model to perform denoising processing on the initial latent representation vector to obtain the denoised sequencing data dimensionality-reduced representation vector.

8. A device for dimensionality reduction representation of single-cell RNA sequencing data based on prior constraints, characterized in that, The prior constraint-based single-cell RNA sequencing data dimensionality reduction representation device includes: a memory and at least one processor, wherein the memory stores instructions; The at least one processor invokes the instructions in the memory to cause the prior-constrained single-cell RNA sequencing data dimensionality reduction representation device to perform the steps of the prior-constrained single-cell RNA sequencing data dimensionality reduction representation method as described in any one of claims 1-6.

9. A computer-readable storage medium storing a computer program / instructions thereon, characterized in that, When the program / instruction is executed by the processor, it implements the steps of the method for dimensionality reduction representation of single-cell RNA sequencing data based on prior constraints as described in any one of claims 1-6.

10. A computer program product, comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method for dimensionality reduction representation of single-cell RNA sequencing data based on prior constraints as described in any one of claims 1-6.

Images