An AI-based evaluation method for human germ cell development and culture systems

Abstract

This invention discloses an AI-based method for evaluating human germ cell development and culture systems, comprising: Step 1, constructing a standardized reference atlas; Step 2, establishing a germ cell development stage prediction model, including a single-cell annotation variational inference prediction module based on transfer learning, a joint prediction module combining single-cell annotation variational inference with k-nearest neighbors, and a cell type classification module optimized by feature selection. These three modules achieve collaborative decision-making through a three-level confidence system; Step 3, evaluating the in vitro culture system, including constructing dual indicators for quantitative evaluation and analyzing the evaluation results. This invention provides a complete solution from single-cell reference atlas construction and development stage prediction to culture system evaluation, solving three key technical bottlenecks in current germ cell research: fragmented reference data, inaccurate developmental stage annotation, and lack of quantitative standards for culture system data.

Description

Technical Field

[0001] This invention relates to an AI-based method for evaluating human germ cell development and culture systems, belonging to the field of single-cell transcriptome data analysis technology. Background Technology

[0002] Currently, three core technical challenges urgently need to be addressed in the evaluation of germ cell developmental stages and in vitro culture systems based on human germ cell single-cell reference atlases: First, existing research data is too fragmented, data processing standards vary, and germ cell classification standards are inconsistent, making it impossible to establish a unified and comparable reference atlas. Second, existing annotation tools lack sufficient precision in identifying germ cell developmental stages, making it difficult to construct highly accurate developmental prediction models. Finally, existing in vitro culture system evaluation methods have significant deficiencies, lacking single-cell resolution and multi-dimensional evaluation indicators comparable to in vivo developmental data. These technical bottlenecks severely restrict the development of human germ cell research and clinical applications. Summary of the Invention

[0003] The purpose of this invention is to provide an AI-based method for evaluating human germ cell development and culture systems. This method realizes a complete solution from single-cell reference atlas construction and developmental stage prediction to culture system evaluation, and solves the three major technical bottlenecks in the current field of germ cell research: fragmented reference data, inaccurate developmental stage annotation, and lack of quantitative standards for culture system data.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is: an AI-based method for evaluating human germ cell development and culture systems, comprising the following steps:

[0005] Step 1: Construct a standardized reference map, including:

[0006] (1.1) Obtain raw transcriptome data from multiple single-cell cells;

[0007] (1.2) Differentiate data based on sequencing technology type and raw data format;

[0008] (1.3) A three-level quality screening system was used to filter the data;

[0009] (1.4) Based on developmental stage and sex characteristics, the germ cell data were divided into three groups: prenatal male germ cell group, prenatal female germ cell group, and postnatal male germ cell group;

[0010] (1.5) Analyze each group independently, remove abnormal cell clusters, and annotate the cell clusters;

[0011] (1.6) Integrate the annotation results with known germ cell development processes to construct a unified developmental hierarchy annotation system;

[0012] Step 2: Establish a prediction model for germ cell development stages, including: a single-cell annotation variational inference prediction module based on transfer learning, a single-cell annotation variational inference combined with k-nearest neighbor joint prediction module, and a cell type classification module optimized by feature selection. The three modules achieve collaborative decision-making through a three-level confidence system.

[0013] Step 3, Evaluation of the in vitro culture system, including:

[0014] (3.1) Construct a dual index for quantitative evaluation, wherein the dual indexes are k-nearest neighbor distance and metacell-Pearson correlation coefficient;

[0015] (3.2) Analysis of evaluation results: Based on the dual-index evaluation, the similarity between cultured cells and normally developing cells in vivo and the quality of the culture system are judged.

[0016] Furthermore, in step 1, the three-level quality screening system rigorously filters the data, including:

[0017] Individual-level screening: retain sample data of individuals under 45 years of age, with a normal BMI, and no abnormalities in gametogenesis;

[0018] Document-level processing: A dual-cell recognition algorithm is used to automatically detect and remove dual-cell interference;

[0019] Cell-level filtering: Differentiated filtering criteria are set according to the type of sequencing technology, as follows:

[0020] For 10x Genomics Chromium sequencing data, cells that simultaneously meet the following conditions are retained: (1) number of gene detections > 1000 and < 8000; (2) mitochondrial gene expression percentage < 20%.

[0021] For SRT-seq sequencing data, cells meeting the following criteria are retained: number of gene sequences > 2000 and < 12000.

[0022] Furthermore, in step 1, the independent analysis process for each group is as follows:

[0023] (1.51) Using Scanpy software, gene variability was calculated on a per-dataset basis, and the genes were sorted from high to low. The top 2000 genes were selected as input features for the single-cell variational inference model.

[0024] (1.52) Data representation learning, the steps are as follows:

[0025] (1.521) A single-cell variational inference model was used, with a counting matrix of 2000 hypervariable genes as input;

[0026] (1.522) Set datasets and libraries as factors to consider in batch calibration;

[0027] (1.523) Train a single-cell variational inference model and extract 32-dimensional latent spatial features;

[0028] (1.524) Perform clustering and uniform manifold approximation and projection dimensionality reduction analysis in the latent space;

[0029] (1.53) In the projection dimensionality reduction analysis, the following abnormal cell clusters are identified and removed according to the set criteria:

[0030] The number of gene detections in the cell cluster is less than the 5th percentile of the entire dataset (calculated in ascending order, the same below);

[0031] The median transcript count in the cell cluster was less than the 5th percentile of the entire dataset.

[0032] The average mitochondrial gene expression in the cell cluster was greater than the 95th percentile of the entire dataset.

[0033] The cell cluster lacks expression of known germ cell marker genes;

[0034] (1.54) After removing abnormal cell clusters, repeat step 1.52 to optimize data quality;

[0035] (1.55) Cell type annotation:

[0036] Cell cluster annotation is performed based on marker gene expression patterns and original annotation information;

[0037] Annotation of male and female germ cell groups before birth to the fine developmental stage;

[0038] After birth, the male group data were first roughly divided into spermatogonia / spermatocytes / sperm cells, and then fine annotation was achieved by iterative analysis and repeating steps 1.52 to 1.55.

[0039] Furthermore, in step 2, the single-cell annotation variational inference prediction module based on transfer learning includes a single-cell variational inference model and a single-cell annotation variational inference model, the training of which includes:

[0040] Pre-training optimization phase: The single-cell variational inference model is pre-trained using optimized hyperparameter configurations, specifically with a latent space dimension of 32, a neuron dropout rate of 0.2, and a training batch size of 1024.

[0041] Hierarchical transfer learning stage: Training a single-cell annotation variational inference model based on a multi-level annotated germ cell development atlas, with a maximum training epoch of 20 epochs.

[0042] Furthermore, in step 2, the execution steps of the joint prediction module combining single-cell annotation variational inference and k-nearest neighbors include:

[0043] Joint embedding space construction: Unknown sample data and reference datasets are jointly mapped to a 32-dimensional latent space to eliminate batch effects;

[0044] k-Nearest Neighbor Classification Prediction: In the latent space, calculate the Euclidean distance between the unknown cell and the reference cell, select the 10 nearest reference cells, count the cell subtype frequencies of the reference cells, and select the cell subtype with the highest frequency as the predicted identity of the unknown cell.

[0045] Furthermore, in step 2, the execution steps of the feature selection optimized cell type classification model include:

[0046] Dual gene screening: The first round uses stochastic gradient descent to screen feature genes, and the second round selects the top 300 genes by cell type for final classification training. The screening process allows the model to focus on the most discriminative features.

[0047] Category-specific modeling: A one-to-many multi-classification strategy is adopted to establish a dedicated classifier for each cell subtype, and finally predict the cell subtype identity of unknown cells by probability.

[0048] Furthermore, in step 2, based on the prediction results of the three modules, a multi-module consensus decision-making mechanism is constructed and a three-level confidence assessment system is established:

[0049] High confidence result: If the three modules predict the cell subtype of the unknown cell in complete agreement, the predicted identity is output and considered a high confidence result.

[0050] Medium confidence result: If any two of the three modules predict the cell subtype of the unknown cell in the same way, the predicted identity is output and considered to be a medium confidence result.

[0051] Low confidence result: If the three modules do not predict the cell subtype of an unknown cell, it is considered a low confidence result and the cell type cannot be output.

[0052] Furthermore, in step 3, the cell stability assessment based on the k-nearest neighbor distance is as follows:

[0053] Joint Embedding Space Construction: The single-cell annotation variational inference prediction module in the germ cell development stage prediction model is used to jointly map in vitro cultured cells and in vivo reference dataset cells to a 32-dimensional latent feature space to ensure data comparability;

[0054] k-Nearest Neighbor Distance Calculation: For each cultured cell, calculate its Euclidean distance to all cells in the reference dataset, select the 10 nearest reference cells, and then calculate the average Euclidean distance between the cultured cell and the 10 nearest reference cells.

[0055] Furthermore, in step 3, the similarity assessment based on the metacell-Pearson correlation coefficient is as follows:

[0056] Gene screening: 2000 highly discriminative characteristic genes were predefined using a germ cell development stage prediction model to ensure the specificity of the analysis;

[0057] Reference expression profile construction: For each cultured cell, its 10 nearest neighbor reference cells are determined in the potential space, and the average expression value of these reference cells on characteristic genes is calculated as the reference expression profile;

[0058] Similarity quantification: The overall similarity between cultured cells and the reference expression profile was assessed using the Pearson correlation coefficient.

[0059] Furthermore, in step 3, the similarity between cultured cells and normally developing cells in vivo is determined based on the following threshold criteria:

[0060] High-quality cell populations: k-nearest neighbor distance < 1.5 and correlation coefficient > 0.5, indicating that the cultured cells are highly similar to normally developing cells in vivo;

[0061] Deviated cell population:

[0062] (1) If the k-nearest neighbor distance is > 1.5 or the correlation coefficient is < 0.5, it indicates that the cultured cells have a low similarity to normally developing cells in vivo;

[0063] (2) If the k-nearest neighbor distance is > 2 or the correlation coefficient is < 0.3, it indicates that the cultured cells have a very low similarity to normally developing cells in vivo;

[0064] Then, the proportion of high-quality cell populations in vitro was calculated, and the quality of the culture system was evaluated: the higher the proportion of high-quality cell populations and the further the developmental stage, the more effective the culture system was considered; otherwise, the culture system was considered to need further optimization.

[0065] The beneficial effects of this invention are as follows:

[0066] (1) Data integration: The most comprehensive human germ cell development atlas has been constructed, covering the entire developmental process from primitive germ cells to mature gametes. The standardized processing procedures make different research data comparable.

[0067] (2) Model prediction: A high-precision model for predicting the developmental stage of germ cells was established. The prediction accuracy rate was 99.5% at the first annotation level and 93% at the finest annotation level. At the same time, it was significantly better than other existing tools in the other three typical model evaluation indicators, namely Cohen-Kappa coefficient, macro-average F1 score and micro-average F1 score.

[0068] (3) Application value: It can complete the whole process analysis from gene expression matrix to culture evaluation results, and can be successfully applied to real in vitro culture data, providing optimization clues. Attached Figure Description

[0069] Figure 1 This is a flowchart of the evaluation method of the present invention;

[0070] Figure 2 This is a schematic diagram of the standardized reference atlas of human germ cells constructed in this invention;

[0071] Figure 3 This is a flowchart of the construction process of the germ cell development stage prediction model of the present invention;

[0072] Figure 4 This is a graph showing the comparison of the predictive performance of the germ cell development stage prediction model of this invention with existing tools;

[0073] Figure 5 This is an example of the results of the in vitro culture data quality assessment using the germ cell development stage prediction model of the present invention. Detailed Implementation

[0074] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0075] like Figures 1 to 5 An AI-based evaluation method for human germ cell development and culture systems mainly includes the following three steps:

[0076] Step 1: Construct a standardized reference map: Establish a unified data processing workflow and a strict three-level quality control system;

[0077] Step 2: Establish a high-precision prediction model for germ cell development stages, including: a single-cell annotation variational inference prediction module based on transfer learning, a joint prediction module combining single-cell annotation variational inference with k-nearest neighbors, and a cell type classification model optimized by feature selection. The three modules achieve collaborative decision-making through a three-level confidence system.

[0078] Step 3: Evaluation of the in vitro culture system. Based on the k-nearest neighbor distance and the metacell-Pearson correlation coefficient, the similarity between the cultured cells and normally developing cells in vivo is determined.

[0079] The specific implementation process of step 1 is as follows:

[0080] 1.1. Obtain multiple sets of raw single-cell transcriptome data from international public databases (GEO, ENA).

[0081] 1.2 A differentiated processing workflow is adopted based on the sequencing technology type and data format:

[0082] 1.2.1 10x Genomics Chromium sequencing technology fastq format data: The official analysis software CellRanger was used for alignment and quantification based on the official 2024-A version transcriptome reference set.

[0083] 1.2.2, 10x Genomics Chromium sequencing technology BAM format data: Convert to FASTQ format using the official tool bam2fastq, and then process according to the procedure in 1.2.1.

[0084] 1.2.3. The unique molecular identifier count format data of 10x Genomics Chromium sequencing technology, based on the correspondence between gene database identifiers and standard gene identifiers, will be uniformly converted to the 2024-A version standard naming.

[0085] 1.2.4, STRT-seq sequencing technology fastq format data: The open-source analysis workflow was used for alignment and quantification based on the same reference genome 2024-A.

[0086] 1.3 A three-level quality screening system is used to strictly filter the data:

[0087] 1.3.1 Individual-level screening: Individual sample data were retained from individuals under 45 years of age, with a normal BMI, and no abnormalities in gamete formation.

[0088] 1.3.2 Library-level processing: For 10x Genomics Chromium sequencing data, a dual-cell recognition algorithm is used to automatically detect and remove dual-cell interference.

[0089] 1.3.3 Cell-level filtering: Differentiated filtering criteria are set according to the sequencing technology type, as follows:

[0090] For 10x Genomics Chromium sequencing data, cells that simultaneously meet the following conditions are retained: (1) number of gene detections > 1000 and < 8000; (2) mitochondrial gene expression percentage < 20%.

[0091] For SRT-seq sequencing data, cells meeting the following criteria are retained: number of gene sequences > 2000 and < 12000.

[0092] 1.4 Based on developmental stage and sex characteristics, germ cell data are divided into the following three groups: prenatal male germ cell group, prenatal female germ cell group, and postnatal male germ cell group.

[0093] 1.5 Perform the following analysis independently for each group:

[0094] 1.5.1 Using Scanpy software, gene variability was calculated on a per-dataset basis, sorted from high to low, and the top 2000 genes were selected as input features for the single-cell variational inference model.

[0095] 1.5.2, Data Representation Learning, the steps are as follows:

[0096] 1.5.2.1, Using a single-cell variational inference model, input a counting matrix of 2000 hypervariable genes;

[0097] 1.5.2.2, Set datasets and libraries as factors to consider in batch calibration;

[0098] 1.5.2.3, Train the single-cell variational inference model and extract 32-dimensional latent spatial features;

[0099] 1.5.2.4, perform clustering and uniform manifold approximation and projection dimensionality reduction analysis in the latent space;

[0100] 1.5.3 In the projection dimensionality reduction analysis, the following abnormal cell clusters are identified and removed according to the set criteria:

[0101] The number of gene detections in the cell cluster is less than the 5th percentile of the entire dataset (calculated in ascending order, the same below);

[0102] The median transcript count in the cell cluster was less than the 5th percentile of the entire dataset.

[0103] The average mitochondrial gene expression in the cell cluster was greater than the 95th percentile of the entire dataset.

[0104] The cell cluster lacks expression of known germ cell marker genes;

[0105] 1.5.4 After removing abnormal cell clusters, repeat step 1.52 to optimize data quality;

[0106] 1.5.5, Cell Type Annotation:

[0107] Cell cluster annotation is performed based on marker gene expression patterns and original annotation information;

[0108] Annotation of male and female germ cell groups before birth to the fine developmental stage;

[0109] After birth, the male group data were first roughly divided into spermatogonia / spermatocytes / sperm cells, and then fine annotation was achieved by iterative analysis and repeating steps 1.52 to 1.55.

[0110] 1.6 Integrating detailed annotation results with known germ cell development processes, a unified developmental hierarchical annotation system was constructed. For male germ cells, a four-level developmental annotation system was constructed. For female germ cells, a three-level developmental annotation system was constructed.

[0111] Step 2, the establishment of the germ cell development stage prediction model includes:

[0112] 2.1 Core module architecture of the prediction model:

[0113] 2.1.1 Single-cell annotation variational inference prediction module based on transfer learning:

[0114] 2.1.1.1 Pre-training Optimization Phase: The single-cell variational inference model was pre-trained using optimized hyperparameter configurations. Specifically, the latent space dimension was 32, the neuron dropout rate was 0.2, and the training batch size was 1024. Testing showed that this parameter combination achieved the best balance between classification accuracy and model generalization ability.

[0115] 2.1.1.2, Hierarchical Transfer Learning Stage: A single-cell annotation variational inference model is trained based on a multi-level annotated germ cell development atlas, with a maximum training epoch of 20 rounds. This design overcomes the limitations of traditional single-level annotation models, simultaneously meeting the needs of coarse-grained phylogenetic analysis and fine-grained developmental stage identification.

[0116] 2.1.2, Joint prediction module combining single-cell annotation variational inference and k-nearest neighbors:

[0117] 2.1.2.1 Joint Embedding Space Construction: First, the unknown sample data and the reference dataset are jointly mapped to a 32-dimensional latent space to effectively eliminate batch effects;

[0118] 2.1.2.2, k-Nearest Neighbor Classification Prediction: Calculate the Euclidean distance between the unknown cell and reference cells, select the 10 nearest reference cells, count the frequency of cell subtypes of the reference cells, and select the cell subtype with the highest frequency as the predicted identity of the unknown cell. This method effectively solves the cell boundary classification problem during the developmental transition period.

[0119] 2.1.3, Cell type classification module with optimized feature selection:

[0120] 2.1.3.1, Dual gene screening: In the first round, feature genes are screened through stochastic gradient descent. In the second round, the top 300 genes by weight are selected according to cell type for final classification training. Through screening, the model focuses on the most discriminative features.

[0121] 2.1.3.2, Category-Specific Modeling. A one-to-many multi-classification strategy is adopted to establish a dedicated classifier for each cell subtype, and finally predict the cell subtype identity of unknown cells through probability.

[0122] 2.2 Based on the prediction results of the three modules, a multi-module consensus decision-making mechanism is constructed and a three-level confidence assessment system is established:

[0123] 2.2.1 High-confidence result: If the three modules predict the cell subtype of the unknown cell in complete agreement, the predicted identity is output and considered a high-confidence result.

[0124] 2.2.2, Medium Confidence Result: If any two of the three modules predict the cell subtype of the unknown cell in the same way, the predicted identity is output and considered to be a medium confidence result;

[0125] 2.2.3 Low Confidence Result: If the prediction results of the cell subtype of the unknown cell are different for the three modules, it is considered a low confidence result and the cell type cannot be output.

[0126] Step 3, in vitro culture evaluation, specifically includes:

[0127] 3.1 Constructing a dual-indicator quantitative evaluation method:

[0128] 3.1.1 Cell stability assessment based on k-nearest neighbor distance:

[0129] 3.1.1.1 Joint Embedding Space Construction: The single-cell annotation variational inference prediction module in the germ cell development stage prediction model is used to jointly map in vitro cultured cells and in vivo reference dataset cells to a 32-dimensional latent feature space to ensure data comparability.

[0130] 3.1.1.2, k-nearest neighbor distance calculation: For each cultured cell, calculate its Euclidean distance with all cells in the reference dataset, select the 10 nearest reference cells, and then calculate the average Euclidean distance between the cultured cell and the 10 nearest reference cells.

[0131] 3.1.2 Similarity assessment based on the metacellular-Pearson correlation coefficient:

[0132] 3.1.2.1 Gene screening: 2000 highly discriminative characteristic genes were predefined using a germ cell development stage prediction model to ensure the specificity of the analysis.

[0133] 3.1.2.2, Construction of reference expression profile: For each cultured cell, determine its 10 nearest neighbor reference cells in the potential space, and calculate the average expression value of these reference cells on the characteristic genes as the reference expression profile.

[0134] 3.1.2.3 Similarity Quantification: The Pearson correlation coefficient was used to assess the overall similarity between cultured cells and the reference expression profile.

[0135] 3.2 Analysis and application of evaluation results:

[0136] 3.2.1 The following threshold criteria are used to determine the similarity between cultured cells and normally developing cells in vivo, as well as the quality of the culture system:

[0137] 3.2.1.1, High-quality cell population: k-nearest neighbor distance < 1.5 and correlation coefficient > 0.5, indicating that the cultured cells are highly similar to normally developing cells in vivo.

[0138] 3.2.1.2, Deviation from cell population:

[0139] (1) If the k-nearest neighbor distance is > 1.5 or the correlation coefficient is < 0.5, it indicates that the cultured cells have a low similarity to normally developing cells in vivo;

[0140] (2) If the k-nearest neighbor distance is > 2 or the correlation coefficient is < 0.3, it indicates that the cultured cells have a very low similarity to normally developing cells in vivo.

[0141] Then, the proportion of high-quality cell populations in vitro was calculated, and the quality of the culture system was evaluated: the higher the proportion of high-quality cell populations and the further the developmental stage, the more effective the culture system was considered; otherwise, the culture system was considered to need further optimization.

[0142] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the above embodiments do not limit the scope of protection of the present invention in any way, and all technical solutions obtained by equivalent substitution or other means fall within the scope of protection of the present invention. Parts not covered in this invention are the same as or can be implemented using existing technology.

Claims

1. An AI-based evaluation method for human germ cell development and culture systems, characterized in that, Includes the following steps: Step 1: Construct a standardized reference map, including: (1.1) Obtain raw transcriptome data from multiple single-cell cells; (1.2) Differentiate data based on sequencing technology type and raw data format; (1.3) A three-level quality screening system was used to filter the data; (1.4) Based on developmental stage and sex characteristics, the germ cell data were divided into three groups: prenatal male germ cell group, prenatal female germ cell group, and postnatal male germ cell group; (1.5) Analyze each group independently, remove abnormal cell clusters, and annotate the cell clusters; (1.6) Integrate the annotation results with known germ cell development processes to construct a unified developmental hierarchy annotation system; Step 2: Establish a prediction model for germ cell development stages, including: a single-cell annotation variational inference prediction module based on transfer learning, a joint prediction module combining single-cell annotation variational inference with k-nearest neighbors, and a cell type classification module optimized by feature selection. These three modules achieve collaborative decision-making through a three-level confidence system. Specifically, based on the prediction results of the three modules, a multi-module consensus decision-making mechanism is constructed, and a three-level confidence evaluation system is established. High confidence result: If the three modules predict the cell subtype of the unknown cell in complete agreement, the predicted identity is output and considered a high confidence result. Medium confidence result: If any two of the three modules predict the cell subtype of the unknown cell in the same way, the predicted identity is output and considered to be a medium confidence result. Low confidence result: If the three modules do not predict the cell subtype of an unknown cell in the same way, it is considered a low confidence result and the cell type cannot be output. Step 3, Evaluation of the in vitro culture system, including: (3.1) Construct a dual index for quantitative evaluation, wherein the dual indexes are k-nearest neighbor distance and metacell-Pearson correlation coefficient; (3.2) Analysis of evaluation results: Based on the dual-index evaluation, the similarity between cultured cells and normally developing cells in vivo and the quality of the culture system are judged; Cell stability assessment based on k-nearest neighbor distance is as follows: Joint Embedding Space Construction: The single-cell annotation variational inference prediction module in the germ cell development stage prediction model is used to jointly map in vitro cultured cells and in vivo reference dataset cells to a 32-dimensional latent feature space to ensure data comparability; k-Nearest Neighbor Distance Calculation: For each cultured cell, calculate its Euclidean distance to all cells in the reference dataset, select the 10 nearest reference cells, and then calculate the average Euclidean distance between the cultured cell and the 10 nearest reference cells. The similarity assessment based on the metacellular-Pearson correlation coefficient is as follows: Gene screening: 2000 highly discriminative characteristic genes were predefined using a germ cell development stage prediction model to ensure the specificity of the analysis; Reference expression profile construction: For each cultured cell, its 10 nearest neighbor reference cells are determined in the potential space, and the average expression value of these reference cells on characteristic genes is calculated as the reference expression profile; Similarity quantification: The overall similarity between cultured cells and the reference expression profile was assessed using the Pearson correlation coefficient.

2. The AI-based evaluation method for human germ cell development and culture system according to claim 1, characterized in that, In step 1, the three-level quality screening system rigorously filters the data, including: Individual-level screening: retain sample data of individuals under 45 years of age, with a normal BMI, and no abnormalities in gametogenesis; Document-level processing: A dual-cell recognition algorithm is used to automatically detect and remove dual-cell interference; Cell-level filtering: Differentiated filtering criteria are set according to the type of sequencing technology, as follows: For 10x Genomics Chromium sequencing data, cells that simultaneously meet the following conditions are retained: (1) number of gene detections > 1000 and < 8000; (2) mitochondrial gene expression percentage < 20%. For STRT-seq sequencing data, cells meeting the following criteria are retained: number of gene sequences > 2000 and <12000.

3. The AI-based evaluation method for human germ cell development and culture system according to claim 1, characterized in that, In step 1, each group is analyzed independently, and the process is as follows: (1.51) Using Scanpy software, gene variability was calculated on a per-dataset basis, and the genes were sorted from high to low. The top 2000 genes were selected as input features for the single-cell variational inference model. (1.52) Data representation learning, the steps are as follows: (1.521) A single-cell variational inference model was used, with a counting matrix of 2000 hypervariable genes as input; (1.522) Set datasets and libraries as factors to consider in batch calibration; (1.523) Train a single-cell variational inference model and extract 32-dimensional latent spatial features; (1.524) Perform clustering and uniform manifold approximation and projection dimensionality reduction analysis in the latent space; (1.53) In the projection dimensionality reduction analysis, the following abnormal cell clusters are identified and removed according to the set criteria: The number of gene sequences detected in the cell cluster was less than the 5th percentile of the entire dataset. The median transcript count in the cell cluster was less than the 5th percentile of the entire dataset. The average mitochondrial gene expression in the cell cluster was greater than the 95th percentile of the entire dataset. The cell cluster lacks expression of known germ cell marker genes; (1.54) After removing abnormal cell clusters, repeat step 1.52 to optimize data quality; (1.55) Cell type annotation: Cell cluster annotation is performed based on marker gene expression patterns and original annotation information; Annotation of male and female germ cell groups before birth to the fine developmental stage; After birth, the male group data were first roughly divided into spermatogonia / spermatocytes / sperm cells, and then fine annotation was achieved by iterative analysis and repeating steps 1.52 to 1.

55.

4. The AI-based evaluation method for human germ cell development and culture system according to claim 1, characterized in that, In step 2, the single-cell annotation variational inference prediction module based on transfer learning includes a single-cell variational inference model and a single-cell annotation variational inference model, the training of which includes: Pre-training optimization phase: The single-cell variational inference model is pre-trained using optimized hyperparameter configurations, specifically with a latent space dimension of 32, a neuron dropout rate of 0.2, and a training batch size of 1024. Hierarchical transfer learning phase: Training a single-cell annotation variational inference model based on a multi-level annotated germ cell development atlas, with a maximum training epoch of 20 epochs.

5. The AI-based evaluation method for human germ cell development and culture system according to claim 1, characterized in that, In step 2, the execution steps of the joint prediction module combining single-cell annotation variational inference and k-nearest neighbors include: Joint embedding space construction: First, unknown sample data and reference dataset are jointly mapped to a 32-dimensional latent space to eliminate batch effects; k-Nearest Neighbor Classification Prediction: Next, in the latent space, calculate the Euclidean distance between the unknown cell and the reference cell, select the 10 nearest reference cells, count the cell subtype frequency of the reference cells, and select the cell subtype with the highest frequency as the predicted identity of the unknown cell.

6. The AI-based evaluation method for human germ cell development and culture system according to claim 1, characterized in that, In step 2, the execution steps of the feature selection optimized cell type classification model include: Dual gene screening: The first round uses stochastic gradient descent to screen feature genes, and the second round selects the top 300 genes by cell type for final classification training. The screening process allows the model to focus on the most discriminative features. Category-specific modeling: A one-to-many multi-classification strategy is adopted to establish a dedicated classifier for each cell subtype, and finally predict the cell subtype identity of unknown cells by probability.

7. The AI-based evaluation method for human germ cell development and culture system according to claim 1, characterized in that, In step 3, the similarity between cultured cells and normally developing cells in vivo is determined based on the following threshold criteria: High-quality cell populations: k-nearest neighbor distance < 1.5 and correlation coefficient > 0.5, indicating that the cultured cells are highly similar to normally developing cells in vivo; Deviated cell population: (1) If the k-nearest neighbor distance is > 1.5 or the correlation coefficient is < 0.5, it indicates that the cultured cells have a low similarity to normally developing cells in vivo; (2) If the k-nearest neighbor distance is > 2 or the correlation coefficient is < 0.3, it indicates that the cultured cells have a very low similarity to normally developing cells in vivo; Then, the proportion of high-quality cell populations in vitro was calculated, and the quality of the culture system was evaluated: the higher the proportion of high-quality cell populations and the later the developmental stage, the more effective the culture system was considered; otherwise, the culture system was considered to need further optimization.

Images