Endometrial implantation window prediction model, construction and use thereof

Abstract

The present disclosure relates to a prediction method, a prediction model based on transcriptome analysis for predicting the window of implantation (WOI) of the endometrium, and its use in patients with recurrent implantation failure (RIF), in particular in patients with endometrium-derived recurrent implantation failure. The present disclosure also relates to a device or system for said method and said use.

Description

Technical Field

[0001] This invention relates to the fields of biotechnology and medicine, and more specifically to a predictive method and model for predicting the endometrial implantation window (WOI) based on transcriptometrial analysis, and its application in patients with recurrent implantation failure (RIF), particularly in patients with endometrial-derived recurrent implantation failure. The invention also relates to apparatus or systems for the methods and applications described herein. Background Technology

[0002] The "implantation window" (also known as the "WOI") is a brief period after ovulation during which the human endometrium transitions from a proliferative to a secretory state. The endometrium in the WOI is receptive, representing the optimal time for embryo implantation. Therefore, the receptivity of the endometrium and the synchronicity between the embryo and the endometrium are closely related to successful pregnancy. Under normal circumstances, for women with regular menstrual cycles, the WOI typically occurs between days 20 and 24 of the menstrual cycle, but the optimal duration of the WOI is very short, lasting only about 24-48 hours in humans. Studies have shown that in assisted reproductive technology (ART) applications, WOI displacement is a significant cause of unsatisfactory pregnancy outcomes in patients with recurrent implantation failure (RIF). Specifically, the patient's "implantation window" may occur earlier or later than usual, resulting in an unreceptive endometrium when embryos are transferred at the conventional time, leading to implantation failure. Therefore, accurate identification and prediction of the WOI are crucial for maximizing the effectiveness of ART in infertile women who have experienced RIF.

[0003] The rsERT method is an RNA sequencing-based method for predicting endometrial receptivity, using 175 genetic biomarkers to predict endometrial receptivity in patients. It has been demonstrated that rsERT-guided personalized embryo transfer (pET) improves pregnancy outcomes in patients with recurrent embryo transfer (RIF). However, this method is costly and time-consuming because it requires three samplings at 48-hour intervals at three time points within the same menstrual cycle (LH+5, LH+7, and LH+9). Furthermore, this multiple invasive sampling approach increases patient pain sensitivity and the risk of secondary injury. In addition, similar to other proposed endometrial receptivity methods (e.g., ERA, ER Map / ERGrade, and Win-Test), rsERT predicts endometrial receptivity but does not provide more precise information on the WOI window duration. Due to the short WOI window, its duration, in addition to WOI displacement, has been shown to be a key factor in clinical implantation failure. Therefore, it is necessary to make more accurate predictions of the WOI window period in order to facilitate the identification of the optimal timing for pET in patients with narrow window periods and further improve the clinical outcomes of implantation.

[0004] In view of the above, there is still an urgent need in the field to establish an improved method and tool that can accurately and efficiently predict WOI through single-time-point sampling, especially an improved method and tool that can provide WOI information with higher temporal accuracy. Invention Overview

[0005] To meet the aforementioned needs, the inventors have developed an improved method and model for predicting the endometrial implantation window through single-timepoint sampling, and demonstrated its clinical benefits in RIF patients through a prospective controlled trial. Based on this, the inventors propose a simpler, less invasive alternative method to determine the optimal timing for embryo transfer in RIF patients. The method and model of this invention not only allow for precise determination of the implantation window (WOI) based on individual patient characteristics, ensuring synchronicity between the embryo and the patient's WOI and improving the clinical outcome of implanted embryos, but also reduce the number of endometrial samplings, increasing patient compliance and reducing the risk of secondary injury. The method and model of this invention are particularly suitable for guiding personalized embryo transfer (pET) in RIF patients in clinical applications, improving their pET clinical outcomes.

[0006] Therefore, in a first aspect, the present invention provides a method for constructing a prediction model for the endometrial implantation window (WOI). In some embodiments, the method includes:

[0007] (i) Obtain or generate a training dataset for model building, wherein each training data vector in the training dataset represents an endometrial tissue sample from a single sampling time point of a human individual and contains the magnitude of each of a set of differentially expressed genes related to the receptivity window extracted from the corresponding sample, and each training data vector also contains a label related to the WOI time of the corresponding sample.

[0008] The quantitative value of the differentially expressed gene characteristic is the expression level of the differentially expressed gene measured on the sample, and

[0009] The WOI time is the time required from the sampling time of the sample to reach the optimal planting window (WOI) of the individual from which the sample originated in the same period;

[0010] (ii) The training data vector set is received on at least one computer-executable processor;

[0011] (iii) On the at least one processor, using the training data vector set, a classifier is trained to generate a WOI prediction model, wherein the WOI prediction model outputs the predicted WOI time with hourly precision.

[0012] In some implementations, the present invention also provides a WOI prediction model generated by the model building method.

[0013] In a second aspect, the present invention provides a method for classifying test data. In some embodiments, the method is an off-site computer-executed method, the method comprising:

[0014] (a) On at least one computer-executable processor, test data is received, wherein the test data contains the expression levels of a set of differentially expressed genes related to receptivity window determined on endometrial tissue samples from a single sampling time point of a human test individual.

[0015] (b) On the at least one processor, the test data is evaluated using a trained classifier.

[0016] (c) Using at least one processor, based on the evaluation of step (b), output the predicted WOI time for the test individual with hourly accuracy.

[0017] The WOI time is the time required from the sampling time of the sample to reach the optimal planting window (WOI) of the individual from which the sample originated in the same period.

[0018] In some preferred embodiments, the trained classifier used to classify the test data is a WOI prediction model generated using the WOI prediction model construction method according to the present invention.

[0019] In a third aspect, the present invention provides methods for predicting the endometrial implantation window (WOI) of an individual and for improving embryo transfer in that individual. In some embodiments, the method includes: predicting the endometrial implantation window time of the individual using the WOI prediction model of the present invention or by means of the test data classification method of the present invention; and optionally, providing a recommendation for embryo transfer timing based on the predicted implantation window time.

[0020] In a fourth aspect, the present invention provides an apparatus or system for predicting an individual's endometrial implantation window and its use. In some embodiments, the apparatus or system includes:

[0021] -Optionally, a module for acquiring differentially expressed gene features related to the receptivity window, wherein the module is capable of extracting the expression level of differentially expressed genes related to the receptivity window from endometrial tissue samples at a single sampling time point of the individual;

[0022] - An endometrial implantation window prediction module, wherein the module comprises a WOI prediction model constructed according to the method of the present invention, or wherein the module is capable of performing a test data classification method according to the present invention to output the predicted endometrial implantation window time of the individual and optionally output the predicted optimal embryo transfer time interval.

[0023] In a fifth aspect, the present invention provides the use of a set of differentially expressed genes related to the receptivity window and their detection reagents in the preparation of a kit for predicting the endometrial implantation window (WOI) time of an individual.

[0024] In some preferred embodiments of any of the foregoing aspects, the WOI time is a number of hours in the range of -96 hours to 96 hours, for example, selected from 0h, 24h, -24h, 48h, -48h, 72h, -72h, 96h and -96h, or a value between any two hour values.

[0025] In some preferred embodiments of any of the foregoing aspects, the receptivity window-related differentially expressed genome comprises Figure 4 At least 100, 150, 170, or preferably 201 differentially expressed genes related to the receptivity window are listed. Brief description of the attached diagram

[0026] Figure 1 The figure shows the correlation between the predicted WOI time distribution and endometrial receptivity status. In the figure, "predicted WOI time" refers to the WOI time predicted using the modified rsERT model according to the present invention (i.e., the time required for the sampling time of the endometrial sample to reach the optimal implantation window (WOI) of the individual from which the sample originates in the same cycle); the pre-receptivity, receptivity period, and post-receptivity periods indicated in the figure represent the receptivity status of the endometrial sample as determined based on pregnancy outcome. Using -18 and 18 hours as cutoff values, samples with predicted WOI times greater than 18 hours and less than -18 hours are defined as pre-receptivity and post-receptivity states, respectively. As shown, the receptivity status determined by predicted WOI time is in high agreement with the receptivity status determined by pregnancy outcome, with only a small deviation.

[0027] Figure 2 The figure shows the statistical distribution of the accurate implantation hour time predicted using the modified rsERT model on endometrial samples from nearly 600 patients.

[0028] Figure 3 The results show the clinical pregnancy rate statistics in patients outside the receptive period and patients within the receptive period.

[0029] Figure 4 The study identified 201 differentially expressed genes related to the acceptance window, which were used as biomarkers for WOI prediction after feature selection.

[0030] Figure 5 Examples of natural and artificial cycle protocols for routine embryo transfer are shown. Invention Details

[0031] Before describing the invention in detail, it should be understood that the invention is not limited to the specific methods and experimental conditions described herein, as these methods and conditions can be modified. Furthermore, the terminology used herein is for illustrative purposes only and is not intended to be restrictive.

[0032] definition

[0033] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art. For the purposes of this invention, the following terms are defined below.

[0034] The term “about” when used in conjunction with a numeric value means to cover a range of numeric values ​​that have a lower limit of 5% less than the specified numeric value and an upper limit of 5% greater than the specified numeric value.

[0035] When the term “and / or” is used to connect two or more options, it should be understood to mean any one of the options or any two or more of the options.

[0036] As used herein, the terms “comprising” or “including” mean to include the stated elements, integers, or steps, but do not exclude any other elements, integers, or steps. In this document, when the terms “comprising” or “including” are used, unless otherwise specified, they also cover situations consisting of the stated elements, integers, or steps.

[0037] As used herein, "feature" refers to any characteristic that can serve as an indicator of an individual's endometrial receptivity, including, for example, objectively measurable biological characteristics. For the purposes of this invention, the biological features include at least a set (e.g., at least 100) of differentially expressed genes identified in endometrial tissue during the pre-receptivity, post-receptivity, and receptivity periods; i.e., receptivity window-related differentially expressed genes. The "quantity" or "feature value" of a biological feature refers to information associated with the biological feature that can be used to characterize the endometrial receptivity of the individual corresponding to the sample. For example, such information could be the expression level of each of the receptivity window-related differentially expressed genes in the endometrial sample, e.g., expression levels determined based on transcriptome sequencing. In this document, each feature can be represented as a dimension of a vector space, where each vector in the vector space is a multi-dimensional vector comprising multiple feature quantities associated with a particular object. Accordingly, the dimension of the vector space corresponds to the size of the feature set. Machine learning algorithms can be trained to find patterns composed of the quantities of multiple features in the feature set to generate a classifier model for predictive classification of test data.

[0038] In this document, "classifier" refers to a machine learning algorithm. For example, in some embodiments of the invention, the classifier used for the predictive model of the invention can be a supervised learning regression algorithm. Supervised learning is a machine learning method that learns and predicts output values ​​from labeled training data. A regression algorithm is a machine learning algorithm used to predict continuous numerical outputs. Supervised learning regression algorithms that can be used in the present invention include, but are not limited to: logistic regression, resilient network regression, neural networks, Bayesian neural networks, linear regression, multinomial regression, support vector regression (SVR), decision tree regression, random forest regression, and ensemble methods (such as gradient boosting trees). These algorithms use labeled training data to learn a model and are used to predict continuous numerical outputs.

[0039] In this paper, "WOI time," used as a label in the training data and as an output feature of the WOI prediction model, refers to the time required from the sampling time of the endometrial sample to reach the optimal implantation window (WOI) of the individual from which the sample originated within the same cycle. In some preferred embodiments of the invention, the optimal implantation window (WOI) of the individual from which the sample originated corresponds to (or is predicted by the model to correspond to) the time during the menstrual cycle when the endometrium is in optimal receptivity, based on the individual's clinical pregnancy outcome after embryo transfer. Therefore, in such embodiments, as those skilled in the art will understand, the optimal implantation window of the individual can be estimated from the embryo transfer time based on the successful clinical pregnancy outcome of the individual undergoing embryo transfer, and the interval between the optimal implantation window and the sampling time of the endometrial sample, i.e., the WOI time corresponding to the sample, can be calculated; conversely, the optimal implantation window (WOI) of the individual from which the sample originated can be calculated based on the (predicted) WOI time corresponding to the sample and the sampling time of the sample, thereby estimating the optimal embryo transfer time interval for the individual.

[0040] Those skilled in the art will understand that, given that the endometrial receptivity period is a window, not a single point in time, the time interval within this window that is close to the optimal receptivity point will have a higher probability of successful embryo implantation. Therefore, as those skilled in the art will understand, in this disclosure, the expressions "optimal implantation window" or "optimal WOI time" related to the individuals from which the sample was sourced refer to, for example, a time period within a cycle that can achieve a relatively high embryo implantation success rate based on model predictions or inferences based on embryonic pregnancy outcomes, and do not necessarily represent the optimal receptivity time point.

[0041] In this article, the terms "cycle," "implantation cycle," and "menstrual cycle" are used interchangeably, including both natural and artificial cycles (also known as hormone replacement therapy (HRT) cycles). In individuals where the implantation window has not shifted, it typically occurs between days 19 and 24 of a normal menstrual cycle, or on day 7 after the luteinizing hormone peak in a natural cycle (LH+7), or on day 5 of progesterone addition in a hormone replacement cycle (P+5). Figure 5 The document shows examples of endometrial preparation protocols for natural and artificial cycles used in routine embryo transfer, based on the timing of the typical implantation window.

[0042] In this article, "WOI shift" refers to a patient's implantation window time (WOI) deviating from the normal implantation window time (i.e., for blastocysts, LH+7 / P+5; or for day 3 cleavage embryos, LH+5 / P+3).

[0043] In this document, the term "training sample set" refers to a set of training samples comprising multiple endometrial samples with known receptivity states. In some embodiments of the invention, the receptivity state is characterized by the WOI time corresponding to the samples as described above. In this invention, the training sample set is used to develop a predictive model to analyze test samples. In some embodiments of the invention, the training sample set comprises samples from at least 100 or more. Features extracted from the training samples are included herein in training data vectors, and a plurality of said training data vectors from a plurality of training samples can constitute a dataset for training a machine learning model. In some embodiments of the invention, the training sample set comprises at least a portion of endometrial tissue samples collected during the pre-receptivity, post-receptivity, and post-receptivity periods. For example, at least a portion of the samples in the training sample set can be collected from a plurality of human individuals undergoing embryo transfer in such a manner that at least three endometrial tissue samples, corresponding to the pre-receptivity, receptivity, and post-receptivity periods respectively, are collected from each of the plurality of human individuals. Preferably, the individual is an individual who has achieved a successful clinical pregnancy after embryo transfer, and preferably, the receptivity state corresponding to the endometrial tissue sample, i.e., pre-receptivity, receptivity period, and post-receptivity period, is determined by the time interval between the embryo transfer time and the endometrial sampling time when the pregnancy outcome is obtained.

[0044] In this paper, the term "training dataset" refers to the collection of data used to train a machine learning model. In some cases, the training dataset may contain training data vectors extracted from all or part of the training samples in the training sample set. In other cases, when cross-validation is used to build a model, the training dataset may be divided into a training set and a validation set to utilize the validation set to evaluate the model's performance and generalization ability, thereby selecting the optimal model parameters or performing model selection.

[0045] In this document, embryo transfer has the meaning commonly known to those skilled in the art, including both fresh and frozen embryo transfer. In some embodiments, the embryo transfer is preferably a frozen embryo transfer.

[0046] In this document, the term "detection tool" refers broadly to any tool (e.g., RNA sequencing devices or reagents, gene expression microarrays, and oligonucleotide chips) that can extract the expression levels of differentially expressed genes related to the receptivity window from an individual's endometrial tissue sample. These tools can be used to form the apparatus or system of this invention, or for use in the methods of this invention (including model building, prediction, and classification of test data methods).

[0047] In this article, the terms “clinical pregnancy” and “intrauterine pregnancy” are used interchangeably, referring to the presence of a gestational sac in the uterine cavity as determined by ultrasound assessment 4-5 weeks after embryo transfer (ET).

[0048] In this article, the term "clinical pregnancy rate" refers to the proportion of successful clinical pregnancies after in-vitro fertilization (IVF). It is one of the important indicators for measuring the success rate of assisted reproductive technologies. The clinical pregnancy rate is usually expressed as a percentage, calculated as the percentage of successful pregnancies out of the total number of cycles. The number of successful pregnancies refers to the number of embryo transfer cycles that resulted in successful pregnancies within a certain period; the total number of cycles refers to the total number of all embryo transfer cycles performed within the same period. Therefore, the IPR can also be calculated as the number of patients who achieved successful pregnancies (visible gestational sac on ultrasound) within a given period divided by the number of patients who underwent embryo transfer.

[0049] As used herein, the term "module" refers to a reagent, reagent kit, component, assembly, and / or device or system that can be used to achieve the function of the module. It should be understood that the composition of a module is not constrained by a particular physical form, as long as it can achieve the desired function. Depending on the intended function, a module can be a combination of reagents and / or devices to achieve the function; or it can be a software object or routine (e.g., as a separate thread) that executes centrally on a single computing system (e.g., a computer program, a tablet computer (PAD), one or more processors). For example, the module for acquiring differentially expressed gene features related to the acceptance window according to the present invention can consist of reagents or reagent kits (e.g., impurity probes) for performing expression level analysis of the gene and optionally devices (e.g., microarrays) that can be used to perform the expression level analysis. As another example, the module for predicting the endometrial implantation window according to the present invention can be a program stored on a computer-readable medium containing computer program logic or code portions for performing the implantation window prediction.

[0050] The following describes the steps and modules / components involved in the method and product of the present invention. It should be understood that the features described in these steps and modules / components may exist in the method and product of the present invention in any combination unless explicitly inappropriate, as if such combinations of features were individually specified.

[0051] I. The method of the present invention

[0052] Compared to previous methods for predicting endometrial receptivity based on multiple sampling, the inventors have proposed a method for predicting the implantation window (WOI) of an individual with hourly precision based on a single sampling point. This method extracts the expression levels of a series of differentially expressed genes related to the receptivity window from endometrial tissue obtained at a single sampling time as input parameters. Combined with a machine learning system, it characterizes the individual's endometrial receptivity with hourly precision, thereby generating the predicted optimal implantation window time for the individual and improving the efficiency of embryo implantation within the individual's endometrial receptivity window. Prospective clinical controlled trials have demonstrated that the method of this invention can accurately predict the implantation window from -96 hours to +96 hours away from the sampling point without requiring secondary or multiple resampling; and it can significantly improve clinical outcomes after embryo implantation in patients with recurrent implantation.

[0053] Therefore, on the one hand, the present invention provides a classification or prediction system and method that can use a classifier, through computer-aided execution, to evaluate the parametric characteristics of differentially expressed genes related to multiple receptivity windows in a subject (i.e., endometrial tissue samples of the individual to be tested), in order to accurately characterize / predict the implantation window time of said individual.

[0054] In another aspect, at least in part based on the classification system and method of the present invention, the inventors provide a method for predicting the endometrial implantation window of a subject (i.e., an endometrial tissue sample of the individual to be tested) and / or improving the clinical outcome of embryo transfer in the individual, the method comprising: obtaining parametric features of a plurality of differentially expressed genes related to receptivity windows from the test individual; and using the classification system or method of the present invention, predicting the implantation window time of the individual based on the parametric features, and optionally providing a recommendation for the timing of embryo transfer.

[0055] Typically, the classification systems and methods according to the present invention include machine learning systems, preferably, regression models based on supervised learning. Therefore, in some aspects, the present invention also provides methods for constructing WOI prediction models and the resulting WOI prediction models. In some embodiments, the prediction model construction method includes at least: extracting magnitudes of differentially expressed gene features related to a plurality of receptivity windows of the present invention from a plurality of training samples (i.e., endometrial tissue samples with known receptivity states), while characterizing the receptivity state of the samples using the corresponding WOI time and using this as a label, constructing a training dataset to train a classifier.

[0056] The various aspects of the method of the present invention are described below.

[0057] I. Extraction of acceptance window-related differentially expressed gene features and their magnitudes for use in the systems and methods of the present invention

[0058] In this paper, differentially expressed genes related to the receptivity window, also known as ER differentially expressed genes, refer to genes that exhibit differential expression levels at different endometrial receptivity states (i.e., pre-receptivity, receptivity period, and post-receptivity). These differentially expressed genes (DEGs) can be used as biomarkers indicating endometrial receptivity (ER). ER differentially expressed genes serving as biomarkers can be identified using RNA-Seq transcriptomic analysis of endometrial tissue samples at different endometrial receptivity states (i.e., pre-receptivity, receptivity period, and post-receptivity); for example, multivariate ANOVA combined with the Tukey HSD test was used, with a screening criterion of q-value < 0.001 for transcriptomic analysis.

[0059] However, as those skilled in the art will understand, in some cases, selectively using independent input variables relevant to the prediction (rather than all available differentially expressed genes) may be more advantageous, as this reduces the problem of "overfitting," simplifies data collection (avoiding the need to measure the expression levels of differentially expressed genes outside the model), and reduces the computational resources required for prediction due to fewer input parameter values ​​needing evaluation. Therefore, in some embodiments, the classification method or system of the present invention also includes a DEG feature selection step. In this document, "feature selection" refers to selecting the most relevant or most important features from the original feature set to improve model performance, reduce overfitting, accelerate training, and improve model interpretability. Feature selection can reduce data dimensionality, decrease model complexity, and improve the model's generalization ability while retaining the most representative and informative features, thereby improving the model's accuracy and efficiency. Commonly used feature selection methods include, but are not limited to, the Filter method, the Wrapper method, and the Embedded method. For example, features can be evaluated and ranked based on statistical relationships between them (such as correlation, variance, etc.), and then the top-ranked features can be selected as input features; or the model performance can be evaluated by trying different subsets of features, and then the best-performing subset can be selected as the final feature set. In some embodiments of the invention, preferably, a set of DEGs for WOI prediction is selected and determined using a machine learning algorithm based on feature importance, from the DEGs determined by transcriptome analysis. In some more preferred embodiments, DEG feature selection is performed using random forest feature importance.

[0060] In some embodiments, the acceptability window-related differentially expressed genes (DEGs) used in the classification / prediction system or method of the present invention include at least 100, 120, 150, 170, 190, 200, or 210 acceptability window-related differentially expressed genes.

[0061] In some preferred embodiments, the tolerance window-related differentially expressed genes (DEGs) used in the classification / prediction system or method of the present invention include Figure 4 The diagram shows at least 100, 150, 170, or 201 differentially expressed genes associated with the receptivity window.

[0062] Various methods are known in the art for obtaining the expression level of a target gene in a sample, such as RNA sequencing, transcriptome mapping analysis, gene expression microarrays, and oligonucleotide probes. These methods are all applicable to extracting the magnitude of the DEG biomarker characteristics according to the present invention from endometrial biopsy tissue samples.

[0063] In some embodiments, the expression level of each of the differentially expressed ER genes of the present invention in an endometrial tissue sample can be extracted by RNA sequencing. For this purpose, methods known in the art can be used to extract total RNA from the endometrial tissue sample to be tested, and after reverse transcription and cDNA amplification, next-generation sequencing can be performed to determine the expression level of the target DEG gene in the sample. In some embodiments, preferably, the RNA expression level of each gene is estimated by normalization using FPKM.

[0064] In other embodiments, the expression level of each of the differentially expressed genes in the ER set of the present invention in an endometrial tissue sample can be extracted using a gene expression microarray. Gene expression microarrays are a high-throughput gene expression analysis technique that can simultaneously detect the expression levels of hundreds or thousands of genes under different conditions as needed. Gene expression microarrays typically involve immobilizing a plurality of oligonucleotide probes on a microarray chip, which can complementaryly pair and hybridize with different gene sequences. Subsequently, the RNA sample to be tested can be transcribed into cDNA and labeled with a detectable marker (e.g., fluorescent label), then added to the microarray chip. The signal intensity of hybridization at each probe on the chip is read by detecting the detectable signal (e.g., fluorescence signal), thereby determining the expression level of the gene corresponding to the probe in the sample. There are no particular limitations on the materials used for chip fabrication. For example, a microarray of approximately 1 cm² can be used. 2 To prepare a chip, fragments of cDNA or oligonucleotides, called probes, are immobilized on a glass slide of a certain size.

[0065] The endometrial tissue samples used to extract the DEG gene characteristic values ​​of the present invention are preferably from the secretory phase of the menstrual cycle. More preferably, the samples are collected between day 5 and day 9 after the individual's luteinizing hormone (LH) peak or between day 3 and day 7 after the individual begins progesterone supplementation.

[0066] II. Determination / Prediction of WOI Time for the System and Method of the Invention

[0067] In the construction of the prediction model according to the invention, it involves extracting the magnitudes of DEG biomarker features from each training set sample and assigning WOI time labels to the samples. Accordingly, in the prediction / classification system and method according to the invention, it involves predicting the WOI time corresponding to the sample from the extracted magnitudes of DEG biomarker features.

[0068] In some advantageous aspects of the invention, the WOI time used in the systems and methods of the invention characterizes the receptivity of the corresponding endometrial samples with hourly precision. Therefore, in some cases used for model construction in the invention, the optimal implantation window (WOI) time corresponding to the individual from whom the training sample originates can be determined based on the pregnancy outcome, and the WOI time label corresponding to that sample can be estimated based on the sampling time of the training sample. Although in individuals with a normal implantation window, the optimal implantation window time can also be estimated from the time of occurrence of the conventional implantation window, this approach is limited for individuals with WOI shift (e.g., individuals with recurrent implantation failure). Considering that successful pregnancy generally indicates that embryo transfer occurred during the receptivity period of the endometrium, in some cases, it is more advantageous to estimate the optimal implantation window of an individual more accurately from the embryo transfer time of a successful pregnancy. Therefore, in some preferred embodiments of the invention, the training sample is derived from individuals who have achieved successful pregnancies through embryo transfer. Subsequently, as those skilled in the art will understand, the sample acceptance status, characterized by the WOI time, can be obtained by calculating the time interval between the sample collection time and the individual's optimal implantation window estimated based on clinical pregnancy outcomes, for example, using the formula WOI time value = estimated optimal implantation window time - sampling time. In this way, those skilled in the art will understand that the WOI time value from endometrial samples in the pre-acceptance phase will be positive, while the WOI time value from endometrial samples in the post-acceptance phase will be negative. In some embodiments, preferably, the WOI time used in this invention is a number of hours in the range of -96 hours to 96 hours, for example, selected from 0h, 24h, -24h, 48h, -48h, 72h, -72h, 96h, and -96h, or any number of hours between any two values. Accordingly, the predictive models, prediction / classification systems, and methods of this invention using this WOI time will allow predictions of individual optimal implantation window times deviating from the sampling point by -96 hours to 96 hours without the need for secondary or multiple repeated sampling.

[0069] III. Training dataset for the system or method of this invention and its construction

[0070] Training datasets typically contain input features and corresponding target variables (or labels), used to train models to learn the relationship between input features and target variables. The quality and diversity of the training dataset play a crucial role in the model's performance and generalization ability.

[0071] The training dataset used to train the WOI prediction model of the present invention, in some embodiments, includes at least the expression levels of a plurality of differentially expressed ER genes extracted from a plurality of endometrial tissue samples with different endometrial receptivity states (i.e., pre-receptivity, receptivity period, and post-receptivity) as input features.

[0072] In addition to the input features mentioned above, each training data vector in the dataset should also contain WOI time labels for the samples in order to construct a suitable training dataset.

[0073] Therefore, in some implementations, the construction of the training dataset according to the present invention includes:

[0074] (a) Obtaining endometrial tissue samples from a number of human individuals at one or more sampling time points;

[0075] (b) Extract the expression level values ​​of each of the set of differentially expressed gene features related to the acceptance window according to the present invention from each tissue sample.

[0076] (c) Based on the optimal planting window time for the individual corresponding to the sample, calculate the corresponding WOI time label for the sample from the sampling time.

[0077] Thus, a training dataset is generated, in which each training data vector represents an endometrial tissue sample from a single sampling time point of a human individual and contains the magnitude of each of a set of differentially expressed gene features related to the receptivity window extracted from the corresponding sample, and each training data vector also contains a WOI time label for the corresponding sample.

[0078] In some preferred embodiments, in step (c), the optimal implantation window for the sample-corresponding individual is estimated based on the clinical pregnancy outcome of embryo transfer in the individuals from whom the sample originated. Preferably, the individual is one who has achieved a successful clinical pregnancy after embryo transfer.

[0079] In other preferred embodiments, in step (b), the set of receptivity window-related differentially expressed genes comprises at least 100, 120, 150, 170, 190, 200, or 210 receptivity window-related differentially expressed genes. More preferably, the receptivity window-related differentially expressed genome comprises differentially expressed genes identified by transcriptomic analysis (e.g., RNA sequencing) from endometrial tissue during the pre-receptivity, post-receptivity, and receptivity periods. Most preferably, the receptivity window-related differentially expressed genome comprises... Figure 4 The list includes at least 100, 150, 170, or 201 differentially expressed genes associated with the receptivity window, or consists of such genes.

[0080] The training samples used to construct the training dataset of this invention can come from healthy individuals with a normal implantation window, or from patients who have undergone embryo transfer, with or without a normal implantation window, such as individuals with recurrent implantation failure. In some embodiments, the training dataset is constructed from a training sample set consisting of a plurality of training samples, wherein the training sample set includes endometrial tissue samples collected at least partially from the pre-receptivity period, at least partially from the post-receptivity period, and at least partially from the receptivity period. In some cases, preferably, the training sample set includes at least 50, 60, 70, 80, 90, 100, 125, 150, 175, 200, 225, 250, 275, 300, or 350 training samples. In some cases, at least a portion of the samples in the training sample set are collected from a plurality of human individuals undergoing embryo transfer in such a manner that at least three endometrial tissue samples, corresponding to the pre-receptivity period, receptivity period, and post-receptivity period, are collected from each of the plurality of human individuals.

[0081] IV. The method and system of the present invention for classifying data using a classification / prediction system.

[0082] Machine learning-based classifiers have been applied in the medical field for data analysis and data mining to extract important information and patterns contained in large datasets. Generally, machine learning involves algorithms; for example, supervised learning algorithms can be trained on data with known labels to achieve generalization. The trained machine learning algorithm can then be applied to new data with unknown labels, evaluating the input features based on the learned patterns and outputting predicted values.

[0083] In this invention, through data analysis and mining, the inventors provide a robust test data classification method. This method evaluates data (test data, i.e., biometric values) obtained from an individual's endometrial tissue sample to predict the corresponding receptivity of the sample, and outputs the prediction results in hourly Word of Interest (WOI) time. The method of this invention generally involves preparing or obtaining training data, training a classifier, generating a prediction model, and using the model to evaluate the test data. In this method, a classifier, such as a machine learning algorithm, is preferably used, including, for example, various regression algorithms. The trained classifier can then output a prediction result for the test sample based on the test data, such as the predicted optimal implantation window time for the individual corresponding to the sample.

[0084] Therefore, in some embodiments, the present invention provides a test data classification method, which includes the following steps:

[0085] The first step is to describe the predefined dataset using a classifier. This is the "learning step" performed on the "training" data.

[0086] The training dataset is a computer-executed data storage that reflects the magnitudes of multiple features of multiple objects with known labels. In this invention, the labels are Word of Interest (WOI) timestamps associated with the object's tolerance state. The data can be stored in a flat file, database, table, or any other retrievable data storage format known in the art. In an exemplary embodiment, the training dataset is stored as multiple vectors, each corresponding to an object and including magnitudes of multiple features of that object and a WOI timestamp associated with the object. Typically, each vector contains a corresponding entry for each of the multiple feature magnitudes. The training dataset may be linked to a network, such as the Internet, so that its content can be remotely accessed by an authorized entity (e.g., an individual user or computer program). Alternatively, the training dataset may reside on a network-isolated computer.

[0087] The second step (optional) involves applying the classifier to the "validation" database, or using cross-validation, and observing model performance metrics, including the ROC curve and AUC value. 2 Values, accuracy, precision, sensitivity, and specificity, etc. For example, in some implementations, only a portion of the training dataset may be used for the learning step, while the remainder of the training dataset may be used as a validation dataset.

[0088] The third step involves submitting the feature values ​​from the test subject to the classifier, which then outputs the prediction results calculated for that test subject (e.g., the predicted WOI time and optionally the predicted individual optimal planting window time).

[0089] Corresponding to the classification method of the present invention, in one aspect, the present invention also provides a classification system and a classifier suitable for performing the method.

[0090] Classification system applicable to the classification method of the present invention

[0091] The classification system of this invention may include computer-executable software, firmware, hardware, or various combinations thereof. For example, the classification system may include references to a processor and tools supporting data storage. Furthermore, the classification system may be implemented on multiple devices or other components that are local to or remote from each other. Moreover, references to software may include non-transitory computer-readable media that, when executed on a computer, cause the computer to perform a series of steps.

[0092] The classification system of this invention may include data storage, such as network-accessible storage, local storage, remote storage, or a combination thereof. Data storage may utilize disk arrays (“RAID”), tape, disks, storage area networks (“SAN”), Internet Small Computer System Interface (“iSCSI”) SAN, Fibre Channel SAN, Universal Internet Archive System (“CIFS”), Network Attached Storage (“NAS”), Network File System (“NFS”), or other computer-accessible storage. In one or more embodiments, data storage may be a database, such as an Oracle database, a Microsoft SQL Server database, a DB2 database, a MySQL database, a Sybase database, a target-oriented database, a hierarchical database, or another database. In one embodiment, data storage may utilize a flat file structure to store data.

[0093] Classifier applicable to the present invention

[0094] As a non-limiting example, classifiers that can be used for classifying test data in this invention include, but are not limited to, logistic regression, elastic network regression, neural networks, Bayesian neural networks, linear regression, multinomial regression, support vector regression (SVR), decision tree regression, random forest regression, and ensemble methods (such as gradient boosting trees). These algorithms use labeled training data to learn a model and are used to predict continuous numerical outputs. In some embodiments of the invention, the classifier used for the prediction model of the invention is selected from linear regression, LASSO regression, and Ridge regression. In some embodiments of the invention, preferably, the classifier used for the prediction model of the invention is selected from ensemble learning-based or decision tree-based regression algorithms; more preferably, the classifier is a random forest regression algorithm.

[0095] V. Exemplary implementations of the predictive model construction method and classification test data method of the present invention

[0096] Based on the aforementioned classification system and classifier, in some implementations, the present invention provides a method for constructing a planting window prediction model and a method for classifying test data.

[0097] Predictive model construction methods

[0098] In some implementations, the method of the present invention for constructing a planting window prediction model includes:

[0099] (i) Obtain a training dataset for model building, wherein each training data vector in the training dataset represents an endometrial tissue sample from a single sampling time point of a human individual and contains the magnitude of each of a set of differentially expressed gene features related to the receptivity window extracted on the corresponding sample, and each training data vector also contains a label related to the WOI time of the corresponding sample.

[0100] The quantitative value of the differentially expressed gene characteristic is the expression level of the differentially expressed gene measured on the sample, and

[0101] The WOI time is the time required from the sampling time of the sample to reach the optimal planting window (WOI) of the individual from which the sample originated in the same period;

[0102] (ii) The training data vector set is received on at least one computer-executable processor;

[0103] (iii) On the at least one processor, using the training data vector set, a classifier is trained to generate a WOI prediction model, wherein the WOI prediction model outputs the predicted WOI time with hourly precision.

[0104] In some implementations, the WOI time is a number of hours in the range of -96 hours to 96 hours, such as hours selected from 0h, 24h, -24h, 48h, -48h, 72h, -72h, 96h, and -96h, or hours between any two values.

[0105] In some implementations, the optimal implantation window time for the sample source individual is estimated based on the individual's clinical pregnancy outcome after embryo transfer, preferably the individual who has achieved a successful clinical pregnancy after embryo transfer.

[0106] In some implementations, the cycle is a natural cycle or an artificial cycle.

[0107] In some implementations, the classifier includes random forest regression.

[0108] In some implementations, step (iii) includes a feature selection step, wherein a random forest is preferably used to select differentially expressed gene features related to the tolerance window based on feature importance ranking.

[0109] In some implementations, step (iii) includes: using a method containing Figure 4 The classifier is trained using a training dataset consisting of at least 100, 150, 170, or 201 differentially expressed genes associated with the tolerance window as features.

[0110] In some implementations, step (iii) further includes: outputting the predicted optimal embryo transfer time interval, wherein the time interval spans 1 hour.

[0111] Test data classification methods

[0112] In some embodiments, the method according to the invention for classifying test data is an off-site computer-executed method, comprising:

[0113] (a) On at least one computer-executable processor, test data is received, wherein the test data contains the expression levels of a set of differentially expressed genes related to receptivity window determined on endometrial tissue samples from a single sampling time point of a human test individual.

[0114] (b) On the at least one processor, the test data is evaluated using a trained classifier to predict the WOI time of the test individual with hourly accuracy.

[0115] (c) Using at least one processor, output the predicted WOI time for the test individual.

[0116] Wherein, the WOI time is the time required from the sampling time of the sample to reach the optimal planting window (WOI) of the individual from which the sample originated in the same period.

[0117] In some implementations, the WOI time is a number of hours in the range of -96 hours to 96 hours, such as hours selected from 0h, 24h, -24h, 48h, -48h, 72h, -72h, 96h, and -96h, or hours between any two values.

[0118] In some implementations, the trained classifier is a classifier generated by the WOI prediction model construction method according to the present invention.

[0119] In some implementations, the method further includes: prior to step (a),

[0120] (a1) Endometrial tissue samples were obtained from the individual through sampling at a single time point.

[0121] Preferably, the sampling is performed at a single time point between day 5 and day 9 after the individual's luteinizing hormone (LH) peak or between day 3 and day 7 after the individual begins progesterone supplementation;

[0122] (a2) Determine the expression level of the acceptance window-related differentially expressed genome on the sample, preferably, the acceptance window-related differentially expressed genome includes... Figure 4 At least 100, 150, 170, or 201 differentially expressed genes associated with the receptivity window are listed.

[0123] In some implementations, the method includes: extracting the expression level of each of the receptivity window-related differentially expressed genomes from an endometrial tissue sample at a single sampling time point of the individual via RNA sequencing.

[0124] In some implementations, the method further includes guiding the individual's embryo transfer based on the endometrial implantation window time predicted in step (c).

[0125] VI. Methods to improve embryo transfer

[0126] In another aspect, the present invention provides a method for improving embryo transfer in an individual, comprising:

[0127] - Using the test data classification method according to the present invention, based on single-sampling-point endometrial samples of the individual, predict the endometrial implantation window time of the individual; and

[0128] - Provide recommendations on the timing of embryo transfer based on the predicted implantation window.

[0129] The method of the present invention can be used for personalized ET, wherein patients undergo pET after receiving a model of the present invention to predict the implantation window time, wherein: the blastocyst transfer time is the optimal WOI time predicted by the model; the cleavage stage embryo transfer time on day 3 is two days earlier than the predicted optimal WOI time, to ensure that the cleavage embryo develops into a blastocyst in vivo and coincides with the implantation window.

[0130] In some preferred embodiments, the Infinitesimal jackknife resampling method can be used to give the distribution of WOI time predicted by the model of the present invention, and the mean and standard deviation (SD) of the predicted WOI time can be calculated from it. The optimal embryo implantation time window is given based on the mean WOI ± 1.96SD (statistical significance p < 0.05).

[0131] In some embodiments, the individual is a patient undergoing a natural or artificial cycle. In some embodiments, the embryo transfer is a fresh or frozen embryo transfer. In some embodiments, the embryo transfer is a cleavage embryo transfer or a blastocyst transfer.

[0132] In some embodiments, the method according to the invention is used to improve embryo implantation and / or clinical pregnancy.

[0133] In some implementations, the individual receiving the personalized ET is a patient with recurrent implantation failure, particularly a patient with recurrent implantation failure of endometrial origin.

[0134] In some implementations, individuals receiving the personalized ET may have a WOI shift or a narrower WOI duration relative to the normal WOI time.

[0135] II. Apparatus, systems, and computer program products for implementing the method of the present invention

[0136] In some aspects, the present invention provides products for implementing the predictive model building and classification / prediction methods of the present invention. The products of the present invention include, but are not limited to, devices, apparatuses, and systems. The devices, apparatuses, and / or systems of the present invention may consist of a plurality of modules or components that implement any method of the present invention.

[0137] In some embodiments, the "modules" used for predictive model building and data classification of the present invention are software objects or routines (e.g., as independent threads) that can be executed centrally on a single computing system (e.g., a computer program, a tablet computer, one or more processors). In other embodiments, programs implementing the methods of the present invention may be stored on a non-transitory computer-readable medium, forming part of the apparatus, device, and / or system of the present invention, wherein the computer-readable medium contains computer program logic or code portions for implementing the methods of the present invention. In one embodiment, the dataset used for model building or data classification of the present invention is stored on a non-transitory computer-readable medium, such as a hard disk or CD-ROM, which may exist independently of or form part of the apparatus, device, and / or system of the present invention. In one embodiment, the apparatus, device, and / or system of the present invention may be configured to communicate with at least one remote device or server, for example, to receive information from a server via a communication network and to transmit information to a server via a network. Furthermore, while the modules and methods described herein are preferably implemented in software in some embodiments, implementation in hardware or a combination of software and hardware is also possible and conceivable to those skilled in the art.

[0138] In addition to the modules / components described above, the apparatus, device and / or system of the present invention may also include other modules or components, such as an apparatus or system for detecting differentially expressed gene expression levels in ER according to the present invention, for example, a module or component for extracting RNA from a sample (such as an RNA extraction kit), a module or component for RNA sequencing, and a component for identifying differentially expressed gene expression level information from sequencing results.

[0139] In other aspects, the present invention also provides a non-transitory computer-readable medium storing computer program instructions that are executed by a computer or computer system to implement the steps of the model building method, data classification method, or prediction method according to any embodiment of the present invention.

[0140] Exemplary embodiments of the apparatus or system of the present invention

[0141] In some embodiments, therefore, the present invention provides an apparatus or system for predicting an individual's endometrial implantation window, comprising:

[0142] -Optionally, a module for acquiring differentially expressed gene features related to the receptivity window, wherein the module is capable of extracting the expression level of differentially expressed genes related to the receptivity window from endometrial tissue samples at a single sampling time point of the individual;

[0143] - An endometrial implantation window prediction module, wherein the module comprises a WOI prediction model constructed according to the method of the present invention, or wherein the module is capable of performing a test data classification method or prediction method according to the present invention to output the predicted endometrial implantation window time of the individual and optionally output the predicted optimal embryo implantation time window.

[0144] In some further embodiments, in the apparatus or system of the present invention, the endometrial implantation window prediction module includes at least one computer-executable processor coupled to an electronic storage device containing an electronic representation of a classifier, the classifier being a WOI prediction model according to the present invention.

[0145] The processor is configured to receive test data from the endometrial tissue sample from the human individual;

[0146] The processor is further configured to evaluate the test data using the electronic representation of the classifier and, based on the evaluation, output the WOI time for the test individual. Based on the predicted WOI time, the optimal implantation window time for the individual can be easily calculated, taking into account the sampling time of the individual sample, and this can guide the individual's embryo transfer.

[0147] As used herein, the term "computer" should be understood to include at least one hardware processor that may use at least one memory. The at least one memory may store a set of instructions. These instructions may be stored permanently or temporarily in the computer's one or more memories. The processor executes the instructions stored in the one or more memories to process data. The set of instructions may include multiple instructions for performing one or more specific tasks (such as the classification and / or rating tasks described herein). The set of instructions for performing the specific task may be represented as a program, a software program, or software. As described above, the computer executes the instructions stored in one or more memories to process data. This data processing may be, for example, in response to a computer user command, in response to previous processing, in response to a request from another computer, and / or any other input. The computer used for at least part of implementing embodiments according to the invention may be a general-purpose computer. However, the computer may also utilize any of a variety of other technologies, including dedicated computers, computer systems including microcomputers, minicomputers, or mainframes, such as programmable microprocessors, microcontrollers, peripheral integrated circuit elements, CSIC (customer application integrated circuit) or ASIC (application application integrated circuit) or other integrated circuits, logic circuits, digital signal processors, programmable logic devices such as FPGAs, PLDs, PLAs, or PALs, or any other device or arrangement of devices capable of implementing at least some of the steps of the methods of the present invention.

[0148] It is understood that, in order to implement the method of the present invention, it is not necessary to physically place the computer's processor and / or memory in the same geographical location. That is, the various processors and memories used by the computer can be located in geographically different locations and connected to communicate in any suitable manner. Furthermore, it should be understood that the various processors and / or memories can consist of different physical components of the device. Therefore, it is not necessary for the processor to be a single device component in one location and the memory to be another single device component in another location. That is, for example, it can be considered that the processor can be two or more device components located in two different physical locations. These two or more different device components can be connected in any suitable manner (such as a network). Furthermore, the memory can include two or more partial memories located in two or more physical locations.

[0149] Various technologies can be used to provide communication between various computers, processors, and / or memories, and to allow the processors and / or memories of this invention to communicate with any other entity; for example, to obtain further instructions or to access and use remote memory. Technologies used to provide such communication may include networks, the Internet, intranets, extranets, LANs, Ethernet, or any client-server system providing communication. Such communication technologies may use any suitable protocol, such as TCP / IP, UDP, or OSI.

[0150] Furthermore, it should be understood that the computer instructions or instruction sets used to implement and operate the present invention will be in the form of suitable computer-readable instructions.

[0151] In some embodiments, various user interfaces can be utilized to allow a user to interact with a computer or machine used to at least partially implement embodiments of the present invention. The user interface may be in the form of a dialog box. The user interface may also include a mouse, touchscreen, keyboard, voice reader, voice recognizer, dialog screen, menu box, list, checkbox, toggle switch, button, or other means that allow the user to receive information about the computer's operation and / or provide information to the computer. Therefore, the user interface can be any means that provides communication between the user and the computer. For example, the information provided by the user to the computer through the user interface may be commands, data, or some other form of input.

[0152] It is also considered that the user interface of the present invention can interact with another computer that is not the user, for example, to transmit and receive information. Therefore, the other computer can be characterized as the user. Furthermore, it is considered that the user interface used in the systems and methods of the present invention can partially interact with one or more other computers while also partially interacting with the user.

[0153] Therefore, in some aspects, the present invention provides systems to assist in carrying out the methods of the present invention. An exemplary system includes a storage device for storing training datasets and / or test datasets and a computer for executing a learning machine (such as a classifier or ensemble classifier as described herein). The computer may also be operable to collect training datasets from a database, preprocess the training datasets, train the learning machine using the preprocessed training datasets, accept input test data, classify the test data using the trained learning machine, and output information about the classification of the test data. The exemplary system may also include a communication device for remotely receiving test datasets and / or training datasets. In such cases, the computer may be used to store the training dataset in the storage device before preprocessing the training dataset, and to store the test dataset in the storage device before preprocessing the training dataset. The exemplary system may also include a display device for displaying the classification of the test data. Example

[0154] Example 1 Model Construction

[0155] Overview

[0156] Patients who achieved successful clinical pregnancy through personalized embryo transfer (pET) under the guidance of three-timepoint rsERT (see He A, J Transl Med 2021, 19(1):176.) were included. From these patients, samples were analyzed to determine the specific acceptance status (pre-acceptance, during-acceptance, and post-acceptance) based on the individual's clinical pregnancy outcome. Differentially expressed gene signatures and WOI time tags were then extracted from these samples to construct a single-timepoint model. This prototype model was used to guide pET in patients undergoing single-timepoint endometrial sampling (e.g., LH+7 or P+5); and patients with clinical pregnancy obtained from this model were incorporated into the prototype model for optimization. Thus, a modified rsERT (i.e., single-timepoint rsERT, also referred to as the "modified rsERT model" in this paper) was finally established.

[0157] Endometrial biopsy, sample collection and processing

[0158] Written informed consent was obtained prior to endometrial sample collection. Endometrial tissue was collected using an endometrial sampler. In the case of a natural cycle, sampling was performed at three time points: day 5, day 7, and day 9 after the luteinizing hormone (LH) peak (i.e., LH+5, LH+7, and LH+9). In the case of a hormone replacement therapy (HRT) cycle, sampling was performed at three time points: day 3, day 5, and day 7 after progesterone supplementation (i.e., P+3, P+5, and P+7). For subsequent model use requiring only single-time-point sampling, endometrial samples were collected only at LH+7 (natural cycle) or P+5 (HRT cycle).

[0159] RNA transcriptional expression data acquisition

[0160] RNA sequencing and gene expression level determination of endometrial biopsy samples included the following steps: total RNA extraction from the samples, RNA quality control, reverse transcription and amplification, and NGS library construction (next-generation sequencing). Single-end sequencing was then performed on a HiSeq 2500 platform (Illumina, San Diego, CA, USA), or an MGISEQ-2000 or MGISEQ-T7 sequencing platform. After raw data filtering, quality control, and mapping, the RNA expression level of each gene was estimated using FPKM (Fragment Per Kilobase of Transcript, per Million Mapped Reads).

[0161] Characteristics of enrolled patients

[0162] Inclusion criteria for model construction: age 20–39 years; body mass index (BMI) 18–25 kg / m2; more than one viable cleavage-stage embryo or blastocyst.

[0163] Table 1. Baseline demographic and clinical characteristics of patients used for rsERT establishment

[0164]

[0165] BMI, Body Mass Index; AMH, Anti-Müllerian Hormone; FSH, Follicle-Stimulating Hormone; LH, Luteinizing Hormone;

[0166] E2, estradiol; T, testosterone; AFC, antral follicle count; IVF, in vitro fertilization; SD: standard deviation; IQR: interquartile range.

[0167] Building a training dataset

[0168] To obtain the optimal WOI time prediction, samples from three time points at different tolerance states are used to find model features, and the corresponding WOI time labels are used to train the WOI time prediction model. To this end, the following training data vectors for model building and optimization are extracted from the training samples: differentially expressed gene (DEG) values ​​and the estimated optimal WOI time.

[0169] During model construction, candidate biomarker genes were identified on training samples sampled at three time points according to the previously published DEG identification scheme [He A, J Transl Med 2021, 19(1):176.]. In short, multivariate analysis of variance combined with the Tukey HSD test was used, with the screening criterion set at q-value < 0.001, thereby identifying differentially expressed genes (DEGs) related to the tolerance window. The expression values ​​of these DEGs in the training samples sampled at three time points were obtained through RNA sequencing and used as input features for model construction. During model optimization, the expression values ​​of each DEG biomarker used in the established model were detected through RNA sequencing on the training samples sampled at a single time point (LH+7 or P+5) and used as input features for optimizing the prototype model.

[0170] Based on the biopsy collection time of the endometrial samples and the subsequent clinical pregnancy outcome of the individual receiving the samples via pET, the proposed optimal implantation window (WOI) time was determined for the training samples, accurate to the hour, and used as the WOI time label for the samples. For example, if a patient achieved a successful pregnancy with a blastocyst transferred at LH+6 (or an embryo at day 3 cleavage stage transferred at LH+4), then for a sample collected from that patient at LH+5, the proposed optimal receptivity time of the endometrium (i.e., the proposed optimal implantation window WOI time) is 24 hours after the collection time, and the corresponding WOI time label for that sample is 24 hours. Similarly, samples collected from the same patient at LH+7 and LH+9 will be assigned the following WOI time labels: -24h (i.e., the proposed optimal receptivity time is 24 hours before the sampling time) and -72h (i.e., the proposed optimal receptivity time is 72 hours before the sampling time), respectively. Similarly, depending on the pregnancy outcome, different combinations of embryo transfer time and different sample collection time will result in WOI time tag values ​​for 0h, 24h, -24h, 48h, -48h, 72h, -72h, 96h and -96h (at 24-hour intervals).

[0171] Building Model

[0172] Using the training dataset, DEG expression values ​​were used as input features. A random forest machine learning method was applied, with the optimal WOI prediction time as the output, to train a regression model. In short, the random forest regression model in the ranger R package (version 0.12.1) generated the optimal WOI prediction time (i.e., the optimal embryo implantation time) with hourly precision. Then, the Boruta R package (version 7.0.0) was used to measure the importance of each feature in the model for feature selection. After feature selection, the model was built using the features whose importance was confirmed (DEG genes).

[0173] The algorithm steps for feature selection are as follows:

[0174] 1. After randomly shuffling all the features in the dataset, a new set of features is created by randomly selecting a set of values ​​from them. These features are called "shadow features".

[0175] 2. Train a random forest model together with the original features and shadow features;

[0176] 3. Compare the importance of features. If the importance score of an original feature is higher than the scores of all shadow features, then the feature is important; otherwise, the feature is removed.

[0177] 4. Repeat steps 2 and 3 until all original features are important;

[0178] 5. The remaining original features are the features obtained through screening.

[0179] Finally, 201 characteristic genes were obtained (such as...). Figure 4 As shown in the figure, it is incorporated into the model training phase.

[0180] Model optimization

[0181] Additional training datasets were added to train the model. During model optimization, random forest regression, linear regression, LASSO regression, and Ridge regression algorithms were tested to construct the WOI time prediction tool. R-squared values ​​were used to evaluate the predictive performance of models trained by different methods. The results showed that the R-squared values ​​were 0.92 (random forest regression), 0.88 (linear regression), 0.75 (LASSO regression), and 0.80 (Ridge regression). This indicates that the random forest regression algorithm performed optimally, and therefore, this regression algorithm was selected for further analysis.

[0182] The predicted WOI time from the statistical model is used to determine the sample's predictability status according to the following classification criteria: predicted time between [-18h, 18h], tolerance period; predicted time > 18h, pre-tolerance period; predicted time < -18h, post-tolerance period. For example... Figure 1 As shown, the acceptance status prediction results match the acceptance status determined based on the clinical pregnancy outcomes of the sample.

[0183] With 10-fold cross-validation, the model achieves an average prediction accuracy of 94.51%. See Table 2 below, which shows the prediction performance of the improved rsERT model with classification criteria.

[0184] Table 2 Performance of the improved rsERT model with classification criteria*

[0185]

[0186] Example 2. Model Application and Effect Verification

[0187] During the model application and efficacy validation phase, the inclusion criteria were appropriately expanded to include: ≤45 years of age; body mass index ≤28 kg / m²; and a history of recurrent intrauterine pregnancy (RIF). RIF was defined as failure to achieve intrauterine pregnancy after two consecutive cycles of fresh or frozen embryo transfer (ET), with a cumulative total of at least 4 morphologically superior embryos transferred from cleavage-stage embryos and at least 2 morphologically superior embryos transferred from blastocysts. The criteria for superior embryos were as follows: Cleavage-stage embryos: ≥7 blastomeres and a fragmentation rate <20% on day 3 post-fertilization; Blastocysts: ≥3 BBs on days 5 and 6, according to the Gardner classification system.

[0188] Table 3. Baseline characteristics of RIF patients

[0189]

[0190] SD: standard deviation

[0191] DOR: Decreased ovarian reserve.

[0192] For patients undergoing endometrial preparation, endometrial samples were collected at LH+7 (natural cycle) or P+5 (HRT cycle) in either a natural cycle or a hormone replacement therapy (HRT) cycle. Gene expression data from the endometrial tissue samples were then collected. RNA sequencing and gene expression level determination of the endometrial biopsy samples, as described in Example 1, included the following steps: total RNA extraction, RNA quality control, reverse transcription and amplification, and NGS library construction (next-generation sequencing). Single-end sequencing was then performed on a HiSeq 2500 platform (Illumina, San Diego, CA, USA), or an MGISEQ-2000 or MGISEQ-T7 sequencing platform. After raw data filtering, quality control, and mapping, the RNA expression level of each gene was estimated using FPKM (Fragment Per Kilobase of Transcript, per Million Mapped Reads).

[0193] The DEG expression values ​​extracted from the samples are input into the modified rsERT model constructed in Example 1 to predict the WOI time window of the samples. If the patient being tested is in a hormone replacement cycle, and the progesterone administration time is 11:00 AM on P+0, and the sampling time is 12:00 PM on P+5, then the sampling time point and the progesterone administration time point differ by 121 hours (that is, sampling is performed 121 hours after progesterone administration). If the model predicts that the WOI time window of the sample is 16 hours, that is, the sample still has 16 hours to reach the optimal receptivity hour window. Therefore, based on the progesterone administration time, it is estimated that the optimal receptivity hour window will be 121h + 16h = 137h later. Therefore, it is recommended that in the embryo transfer cycle, the blastocyst transfer time be converted to days, based on the progesterone administration time, which is P+5 days + 17 hours.

[0194] Using a modified rsERT model, precise embryo transfer time was statistically analyzed on endometrial samples from nearly 600 patients, with a mean of 138 hours. The overall distribution is as follows: Figure 2As shown in the figure, the data distribution indicates that most RIF patients exhibited delays in the optimal receptivity window at different times. After the model predicted the endometrial WOI time window in the RIF population, embryo transfer timing intervention was performed according to the model prediction results, clinical outcomes were collected, and statistical analysis was performed. Table 4 below shows the baseline statistics for non-receptivity period patients (nWOI) and receptivity period patients (sWOI) based on the model prediction. The clinical pregnancy rate statistics showed that non-receptivity period patients (61.6%) benefited by approximately 7% compared to receptivity period patients (54.5%). Figure 3 As shown in the figure, the data indicate that the embryo implantation rate in the group undergoing transfer based on the receptivity predicted by the modified rsERT was significantly higher than that in the control group (43.21 vs. 30.87, p = 0.032). The embryo implantation rate (48.36% vs. 30.87%, p < 0.001), intrauterine pregnancy rate (52.54 vs. 36.33, p < 0.001), and sustained pregnancy rate (48.02 vs. 29.07, p < 0.001) in the group undergoing transfer based on the precise hour predicted by the modified rsERT were all significantly higher than those in the control group, while the miscarriage rate (8.60 vs. 20.00, p = 0.024) was significantly lower in the group.

[0195] Table 4. Baseline statistics of patients in the non-receptivity period and the receptivity period

[0196]

[0197] To determine whether natural cycles were superior to HRT cycles, we conducted a subgroup analysis of the experimental groups. The natural cycle group included 146 individuals, and the HRT group included 408 individuals. Baseline characteristics and pregnancy outcomes of these women are shown in Table 5. Age, BMI, duration of infertility, type of infertility, number of previous failed cycles, primary cause of infertility, PGS or PGD ratio, endometrial thickness, endometrial type, embryo stage transferred, rate of high-quality embryos transferred, and number of embryos transferred were comparable between the two groups (P > 0.05). The subgroup analysis results showed no significant difference in pregnancy outcomes between the two groups, indicating that the endometrial preparation protocol (natural cycle or HRT cycle) had no significant impact on model-guided pET pregnancy outcomes.

[0198] Table 5. Baseline characteristics and pregnancy outcomes of natural and HRT cycles in the experimental group.

[0199]

[0200]

[0201] t: t-test, χ2: Chi-square test

[0202] SD: standard deviation

[0203] In summary, the improved rsERT model according to the present invention offers at least the following advantages: it not only allows WOI prediction via a single endometrial biopsy, but also provides endometrial WOI prediction results with hourly accuracy, thus making it more suitable for the personalized and precise requirements of pET.

Claims

1. A method for constructing a planting window WOI prediction model, the method comprising: (i) Obtain a training dataset for model building, wherein each training data vector in the training dataset represents an endometrial tissue sample from a single sampling time point of a human individual and contains the magnitude of each of a set of differentially expressed gene features related to the receptivity window extracted on the corresponding sample, and each training data vector also contains a label related to the WOI time of the corresponding sample. The quantitative value of the differentially expressed gene characteristic is the expression level of the differentially expressed gene measured on the sample, and The WOI time stamp is the time value required from the sampling time of the sample to reach the optimal WOI of the individual from which the sample originated in the same period; (ii) The training data vector set is received on at least one computer-executable processor; (iii) On the at least one processor, a regression model is trained using the training data vector set to generate a WOI prediction model, wherein the WOI prediction model is used to predict the WOI time within a range of -96 hours to +96 hours relative to the sampling time point, and outputs the predicted WOI time in hours. Step (iii) includes: training a prediction model using a training dataset containing 201 differentially expressed genes related to the tolerance window as features. The 201 differentially expressed genes related to the acceptance window are: C2CD4A, SCGB2A2, SDCBP2, CXCL14, DUSP2, DEFB1, KLF6, PLA2G16, PAEP, IER3, SLPI, C10orf10, RABGAP1L, AC073218.2, C4BPA, TWF2, OGFOD1, CTD-2008A1.2, STX19, FXYD3, OPRK1, GAL, PCCA, NU PR1, MFSD4, RHPN2, SCML1, SNRNP25, RP11-49I11.1, ALPL, ANK3, PRR15, LRRC26, ATP6V0E1, CD55, MYL12A, HSD 11B2, PIKFYVE, RIMKLB, STC1, DCXR, SORD, AC006116.17, TP53I3, GRB7, PGR, GPX3, TRAK1, TSPO, ETHE1, CSTB, SPATS2, SAMHD1, LGALS3, EZR, SORBS2, FAM84B, ID1, RP1-193H18.2, NUDT16, CBLC, RASSF4, MFSD3, GGTA1P, HM GN5, MFAP5, ARG2, GDF15, MOCOS, TM7SF2, SOD2, FGL1, EFNA1, ANXA2, ARRDC1, FZD5, FAM134B, CKB, C19orf77, R NASET2, SERTAD1, NMRAL1, RP11-481J2.2, SAT1, GJB2, HEY2, WASIR2, CLCF1, ENAH, MAP9, TMED3, CBR3, HPRT1, HLA-DMB, MAP2K6, SOX17, HTATIP2, KRT23, DSC2, PRR15L, GMPR, ENSG00000278053, HK2, AMOTL1, CTC-490G23.2、PKHD1L1、SORT1、CTNNAL1、MAP1LC3A、AGPAT2、TPD52L1、MT1M、CP、METTL21A、GPX1、STXBP6、A TP1B1、MAOA、GREM2、NFIL3、NAAA、CCDC159、SLC35B4、DUSP23、SOX7、ID3、HLA-DOB、CEBPD、SLC1A 1、GNG11、HLA-DMA、EPAS1、NOV、ADK、GPR160、NR4A2、SLC12A7、G0S2、GPT2、TMEM140、KBTBD3、GL YCTK、JUNB、ATF7IP2、FAM13A、MYL12B、ANG、CFDP1、NUP153、CD81、DUSP1、TMEM139、TFCP2L1、ANX A1、OSER1、NPC2、TMEM144、USP22、CTA-293F17.1、CRIP1、SUDS3、PRKAG2-AS1、BVES、CMBL、PHB2 、BTG3、PPA1、FKBP14、PLA2G4F、CSRP2、UCA1、LRRC1、ATP6V0E2、C6orf1、GALE、EPT1、CLDN3、CCAT 1、CYSTM1、SLC18A2、DEGS2、NTHL1、CAPN6、HN1、PPM1M、MPC1、B3GALT4、ACSL3、MALL、C8orf4、RP1 1-622K12.1、PHLDA2、TSPYL2、ZNF566、ARL4D、BLVRA、ISG20、KRTCAP3、DUOX1、IQGAP2、B3GNT3。.

2. The method as described in claim 1, characterized in that, The WOI time tag value is the number of hours in the range of -96 hours to 96 hours.

3. The method as described in claim 1, characterized in that, The optimal implantation window time for the individuals from whom the sample is derived is estimated based on the clinical pregnancy outcome of the individuals after embryo transfer, wherein the individuals are those who have achieved a successful clinical pregnancy after embryo transfer.

4. The method as described in claim 1, characterized in that, The cycle can be a natural cycle or an artificial cycle.

5. The method as described in claim 1, characterized in that, The regression models include LASSO regression, Ridge regression, linear regression, or random forest regression.

6. The method as described in claim 5, characterized in that, The regression model includes random forest regression.

7. The method as described in claim 1, characterized in that, Step (iii) includes a feature selection step.

8. The method as described in claim 7, characterized in that, Random forests are used to select differentially expressed gene features related to the tolerance window based on feature importance ranking.

9. The method according to any one of claims 1-8, characterized in that, Step (iii) also includes: outputting the predicted optimal embryo transfer time interval, wherein the time interval spans 1 hour.

10. The method as described in claim 1, characterized in that, The training dataset is obtained from a training sample set, wherein the training sample set includes tissue samples of the endometrium collected during the pre-receptive, post-receptive, and receptive phases.

11. The method as described in claim 10, characterized in that, The training sample set contains at least 50, 60, 70, 80, 90, 100, 125, 150, 175, 200, 225 or 250 training samples.

12. The method as described in claim 10, characterized in that, At least a portion of the samples in the training sample set were collected from a plurality of human individuals undergoing embryo transfer in the manner in which at least three endometrial tissue samples were collected from each of the plurality of human individuals, corresponding to the pre-receptivity period, receptivity period, and post-receptivity period of the same cycle.

13. A method for classifying test data, characterized in that, The method described is a computer-executed method outside the body. The test data contained the expression levels of a set of differentially expressed genes related to receptivity window, measured on endometrial tissue samples from a single sampling time point of a human test individual. The method includes: (a) The test data is received on at least one computer-executable processor; (b) On the at least one processor, the test data is evaluated using a trained prediction model to predict the WOI time of the test individual in hourly increments. (c) Using at least one processor, output the predicted WOI time for the test individual. Wherein, the WOI time is the time required from the sampling time of the sample to reach the optimal WOI of the individual from which the sample originated in the same period. The trained prediction model is a WOI prediction model constructed according to any one of claims 1-12.

14. The method as described in claim 13, characterized in that, The method further includes: before step (a), (a1) Endometrial tissue samples were obtained from the individual through sampling at a single time point. (a2) Determine the expression level of the receptivity window-related differentially expressed genome on the sample, wherein the receptivity window-related differentially expressed genome comprises the 201 receptivity window-related differentially expressed genes as defined in claim 1.

15. The method as described in claim 14, characterized in that, In step (a1), the sampling is performed at a single time point between day 5 and day 9 after the individual's luteinizing hormone (LH) peak or between day 3 and day 7 after the individual begins progesterone supplementation.

16. The method as described in claim 13, characterized in that, The expression levels of each of the receptivity window-related differentially expressed genomes were extracted from endometrial tissue samples at a single sampling time point of the individual using RNA sequencing.

17. A method for providing advice on the timing of embryo transfer, the method comprising: - The method according to any one of claims 13-16, predicting the endometrial implantation window time for the individual; and - Provide recommendations on the timing of embryo transfer based on the predicted implantation window.

18. The method as described in claim 17, characterized in that: - The individuals are patients receiving either natural or artificial cycles; and / or - The embryo transfer mentioned above is either a fresh or frozen embryo transfer.

19. The method as described in claim 17, characterized in that, The embryo transfer is either a cleavage embryo transfer or a blastocyst transfer.

20. The method according to any one of claims 17-19, characterized in that, The individual in question is a patient with repeated implantation failures.

21. The method as described in claim 20, characterized in that, The individual in question is a patient with recurrent implantation failure due to endometrial origin.

22. The method according to any one of claims 17-19, characterized in that, Compared to the normal WOI time, the individual has a WOI shift or a narrower WOI duration.

23. A system for predicting an individual's endometrial implantation window, comprising: - An endometrial implantation window prediction module, wherein the module comprises a WOI prediction model constructed according to any one of claims 1-12, or wherein the module is capable of performing the method according to any one of claims 13-22 to output the predicted endometrial implantation window time for the individual or to output the predicted optimal embryo transfer time interval.

24. The system as claimed in claim 23, characterized in that, The system also includes: A module for acquiring differentially expressed gene features related to the receptivity window, wherein the module is capable of extracting the expression levels of differentially expressed genes related to the receptivity window from endometrial tissue samples at a single sampling time point of the individual.

25. Use of the system as described in claim 23 or 24 for predicting an individual's endometrial implantation window time and / or optimal embryo transfer time interval.

26. Use of the system of claim 23 or 24 in the preparation of products for predicting an individual's endometrial implantation window time and / or optimal embryo transfer time interval, or for improving embryo transfer, or for improving embryo implantation rate and / or clinical pregnancy rate.

Images