A method for removing parent source pollution based on machine learning and related equipment

Abstract

The present invention provides a method for removing maternal contamination based on machine learning and related equipment, which are applied to the field of data processing technology. The present application obtains target user data and a training sample set, wherein the target user data includes culture fluid sample information of in vitro cultured embryos of the target user, and the training sample set includes culture fluid sample information of in vitro cultured embryos of other users; pre-processes the training sample set to generate a training data set with target feature data, wherein the target feature data is methylation sequencing data of a single cell whole genome, and the methylation sequencing data includes read information; obtains an initial maternal contamination warning model that matches the target feature data; trains the initial maternal contamination warning model based on the training data set with the target feature data to generate a target maternal contamination warning model; processes the target user data based on the target maternal contamination warning model to generate a target chromosome ploidy.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method for removing parent source pollution based on machine learning and related equipment. Background Art

[0002] The rate of embryonic aneuploidy increases exponentially with the age of the mother, and is the main cause of implantation failure, miscarriage and congenital abnormalities in natural conception and in vitro fertilization. Accurate identification of high-quality embryos is a hot issue in the field of reproductive medicine. Preimplantation genetic testing for aneuploidies (PGT-A) has become an important tool for evaluating embryonic chromosomal ploidy abnormalities. A systematic study of 11 randomized controlled trials showed that PGT-A can significantly improve the live birth rate of older pregnant women (over 35 years old). Currently, the commonly used PGT-A techniques in clinical practice include polar body, blastomere or trophoblast cell (TE) biopsy. All of the above methods are invasive operations on embryos, and have many disadvantages in terms of convenience and potential adverse effects on embryos.

[0003] Recent scientific studies have used cell-free DNA (cfDNA) in the culture medium (SECM) of in vitro cultured embryos for noninvasive preimplantation genetic testing for aneuploidies (niPGT-A), which has opened up a new path for noninvasive assessment of embryo quality. However, challenges still exist. There is a large amount of maternal DNA contamination in SECM, which can lead to gender inconsistency and false-negative results. Single-nucleotide polymorphism (SNP) studies have shown that there is serious maternal DNA contamination (86-94%) in SECM; whole-genome DNA methylation sequencing accurately identified that cfDNA in the culture medium came from blastocysts, maternal cells, and polar body cells. SECM samples with severe maternal contamination (>60%) account for one-third of the total samples. As maternal contamination increases, the gender inconsistency rate and false-negative rate increase accordingly. Once maternal contamination exceeds 60%, the embryonic chromosome ploidy inferred from SECM is inaccurate. Severe maternal contamination will greatly challenge the reliability of SECM chromosome ploidy detection, so this highlights the necessity and importance of removing maternal contamination in niPGT.

[0004] At present, the method for assessing the level of maternal contamination is mainly to use DNA methylation marker detection. However, this method can only estimate the level of maternal contamination but cannot remove it at the same time. At this stage, measures to reduce the level of maternal contamination are mainly focused on optimizing the sample collection process. For example, changing the SECM collection time, adjusting the embryo flushing scheme, culture medium renewal and stripping of residual maternal cells. Prolonging the culture time will increase the amount of cfDNA in the culture medium. SECM of embryos cultured to day 6 / 7 showed more information and higher consistency than SECM of embryos cultured to day 5. Compared with the one-step embryo flushing scheme, the sequential flushing method showed significantly superior results. Replacing the embryo culture medium on day 4 increased the consistency between niPGT and TE biopsy. In addition, stripping the residual granulosa cells again on day 3 improved the accuracy of cfDNA detection in niPGT. However, these methods have limited effects in removing maternal contamination and require additional complicated experimental steps, which increases the workload of IVF laboratory staff and is challenging to promote on a large scale.

[0005] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0006] The purpose of this application is to provide a method and related equipment for removing maternal contamination based on machine learning, which at least overcomes the problems existing in the prior art to a certain extent, analyzes the free DNA of in vitro cultured embryos through single-cell whole-genome methylation sequencing, uses deep learning model training to distinguish the embryonic or maternal origin of the reads, estimates the proportion of maternal contamination, and removes highly contaminated reads, so as to accurately detect the chromosome ploidy of the embryo and reduce false negative results.

[0007] Other features and advantages of the present application will become apparent from the following detailed description, or may be learned in part by the practice of the present invention.

[0008] According to one aspect of the present application, a method for removing maternal contamination based on machine learning is provided, comprising: obtaining target user data and a training sample set, wherein the target user data includes culture medium sample information of in vitro cultured embryos of the target user, and the training sample set includes culture medium sample information of in vitro cultured embryos of other users; preprocessing the training sample set to generate a training data set with target feature data, wherein the target feature data is methylation sequencing data of a single-cell whole genome, and the methylation sequencing data includes read information; obtaining an initial maternal contamination warning model that matches the target feature data; training the initial maternal contamination warning model based on the training data set with the target feature data to generate a target maternal contamination warning model; processing the target user data based on the target maternal contamination warning model to generate a target chromosome ploidy.

[0009] Another aspect of the present application is a maternal contamination removal device based on machine learning, characterized in that it includes: an acquisition module, used to acquire target user data and a training sample set, wherein the target user data includes culture medium sample information of in vitro cultured embryos of the target user, and the training sample set includes culture medium sample information of in vitro cultured embryos of other users; acquiring an initial maternal contamination warning model that matches the target feature data; a processing module, used to preprocess the training sample set to generate a training data set with target feature data, wherein the target feature data is methylation sequencing data of a single-cell whole genome, and the methylation sequencing data includes read information; training the initial maternal contamination warning model based on the training data set with the target feature data to generate a target maternal contamination warning model; processing the target user data based on the target maternal contamination warning model to generate a target chromosome ploidy.

[0010] According to another aspect of the present application, an electronic device is characterized in that it includes: a first processor; and a memory for storing executable instructions of the first processor; wherein the first processor is configured to execute the above-mentioned machine learning-based maternal source pollution removal method by executing the executable instructions.

[0011] According to another aspect of the present application, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a second processor, the above-mentioned method for removing parent source pollution based on machine learning is implemented.

[0012] According to another aspect of the present application, a computer program product is provided, including a computer program, characterized in that when the computer program is executed by a third processor, the above-mentioned method for removing parent source pollution based on machine learning is implemented.

[0013] The present application provides a method for removing maternal contamination based on machine learning and related equipment, in which a server performs single-cell whole-genome methylation library construction and sequencing on cfDNA (free DNA) in SECM (culture medium for in vitro cultured embryos). A series of data preprocessing steps are then performed, including removing sequencing adapters, primers, removing low-quality bases from the original sulfite sequencing reads, and aligning high-quality reads to the reference genome. Training data and deep learning models are then constructed for training, with the aim of determining the classification probability that each read in the data belongs to the embryo or the mother. After training, the proportion of maternal contamination in the sample is estimated from the classification probability by constructing a mathematical model based on maximum likelihood estimation. At the same time, according to the classification probability of the reads, reads with high maternal probabilities, i.e., maternal contamination, are removed, and the embryo's cfDNA is retained for copy number variations (CNVs) detection to restore the true chromosome ploidy of the embryo and remove false negative results.

[0014] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A flow chart showing a method for removing maternal pollution based on machine learning provided in one embodiment of the present application is shown;

[0016] Figure 2 A schematic structural diagram of a maternal source pollution removal device based on machine learning provided in one embodiment of the present application is shown;

[0017] Figure 3 A schematic diagram of the structure of an electronic device provided by an embodiment of the present application is shown;

[0018] Figure 4 A schematic diagram of a storage medium provided by an embodiment of the present application is shown;

[0019] Figure 5 A schematic diagram of screening embryo copy number provided in one embodiment of the present application is shown. DETAILED DESCRIPTION

[0020] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0021] Combine the following Figure 1To describe the method for removing maternal pollution based on machine learning according to an exemplary embodiment of the present application. It should be noted that the following application scenarios are only shown to facilitate understanding of the spirit and principles of the present application, and the embodiments of the present application are not limited in this regard. On the contrary, the embodiments of the present application can be applied to any applicable scenario.

[0022] In one embodiment, the present application also proposes a method for removing maternal pollution based on machine learning and related equipment. Figure 1 The following schematically shows a flow chart of a method for removing maternal pollution based on machine learning according to an embodiment of the present application. Figure 1 As shown, the method is applied to a server, comprising:

[0023] S101, obtaining target user data and a training sample set.

[0024] In one embodiment, the target user data includes the culture medium sample information of the target user's in vitro cultured embryos, and the training sample set includes the culture medium sample information of other users' in vitro cultured embryos.

[0025] Patients with more than 10 embryos on the third day of in vitro culture for blastocyst culture were selected for PGT-A (Preimplantation Genetic Testing for Aneuploidies). The discarded culture medium of embryos developed to the expanded blastocyst stage (usually on the 5th to 6th day) was collected using polymerase chain reaction (PCR) tubes. These culture medium samples contained free DNA from embryonic and maternal cells. The collected samples were stored in a -20°C refrigerator to maintain sample stability and prevent DNA degradation.

[0026] DNA is extracted from the stored culture medium samples, including fetal cfDNA (cell-free DNA) and possible maternal DNA, and the extracted DNA is quality tested to ensure that it is suitable for subsequent methylation sequencing and analysis. The extracted DNA is subjected to bisulfite sequencing to determine the methylation status of the DNA. This helps to distinguish between fetal and maternal DNA, as their methylation patterns are usually different.

[0027] Through this process, the genetic information of embryos can be effectively extracted and analyzed from culture fluid samples, while reducing the impact of maternal contamination and improving the accuracy and reliability of PGT-A. This is of great significance for improving the success rate of in vitro fertilization and preventing genetic diseases.

[0028] S102, preprocessing the training sample set to generate a training data set with target feature data.

[0029] In one embodiment, the target feature data is methylation sequencing data of a single cell whole genome, and the methylation sequencing data includes read information.

[0030] The training sample set is subjected to feature extraction processing to generate a sequencing database, wherein the sequencing database includes a number of original bisulfite sequencing information, and the original bisulfite sequencing information is processed respectively to generate target feature data, wherein the target feature data includes target bisulfite sequencing information, and the target bisulfite sequencing information is methylation sequencing data of a single-cell whole genome whose methylation read information is greater than a preset threshold.

[0031] When processing cfDNA (free DNA) in SECM (culture medium of in vitro cultured embryos) for single-cell whole-genome methylation library construction and sequencing, data preprocessing is a crucial step to ensure accurate and reliable analysis results. The following are the key steps of data preprocessing:

[0032] Removal of sequencing adapters and primers: During the sequencing process, adapters and primers are added to the DNA fragments to facilitate sequencing. These sequences need to be removed before data analysis because they do not contain genetic information related to the research target.

[0033] Removing low-quality bases: Some errors may occur during the sequencing process, resulting in low quality of some bases. Removing these low-quality bases helps reduce the impact of sequencing errors on subsequent analysis.

[0034] Alignment to the reference genome: Align the sequenced reads to the human reference genome to determine their locations in the genome. This step is the basis for subsequent analysis, such as variant detection and gene expression analysis.

[0035] Handling unaligned reads: For reads that fail to align successfully to the reference genome, it may be necessary to use more flexible alignment strategies, such as local alignment, to identify reads that may not be identified due to genomic polymorphism or structural variation.

[0036] Removal of duplicate reads: During the PCR amplification process, due to the characteristics of the amplification reaction, duplicate DNA fragments, i.e., duplicate reads, may be generated. The presence of these duplicate reads will cause many problems. They may interfere with the subsequent accurate analysis of embryonic and maternal DNA, thereby affecting the accurate assessment of the proportion of maternal contamination and the judgment of embryonic chromosome ploidy, and ultimately leading to the appearance of false positive results. In order to effectively remove these duplicate reads, researchers will use specific algorithms. These algorithms usually identify duplicate reads based on the analysis of sequence features, alignment positions and other information of the reads. For example, some algorithms compare the similarity between reads. If two or more reads are almost identical in sequence and the positions aligned to the genome are also the same or very close, then they are likely to be duplicate reads. The algorithm will mark these duplicate reads and remove them in subsequent data processing. Specifically, the algorithm is an alignment-based algorithm - Picard MarkDuplicates. This tool aligns the sequencing reads with the reference genome, identifies the reads with exactly the same alignment positions, and marks them as duplicate reads. It considers the sequence content of the reads and the alignment position on the genome to determine the duplication. First, the sequencing data is aligned with the reference genome to generate an alignment result file (such as BAM format). Then, the PicardMarkDuplicates tool reads the alignment result file, determines whether the read is a duplicate read according to the set parameters (such as minimum alignment quality, etc.), and marks the duplicate read in the file. By removing duplicate reads, redundant information in the data can be reduced, and the quality and reliability of the data can be improved. In this way, when the data is processed based on the target maternal contamination early warning model, the source of the read can be more accurately determined, and the proportion of maternal contamination can be more accurately calculated, so as to more reliably generate the target chromosome ploidy, which provides a more accurate basis for embryo selection and genetic disease screening in assisted reproductive technology, helps to improve the success rate of embryo transplantation, and reduce the risk of genetic diseases.

[0037] Calculation of DNA methylation level: For methylation sequencing data, it is necessary to calculate the methylation level of each CpG site, that is, the ratio of the number of methylated C to the total number of C (methylated and unmethylated) at the site.

[0038] Data filtering: Filter the data based on criteria such as the number of reads, quality scores, alignment positions, etc. to ensure that the analyzed dataset has sufficient quality and coverage.

[0039] Through these steps, researchers can ensure that they obtain high-quality data and provide accurate input for subsequent deep learning model training and identification and removal of maternal contamination. Sequencing adapters, primers, and low-quality bases are removed from the original bisulfite sequencing reads, and high-quality reads are aligned to the reference genome. After that, training data and deep learning models are constructed for training, with the goal of determining the classification probability of each read in the data belonging to the embryo or the mother.

[0040] In another embodiment, raw data is collected from a single-cell whole-genome methylation sequencing experiment, which is usually derived from a DNA sample processed using bisulfite sequencing technology. The collected raw bisulfite sequencing information is integrated into a database, which contains sequencing data of multiple samples, and the data of each sample includes a large amount of read information.

[0041] The raw data in the sequencing database is processed for feature extraction, which may include identifying the methylation status of CpG sites, calculating methylation levels, identifying methylation patterns, etc., and further screening the target feature data from the extracted features. These data include methylation read information, and the methylation levels of these reads are higher than a preset threshold, which is used to distinguish between high-methylation and low-methylation regions, and is usually set based on biological significance or preliminary data analysis results.

[0042] The raw bisulfite sequencing information of each sample is processed to generate target bisulfite sequencing information, which may include removing low-quality reads, aligning to the reference genome, identifying and quantifying methylation sites, etc. Determine the target bisulfite sequencing information, which is the single-cell whole-genome methylation sequencing data that has been screened and represents the highly methylated regions, and these data will be used for subsequent machine learning model training.

[0043] In another embodiment, whole genome DNA methylation sequencing of embryo culture fluid, specifically, single cell whole genome methylation sequencing is used to detect DNA methyl groups in blastocyst culture fluid. The culture fluid is supplemented to a volume of 20 μL with nuclease-free water and lysed with a corresponding volume of lysis buffer (20 mM Tris-EDTA, 20 mM KCl, 0.3% TritonX-100 and 1 mg / mL proteinase K lysis). Place at 50°C for 1.5 hours, and then use the EZ-96 DNAMethylation-Direct MagPrep kit for bisulfite treatment. After purification, the first DNA is synthesized using random primers P5-N9 (5'-CTACACGACGCTCTTCCGATCTNNNNNNNNN-3') (SEQ ID NO.1) with Klenow polymerase, and this step is repeated four times.

[0044] The second DNA was synthesized using the P7-N9 primer (5'-AGACGTGTGCTCTTCCGATCTNNNNNNNNN-3') (SEQ ID NO. 2). PCR amplification was performed using the index primer and Illumina universal PCR primer to obtain a library for sequencing, which was sequenced on a machine, with 5G data for each sample.

[0045] S103, obtaining an initial maternal source pollution warning model that matches the target characteristic data.

[0046] In one embodiment, the initial maternal pollution warning model is composed of the following modules. The structure of the maternal pollution warning model is as follows: Overall architecture: A deep neural network architecture is adopted, such as a structure combining a convolutional neural network (CNN) and a fully connected layer. The network as a whole includes an input layer, multiple hidden layers, and an output layer. Each layer is connected by weights for transmitting and processing data.

[0047] Input layer: The input data is the read information in the preprocessed single-cell whole-genome methylation sequencing data, and its format can be a vector that encodes the methylation status of each read. For example, if the methylated site is represented by 1 and the unmethylated site is represented by 0, the sequence of a read is encoded to form a fixed-length vector as a sample of the input layer.

[0048] Hidden layer - convolution layer (CNN part): contains multiple convolution layers, each with a set of convolution kernels. For example, the first convolution layer sets 32 convolution kernels, and the convolution kernel size is 3x3 (here 3x3 represents the sliding window size on the methylation state encoding vector). The function of the convolution layer is to automatically extract local features in the input data, and perform convolution operations by sliding the convolution kernel on the input data to generate a feature map. For example, the convolution kernel slides on the methylation state encoding vector to detect the combination of methylation sites in a specific pattern, and generates a feature map reflecting the frequency and location of these patterns. Each convolution layer can be followed by an activation function layer, such as the ReLU function layer, which is used to introduce nonlinear factors and enhance the expressiveness of the model. The ReLU function sets the negative values ​​output by the convolution layer to 0, retaining only the positive value part, so that the model can learn more complex feature relationships.

[0049] Hidden layer - pooling layer: A pooling layer can be set after some convolutional layers, such as a maximum pooling layer with a pooling window size of 2x2. The role of the pooling layer is to downsample the feature map, reduce the amount of data, reduce the computational complexity, and retain the main feature information. For example, the maximum pooling layer will select the maximum value in a 2x2 window as the output, which can reduce the size of the feature map while highlighting the significant information in the feature.

[0050] Hidden layer - fully connected layer: After the convolution layer and pooling layer (if any), connect one or more fully connected layers. For example, set two fully connected layers, the first fully connected layer has 128 neurons, and the second fully connected layer has 64 neurons. The fully connected layer integrates and maps the features extracted by the previous convolution layer and pooling layer. Each neuron is fully connected to all neurons in the previous layer, and information is transferred and transformed through the weight matrix. The fully connected layer can learn the global relationship between different features, further abstract and combine the features, so as to finally classify and judge the source of the read segment.

[0051] Output layer: The output layer contains two neurons, one for each of the two classification results: maternal and embryonic. The softmax function is used as the activation function to convert the output values ​​of the fully connected layer into probability values, indicating the probability that the read belongs to the maternal or embryonic source. For example, the softmax function normalizes the output values ​​of the two neurons to between 0 and 1, and the sum of the two probability values ​​is 1. The model determines the source classification of the read based on these probability values.

[0052] When constructing and training this maternal pollution warning model, a large number of training samples with known read source labels (maternal or embryonic) are used to adjust the weight parameters of each layer of the model so that the model can accurately predict the source classification based on the input methylation sequencing read information, thereby realizing the early warning of maternal pollution. At the same time, a variety of optimization algorithms (such as the Adam optimization algorithm) and regularization techniques (such as L2 regularization) can be used in the training process to prevent overfitting and improve the generalization ability of the model. In addition, by evaluating the model performance on the validation set, the model structure and parameters are adjusted to obtain the initial maternal pollution warning model that best matches the target feature data and has the best performance. .

[0053] S104, training the initial maternal source pollution warning model based on the training data set with target feature data to generate a target maternal source pollution warning model.

[0054] In one embodiment, a training data set with target feature data is processed, which includes selecting two types of reads from maternal and embryonic sources in different samples to generate a training set and a validation set, and randomly dividing the training data set with target feature data into a training set and a validation set. The ratio is usually 70:30 or 80:20 to ensure that the model is tested on independent data. Multiple data groups are extracted from the training set, each of which contains a preset number of data samples, wherein at least one data sample includes identification information. Multiple data groups are randomly selected from the training set, each of which contains a preset number of data samples. This can increase the diversity and robustness of model training. Each data group contains at least one data sample with identification information, which may include category labels or other key metadata of the sample.

[0055] The initial maternal pollution warning model is trained based on data samples in multiple data sets to generate a trained maternal pollution warning model, and the extracted data sets are used to train the initial maternal pollution warning model. This process involves adjusting model parameters, learning feature representations from the data, and ultimately giving the probability that each read in the sample is from the maternal source. A variety of machine learning algorithms may be used during the training process. This application does not limit this, and the applicant can choose according to actual needs. The trained maternal pollution warning model is processed based on the validation set to generate a verification result. The trained maternal pollution warning model is applied to the validation set to generate a verification result. The verification result includes the model's predictions for the reads in the validation set samples. These predictions will be compared with the actual labels to evaluate the accuracy and reliability of the model. The maternal pollution proportion is then predicted based on the probability distribution of the reads in the sample. The prediction formula for the maternal proportion is: ; Among them, r is the proportion of maternal source pollution, represents the probability that the i-th read segment comes from the embryo, and N is the total number of read segments.

[0056] If the data sample containing identification information in the verification result has a maternal pollution ratio higher than the preset threshold, the trained maternal pollution warning model is used as the target maternal pollution warning model, and the data sample containing identification information in the verification result is checked to determine whether its predicted maternal pollution ratio is higher than the preset threshold. The preset threshold is set based on business needs or previous research and is used to determine when the warning model should trigger an alarm. If the verification result shows that the maternal pollution ratio of the data sample with identification information is higher than the preset threshold, the trained maternal pollution warning model is confirmed as the final target model, which will be used in practical applications to predict and warn maternal pollution. The confirmed model is deployed to the production environment for prediction on actual data, and the performance of the model is continuously monitored to ensure its prediction accuracy and timeliness. The model is retrained and optimized as needed. The scientificity and practicality of the maternal pollution warning model are ensured, which helps to improve the accuracy and efficiency of related medical tests.

[0057] S105, processing the target user data based on the target maternal contamination warning model to generate a target chromosome ploidy.

[0058] In one embodiment, the target user data is processed based on the target maternal contamination warning model to generate the target user's bisulfite sequencing information, and the target maternal contamination warning model is used to process the target user's raw data to extract or generate bisulfite sequencing information. This information includes the methylation status of DNA, which is the basis for subsequent analysis.

[0059] Specifically, the target user samples refer to samples collected from embryo culture medium, which contain free DNA (cfDNA) released by the embryo and possible maternal DNA contamination. DNA is extracted from the culture medium using a specific biochemical method. This step needs to ensure the quality and quantity of the DNA for subsequent bisulfite sequencing. For details, please refer to the above-mentioned public content, which will not be repeated here. The extracted DNA is treated with bisulfite, a chemical that can convert unmethylated cytosine (C) into uracil (U), while methylated cytosine is not affected. This step is the key to distinguishing between DNA methylation and unmethylation states. The bisulfite-treated DNA is PCR amplified using specific primers to obtain sufficient DNA for sequencing, and the amplified DNA fragments are connected to the adapter compatible with the sequencing platform to construct a sequencing library. Quality control is performed before sequencing to ensure that the quality of the library meets the sequencing requirements. The library is sequenced using high-throughput sequencing technology (such as the Illumina platform) to generate a large number of short reads, and the raw data generated by sequencing is quality controlled, including removing adapters, primers, low-quality bases, and aligning to the reference genome. Analyze the processed data to determine the methylation status of each CpG site. This typically involves aligning the sequencing reads to a reference genome and inferring the methylation status based on the results of bisulfite treatment.

[0060] Based on the target maternal contamination warning model, the target user's bisulfite sequencing information is processed to generate maternal contamination ratio information. Using the trained model, each read segment is analyzed to estimate its probability of coming from the maternal source. By counting the probability of all read segments being judged as maternal, the proportion of maternal contamination in the entire sample is calculated. According to the calculated maternal contamination ratio, the degree of contamination of maternal DNA in the sample is evaluated. A high contamination ratio may affect the accuracy of embryo genetic testing. The maternal contamination ratio is a key indicator for determining whether the sample needs to be further processed to remove maternal contamination. If the contamination ratio is too high, additional bioinformatics methods or experimental techniques may be required to remove or correct the contamination.

[0061] The maternal contamination ratio information is processed to generate sequencing depth, and the required sequencing depth is calculated based on the maternal contamination ratio information. Sequencing depth refers to the coverage of the target DNA region sequenced, which is usually inversely proportional to the contamination ratio. The higher the contamination, the greater the sequencing depth required to obtain more accurate data. In the case of a high contamination ratio, deeper sequencing may be required to obtain sufficient fetal DNA signals.

[0062] The target user's bisulfite sequencing information is processed based on the maternal contamination ratio information and sequencing depth to generate target embryo reads. Specifically, the processing of bisulfite sequencing information includes quality control, such as removing low-quality reads, removing adapters and primer sequences, and aligning to the reference genome. Bisulfite sequencing data is used to determine the methylation state of DNA, which is particularly important for distinguishing between maternal and fetal DNA because they may differ in methylation patterns. Based on the maternal contamination ratio and sequencing depth, select those reads that are most likely to be from the embryo, which may involve removing those reads that are highly suspected of being from the mother. By using a machine learning model to estimate the probability that each read is from the embryo, the most reliable reads are selected based on these probabilities to classify and identify those reads that are most likely to be from the embryo.

[0063] Based on the target maternal contamination warning model, the target embryo reads are processed to generate the target copy number. The target maternal contamination warning model is used to screen high-confidence reads from the sequencing data of the target embryo. The methylation patterns of these reads are consistent with the methylation patterns of the embryos, and the maternal contamination is low. The reads selected by the deep learning algorithm are analyzed for chromosome copy number variation (CNV).

[0064] For each specific gene or chromosome region, the number of times it is repeated in the genome is calculated. This is usually achieved by comparing sequencing coverage to normal diploid levels, and the copy number is normalized to reduce the impact of sequencing depth and experimental variation. This can be achieved by calculating the ratio of the copy number of each region to the copy number of adjacent regions. Based on the calculated copy number, the ploidy state of the embryonic chromosomes is assessed. For example, the diploid (normal) state means that the copy number of each chromosome region is 2, while the triploid state means that the copy number of certain chromosome regions is 3, identifying chromosome regions with abnormal copy numbers, which may be associated with chromosomal aneuploidy, such as Down syndrome (trisomy 21) or other genetic diseases.

[0065] refer to Figure 5 As shown, the target copy number is processed to generate the target chromosome ploidy, and the target maternal contamination warning model is used to analyze the target embryo reads to generate the target copy number. The copy number refers to the number of times a specific gene or chromosome region is repeated in the genome, which can reflect the ploidy state of the chromosome. The target copy number is analyzed to determine the ploidy state of the target chromosome. The chromosome ploidy state refers to whether the number of chromosomes in the cell is a normal multiple (such as diploid, triploid, etc.).

[0066] Use the filtered reads for further genetic analysis, such as copy number variation (CNV) analysis or single nucleotide polymorphism (SNP) analysis. These data can be used for embryo selection in the assisted reproductive process. Through step-by-step and refined processing, useful biological information can be extracted from the original sequencing data. The ultimate goal is to evaluate the chromosomal ploidy status of the embryo, which is very important for embryo selection and genetic disease screening in assisted reproductive technology. Helping to select the most genetically healthy embryos for transplantation can minimize the impact of maternal contamination and improve the accuracy of the assessment of the genetic status of the embryo, thereby improving the success rate and safety in assisted reproductive technology, helping to improve the success rate of embryo transplantation and reduce the risk of genetic diseases.

[0067] In another embodiment, the method includes a calculation formula for calculating the maternal source pollution ratio information, and the calculation formula is:

[0068]

[0069] Among them, r is the proportion of parent source pollution, represents the probability that the i-th read segment comes from the embryo, and N is the total number of read segments. The proportion of maternal contamination r is calculated by maximizing a product function of the probability that the read segment comes from the embryo. Each read segment i has an associated probability , which represents the probability that it comes from the embryo. By maximizing this product function, the optimal r value can be found, which is the value that is most likely to reflect the actual maternal contamination ratio. An accurate maternal contamination ratio is essential to ensure the accuracy and reliability of sequencing data, especially when performing embryonic chromosomal abnormality detection, thereby helping researchers evaluate and control the impact of maternal DNA contamination on sequencing results.

[0070] Suppose we have sequencing data of an embryo culture fluid sample, with a total number of reads. Here are the hypothetical values ​​of the probability that each read comes from an embryo:

[0071]

[0072] First, we substitute different values ​​of r (assuming r ranges from 0 to 1 in steps of 0.1) into the formula Calculation is performed in .

[0073] When r=0, calculate .

[0074] For read segment 1: ;

[0075] For read segment 2: ;

[0076] Similarly, calculate the results of all read segments and multiply them to get a product value (The calculation process is omitted here, and the actual calculation can be done by multiplying them one by one).

[0077] When r=0.1, calculate .

[0078] For example, for read segment 1: ;

[0079] Similarly, all read segments are calculated and multiplied to obtain .

[0080] Continue to calculate the product value when r=0.2, 0.3, ..., 1 .

[0081] Find argmax, compare The size of these values, assuming that after calculation, is the largest value among them, then according to the formula, r=0.4 is the proportion of parent source pollution we are looking for.

[0082] In this example, the maternal contamination ratio means that in this embryo culture fluid sample, approximately of the DNA may come from maternal contamination, while it may come from the embryo itself. In the detection of embryonic chromosomal abnormalities, an accurate maternal contamination ratio can help researchers judge the reliability of the test results. For example, if the maternal contamination ratio is too high, further measures (such as increasing sequencing depth, optimizing data processing methods, etc.) may be needed to reduce the impact of maternal contamination on the true chromosomal status of the detected embryo, thereby more accurately assessing whether the embryo has chromosomal abnormalities and providing a more reliable basis for embryo selection in assisted reproductive technology.

[0083] In another embodiment, the target user data is processed based on the target maternal pollution warning model to generate the target chromosome ploidy, further comprising:

[0084] The method includes a calculation formula for calculating sequencing depth, which is:

[0085]

[0086] Among them, r is the proportion of maternal contamination, and Counts represents the required sequencing depth, that is, the total number of reads required to achieve sufficient sequencing coverage.

[0087] The CNV detection results will be affected by the sequencing depth. Too few reads will make the CNV detection results too noisy and reduce the effect of removing pollution. Therefore, it is necessary to specify an appropriate sequencing depth to apply the algorithm. For a sample, assuming that its pollution ratio is r, the sequencing depth recommended by the present invention is: Therefore, when it happens to be severely contaminated (60%), the required sequencing depth is about 3 million reads. Therefore, the present invention recommends that the sequencing depth of all samples be 3 million reads at the beginning, and then the contamination ratio is estimated through the deep learning grid. If it is less than or equal to 60%, you can choose to use the algorithm to screen the reads and then do CNV detection or directly do CNV detection, and compare the test results. For severe contamination (greater than or equal to 60%), an algorithm must be used to remove contamination. The present invention recommends deepening the sequencing after estimating the sequencing depth required for the current contamination according to the calculation formula, and then applying the algorithm to remove contamination for CNV detection.

[0088] Sequencing depth is a key factor when performing copy number variation (CNV) detection, which directly affects the accuracy and reliability of the test results. Sequencing depth refers to the average number of times the target DNA sequence is sequenced. The following is a detailed description of how to estimate the sequencing depth required for the current contamination based on the calculation formula and adjust the sequencing depth accordingly: A high contamination ratio means that more reads come from the mother rather than the embryo, which will interfere with the accurate assessment of the genetic status of the embryo. Therefore, when the contamination ratio is high, a deeper sequencing depth is required to obtain sufficient fetal DNA signals to improve the signal-to-noise ratio. According to the recommendations of this invention, the initial sequencing depth of all samples is set to 3 million reads. This is an empirical value designed to ensure that sufficient data can be obtained for analysis even in the case of high contamination. By applying a deep learning model, the actual contamination ratio r of each sample is estimated. This model may predict the contamination ratio based on methylation patterns, read distribution, or other bioinformatics features. If the initial sequencing depth is not enough to achieve the required signal-to-noise ratio, the sequencing depth is increased according to the calculation results. This may involve additional sequencing work to ensure more accurate CNV detection results. After increasing the sequencing depth, CNV analysis is repeated, the algorithm is used to remove contamination, and the genetic status of the embryo is evaluated. The CNV detection results before and after adjusting the sequencing depth were compared to verify whether increasing the sequencing depth effectively improved the accuracy and reliability of the results. Based on the feedback from the results, the sequencing depth calculation formula and CNV detection algorithm were continuously optimized to adapt to different samples and contamination conditions.

[0089] This approach ensures that even in the case of high contamination, accurate CNV detection results can be obtained through appropriate sequencing depth and data processing algorithms, which is critical for the genetic health assessment of embryos and the success of assisted reproductive technology.

[0090] In another embodiment, the target user data is processed based on the target maternal pollution warning model to generate the target chromosome ploidy, further comprising:

[0091] The method includes a calculation formula for calculating embryonic copy number, which is:

[0092]

[0093] in, Represents the normalized copy number, which is used to reduce the fluctuation and noise caused by sequencing depth. Represents the average sequencing coverage within a specific genomic region (bin), Represents the raw copy number of a specific genomic region (bin).

[0094] Assume that within the specific genomic region (bin) we are studying, the average sequencing coverage is Original copy number .

[0095] First calculate the numerator: .

[0096] Then calculate the denominator: .

[0097] Then calculate the fractional part: .

[0098] Finally, calculate the normalized embryo copy number :

[0099] .

[0100] In embryonic chromosome ploidy detection, sequencing depth may fluctuate in different regions. This fluctuation may lead to inaccurate copy number calculation and produce false positive or false negative results. This normalization formula can reduce the fluctuation and noise effects caused by sequencing depth. For example, when evaluating whether an embryo has abnormal chromosome copy number (such as trisomy or monosomy), a more accurate copy number calculation can help to more accurately determine the chromosome ploidy status of the embryo. Without normalization, the copy number in certain regions may be misjudged due to uneven sequencing depth, which in turn affects the assessment of the chromosome health of the entire embryo. The normalized copy number can provide more stable and reliable results, helping researchers to better select genetically healthy embryos for transplantation in assisted reproductive technology, improve the success rate of embryo transplantation, and reduce the risk of genetic diseases.

[0101] By dividing the original copy number Average sequencing coverage Compare and divide by the normalization factor , to calculate the normalized copy number , normalization processing helps to correct false positive results of copy number variation caused by uneven sequencing depth, making copy number estimation more accurate and reliable. It is suitable for non-invasive preimplantation genetic testing of embryos, especially when evaluating the chromosomal ploidy status of embryos, and can provide more accurate copy number data. Through this calculation method, researchers can better understand the genetic status of embryos and make more informed decisions in assisted reproductive technology. It is an effective tool for processing sequencing data and evaluating the copy number status of embryonic chromosomes, and is of great significance for improving the accuracy of genetic testing.

[0102] In this application, the server performs single-cell whole-genome methylation library construction and sequencing on cfDNA (free DNA) in SECM (culture medium of in vitro cultured embryos). A series of data preprocessing steps are then performed, including removing sequencing adapters, primers, and low-quality bases from the original sulfite sequencing reads, and aligning high-quality reads to the reference genome. Training data and deep learning models are then constructed for training, with the goal of determining the classification probability of each read in the data belonging to the embryo or the mother. After training, the proportion of maternal contamination in the sample is estimated from the classification probability by constructing a mathematical model based on maximum likelihood estimation. At the same time, according to the classification probability of the reads, reads with a high maternal probability, i.e., maternal contamination, are removed, and the embryo's cfDNA is retained for copy number variation detection (CNVs) to restore the true chromosome ploidy of the embryo and remove false negative results.

[0103] In one embodiment, if Figure 2 As shown, the present application also provides a maternal source pollution removal device based on machine learning, comprising:

[0104] The acquisition module 201 is used to acquire target user data and a training sample set, wherein the target user data includes the culture fluid sample information of the target user's in vitro cultured embryos, and the training sample set includes the culture fluid sample information of other users' in vitro cultured embryos; and acquire an initial maternal source pollution warning model that matches the target feature data;

[0105] Processing module 202 is used to preprocess the training sample set to generate a training data set with target feature data, wherein the target feature data is methylation sequencing data of the whole genome of a single cell, and the methylation sequencing data includes read information; based on the training data set with the target feature data, the initial maternal contamination warning model is trained to generate a target maternal contamination warning model; based on the target maternal contamination warning model, the target user data is processed to generate a target chromosome ploidy.

[0106] The present application embodiment provides an electronic device, such as Figure 3As shown, the electronic device 3 includes a first processor 300, a memory 301, a bus 302 and a communication interface 303, and the first processor 300, the communication interface 303 and the memory 301 are connected via the bus 302; the memory 301 stores a computer program that can be run on the first processor 300, and when the first processor 300 runs the computer program, it executes the maternal source pollution removal method based on machine learning provided in any of the aforementioned embodiments of the present application.

[0107] The memory 301 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk memory. The communication connection between the system network element and at least one other network element is realized through at least one communication interface 303 (which may be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. may be used.

[0108] The bus 302 may be an ISA bus, a PCI bus, or an EISA bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The memory 301 is used to store a program, and the first processor 300 executes the program after receiving an execution instruction. The method for removing maternal pollution based on machine learning disclosed in any implementation of the embodiment of the present application may be applied to the first processor 300, or implemented by the first processor 300.

[0109] The first processor 300 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the first processor 300. The above-mentioned first processor 300 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a readily available programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be embodied as a hardware decoding processor to execute, or a combination of hardware and software modules in the decoding processor to execute. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 301, and the first processor 300 reads the information in the memory 301 and completes the steps of the above method in combination with its hardware.

[0110] The electronic device provided in the above-mentioned embodiments of the present application and the method for removing maternal pollution based on machine learning provided in the embodiments of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

[0111] The present application embodiment provides a computer-readable storage medium, such as Figure 4 As shown, the computer-readable storage medium 401 stores a computer program, and when the computer program is read and executed by the second processor 402, the aforementioned method for removing maternal pollution based on machine learning is implemented.

[0112] The technical solution of the embodiment of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling an electronic device (which may be an air conditioner, a refrigeration device, a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the method described in the embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a ROM, a RAM, a magnetic disk, or an optical disk.

[0113] The computer-readable storage medium provided in the above-mentioned embodiments of the present application and the method for removing maternal pollution based on machine learning provided in the embodiments of the present application are based on the same inventive concept and have the same beneficial effects as the method adopted, run or implemented by the application program stored therein.

[0114] An embodiment of the present application provides a computer program product, including a computer program, wherein the computer program is executed by a third processor to implement the method described above.

[0115] The computer program product provided in the above-mentioned embodiments of the present application and the method for removing maternal pollution based on machine learning provided in the embodiments of the present application are based on the same inventive concept and have the same beneficial effects as the method adopted, run or implemented by the application program stored therein.

[0116] It should be noted that, in this application, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or still includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise one..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0117] Each embodiment in the present application is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the evaluation of the maternal source pollution removal method based on machine learning, the electronic device, the electronic device, and the readable storage medium embodiment, since they are basically similar to the above-mentioned embodiment of the maternal source pollution removal method based on machine learning, the description is relatively simple, and the relevant parts can be referred to the partial description of the embodiment of the maternal source pollution removal method based on machine learning.

[0118] Although the present application is disclosed as above, the present application is not limited thereto. Any person skilled in the art can make various changes and modifications without departing from the spirit and scope of the present application. Therefore, the protection scope of the present application shall be subject to the scope defined by the claims.

Claims

1. A method for removing maternal pollution based on machine learning, characterized in that: include: Acquire target user data and a training sample set, wherein the target user data includes culture fluid sample information of in vitro cultured embryos of the target user, and the training sample set includes culture fluid sample information of in vitro cultured embryos of other users; Preprocessing the training sample set to generate a training data set with target feature data, wherein the target feature data is methylation sequencing data of a single cell whole genome, and the methylation sequencing data includes read segment information; Obtaining an initial parent source pollution warning model that matches the target characteristic data; Training the initial maternal source pollution warning model based on the training data set with target feature data to generate a target maternal source pollution warning model; Processing the target user data based on the target maternal pollution warning model to generate a target chromosome ploidy; The initial maternal pollution warning model adopts a structure combining a convolutional neural network and a fully connected layer. The model includes an input layer, multiple hidden layers and an output layer. The layers are connected by weights for transmitting and processing data. The input data of the input layer is the read information in the preprocessed single-cell whole genome methylation sequencing data, and its format is a vector form after encoding the methylation state of each read segment. Hidden layer-convolution layer: contains multiple convolution layers, each convolution layer has a set of convolution kernels, where the first convolution layer is set with 32 convolution kernels, the convolution kernel size is 3x3, and each convolution layer is followed by an activation function layer; Hidden layer-pooling layer: a pooling layer is set after some convolution layers, and the pooling window size is 2x2; Hidden layer - fully connected layer: After the convolutional layer and the pooling layer, two fully connected layers are connected. The first fully connected layer has 128 neurons and the second fully connected layer has 64 neurons. The output layer contains two neurons, corresponding to the two classification results of the read segment coming from the mother and the embryo respectively. The softmax function is used as the activation function to convert the numerical value output by the fully connected layer into a probability value, indicating the probability that the read segment belongs to the mother or the embryo.

2. The method according to claim 1, characterized in that Preprocessing the training sample set to generate a training data set with target feature data includes: Performing feature extraction processing on the training sample set to generate a sequencing database, wherein the sequencing database includes a plurality of original bisulfite sequencing information; The original bisulfite sequencing information is processed respectively to generate target feature data, wherein the target feature data includes target bisulfite sequencing information, and the target bisulfite sequencing information is methylation sequencing data of a single cell whole genome whose methylation read information is greater than a preset threshold.

3. The method according to claim 2, characterized in that The initial maternal source pollution warning model is trained based on the training data set with target feature data to generate a target maternal source pollution warning model, including: Processing the training data set with target feature data to generate a training set and a validation set; Extracting a plurality of data groups from the training set, wherein each data group includes a preset number of data samples, wherein at least one data sample includes identification information; Training the initial maternal pollution warning model based on data samples in multiple data groups to generate a trained maternal pollution warning model; Processing the trained maternal source pollution early warning model based on the verification set to generate a verification result; If the data sample containing identification information in the verification result has a maternal source pollution ratio higher than a preset threshold, the trained maternal source pollution warning model is used as the target maternal source pollution warning model.

4. The method according to claim 1, characterized in that The target user data is processed based on the target maternal pollution warning model to generate a target chromosome ploidy, including: Processing the target user data based on the target maternal contamination early warning model to generate bisulfite sequencing information of the target user; Processing the bisulfite sequencing information of the target user based on the target maternal contamination early warning model to generate maternal contamination ratio information; Processing the maternal contamination ratio information to generate sequencing depth; Processing the bisulfite sequencing information of the target user based on the maternal contamination ratio information and the sequencing depth to generate a target embryo read segment; Processing the target embryo read segment based on the target maternal contamination warning model to generate a target copy number; The target copy number is processed to generate a target chromosome ploidy.

5. The method according to claim 4, characterized in that The target user data is processed based on the target maternal pollution warning model to generate a target chromosome ploidy, further comprising: The method includes a calculation formula for calculating the maternal source pollution ratio information, and the calculation formula is: ； In the equation, r is the proportion of parent source pollution, represents the probability that the i-th read segment comes from the embryo, and N is the total number of read segments.

6. The method according to claim 4, characterized in that The target user data is processed based on the target maternal pollution warning model to generate a target chromosome ploidy, further comprising: The method includes a calculation formula for calculating the sequencing depth, and the calculation formula is: ； Among them, r is the proportion of maternal contamination, and Counts represents the required sequencing depth, that is, the total number of reads required to achieve sufficient sequencing coverage.

7. The method according to claim 4, characterized in that The target user data is processed based on the target maternal pollution warning model to generate a target chromosome ploidy, further comprising: The method includes a calculation formula for calculating the embryo copy number, which is: ； in, Represents the normalized copy number, which is used to reduce the fluctuation and noise caused by sequencing depth. Shows the average sequencing coverage within a specific genomic region (bin). Represents the raw copy number of a specific genomic region (bin).

8. A device for removing parent source pollution based on machine learning, characterized in that: The device comprises: An acquisition module is used to acquire target user data and a training sample set, wherein the target user data includes the culture fluid sample information of the target user's in vitro cultured embryos, and the training sample set includes the culture fluid sample information of other users' in vitro cultured embryos; and acquire an initial maternal source contamination early warning model that matches the target feature data; A processing module is used to pre-process the training sample set to generate a training data set with target feature data, wherein the target feature data is methylation sequencing data of a single cell whole genome, and the methylation sequencing data includes read information; based on the training data set with target feature data, the initial maternal pollution warning model is trained to generate a target maternal pollution warning model; based on the target maternal pollution warning model, the target user data is processed to generate a target chromosome ploidy; The initial maternal pollution warning model adopts a structure combining a convolutional neural network and a fully connected layer. The model includes an input layer, multiple hidden layers and an output layer. The layers are connected by weights for transmitting and processing data. The input data of the input layer is the read information in the preprocessed single-cell whole genome methylation sequencing data, and its format is a vector form after encoding the methylation state of each read segment. Hidden layer-convolution layer: contains multiple convolution layers, each convolution layer has a set of convolution kernels, where the first convolution layer is set with 32 convolution kernels, the convolution kernel size is 3x3, and each convolution layer is followed by an activation function layer; Hidden layer-pooling layer: a pooling layer is set after some convolution layers, and the pooling window size is 2x2; Hidden layer - fully connected layer: After the convolutional layer and the pooling layer, two fully connected layers are connected. The first fully connected layer has 128 neurons and the second fully connected layer has 64 neurons. The output layer contains two neurons, corresponding to the two classification results of the read segment coming from the mother and the embryo respectively. The softmax function is used as the activation function to convert the numerical value output by the fully connected layer into a probability value, indicating the probability that the read segment belongs to the mother or the embryo.

9. An electronic device, characterized in that: include: a first processor; and a memory for storing executable instructions of the first processor; Wherein, the first processor is configured to execute the maternal source pollution removal method based on machine learning as described in any one of claims 1 to 7 by executing the executable instructions.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the second processor, the method for removing parent source pollution based on machine learning described in any one of claims 1 to 7 is implemented.

Images