Cell differentiation trajectory inference method based on deep hyperbolic manifold embedding

Abstract

The invention relates to a cell differentiation trajectory inference method based on deep hyperbolic manifold embedding. The method comprises the following steps: acquiring single-cell RNA sequencing data; introducing hyperbolic space embedding with negative curvature, performing feature extraction on the single-cell RNA sequencing data through a deep feature extraction network, and mapping extracted high-dimensional features to a low-dimensional hyperbolic space; through track inference in a low-dimensional hyperbolic space, transformation paths of cells from progenitor cells to various differentiation end cell types are constructed, and key genes and pathways are evaluated by utilizing a sensitivity analysis method. The method has the beneficial effects that by introducing hyperbolic space embedding with negative curvature, the model can better capture the hierarchical structure of data, and the accuracy and consistency of differentiation tracks are enhanced. Moreover, high-dimensional single-cell sequencing data can be processed, the limitation of a traditional method under the high-dimensional condition is overcome, and the applicability of the model is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of bioinformatics, and more specifically, to a cell differentiation trajectory inference method based on deep hyperbolic manifold embedding. Background Art

[0002] With the development of single-cell RNA sequencing technology, more and more cell types and their differentiation processes have been studied in depth. However, due to the high dimensionality and complexity of single-cell data, traditional cell differentiation trajectory inference methods have obvious shortcomings in capturing the hierarchical structure between cells. Existing methods, such as dimensionality reduction methods based on PCA or t-SNE, although they can map data to a low-dimensional space, are difficult to effectively retain the hierarchical information of the data and the global relationship of cell differentiation. In addition, traditional methods are mostly based on Euclidean space and cannot accurately describe the inherent hierarchical structural characteristics of the cell differentiation process. Therefore, it is of great significance to propose a deep learning method based on hyperbolic manifold embedding to better represent the hierarchical relationship between cells and infer differentiation trajectories. Summary of the Invention

[0003] The purpose of the present invention is to address the deficiencies of the existing technology and propose a cell differentiation trajectory inference method based on deep hyperbolic manifold embedding.

[0004] First, a cell differentiation trajectory inference method based on deep hyperbolic manifold embedding is provided, including: Step 1: Obtain single-cell RNA sequencing data; Step 2: Introducing negative curvature hyperbolic space embedding, extracting features from the single-cell RNA sequencing data through a deep feature extraction network, and mapping the extracted high-dimensional features into a low-dimensional hyperbolic space; Step 3: Through trajectory inference in low-dimensional hyperbolic space, construct the transformation path of cells from progenitor cells to various differentiated terminal cell types, and use sensitivity analysis methods to evaluate key genes and pathways.

[0005] Preferably, in step 2, the deep feature extraction network consists of multiple fully connected layers and activation functions; the last layer of the deep feature extraction network consists of a hyperbolic mapping layer and a hyperbolic activation function.

[0006] Preferably, step 3 includes: Step 3.1, calculate the hyperbolic distance between cells; Step 3.2: constructing a cell differentiation tree using a hierarchical clustering algorithm based on the hyperbolic distances between the cells to clarify the transformation pathways of cells from progenitor cells to various terminal cell types; Step 3.3: Use sensitivity analysis methods to perturb the trajectories in the hyperbolic space and identify genes and signaling pathways that play a key role in the differentiation process.

[0007] Preferably, step 2 further comprises: A reconstruction loss function is calculated, which is used to keep the distances of neighboring cells close in the low-dimensional hyperbolic space.

[0008] Preferably, step 2 further comprises: A hyperbolic regularization loss function is calculated, which is used to regularize the embedded points to prevent the embedded points from deviating from the boundary of the hyperbolic space.

[0009] In a second aspect, a cell differentiation trajectory inference system based on deep hyperbolic manifold embedding is provided, which is used to execute any of the methods described in the first aspect, including: Acquisition module, used to obtain single-cell RNA sequencing data; An extraction module, configured to introduce a negative curvature hyperbolic space embedding, extract features from the single-cell RNA sequencing data through a deep feature extraction network, and map the extracted high-dimensional features into a low-dimensional hyperbolic space; A building module is used to construct the transition path of cells from progenitor cells to various terminal differentiated cell types through trajectory inference in low-dimensional hyperbolic space, and to evaluate key genes and pathways using sensitivity analysis methods.

[0010] According to a third aspect, a computer storage medium is provided, wherein a computer program is stored in the computer storage medium; when the computer program is executed on a computer, the computer executes any one of the methods described in the first aspect.

[0011] In a fourth aspect, an electronic device is provided, including: Memory, used to store computer programs; A processor is used to execute the computer program to implement any method as described in the first aspect.

[0012] The beneficial effects of the present invention are: 1. By introducing negative curvature hyperbolic space embedding, the model can better capture the hierarchical structure of the data and enhance the accuracy and consistency of the differentiation trajectory.

[0013] 2. This invention can process high-dimensional single-cell sequencing data, overcome the limitations of traditional methods in high-dimensional situations, and improve the applicability of the model.

[0014] 3. This invention combines deep learning technology to quickly extract nonlinear features, improve data processing efficiency, and shorten the research cycle.

[0015] 4. The present invention is applicable to a variety of single-cell biology research and medical applications, such as disease mechanism research and new drug development, and has broad practical application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is the overall flow chart of a cell differentiation trajectory inference method based on deep hyperbolic manifold embedding; Figure 2 This is the overall framework of a cell differentiation trajectory inference method based on deep hyperbolic manifold embedding; Figure 3 A schematic diagram of a cell differentiation tree provided in an embodiment of the present invention; Figure 4 Schematic diagram of the structure of a cell differentiation trajectory inference system based on deep hyperbolic manifold embedding. DETAILED DESCRIPTION

[0017] The present invention will be further described below with reference to the following examples. The following examples are provided only to facilitate understanding of the present invention. It should be noted that, without departing from the principles of the present invention, it is possible for a person skilled in the art to make various modifications to the present invention, and such improvements and modifications fall within the scope of the claims of the present invention.

[0018] Example 1: To address the limitations of existing methods in inferring cell differentiation trajectories, e.g. Figure 1 As shown, the present invention provides a cell differentiation trajectory inference method based on deep hyperbolic manifold embedding, comprising: Step 1: Obtain single-cell RNA sequencing data.

[0019] For example, the single-cell RNA sequencing data is recorded as ,in is the number of cells, is the number of genes.

[0020] Step 2: Introduce negative curvature hyperbolic space embedding, extract features from the single-cell RNA sequencing data through a deep feature extraction network, and map the extracted high-dimensional features to a low-dimensional hyperbolic space.

[0021] Specifically, the deep feature extraction network consists of multiple fully connected layers and activation functions. The nonlinear transformation of each layer can be expressed as:

[0022] in, and Respectively The layer's weight matrix and bias vector, is the activation function, For the The output features of the layer. The last layer is the hyperbolic mapping layer, as shown below, which maps the high-dimensional features to the hyperbolic space and uses the hyperbolic activation function To ensure the validity of the mapping:

[0023] in, For the Representation of cells in hyperbolic space.

[0024] In addition, by introducing negative curvature hyperbolic space embedding, the model can better capture the hierarchical structure of the data and enhance the accuracy and consistency of the differentiation trajectory.

[0025] Specifically, features are extracted through deep neural networks, and high-dimensional features are mapped to low-dimensional hyperbolic space, which is recorded as ,in is the dimension of the embedding space:

[0026] in, Indicated by the parameter Deep neural networks for control, is an embedding representation in hyperbolic space.

[0027] Step 3: Through trajectory inference in low-dimensional hyperbolic space, construct the transformation path of cells from progenitor cells to various differentiated terminal cell types, and use sensitivity analysis methods to evaluate key genes and pathways.

[0028] Trajectory inference uses algorithms to infer the continuous paths of change in dynamic biological processes, such as cell differentiation, development, or disease progression. Based on single-cell data, it constructs similarity relationships between cells, arranging the order of cell changes over time or state, thereby reconstructing the dynamic trajectory from the starting state to the final state.

[0029] Specifically, the present application constructs a cell differentiation trajectory inference module to execute step 4. The cell differentiation trajectory inference module includes a trajectory inference unit, a differentiation tree construction unit, and a key gene and pathway identification unit.

[0030] Step 3 includes: Step 3.1: Calculate the hyperbolic distance between cells through the trajectory inference unit.

[0031] The calculation formula for step 3.1 is:

[0032] in, Represents cells and cells The hyperbolic distance between Represents the Euclidean norm of a vector.

[0033] Step 3.2: Based on the hyperbolic distances between the cells, the differentiation tree construction unit uses a hierarchical clustering algorithm to construct a cell differentiation tree, clarifying the transformation path of cells from progenitor cells to each terminal cell type. In this way, the hierarchical structure of the cell differentiation process can be clearly revealed.

[0034] Step 3.3: Using the sensitivity analysis method, the key gene and pathway identification unit perturbs the trajectory in the hyperbolic space to identify genes and signaling pathways that play a key role in the differentiation process.

[0035] The calculation formula for step 3.3 is:

[0036] in, Indicates gene Differentiation trajectory By calculating the sensitivity of the pathways that guide the dynamic transition of cells from progenitor cells to their terminal states, we can determine which genes play an important role in the cell differentiation process.

[0037] Pathway sensitivity analysis: targeting a signaling pathway (Contains multiple genes , ,..., ), by comprehensively evaluating the sensitivity of all genes in the pathway to the trajectory Contributions:

[0038] in, This reflects the overall impact of the signaling pathway on the differentiation trajectory.

[0039] Example 2: Based on Example 1, Example 2 of the present application provides a more specific method for inferring cell differentiation trajectories based on deep hyperbolic manifold embedding, including: Step 1: Obtain single-cell RNA sequencing data.

[0040] Step 2: Introduce negative curvature hyperbolic space embedding, extract features from the single-cell RNA sequencing data through a deep feature extraction network, and map the extracted high-dimensional features to a low-dimensional hyperbolic space.

[0041] In step 2, the embodiment of the present application also introduces a hyperbolic embedding loss function module to ensure that high-dimensional single-cell data can be effectively mapped to a low-dimensional hyperbolic space and retain the hierarchical relationship and proximity structure of the data; the loss function module includes a reconstruction loss unit and a hyperbolic regularization loss unit.

[0042] Specifically, the reconstruction loss unit is used to calculate the reconstruction loss function so that the distances between neighboring cells in the hyperbolic space are kept as close as possible, which is achieved as follows:

[0043] in, Represents cells and cells The weight between is the distance in hyperbolic space, is the distance in the original space.

[0044] The hyperbolic regularization loss unit is used to calculate the hyperbolic regularization loss function to ensure the stability of the data in the hyperbolic space and the validity of the embedding. The hyperbolic regularization loss function requires regularizing the embedded points to prevent them from deviating from the boundaries of the hyperbolic space; this is achieved by:

[0045] This loss term is used to constrain the embedding points The modulus of is close to 1, ensuring that they are within a reasonable range of hyperbolic space.

[0046] The total loss function is the weighted sum of the above loss functions:

[0047] in, , is the weight coefficient of the loss term, which is used to balance the impact of each part of the loss.

[0048] Step 3: Through trajectory inference in low-dimensional hyperbolic space, construct the transformation path of cells from progenitor cells to various differentiated terminal cell types, and use sensitivity analysis methods to evaluate key genes and pathways.

[0049] In summary, the present invention provides an efficient and accurate method for inferring cell differentiation trajectories, which can effectively process high-dimensional single-cell data, capture the hierarchical structure and key genes in the cell differentiation process, and provide a powerful tool for cell biology research.

[0050] For example, in the study of embryonic development, single-cell RNA sequencing data at different developmental stages can be input into the method of the present invention. Through a deep feature extraction network, the characteristics of the cells are extracted and embedded in a hyperbolic space. Then, based on the hyperbolic distance, the differentiation trajectory between cells is inferred, and a differentiation tree of embryonic cells is constructed to clarify the differentiation paths of different cell types during development. At the same time, through sensitivity analysis, genes and signaling pathways that play a key role in the differentiation process are identified, providing important clues for revealing the molecular mechanisms of embryonic development.

[0051] For example, a single-cell RNA sequencing dataset from zebrafish larvae was used as input. This dataset, derived from five-day-old zebrafish larvae, encompasses over 70,000 cells that were isolated and single-cell captured for transcriptome sequencing. Single-cell capture was performed using the 10X Genomics Chromium System, coupled with Droplet Microfluidics technology. Each cell was encapsulated in a microdroplet, lysed, and reverse-transcribed to generate cDNA with a cell-specific barcode and unique molecular identifier (UMI).

[0052] like Figure 3 As shown, Figure 3 This is a cell differentiation tree generated using the method of the present invention on a single-cell RNA sequencing dataset of zebrafish larvae. The scale bar in the lower left corner represents the ratio of branch length to differentiation distance.

[0053] It should be noted that the parts in this embodiment that are the same or similar to those in Example 1 can be referenced to each other and will not be described in detail in this application.

[0054] Example 3: Based on Examples 1 and 2, Example 3 of the present application provides a cell differentiation trajectory inference system based on deep hyperbolic manifold embedding, such as Figure 4 As shown, including: Acquisition module, used to obtain single-cell RNA sequencing data; An extraction module, configured to introduce a negative curvature hyperbolic space embedding, extract features from the single-cell RNA sequencing data through a deep feature extraction network, and map the extracted high-dimensional features into a low-dimensional hyperbolic space; A building module is used to construct the transition path of cells from progenitor cells to various terminal differentiated cell types through trajectory inference in low-dimensional hyperbolic space, and to evaluate key genes and pathways using sensitivity analysis methods.

[0055] Specifically, the system provided in this embodiment is a system corresponding to the method provided in Examples 1 and 2. Therefore, the parts in this embodiment that are the same or similar to Examples 1 and 2 can be referenced to each other and will not be repeated in this application.

Claims

1. A cell differentiation trajectory inference method based on deep hyperbolic manifold embedding, characterized in that: include: Step 1: Obtain single-cell RNA sequencing data; Step 2: Introduce negative curvature hyperbolic space embedding, extract features from the single-cell RNA sequencing data through a deep feature extraction network, and map the extracted high-dimensional features to a low-dimensional hyperbolic space; Step 3: Through trajectory inference in low-dimensional hyperbolic space, construct the transformation path of cells from progenitor cells to various terminal cell types, and use sensitivity analysis methods to evaluate key genes and pathways.

2. The cell differentiation trajectory inference method based on deep hyperbolic manifold embedding according to claim 1, characterized in that: In step 2, the deep feature extraction network is composed of multiple fully connected layers and activation functions; the last layer of the deep feature extraction network is composed of a hyperbolic mapping layer and a hyperbolic activation function.

3. The cell differentiation trajectory inference method based on deep hyperbolic manifold embedding according to claim 2, characterized in that: Step 3 includes: Step 3.1, calculate the hyperbolic distance between cells; Step 3.2, constructing a cell differentiation tree using a hierarchical clustering algorithm based on the hyperbolic distance between the cells, and clarifying the transformation path of cells from progenitor cells to each terminal cell type; Step 3.3: Use sensitivity analysis methods to perturb the trajectories in the hyperbolic space and identify genes and signaling pathways that play a key role in the differentiation process.

4. The cell differentiation trajectory inference method based on deep hyperbolic manifold embedding according to claim 3 is characterized in that: Step 2 also includes: A reconstruction loss function is calculated, which is used to keep the distances of neighboring cells close in the low-dimensional hyperbolic space.

5. The deep hyperbolic manifold embedded cell differentiation trajectory inference method according to claim 4, characterized in that: Step 2 also includes: A hyperbolic regularization loss function is calculated, where the hyperbolic regularization loss function is used to regularize the embedded points to prevent the embedded points from deviating from the boundary of the hyperbolic space.

6. A cell differentiation trajectory inference system based on deep hyperbolic manifold embedding, characterized by: The method for executing any one of claims 1 to 5 comprises: Acquisition module, used to acquire single-cell RNA sequencing data; An extraction module, used for introducing a hyperbolic space embedding with negative curvature, extracting features from the single-cell RNA sequencing data through a deep feature extraction network, and mapping the extracted high-dimensional features to a low-dimensional hyperbolic space; A building module is used to construct the transformation path of cells from progenitor cells to various terminal cell types through trajectory inference in low-dimensional hyperbolic space, and to evaluate key genes and pathways using sensitivity analysis methods.

7. A computer storage medium, characterized in that: The computer storage medium stores a computer program; when the computer program is executed on a computer, the computer executes any one of the methods described in claims 1 to 5.

8. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the method according to any one of claims 1 to 5.

Images