Compositions, systems and methods for detecting ovarian cancer

Abstract

Compositions and methods for diagnosing epithelial ovarian cancer (EOC) and assessing the prognosis thereof are provided. Development of an artificial intelligence method named MethylBert and its use in the identification of methylation sites useful for the identification of EOC in free DNA obtained from serum or plasma are shown. A PCR-based method for the methylation status of the OV-1 site identified using the McthylBERT method can be used to screen EOC using free DNA obtained from a blood sample, and can be combined with the measurement of EOC-related tumor markers.

Description

[0001] This application claims the benefit of U.S. Provisional Application No. 63 / 409,986, filed Sep. 26, 2022. These and all other cited external materials are incorporated herein by reference in their entirety. When the definition or use of a term in a reference incorporated by reference is inconsistent with or contrary to the definition of that term provided herein, the definition of that term provided herein shall govern. Technical Field

[0002] The technical field of the present invention is methods for diagnosing cancer, particularly ovarian cancer. Background Art

[0003] The background description includes information that may be useful for understanding the present invention. This is not an admission that any of the information provided herein is prior art or relevant to the currently claimed invention, or that any of the publications specifically or implicitly referenced are prior art.

[0004] All publications herein are incorporated by reference to the same extent as if each individual publication or patent application was specifically and individually indicated to be incorporated by reference. When the definition or use of a term in an incorporated reference is inconsistent with or contrary to the definition of that term provided herein, the definition of that term provided herein shall apply rather than the definition of that term in the reference.

[0005] Ovarian cancer is a leading cause of cancer death in women, with an incidence of 313,000 cases in 2020 and over 200,000 deaths. 85%-95% of ovarian cancers arise from epithelial cells. Although breast and cervical cancers are more common, the 5-year survival rate of epithelial ovarian cancer (EOC) is much lower, making it more lethal to women compared to breast and cervical cancers. The high mortality and low 5-year survival rate of EOC are mainly related to late diagnosis, with over 80% of patients being diagnosed at an advanced stage. Based on currently available data, if EOC is diagnosed at stage I, the 5-year survival rate is approximately 90%. However, if diagnosed at the advanced stage III / IV, it rapidly drops to approximately 20%.

[0006] Currently, the serum biomarker CA125 and transvaginal ultrasound examination (TVU) are the two most commonly used EOC screening tests. Serum HE4 has also become an important serum biomarker for EOC diagnosis and is associated with recurrence detection. The sensitivity of using CA125 alone for EOC detection is low, while TVU is highly sensitive and accurate for EOC detection. However, due to its inconvenience, time-consuming nature, and the fact that the conclusions largely depend on the experience of the sonographer, routine TVU application is not suitable for first-line large-scale EOC screening in clinical practice. However, two large clinical studies found that annual evaluation using CA125 and TVU alone or in combination did not reduce EOC-related mortality. These findings highlight the urgent need for a highly sensitive and specific EOC test that is effective for early detection of EOC.

[0007] Circulating cell-free DNA (cfDNA) is extracellular nucleic acid fragments found in liquid biopsies. When tumor cells shed cfDNA, for example during apoptosis, they may be useful for cancer diagnosis because they contain the same genetic and epigenetic alterations as the tumor cells from which they originated. The potential use of cfDNA in EOC screening has shown some promising results. However, these studies have been limited by small sample sizes and sample bias towards advanced-stage EOC. Therefore, the utility of these tests for early EOC diagnosis has not been well characterized. In addition, there is a lack of comprehensive and systematic genome-wide analysis of EOC-specific methylation markers.

[0008] Therefore, there is still a need for sensitive and accurate methods for diagnosing epithelial ovarian cancer (EOC) using readily available samples. Summary of the Invention

[0009] The subject matter of the present invention provides devices, systems, and methods for diagnosing epithelial ovarian cancer (EOC) and / or providing EOC prognosis, wherein the methylation status of one or more sites identified by artificial intelligence screening using cell-free DNA (cfDNA) samples provided by healthy individuals and individuals with EOC is evaluated.

[0010] Some embodiments of the inventive concept include methods for assisting in the diagnosis of epithelial ovarian cancer (EOC), wherein cell-free genetic material is isolated from an individual, followed by characterizing the nucleic acid methylation of one or more genetic markers from the group of OV1, OV2, OV6, OV3, OV4, OV5, OV6, OV6, OV6, OV6, OV7, OV8, OV9, OV10, OV11, OV12, OV13, OV14, OV15, OV16, OV17, OV18, OV19, OV20, OV21, and OV22. Aberrant methylation of one or more genetic markers indicates epithelial ovarian cancer. In some embodiments, the cell-free genetic material is cell-free DNA. Such methods may include the step of treating at least a portion of the cell-free genetic material with bisulfite, and may include contacting the cell-free genetic material with nucleic acid primers and performing nucleic acid amplification on the bisulfite-treated and untreated samples to generate amplification products, the primers being complementary to a portion of the genetic material adjacent to at least one of OV1, OV2, OV6, OV3, OV4, OV5, OV6, OV6, OV6, OV6, OV7, OV8, OV9, OV10, OV11, OV12, OV13, OV14, OV15, OV16, OV17, OV18, OV19, OV20, OV21, and OV22. Such nucleic acid amplification steps may include contacting the resulting amplification products with a probe that is complementary to at least one of OV1, OV2, OV6, OV3, OV4, OV5, OV6, OV6, OV6, OV6, OV7, OV8, OV9, OV10, OV11, OV12, OV13, OV14, OV15, OV16, OV17, OV18, OV19, OV20, OV21, and OV22. In some embodiments, such testing is directed to OV1. In some embodiments, CA125 is also characterized, wherein an abnormally high level of CA125 indicates epithelial ovarian cancer.

[0011] Some embodiments of the inventive concept include methods for assisting in the assessment of the prognosis of an individual with epithelial ovarian cancer, wherein cell-free genetic material is isolated from the individual, followed by characterizing the nucleic acid methylation of one or more genetic markers from OV1, OV2, OV6, OV3, OV4, OV5, OV6, OV6, OV6, OV6, OV7, OV8, OV9, OV10, OV11, OV12, OV13, OV14, OV15, OV16, OV17, OV18, OV19, OV20, OV21, and OV22. Aberrant methylation of one or more genetic markers indicates a poor prognosis and / or advanced disease. In some embodiments, the cell-free genetic material is cell-free DNA. Such methods can include the step of treating at least a portion of the cell-free genetic material with bisulfite, and can include contacting the cell-free genetic material with nucleic acid primers and performing nucleic acid amplification on bisulfite-treated and untreated samples to generate amplification products, the primers being complementary to a portion of the genetic material adjacent to at least one of OV1, OV2, OV6, OV3, OV4, OV5, OV6, OV6, OV6, OV6, OV7, OV8, OV9, OV10, OV11, OV12, OV13, OV14, OV15, OV16, OV17, OV18, OV19, OV20, OV21, and OV22. Such nucleic acid amplification steps can include contacting the resulting amplification products with a probe that is complementary to at least one of OV1, OV2, OV6, OV3, OV4, OV5, OV6, OV6, OV6, OV6, OV7, OV8, OV9, OV10, OV11, OV12, OV13, OV14, OV15, OV16, OV17, OV18, OV19, OV20, OV21, and OV22. In some embodiments, such testing is directed to OV1. In some embodiments, CA125 is also characterized, wherein an abnormally high level of CA125 indicates a poor prognosis or advanced epithelial ovarian cancer.

[0012] Some embodiments of the present invention concept include the implementation of artificial intelligence algorithms to identify methylation patterns associated with EOC diagnosis and / or EOC prognosis. Such artificial intelligence algorithms can be implemented prior to the step of characterizing methylation in the above methods. In some such embodiments, such artificial intelligence algorithms can include correlations with chromosome embedding, positional embedding, methylation level embedding, and gene embedding, and wherein the methylation pattern comprises one or more genetic markers that exhibit methylation differences between individuals with EOC and individuals without EOC. Such artificial intelligence algorithms can include a matrix decomposition-based Transformer model, which reduces the computational complexity from quadratic to linear (O(L 2 )) to O(L)), which in turn enables efficient processing of large-scale data. In some embodiments, such a Transformer model comprises a Performer, which can include an attention mechanism, the formula of which is Att(Q,K,V) = softmax(Φ(Q)Φ(K) T / sqrt(d k ))V, where Q, K, and V represent query, key, and value matrices respectively, and Φ represents a feature map function that projects the input into a new space. Such a Performer can provide an effective approximation of dot-product attention. Such artificial intelligence algorithms can be applied to a training database comprising individuals identified as having EOC and individuals without EOC, and identify methylation patterns associated with individuals with EOC, wherein the artificial intelligence algorithms identify such methylation patterns that comprise one or more genetic markers that exhibit methylation differences between individuals with EOC and individuals without EOC.

[0013] Another embodiment of the inventive concept is a composition for the diagnosis or prognosis of epithelial ovarian cancer, comprising a first primer that is complementary to a first portion of a nucleic acid of a genetic marker adjacent to OV1, OV2, OV6, OV3, OV4, OV5, OV6, OV6, OV6, OV6, OV7, OV8, OV9, OV10, OV11, OV12, OV13, OV14, OV15, OV16, OV17, OV18, OV19, OV20, OV21, and OV22. Such a composition further comprises a probe that is complementary to at least one genetic marker of OV1, OV2, OV6, OV3, OV4, OV5, OV6, OV6, OV6, OV6, OV7, OV8, OV9, OV10, OV11, OV12, OV13, OV14, OV15, OV16, OV17, OV18, OV19, OV20, OV21, and OV22 and comprises a first dye or a first fluorophore. In some embodiments, the genetic marker is OV1. In some embodiments, the composition comprises a second primer that targets a second genetic marker of one of OV1, OV2, OV6, OV3, OV4, OV5, OV6, OV6, OV6, OV6, OV7, OV8, OV9, OV10, OV11, OV12, OV13, OV14, OV15, OV16, OV17, OV18, OV19, OV20, OV21, and OV22. In some such embodiments, the composition may comprise a second probe that is complementary to OV1, OV2, OV6, OV3, OV4, OV5, OV6, OV6, OV6, OV6, OV7, OV8, OV9, OV10, OV11, OV12, OV13, OV14, OV15, OV16, OV17, OV18, OV19, OV20, OV21, or OV22. In some such embodiments, the second probe may comprise a second dye or a second fluorophore.

[0014] The various objects, features, aspects, and advantages of the subject matter of the present invention will become more apparent from the following detailed description of some preferred embodiments and the accompanying drawings in which like numerals represent like elements. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 : Figure 1 A flowchart is provided that describes the use of LASSO to identify methylation sites useful for the diagnosis and prognosis of EOC status.

[0016] Figures 2A to 2F : Figure 2A A confusion table of the OCDP binary results in the training dataset is provided. Figure 2BA confusion table of the OCDP binary results in the validation dataset is provided. Figure 2C A typical ROC curve of the diagnostic prediction model with methylation markers in the training dataset is shown. Figure 2D A typical ROC curve of the diagnostic prediction model with methylation markers in the validation dataset is shown. Figure 2E A typical unsupervised hierarchical clustering of the seven methylation markers selected for the diagnostic prediction model is shown. Figure 2F A typical unsupervised hierarchical clustering of the seven methylation markers selected for the diagnostic prediction model in the validation dataset is shown.

[0017] Figures 3A to 3E : Figure 3A To 3 show the research results of OCDP and CA125 in different stages of EOC. Figure 3A A confusion table of the binary results of the diagnostic prediction model for different cancer stages of ovarian cancer and healthy women is provided. Figure 3B A typical ROC curve showing OCDP for differentiating early-stage (red) or late-stage (blue) EOC patients from healthy women is shown. Figure 3C A typical ROC curve showing OCDP (red) and the combination of OCDP and CA125 (blue) for differentiating early-stage EOC from healthy women is shown. Note: Only samples with CA125 information are summarized in this figure. Figure 3D Describes the CA12 levels of healthy women samples and EOC samples at different stages. Figure 3E Describes the OCDP cd-score values of CA12 in healthy women samples and EOC samples at different stages.

[0018] Figures 4A to 4D : Figures 4A to 4D Shows the research results of the utility of OCPP for predicting the prognosis of EOC. Figure 4A An exemplary Kaplan-Meier plot of the overall survival of EOC patients in the low-risk and high-risk groups determined by OCPP in the training dataset is shown. Figure 4B An exemplary Kaplan-Meier plot of the overall survival of EOC patients in the low-risk and high-risk groups determined by OCPP in the validation dataset is shown. Figure 4C A typical ROC curve and the corresponding AUC for predicting the 5-year survival rate of early EOC by OCPP cp score and CA125 are shown. Figure 4D The results of a multivariate analysis of the survival rate of early EOC with covariates including OCPP cp score and CA125 are shown.

[0019] Figures 5A to 5G : Figure 5E Shows data related to the performance of ddPCR assays using OV1 in differentiating ovarian cancer from healthy women. Figure 5A Shows the typical ROC curves of OV1 in the training (red) and validation (blue) datasets of the ddPCR cohort. Figure 5B Provides a confusion table of the binary results of the OV1 prediction model in the training dataset. Figure 5C Provides a confusion table of the binary results of the OV1 prediction model in the validation dataset. Figure 5D Shows the typical ROC curves of OV1 and the OV1-CA125 combination in differentiating early (red and green) and late (blue and purple) EOC from healthy women in the ddPCR cohort. Figure 5E Provides a confusion table summarizing OV1 in differentiating early and late EOC. Figure 5F Provides a summary of the sensitivity and specificity results of CA125, OVA-1 ddPCR, and the CA125 and OVA-1 PCR combination in healthy, early EOC, and late EOC. Figure 5G Shows a Beewarm plot presenting the OV1 methylation levels between ovarian cancer and healthy women in the ddPCR cohort, with the red plot representing healthy female samples and the blue plot representing EOC samples.

[0020] Figure 6 : Figure 6 Provides a flow chart that describes the identification of methylation sites useful for EOC status diagnosis and prognosis using MethylBERT.

[0021] Figures 7A to 7G : Figures 7A to 7G Shows the cfDNA methylation analysis of the MethylBERT-EOC diagnostic model. Figure 2A Provides a schematic overview of the MethylBERT model. Figure 7B Provides a confusion table of the binary results of the MethylBERT-EOC diagnostic model in the training dataset. Figure 7C Provides a confusion table of the binary results of the MethylBERT-EOC diagnostic model in the validation dataset. Figure 7D Shows the typical ROC curves of the MethylBERT-EOC diagnostic model in the EOC diagnosis prediction training and validation datasets. Figure 7E Shows the typical ROC curves of the MethylBERT-EOC diagnostic model in the training dataset for early and late EOC diagnosis prediction. Figure 7F Shows the typical ROC curves of the MethylBERT-EOC diagnostic model in the validation dataset for early and late EOC diagnosis prediction. Figure 7GProvided are EOC prediction scores for healthy female samples and EOC samples at different stages based on MethylBERT-EOC diagnosis.

[0022] Figure 8 : Figure 8 Shows typical data of studies characterizing the relative sensitivity of CA125 measurement and the application of the MethylBERT model for EOC.

[0023] Figures 9A to 9F : Figures 9A to 9D Shows the results of cfDNA methylation analysis using the LASSO-EOC diagnostic model in an individual cohort. Figure 9A Provides a confusion table of the binary results of the LASSO-EOC diagnostic model in the training dataset. Figure 9B Provides a confusion table of the binary results of the LASSO-EOC diagnostic model in the validation dataset. Figure 9C Shows typical ROC curves of the LASSO-EOC diagnostic model in EOC diagnostic prediction training and validation datasets. Figure 9D Shows typical ROC curves of the LASSO-based EOC diagnostic model in early and late EOC diagnostic prediction training datasets. Figure 9E Shows typical ROC curves of the LASSO-based EOC diagnostic model in early and late EOC diagnostic prediction in the validation dataset. Figure 9F Provides the results of EOC prediction scores for healthy female samples and EOC samples at different stages based on LASSO-EOC diagnosis. Detailed implementation

[0024] The subject matter of the present invention provides devices, systems, and methods in which circulating free DNA obtained from conveniently accessible serum or plasma samples can be used to diagnose epithelial ovarian cancer (EOC) in a sensitive and accurate manner. The inventors have identified a panel of DNA methylation modifications that can be characterized using appropriate methods (e.g., DNA sequencing, hybridization, PCR, etc.) to determine whether an individual has early-stage (grade I or II) or late-stage (grade III or IV) epithelial ovarian cancer. This panel is particularly useful in identifying early-stage diseases that are more treatable but relatively asymptomatic. By combining with the test results of cancer markers associated with EOC (e.g., CA-125 and / or HE4), the accuracy and sensitivity of this panel can be improved.

[0025] In addition, the inventors have determined that characterizing methylation at a single locus (e.g., at the designated OV1 locus), such as by DNA amplification using methylation-specific primers and / or probes, can provide a simplified detection method that can provide results useful for accurate and sensitive diagnosis of EOC at early and late stages, particularly when paired with the detection of EOC-related cancer markers (e.g., CA125 and / or HE4).

[0026] The inventors performed an AI-assisted genome-wide survey by screening 3.3 million methylated CpG sites to identify a panel of EOC-specific methylation markers using cell-free DNA (cfDNA) samples from EOC patients and healthy female subjects.

[0027] In an exemplary study, the inventors used eight selected methylation markers as a diagnostic panel based on a training dataset of 471 EOC patients and 742 healthy female subjects that showed 86.62% sensitivity and 89.08% specificity (AUC = 0.933) and a validation dataset (283 EOC patients and 376 healthy females) that showed 84.45% sensitivity and 87.77% specificity (AUC = 0.926). Importantly, the detection rate was higher for early-stage EOC (I / II stage) (72.68%, n = 205) compared to all-stage EOC patients (85.81%, n = 754). In contrast, the sensitivities of the conventional CA-125 and HE-4 markers were 48.81% and 46.85% respectively (combined sensitivity of 60.6%). Additionally, among the 282 EOC patients missed by CA-125 / HE4, the panel was able to detect 235 of these patients (83.33%). When the cfDNA methylation diagnostic panel was combined with CA125, the detection sensitivity was further improved (from 73.22% to 84.15% for early EOC patients and from 91.01% to 95.51% for advanced stage III / IV EOC patients). The most important methylation site OV1 in our panel demonstrated the potential adaptability of this method for cost-effective and time-saving droplet digital PCR (ddPCR) assays. These results document the potential of using cfDNA methylation patterns as a panel of first-line screening assays for early detection and diagnosis of EOC.

[0028] Various objects, features, aspects, and advantages of the subject matter of the present invention will become more apparent from the following detailed description of some preferred embodiments and the accompanying drawings in which like numerals represent like elements.

[0029] The following description includes information that may be helpful in understanding the present invention. This does not mean an admission that any of the information provided herein is prior art or information relevant to the presently claimed invention, or that any publication specifically or implicitly referenced is prior art.

[0030] In some embodiments, the numbers used to describe and claim certain properties of the present invention, such as amounts of ingredients, concentrations, reaction conditions, etc., should be understood to be modified in some instances by the term "about". Thus, in some embodiments, the numerical parameters set forth in the written description and the appended claims are approximations that can vary depending upon the desired properties sought to be obtained by a particular embodiment. In some embodiments, such numerical parameters should be construed in accordance with the significant digits reported and by applying ordinary rounding techniques. Although the numerical ranges and parameters setting forth broad ranges of the present invention are approximations, the numerical values set forth in the specific examples are reported as precisely as possible. The numerical values given in some embodiments of the present invention may contain certain errors necessarily resulting from the standard deviation found in their respective testing measurements.

[0031] As used herein and throughout the appended claims, unless the context clearly dictates otherwise, a noun in the singular without a quantifier means one or more. Further, as used herein, unless the context clearly dictates otherwise, the meaning of "in" includes "in" and "on".

[0032] The recitation of numerical ranges herein is merely intended to be a shorthand method of referring individually to each separate numerical value falling within the range. Unless otherwise indicated herein, each separate value is incorporated into the specification as if it were individually recited herein. Unless otherwise indicated or clearly contradicted by context, all methods described herein may be performed in any suitable order. The use of any and all examples, or exemplary language (e.g., "such as") provided herein with respect to certain embodiments is merely intended to better illuminate the invention and does not pose a limitation on the scope of the invention unless otherwise claimed. No language in the specification should be construed as indicating any non-claimed element as essential to the practice of the invention.

[0033] The grouping of alternative elements or embodiments of the invention disclosed herein should not be construed as a limitation. Each member of the group can be referred to and claimed individually or in any combination with other members of the group or other elements found herein. For reasons of convenience and / or patentability, one or more members of the group may be included in or deleted from the group. When any such inclusion or deletion occurs, the specification is deemed to include the modified group in order to satisfy the written description of all Markush groups used in the appended claims.

[0034] It should be noted that any language directed to a computer should be understood to include any suitable combination of computer devices, including servers, interfaces, systems, databases, agents, peers, engines, controllers, or other types of computer devices that operate individually or in combination. It should be understood that the computer device includes a processor configured to execute software instructions stored on a tangible, non-transitory computer-readable storage medium (e.g., hard disk, solid-state drive, RAM, flash memory, ROM, etc.). The software instructions preferably configure the computer device to provide the functions, tasks, or other operations of the disclosed device as discussed below. In a particularly preferred embodiment, multiple servers, systems, databases, or interfaces exchange data using standardized protocols or algorithms, possibly based on HTTP, HTTPS, AES, public-private key exchange, web service APIs, known financial transaction protocols, or other methods of electronic information exchange. The data exchange preferably occurs over a packet-switched network, the Internet, a LAN, a WAN, a VPN, or other types of packet-switched networks.

[0035] It should be understood that the disclosed technology provides a number of advantageous technical effects, including providing a sensitive and accurate diagnosis of the presence and prognosis of early-stage and late-stage epithelial ovarian cancer using readily available serum or plasma samples.

[0036] The following discussion provides many exemplary embodiments of the subject matter of the present invention. Although each embodiment represents a single combination of elements of the invention, the subject matter of the present invention is considered to include all possible combinations of the disclosed elements. Thus, if one embodiment includes elements A, B, and C, and a second embodiment includes elements B and D, then the subject matter of the present invention is also considered to include the other remaining combinations of A, B, C, or D, even if not explicitly disclosed.

[0037] Some embodiments of the present inventive concept include compositions, methods, or systems that use a set of DNA methylation markers (OV1, OV2, OV36, OV4, OV5, OV6, OV7, OV8, OV9, OV10, O11, OV12, OV13, OV14, OV15, OV16, OV17, OV18, OV19, OV20, OV21, and / or OV22), a subset of these methylation markers, or a single methylation marker in a detection panel, which have specific utility in diagnosing epithelial ovarian cancer (in early and late stages) and / or its prognosis. In a preferred embodiment, cell-free DNA obtained from a patient, such as cell-free DNA obtained from a blood, serum, or plasma sample, is used.

[0038] In some embodiments, such detection can comprise the steps of: dividing a blood, serum, or plasma sample into at least two parts, subjecting the cfDNA of one of these parts to bisulfite treatment, and selectively amplifying the cfDNA of the untreated sample and the bisulfite-treated sample at one or more loci using primers and / or probes specific for methylation and bisulfite-treated DNA, wherein the methylation differences between the EOC and control groups are associated with EOC (e.g., OV1, OV2, OV36, OV4, OV5, OV6, OV7, OV8, OV9, OV10, O11, OV12, OV13, OV14, OV15, OV16, OV17, OV18, OV19, OV20, OV21, and / or OV22). The results of these studies can be combined with the results of tests (e.g., immunoassays) for EOC-related cancer markers (e.g., CA125 and / or HE4).

[0039] Such methods rely on the identification of the methylation status of these markers, which can be characterized using any suitable technique. Such techniques include, but are not limited to: sequencing (before and after bisulfite treatment), DNA amplification using methylation-specific primers (e.g., PCR and related methods), amplification using primers that hybridize proximal (e.g., "upstream") to the marker being characterized and detection using methylation-specific probes (e.g., probe sequences including dyes or fluorophores), amplification using primers that hybridize proximal (e.g., "upstream") to the marker being characterized followed by hybridization to a methylation-specific probe sequence coupled to a solid phase, amplification using primers that hybridize proximal (e.g., "upstream") to the marker being characterized followed by electrophoresis and transfer to a membrane, amplification using primers that hybridize proximal (e.g., "upstream") to the marker being characterized followed by characterization of the amplification product using mass spectrometry, etc.

[0040] The compositions of the inventive concept may comprise one or more primer sequences for DNA amplification reactions. Such primer sequences may be located proximal to the methylation marker site and may be derived from sequence data based on the position of the methylation site in the genome (once that position is identified) and the known effects of bisulfite treatment using methods known in the art. In some embodiments, the primer sequences may be methylation-specific primers and may be applied to untreated and / or bisulfite-treated samples. Such primer sequences may be provided as a pair of primers, including a primer complementary to the DNA strand complementary to the methylation marker site, such that repeated amplification (e.g., as in PCR) results in amplified DNA of a characteristic size. The compositions of the inventive concept may comprise one or more probe sequences, including a portion complementary to the methylation marker site. In some embodiments, such probe sequences may comprise a dye, fluorophore, or other observable label that provides a characteristic signal. In some embodiments, two or more such probe sequences carrying different observable labels may be provided, which may support multiplex detection. In some embodiments, such probe sequences may be coupled to an insoluble support (e.g., microarray, microparticle, etc.). In some such embodiments, multiple probe sequences may be provided on a solid support that is encoded (e.g., by position within the array, dye content, and / or size of the microparticle, etc.), allowing identification of the relevant probe sequences.

[0041] Some embodiments of the inventive concept, in addition to the components for immunoassays directed against one or more tumor markers associated with EOC (e.g., CA125 and / or HE4), further comprise components selected to identify the methylation status of a specified site in cfDNA (as described above). These components include, but are not limited to, a primary antibody directed against the tumor marker, a secondary antibody directed against the primary antibody, radiolabeled conjugates, enzyme conjugates, fluorescent conjugates, and / or luminescent conjugates.

[0042] The system of the present inventive concept may comprise reagents including the above-described composition and auxiliary reagents (e.g., buffers, enzymes, etc.). In some embodiments, such a system may comprise liquid handling equipment (pipettes, etc.) for measuring and dispensing reagents and / or test samples into test containers. In some embodiments, such a system may comprise a subsystem for performing a test reaction (e.g., a thermal cycler, a hybridization incubator, a washing subsystem, etc.) and / or an analysis subsystem for detecting or characterizing signals obtained as a result of the characterization reaction (e.g., a colorimeter, a spectrophotometer, a fluorometer, a particle sorter, etc.). These systems may comprise a controller that communicates with these subsystems and controls their actions. Such a controller may comprise a database of test protocols and may record, store, and / or report test results. In some embodiments, such a controller may comprise an algorithm for analyzing test results and may report the probability that an individual has early and / or late-stage EOC and, in some embodiments, may report the prognosis of the individual. In some embodiments, such an algorithm may comprise or be implemented by an artificial intelligence (AI) algorithm.

[0043] Identifying cfDNA Methylation Markers for EOC Using a Sample Pool Strategy

[0044] As Figure 1 shown, the inventors employed a sensitive detection method (TMC- Kit) for the discovery of EOC methylation site markers. One limitation of the current technology is that at least 500 ng of DNA is required for library construction, which is difficult to obtain from a single serum sample. To address this issue, the inventors constructed several libraries using pooled cfDNA samples, obtained 5 early-stage pooled samples and 6 late-stage pooled samples from a total of >200 - 220 EOC patients, and obtained cfDNA from healthy subjects from 10 healthy female pooled samples (total number of healthy subjects included >200). These pooled sample libraries are referred to as the pool cohort in the following (Table 1).

[0045] Table 1: Demographic characteristics of three retrospective cohorts

[0046]

[0047] For each of the pooled samples, over 3.3 million CpG sites were examined and our sequencing results gave an average of 63.3 (21.7 - 126.2) reads per CpG position. Thus, when comparing cfDNA sample pools of EOC patients and healthy females, the inventors identified over 268,039 differentially methylated loci (DMLs) or 21,104 differentially methylated regions (DMRs) with a methylation value difference > 10%.

[0048] Verification of Methylation Markers in Individual cfDNA Samples

[0049] Since these markers were identified from the pooled cohort, it was crucial to validate them in individual samples to ensure that the representation in the pooled samples was relevant to the individuals. Due to the limited plasma samples for each individual, 500 CpG sites were selected as candidate markers for individual validation based on the methylation differences and P-values between the EOC and healthy pooled samples. Thus, the corresponding probes for these markers were designed and synthesized and applied to detect cfDNA samples of 1909 individuals, all of which were completely different from the samples that constituted the pooled cohort. After combining the sequencing results with UMIs, the probes successfully captured 493 out of 500 candidate markers in these individual samples, and for 1872 samples (754 EOC patients and 1118 healthy females), there were over 10 reads per CpG position on average, and these samples were retained for subsequent analysis. These individual samples are referred to as the individual cohort in the following text (Table 1). The results showed good consistency in methylation changes between the individual cohort and the pooled cohort, and the inventors confirmed 165 markers with significant methylation differences (difference > 10%, P < 0.05).

[0050] The inventors also studied the methylation status of 500 CpG positions in seven ovarian tumor cell lines and one ovarian epithelial cell line. Positions with methylation differences > 10% between the epithelial cell line and more than four of the seven tumor cell lines were retained and overlapped with the 165 markers confirmed in the individual samples. As a result, 33 additional markers were identified as potential cfDNA markers.

[0051] Construction of an EOC Diagnostic Panel (OCDP)

[0052] LASSO-based selection: In some embodiments, the inventors randomly divided the above-detected individual cohort data into a training data set (471 EOC and 742 healthy female samples) and a validation data set (283 EOC and 376 healthy female samples). In the training data set, the inventors analyzed 33 ctDNA markers by least absolute shrinkage and selection operator (LASSO) and random forest to distinguish EOC patient samples from healthy female subject samples. In more than 90 out of 100 LASSO runs, the identified markers overlapped with those given by random forest analysis, and the retained markers were applied to logistic regression for diagnostic modeling. Based on this method, eight markers (OCDP, Table 2) that may be suitable for an EOC diagnostic test panel were selected. In some embodiments, all eight markers are used as at least part of an EOC diagnostic test panel. In some embodiments, a portion (e.g., two or more) of these markers is used as at least part of an EOC diagnostic test panel.

[0053] Table 2: Characteristics of eight methylation markers in OCDP and their coefficients in EOC diagnosis

[0054]

[0055] In some embodiments, single markers (e.g., OV1) among the markers listed in Table 2 can be used as at least part of an EOC diagnostic test panel. Such a diagnostic test panel can be used for EOC screening studies, EOC staging, and / or monitoring the treatment of EOC. Such a diagnostic test panel can incorporate other cancer markers known in the art related to EOC. Such a diagnostic test panel can be implemented using any suitable DNA detection and / or identification technique, such as DNA amplification, multiplex DNA amplification, etc. In the training data set, the inventors observed that the sensitivity of OCDP for diagnosing EOC was 86.62% and the specificity was 89.08% (AUC = 0.933), and when applied to the validation data set, very similar sensitivities (84.45%) and specificities (87.77%) (AUC = 0.9257) were also observed ( Figures 2A to 2F ). Figure 2A and 2B respectively provide the confusion tables of the binary results of OCDP in the training and validation data sets. Figure 2C and 2D respectively show the typical ROC curves of the diagnostic prediction models with methylation markers in the training and validation data sets. Figure 2E and 2FUnsupervised hierarchical clustering of seven methylation markers selected for the diagnostic prediction model in the training and validation datasets is shown separately.

[0056] In addition, in the training dataset, the positive predictive value (PPV), negative predictive value (NPV), and false positive rate (FPR) of OCDP were 0.83, 0.91, and 0.11, respectively, and in the validation dataset, they were 0.84, 0.89, and 0.12, respectively. Since the EOC and healthy individual datasets were not strictly age-matched, to evaluate whether our OCDP would be affected by age, the inventors evaluated the cd-score of OCDP in different age groups in a cohort of healthy female subjects and no significant differences were observed between different age groups.

[0057] Another important observation is that significant differences in methylation were also found at CpG positions (differentially methylated regions or DMRs) around these identified markers, which implies a functional correlation between the methylation status of these markers and gene expression. In fact, among the 6 DMRs that could be annotated to genes or gene regulatory regions, the expression of most of the corresponding genes was significantly different in ovarian tissues from 426 ovarian cancer patients and 88 healthy women.

[0058] To determine the performance characteristics of the ovarian cancer diagnostic detection panel (OCDP) as shown in Table 2 with reference to disease stage, the inventors first reviewed early and late EOC in the training and validation datasets separately, and found that the ROC curves for any stage were relatively consistent between the training and validation datasets. The results are as Figures 3A to 3E shown. Figure 3A To Figure 3 shows the results of OCDP and CA125 for different stages of EOC. Figure 3A A confusion table of the binary results of the diagnostic prediction model for different stages of ovarian cancer and healthy women is provided. Figure 3B Shows the typical ROC curve of OCDP differentiating early-stage (red) or late-stage (blue) EOC patients from healthy women. Figure 3C Shows the typical ROC curve of OCDP (red) and the combination of OCDP and CA125 (blue) differentiating early-stage EOC and healthy women. Note: This figure only summarizes samples with CA125 information. Figure 3D Describes the CA12 levels of healthy female samples and EOC samples at different stages. Figure 3E Describes the CA12 OCDP cd-score values of healthy female samples and EOC samples at different stages.

[0059] Overall, in the individual cohort, the OCDP shown in Table 2 successfully identified 149 out of 205 early-stage EOCs (sensitivity of 72.68%) and 498 out of 549 late-stage EOCs (sensitivity of 90.71%), with an overall specificity of 88.64%( Figure 3A and 3B ). Additionally, the inventors estimated the corresponding sensitivities of the OCDP at different specificities (85%, 90%, 95%) in the individual cohort and found that if the inventors adjusted the specificity of the OCDP to >95%, its sensitivity would still be as high as over 75% (Table 3).

[0060] Table 3: Corresponding sensitivities of the OCDP for differentiating EOC from healthy women when the specificity exceeds 85%, 90%, and 95%

[0061]

[0062] When the OCDP shown in Table 2 was used to evaluate its diagnostic ability in EOC and endometriosis samples, there was not only a significant difference in the cd-score between the EOC samples (n = 754) in the individual cohort and the endometriosis samples (n = 96) in the independent cohort, but also it was observed that when the OCDP differentiated EOC and endometriosis samples, the specificity was 75% and the sensitivity was 85.81%.

[0063] Performance comparison of the OCDP identified in the LASSO analysis with conventional serum biomarkers: The CA125 and HE4 tests are commonly used EOC screening tests in clinical practice, although their sensitivities are not satisfactory. When the inventors compared the diagnostic performance of CA125 and HE4 with the OCDP shown in Table 2 in 715 samples of the individual cohort with complete CA125 and HE4 information, the sensitivities provided by CA125 alone or HE4 alone were only 48.81% and 46.85% respectively. When CA125 and HE4 were used in combination, the sensitivity increased to 60.6%. In contrast, our OCDP showed a sensitivity of 86.71% in these samples. Importantly, among the 282 EOC samples missed by the CA125 and HE4 tests, the OCDP correctly identified / diagnosed 235 cases (83.33%).

[0064] The inventors did not observe a significant difference in the CA125 biomarker levels between early and late-stage EOCs( Figure 3D ). However, compared with early-stage EOC, the inventors observed a higher cd score level in the late stage with the OCDP( Figure 3E ), which indicates a correlation with tumor burden.

[0065] Next, the inventors evaluated whether the combination of OCDP shown in Table 2 with CA125 could improve the sensitivity of the resulting combined test assay. In a cohort sample of 1,058 individuals with CA125 information, compared to OCDP alone, the sensitivity of our combination of OCDP and CA125 increased by 6.14% (from 86.47% to 92.61%, n = 717), but the specificity decreased by 3.52% (from 90.62% to 87.1%, n = 341). Interestingly, for patients with early-stage EOC (n = 183), the specificity of the combined model was 87.1%, and the sensitivity increased from 73.22% to 84.15%. More importantly, when the specificity exceeded approximately 90% or 95%, this sensitivity could be as high as nearly 80% or 70% ( Figure 3C , Tables 4 and 5).

[0066] Table 4: Specificity and sensitivity of OCDP and CA125 in differentiating ovarian cancer and healthy female samples

[0067]

[0068] Note: This table only summarizes samples with CA125 information. Early: Early-stage EOC; Advanced: Advanced-stage EOC. Table 5: Corresponding sensitivities of the OCDP + CA125 combined model in differentiating EOC and healthy females when the specificity exceeds 85%, 90%, and 95%

[0069]

[0070] Prognostic prediction of EOC: To study the prognostic prediction potential of 500 CpG positions selected from the pooled cohort, the inventors investigated their methylation status in the above cell lines. The results are as Figures 4A to 4D shown. Figure 4A Shows an exemplary Kaplan-Meier plot of the overall survival of EOC patients in the low-risk and high-risk groups determined by OCDP shown in Table 2 in the training dataset. Figure 4B Shows a typical Kaplan-Meier plot of the overall survival of EOC patients in the low-risk and high-risk groups determined by OCPP in the validation dataset. Figure 4C Shows a typical ROC curve and corresponding AUC for predicting the 5-year survival rate of OCPP cp score, CA125, and early EOC. Figure 4D Shows the results of a multivariate analysis of the survival rate of early EOC with covariates including OCPP cp score and CA125.

[0071] One hundred and twenty-four markers were selected for prognostic analysis because they showed more than 10% methylation changes between EOC and healthy women in the individual cohorts, and this change was consistent in more than four of the seven ovarian tumor lines. Four hundred and thirty-seven EOC patients in the individual cohort with complete survival information were randomly divided into training and validation data sets at a ratio of 2:1. UniCox and LASSO were applied to reduce the dimensions to three markers, and a Cox model for the EOC prognostic test panel (OCPP) was constructed (Table 6). Kaplan-Meier curves ( Figure 4A and 4B ) were generated using the combined prognosis score (cp-score) of OCPP in the training and validation data sets. The high-risk group had 135 observations with 61 events in the training data set and 55 observations with 24 events in the validation data set; while the low-risk group had 157 observations with 24 events in the training data set and 90 observations with 23 events in the validation data set. By performing log-rank tests in the training data set (P < 0.0001) and the validation data set (P = 0.012), the median survival time of the low-risk group was significantly shorter than that of the high-risk group ( Figure 4A and 4B ).

[0072] Table 6: Characteristics of three methylation markers and their coefficients in EOC prognosis

[0073]

[0074] In some embodiments, a single marker among the markers listed in Table 6 can be used as at least part of an EOC diagnostic or prognostic test panel. In some embodiments, two or more markers among the markers listed in Table 6 can be used as at least part of an EOC diagnostic or prognostic test panel. Such a test panel can incorporate other cancer markers known in the art related to EOC. Such a diagnostic or prognostic test panel can be implemented using any suitable DNA detection and / or identification technique, such as DNA amplification, multiplex DNA amplification, etc.

[0075] Meanwhile, CA125 and cancer stage are also effective in predicting the prognosis of EOC. In terms of AUC, HR, and the separation of the KM curves, compared with OPCC, these two factors showed similar abilities in predicting the prognosis of EOC, and in combination with OPCC, CA125 and cancer stage significantly improved the prognosis prediction efficiency (Figures S10A-D, S11A-D). On the other hand, when the inventors observed early-stage EOC (17 events out of 126 observations), OPCC was still effective in predicting its prognosis, with an AUC of 0.684 and an HR of 2.88 (95% CI: 1.04 - 7.95), although the effectiveness of CA125 in predicting the prognosis of these observations was reduced.

[0076] Prognosis prediction of EOC using the markers identified by LASSO and UniCox: The inventors also determined the prognostic prediction potential of 493 methylation markers selected from the pooled cohort. 151 markers were used for prognostic analysis because they showed more than 10% methylation changes between EOC and healthy women in the individual cohort. 437 EOC patients with complete survival information in the individual cohort were randomly divided into training and validation datasets at a ratio of 2:1. UniCox and LASSO were applied to reduce the dimensions to three markers, and a Cox model for the EOC prognostic detection panel (OCPP) was constructed ( Figure 4A ). The Kaplan-Meier curves were generated in the training and validation datasets using the combined prognostic score (cp-score) of OCPP ( Figure 4B and Figure 4C ). The high-risk group had 135 observations with 61 events in the training dataset and 55 observations with 24 events in the validation dataset. The low-risk group had 157 observations with 24 events in the training dataset and 90 observations with 23 events in the validation dataset. By performing log-rank tests in the training dataset (P < 0.0001) and the validation dataset (P = 0.012), the median survival time of the high-risk group was significantly shorter than that of the low-risk group ( Figure 4B and Figure 4C ).

[0077] Selection Based on MethylBERT

[0078] Binary classification logistic regression after dimension reduction based on LASSO is a classic diagnostic model construction strategy, but it has certain limitations. Due to the limitation of the number of events per variable (EPV) (the ratio of the number of features to the sample size), this method is restricted by the number of biomarker that can be used for modeling. For example, a prediction model based on logistic regression analysis usually requires >10 EPV to obtain good prediction. Therefore, in a study of a logistic regression model with a sample size of 754 (such as the EOC cohort described in this article), the number of methylation markers that can be selected is no more than 75. In addition, when sequencing cfDNA samples extracted from 2 mL of plasma to saturation, it was found that less than 35% of the DNA base depths in the whole genome exceeded 10X. This means that analyzing cfDNA sequencing data in this way may not expose a large number of methylation markers. Therefore, a method that is not restricted by the number of input markers and can also simulate undetected methylation sites from detected methylation sites would be more suitable for a diagnostic model relying on cfDNA methylation markers.

[0079] The inventors developed a method for discovering methylation markers using a deep learning language model, which can overcome the above limitations of the conventional LASSO method. Figure 6 An overview of the derivation of the model is provided. The inventors adopted bidirectional encoder representations from transformers (BERT), a Transformer-based language pre-training model that can learn a wide range of clinical and biological knowledge as well as feature representations, and applied the BERT paradigm to analyze all available cancer DNA methylation datasets to utilize the extensive knowledge and interactions between chromosomes, positions, methylation levels, and gene functions in more than 90,000 cancer samples from GEO and TCGA. The inventors then constructed a model that enables the system to learn the representation of individual methylated CpG sites and the relationships between multiple CpG-CpG sites. This is named MethylBERT( Figure 7A ). Thereafter, MethylBERT was applied to analyze the training dataset randomly selected from 2 / 3 of the individual cohort, using a fully connected layer and a sigmoid activation function for binary classification, assigning each sample an EOC probability, and thus constructing the MethylBERT-EOC diagnostic model. Figures 7A to 7D Aspects of the MethylBERT model, the analysis using MethylBERT, and typical results are described.

[0080] Figure 7AA schematic overview of the MethylBERT model is provided. More than 110,000 WGBS and RRBS data are collected from GEO and TCGA datasets (upper left). Their chromosomal embeddings, positional embeddings, methylation level embeddings, and gene embeddings are combined into a CpG site embedding scheme (upper right), which is input into a Transformer model based on matrix factorization. A certain proportion of CpG sites are randomly masked and pre-trained to predict the methylation levels of the masked CpG sites given the context of their surrounding CpG sites (lower left). The pre-trained model (MethylBERT) is fine-tuned to process the methylation data of input samples (lower right). Figures B and C respectively provide confusion tables of the binary results of the MethylBERT-EOC diagnostic model in the training and validation datasets. Figure 7D Describes the typical ROC curves of the MethylBERT-EOC diagnostic model in the EOC diagnostic prediction training and validation datasets. Figure 7E and 7F Provides the typical ROC curves of the training and validation datasets of the MethylBERT-EOC diagnostic model in the early and late EOC diagnostic predictions. Figure 7G Shows the EOC prediction scores of healthy female samples and EOC samples at different stages based on the typical MethylBERT-EOC diagnosis. (**P < 0.01, ***P < 0.001, ****P < 0.0001; NS, not significant)

[0081] The MethylBERT-EOC diagnostic model showed 93.24% sensitivity and 95.3% specificity (AUC = 0.98) in the training dataset (consisting of 503 EOC and 744 healthy female samples), and 89.24% sensitivity and 94.39% specificity (AUC = 0.97) in the validation dataset (consisting of 251 EOC and 374 healthy female samples) ( Figures 7B to 7D ). In the training dataset, the positive predictive value (PPV), negative predictive value (NPV), and false positive rate (FPR) were 93.06%, 95.3%, and 4.71% respectively, and in the validation dataset were 91.43%, 94.39%, and 5.53% respectively. In addition, in the training dataset, there were 132 early EOC samples, and the MethylBERT-EOC diagnostic model could successfully diagnose 111 of them (84.09% sensitivity), while in the 73 early EOC samples in the validation dataset, the model diagnosed 58 of them (79.45% sensitivity) ( Figure 7C 、 7E and 7F).

[0082] Since the EOC and healthy subjects in the individual cohorts were not strictly age-matched, the potential impact of age differences on the MethylBERT-EOC diagnostic model was evaluated. No significant differences were observed between different age groups. In addition, the corresponding estimated sensitivities of the MethylBERT-EOC diagnostic model at different specificities (85%, 90%, 95%, 99%) showed that when the specificity was greater than 99%, the sensitivity was similar to the estimate when the sensitivity in the validation dataset exceeded 70% (Table 7).

[0083] Table 7

[0084]

[0085] To determine whether the MethylBERT-EOC diagnostic model could distinguish EOC from other gynecological diseases, the inventors detected 493 markers in 96 cfDNA samples of endometriosis and applied their methylation data to the model. Compared with the EOC samples in the validation dataset, the EOC probability in these endometriosis samples was not only significantly reduced, but also a sensitivity of 89.24% and a specificity of 91.66% were observed in distinguishing EOC samples from endometriosis samples.

[0086] The inventors believe that the MethylBERT artificial intelligence method can be used in the EOC diagnostic model to accurately and specifically identify cfDNA samples obtained from individuals with EOC, monitor the EOC treatment of individuals with the disease, and / or assist in determining the prognosis of individuals with EOC. The inventors further believe that the MethylBERT artificial intelligence method is generally applicable to disease states believed to be related to methylation at specific loci within the genome. The MethylBERT artificial intelligence algorithm allows the analysis of larger datasets than previous methods, providing higher sensitivity, higher sensitivity and / or reducing the time required for analysis and / or the time required to identify affected individuals on a given computational platform related to current DNA diagnostic methods.

[0087] Performance Comparison of MethylBERT Analysis with Conventional Serum Biomarkers: As mentioned above, CA125 and HE4 tests are commonly used EOC screening tests in clinical practice, although their sensitivities are not satisfactory. The inventors compared the diagnostic sensitivities of CA125 and HE4 with the MethylBERT-EOC diagnostic model (shown in Table 10) using 715 EOC samples with complete CA125 and HE4 information in an individual cohort. For these samples, the sensitivities provided by CA125 alone or HE4 alone were 48.81% and 46.85% respectively. When CA125 and HE4 were used in combination, the sensitivity increased to 60.6%. In contrast, the sensitivity of the MethylBERT-EOC diagnostic model in these samples was 92.73%. Importantly, among the 282 EOC samples missed by the CA125 and HE4 tests, the MethylBERT-EOC diagnostic model correctly diagnosed 255 cases (90.43%) ( Figure 8 ).

[0088] In an individual cohort (n = 1058, with CA125 information), the inventors did not observe a significant difference between early and late stage EOC. However, applying the MethylBERT-EOC diagnostic model to the same EOC samples showed a higher likelihood of EOC in late stage EOC ( Figure 7G ), indicating a potential correlation with tumor burden. This suggests that the MethylBERT-EOC diagnostic model can be used for tracking disease progression and efficacy monitoring, and can also play a role in monitoring disease recurrence.

[0089] The inventors also evaluated the effect of combining the MethylBERT-EOC diagnostic model with CA125 on the detection sensitivity of EOC. Among 1,058 samples with CA125 information in an individual cohort, compared with the MethylBERT-EOC diagnostic model alone, the sensitivity of the combination of the MethylBERT-EOC diagnostic model and CA125 detection increased by 3.21% (from 92.47% to 95.68%), but the specificity decreased by 3.81% (from 97.36% to 93.55%). However, for early stage EOC samples (n = 183), the sensitivity of the combined model increased sharply from 82.51% to 89.62% at a specificity of 93.55%. Notably, when the specificity exceeded approximately 95%, this sensitivity could be as high as approximately 85% (Tables 8 and 4).

[0090] Table 8

[0091]

[0092] Table 9

[0093]

[0094] Comparison of the MethylBERT-EOC Diagnostic Model with the LASSO Logistic Regression EOC Diagnostic Model: Using the same training dataset for the MethylBERT-EOC diagnostic model, the inventors developed a diagnostic model using the above-described conventional LASSO-logistic regression strategy. To reduce the number of markers, 500 LASSO analyses were performed on 493 markers. Subsequently, 21 markers identified in more than 450 LASSO analyses were applied to the logistic regression for diagnostic modeling. In this way, the LASSO logistic regression EOC diagnostic model based on the same training set as the MethylBERT-EOC diagnostic model was obtained (Table 10).

[0095] Table 10

[0096]

[0097] In some embodiments, a single marker among the markers listed in Table 10 (e.g., OV1) can be used as at least a part of a diagnostic assay panel for EOC. Such a diagnostic assay panel can be used for EOC screening studies, EOC staging, and / or monitoring the treatment of EOC. Such a diagnostic assay panel can incorporate other cancer markers related to EOC known in the art. Such a diagnostic assay panel can be implemented using any suitable DNA detection and / or identification technique, such as DNA amplification, multiplex DNA amplification, etc. Exemplary primer and probe sequences suitable for amplifying and identifying the sites described in Table 10 are provided in Table 11. The applicant points out that suitable amplification primer (e.g., forward and / or reverse) sequences and probe sequences can be obtained from the specific methylation site positions listed in Table 10 and the available human genome sequence data using conventional methods and tools.

[0098] Figures 9A to 9F The application results of the LASSO logistic regression model are provided. Figure 9A A confusion table of the binary results of the LASSO-EOC diagnostic model in the training dataset is provided. Figure 9B A confusion table of the binary results of the ASSO-EOC diagnostic model in the validation dataset is provided. Figure 9C Typical ROC curves of the LASSO-EOC diagnostic model in the training and validation datasets for EOC diagnostic prediction are shown. Figure 9D Typical ROC curves of the LASSO-based EOC diagnostic model in the training dataset for early and late EOC diagnostic prediction are shown. Figure 9E Typical ROC curves of the LASSO-based EOC diagnostic model in the validation dataset for early and late EOC diagnostic prediction are shown. Figure 9FResults of EOC prediction scores for healthy female samples and EOC samples at different stages based on LASSO-EOC diagnosis are provided.

[0099] In the training dataset, the LASSO logistic regression EOC diagnostic model showed 88.27% sensitivity and 93.82% specificity (AUC = 0.97) for EOC diagnosis. When applied to the same validation dataset used by the MethylBERT-EOC diagnostic model, the sensitivity of the LASSO-logistic regression EOC diagnostic model was 83.67% and the specificity was 89.04% (AUC = 0.92)( Figures 9A to 9C ). Both of these results were lower than those given by the MethylBERT-EOC diagnostic model. In addition, the LASSO logistic regression EOC diagnostic model also showed a significant difference in the combined prognostic score (cd-score) between early and late stage EOC samples. However, in the validation dataset, it only identified 49 out of 73 early EOC cases, with a sensitivity of 67.12%( Figure 9E ), which was 12% lower than the sensitivity given by the MethylBERT-EOC diagnostic model.

[0100] Development of a Simple and Cost-Effective ddPCR Assay

[0101] As described above, the inventors believe that the MethylBERT artificial intelligence method using cfDNA obtained from blood samples can be used to identify individuals with EOC and / or assist in providing a prognosis for individuals with EOC. However, in some cases, there may not be resources to support this method. Therefore, considering that the MethylBERT-derived OCDP has good performance characteristics, the inventors evaluated the potential of developing a simple and cost-effective PCR-based assay for the diagnosis and / or prognosis of EOC. The inventors selected the statistically most important methylation marker OV1 to design a ddPCR assay and validated its utility in an independent cohort consisting of 305 EOC patients and 480 healthy female subjects (referred to as the ddPCR cohort, Table 1). As described above, suitable primer and probe sequences can be obtained from known sequence data and the identification of the OV1 methylation marker position shown above. Table 11 shows the suitable methylation sequences and bisulfite-treated proof sequences as well as forward and reverse primers for this exemplary ddPCR assay, and exemplary primer and probe sequences for amplifying the methylation sites related to EOC identified herein are also provided.

[0102] Table 11

[0103]

[0104]

[0105]

[0106] The results are as Figures 5A to 5E shown. Figure 5A The typical ROC curves of OV1 in the training (red) and validation (blue) datasets of the ddPCR cohort are shown. Figure 5B A confusion table of the binary results of the OV1 prediction model in the training dataset is provided. Figure 5C A confusion table of the binary results of the OV1 prediction model in the validation dataset is provided. Figure 5D The typical ROC curves of OV1 and the OV1-CA125 combination in differentiating early (red and green) and late (blue and purple) EOC from healthy women in the ddPCR cohort are shown. Figure 5E A confusion table summarizing the differentiation of early and late EOC by OV1 is provided. Figure 5F A summary of the sensitivity and specificity results of CA125, OVA-1 ddPCR, and the CA125 and OVA-1 PCR combination in healthy, early EOC, and late EOC is provided. Figure 5G The Beewarm plot shows the OV1 methylation levels between ovarian cancer and healthy women in the ddPCR cohort. The red plot represents healthy female samples, and the blue plot represents EOC.

[0107] Overall, ddPCR showed a significant methylation difference of OV1 between EOC and healthy female samples ( Figure 5G ), so this cohort was randomly divided into training and validation datasets at a ratio of 2:1, and logistic regression was applied to the training dataset to determine the threshold. The results showed that OV1 achieved 77.4% sensitivity and 92.59% specificity (AUC = 0.912) in the training dataset and 72.16% sensitivity and 92.95% specificity (AUC = 0.877) in the validation dataset ( Figures 5A to 5C ). In contrast, CA125 showed only 48.85% sensitivity and 95% specificity in this ddPCR cohort. For the diagnosis of different stages of EOC in this cohort, OV1 and CA125 achieved 57.66% sensitivity and 92.71% specificity, and 38.74% sensitivity and 95% specificity, respectively, in early EOC (n = 111), and 86.08% sensitivity and 92.71% specificity, and 54.64% sensitivity and 95% specificity, respectively, in late EOC (n = 194) ( Figure 5D and Figure 5F, Figure S13). Although the diagnostic sensitivities of OV1 and CA125 for early EOC are not satisfactory, the combination of OV1 and CA125 increases the sensitivity sharply to 72.07%, while the specificity remains as high as 88.12% ( Figure 5G ).

[0108] More importantly, when evaluating different sensitivity values of OV1 (80%, 85%, 90%, 95%), the inventors found that the corresponding specificity values between ddPCR detection and sequencing results had very good consistency (Tables 12 and 13), indicating excellent ddPCR performance and that this platform could be an ideal alternative to sequencing strategies.

[0109] Table 12: Specificity of OV1 in differentiating EOC from healthy women at its corresponding sensitivities (∼80%, ∼85%, ∼90%, ∼95%) in the individual cohort

[0110]

[0111] Table 13: Specificity of OV1 in differentiating EOC from healthy women at its corresponding sensitivities (∼80%, ∼85%, ∼90%, ∼95%) in the ddPCR cohort

[0112]

[0113] The inventors also evaluated the utility of OV1 as a methylation marker in a longitudinal cancer screening cohort consisting of 2,117 high-risk EOC participants. The OV1 values measured by ddPCR testing were used as the first-line test for this high-risk EOC cohort. The inventors also measured the CA125 values. Study participants who obtained a positive predictive score for EOC (first-line test) from the OV1 and CA125 combined model underwent a TVU imaging study (second-line test) performed by experienced senior sonographers. If no adnexal mass was detected, the participants would have two more TVU tests within the next 6 months to verify the results. If a mass was detected, the participants would undergo a TVU test monthly for the next 6 months. Any participant suspected of malignancy or with an indeterminate TVU test result as judged by the subjective assessment of the senior sonographer would undergo further abdominal nuclear MRI, and positive participants would proceed to biopsy for histological confirmation.

[0114] In this study, the inventors identified 314 EOC-positive participants from the first-line tests, of which 4 were diagnosed with EOC, 3 of them in stage I and 1 in stage III. Additionally, after following up this cohort for more than 18 months, during which each participant had at least two TVU tests per year within these 18 months, only one negative predictive participant was reported as EOC. Therefore, in this prospective cohort, the method of combining OV1 and CA125 gave a sensitivity of 80.0% and a specificity of 85.3%, which was highly consistent with the results of our retrospective ddPCR study cohort.

[0115] The global cancer burden lies mainly in late detection. This is especially true for EOC, which has a high mortality rate at the time of late-stage detection. Reducing the death burden of EOC patients depends heavily on early detection. Genetic and epigenetic analysis of cfDNA obtained from liquid biopsy is a promising method to obtain diagnostic information from only blood samples. ctDNA can be shed by tumor cells, and importantly, they retain the same copy number alterations, mutations, and epigenetic marks. Therefore, genetic analysis of these cfDNA can detect early epigenetic changes associated with malignant transformation. Promising results have emerged in using cfDNA to diagnose different cancers. The conventional search strategy for cancer cfDNA markers is to utilize normal and cancer tissue samples or cell lines to compare mutational and / or epigenetic differences.

[0116] Traditional CA125 biomarker is an effective indicator for EOC. Our individual and ddPCR cohorts showed that its sensitivity exceeded 50% in all stages of EOC and was close to 40% in early EOC detection. In fact, a clinical trial with over 200,000 participants showed that measuring CA125 annually increased the incidence of early EOC by 39.2% and decreased the incidence of advanced EOC by 10.2%. However, as an ideal first-line method for general population screening, this performance is still not effective enough. In fact, in the clinical trial, the mortality rate related to EOC was not significantly reduced. Achieving the goal of reducing mortality will require a more sensitive screening strategy. At the same time, this trial also showed that although it has high accuracy, annual TVU is not suitable as a first-line screening strategy because its performance in early EOC detection is not as good as CA125 (23). Interestingly, in the clinical trial, CA125 increased the incidence of early EOC by 39.2%, and this figure is close to the sensitivity of CA125 in early EOC detection in the individual and OV-1 ddPCR cohorts (44.2% and 38.7% respectively). If the ~40% increase in the incidence of early EOC in the trial is achieved by the ~40% sensitivity of CA125 in early EOC, then combining CA125 with our OCPC will increase the early incidence to over 80%. In summary, the OCDP of the present invention concept can be an excellent alternative or supplement to the CA125 test because they can increase the diagnostic sensitivity of early EOC to over 80% while the specificity is as high as 86%.

[0117] Although large databases of mutation markers are available, the scale of the epigenetics marker database is limited. Most of the data in the epigenetics database are generated by HumanMethylation27 or 450, and most of them are from tumor tissues, which cover less than 0.03% of the human methylome. In this study, the inventors used the TMC- Kit, which can evaluate the methylation status of nearly 8 times more loci and regions. In fact, the method of the present invention concept reveals many new DMLs and DMRs related to EOC, and on this basis, diagnostic and prognostic prediction models for EOC are developed, which show excellent performance not only in advanced EOC but also in early EOC. However, the TMC- Kit covers less than 1% of the entire methylome, and further expanding the CpG candidate sites should have great potential for improving EOC screening.

[0118] The ideal EOC diagnostic test has three characteristics: high sensitivity, high PPV, and low FPR. A highly sensitive EOC diagnostic test, especially in the early EOC stage domain, will lead to an increase in cancer mortality. Computer simulations have shown that by increasing the sensitivity of current CA125-dependent EOC detection, the overall mortality rate can be reduced by up to 25%. An EOC test with high PPV will help relieve patients' anxiety while waiting for confirmatory TVU results. Finally, a test with low FPR will be truly beneficial to the healthcare system as the number of unnecessary TVU tests will be kept to a minimum. In an individual cohort consisting of 754 EOC patients and 1118 healthy women, the sensitivity, PPV, and FPR of the OCDP of the present invention concept were estimated to be 85.81%, 0.84, and 0.11, respectively. Compared with CA125 in the same cohort, the sensitivity was increased by more than 30%, and the PPV and FPR were at acceptable levels. In addition, combining the OCDP of the present invention concept with CA125 can sharply increase the sensitivity and PPV to 92.61% and 0.93, respectively, although the FPR increased slightly to 0.13.

[0119] It is known that DNA methylation is related to gene expression. Gene promoter methylation generally leads to the inhibition of gene expression, while the function of gene body methylation is still unclear. In methylated gene bodies, both over-regulation and down-regulation of gene expression have been reported. Despite the regulatory relationship, abnormal methylation status usually reflects dysregulated gene expression. In an OCDP of the present invention concept, OV18 was annotated to be located on the promoter of TRIM15, which is hypomethylated in EOC cfDNA and overexpressed in ovarian cancer tissues. In addition, most of the other OCDP markers on the gene body are related to dysregulated gene expression, supporting the model that these markers are derived from tumors. Another interesting finding of this study is that the most powerful predictive marker, OV1, is located in an intergenic region and mapped to the microRNA host gene MIR3681HG. Since its cfDNA shows a methylation difference of more than 20% between EOC and healthy women's cfDNA, this indicates (from the inventors' experience) that the associated gene expression difference is more than three times. If its source is microRNA, its impact on target DNA will be further amplified. However, no studies on MIR3681HG have been found, and there are few predicted targets for MIR3681.

[0120] Examples

[0121] The research described in this paper was approved by the Research Ethics Committee of Guangzhou Women and Children's Medical Center, and the prospective study was approved by the Research Ethics Committee of Zhuhai People's Hospital. The inventors studied three retrospective cohorts of EOC and healthy female plasma samples: the pool cohort, the individual cohort, and the ddPCR cohort. The pool cohort consisted of pooled samples each containing >20 individual samples, and the other two cohorts consisted of individual samples. In addition, 96 independent plasma samples from endometriosis patients and a prospective cohort of >2000 plasma samples were also included in this study.

[0122] cfDNA methylation markers were mainly screened for more than 3.3 million CpG sites in the pool cohort by the TMC- Kit. In this way, 500 markers were screened based on their methylation differences and P values, and these markers were further screened using its custom probes in the individual cohort and seven EOC and one ovarian epithelial cell line. The retained markers were used for OCDP and OCPP construction in the individual cohort, and the best marker OV1 was further detected on the ddPCR platform of the ddPCR cohort. Finally, OV1 was detected on the ddPCR platform of the prospective cohort, and the results were confirmed by imaging. Before extracting cfDNA, 1.5 - 2 mL of plasma was collected from each subject and stored at -80 °C.

[0123] EOC Screening

[0124] The inventors conducted a prospective EOC screening cohort study to evaluate the utility of the OV1 methylation marker compared with conventional screening methods. All participants were enrolled due to increased EOC risk, including: (i) women, (ii) postmenopausal, (iii) history of breast cancer or family history of cancer, (iv) BRCA1 / 2 mutations.

[0125] For the high-risk prospective cohort, 2117 subjects were screened by OV1 ddPCR and CA125 as the first-line test, and samples predicted positive by the OV1 and CA125 combined model were further detected by TVU as the second-line test. Any positive or suspicious TVU results were then verified by abdominal MRI imaging. Verified participants were referred to a gynecologist and followed up continuously, and their tissue samples were obtained and sent to a pathologist for histological confirmation.

[0126] Cell-Free DNA Extraction

[0127] Use Magen cfDNA extraction (D3182-04) to isolate cell-free DNA from plasma according to the manufacturer's instructions. Use Qubit A fluorometer (Invitrogen, Life Technologies) and Qubit dsDNA High Sensitivity (Invitrogen) were used to measure the amount of cfDNA.

[0128] TMC EPIC Methylation Library Preparation and Sequencing

[0129] Twenty EOC cfDNA samples at a similar stage (early or late stage) were pooled together. The Qubit fluorometer (Invitrogen, Life Technologies) was used to estimate the amount of DNA in the pooled sample. If the total amount exceeded 500 ng, the pooled sample was applied to subsequent library construction. If the amount was less than 500 ng, more cfDNA samples were added to the pooled sample until 500 ng was reached. The pooled cfDNA sample from healthy women was obtained in the same way.

[0130] The methylation library of each pooled sample was prepared using TruSeq Methyl Capture EPIC Library Prep (FC-151-1003, Illumina), except that the fragmentation step was not performed, and it was carried out according to the manufacturer's instructions. The Qubit fluorometer (Invitrogen, Life Technologies) was used to measure the concentration of the prepared library, and the quality of the library was evaluated by capillary electrophoresis (Qseq Bioptic). The qualified libraries were sequenced on the Illumina HiSeq platform (Illumina).

[0131] Targeted EPIC Library Using Custom Probes

[0132] The cfDNA extracted from plasma samples was processed using NEBNext Ultra TM II DNA LibraryPrep from NEB for Illumina (NEB#7645L) is ligated to a methylation adaptor. The methylation adaptor is custom-made by Integrated DNA Technologies and consists of an 8-bp index and an 8-bp index linked to a sequence of a 9-bp UMI (Reference No.: 04099708Q). The cfDNA ligated with the adaptor is mixed at a ratio of 12 to 1 and the DNA libraries are captured by hybridization using custom probes (Integrated DNA Technologies). (Integrated DNA Technologies) for hybridization. The hybridization mixture samples are eluted by using the reagents and steps of the "second elution" part of TruSeq Methyl Capture EPIC Library Prep (FC-151-1003, Illumina), and then bisulfite conversion is performed by using EZ-96 DNA Methylation-Lightning MagPrep (D5047, Zymo Research). The bisulfite-converted samples are amplified by using the reagents and steps of the "amplify enriched library" part of TruSeq Methyl Capture EPIC Library Prep (FC-151-1003, Illumina). The concentration of the prepared libraries is measured by using a Qubit fluorometer (Invitrogen, Life Technologies), and the quality of the libraries is evaluated by capillary electrophoresis( Bioptic). The qualified libraries are sequenced on the Illumina Nova- platform (Illumina).

[0133] Sequencing Data Analysis

[0134] For the pool cohort, the raw methylation data is preprocessed using Fastp (version 0.20.0) with default parameters. The clean reads are then aligned to the human genome build hg19 using (version 1.0.2.3) with default parameters, and the bam-format results are sorted using (version 0.7.0). DNA methylation calls are extracted using (version 0.4.0) with default parameters, and the DNA methylation calls of the methylated and unmethylated controls are extracted from the alignment files. Using (Version 2.29.0) Extract methylation values located in the target region.

[0135] For the individual cohort, the raw methylation data was processed by Umi-tools (Version 1.0.1) and the extraction program (extractprogram), and the reads were preprocessed using (Version 0.20.0) and default parameters. The clean reads were then aligned to the human genome build hg19 in "pbat" mode using (Version 1.0.2.3), and the bam format results were sorted using (Version 0.7.0). The aligned reads were deduplicated based on UMIs using the umi-tools dedup program. DNA methylation calls were extracted using (Version 0.4.0) with the "--keepDups" parameter, and DNA methylation calls for methylated and unmethylated controls were extracted from the alignment files. (Version 2.29.0) Extract methylation values located in the target region. (Version 2.29.0) Extract methylation values located in the target region.

[0136] Statistical Analysis

[0137] For TMC EPIC and targeted EPIC methylation sequencing, (Version 0.3) and beta-binomial hierarchical modeling and Wald test were used to identify differentially methylated CpGs between healthy and tumor samples, with significance cutoff at p < 0.05 and |Dif| > 0.1. ROC analysis was performed using the pROC package to evaluate the diagnostic performance. The cd-score between clinical characteristics was evaluated by the Wilcoxon rand-sum test, and a p-value < 0.05 was considered statistically significant.

[0138] Diagnostic Model

[0139] MethylBERT EOC model

[0140] Large-scale unlabeled data was used for pre-training: To facilitate the pre-training of MethylBERT, we collected extensive DNA methylation data from two main sources: the GEO-methyl dataset and the TCGA-methyl dataset. We collected a total of over 110,000 samples, with a data volume of over 3TB. Both datasets contain comprehensive genome-wide methylation data from different tissue types and conditions, providing rich resources for training and evaluating the performance of the model.

[0141] The GEO-methyl dataset is derived from the Gene Expression Omnibus (GEO) repository of the National Center for Biotechnology Information (NCBI), which houses a large collection of high-throughput sequencing and microarray datasets. The inventors focused on providing whole-genome bisulfite sequencing (WGBS) and reduced representation bisulfite sequencing (RRBS) datasets of genome-wide methylation data across various tissues, diseases, and conditions. We retrieved access data from several platforms, including GPL13534, GPL21145, GPL8490, GPL23976, and GPL9183, and extracted beta value files for further analysis. After data cleaning and preprocessing, the GEO-methyl dataset contains methylation data from 95,995 samples, covering a wide range of species, including humans, mice, and plants. This dataset provides diverse and extensive resources for pre-training the MethylBERT model.

[0142] The TCGA-methyl dataset is derived from The Cancer Genome Atlas (TCGA), a comprehensive resource containing multi-omics data for different cancer types. For MethylBERT, the inventors focused on using DNA methylation data generated by the Illumina Infinium HumanMethylation450 platform. The TCGA-methyl dataset includes methylation data from 15,439 human samples, including tumor samples and adjacent normal tissues. By incorporating this dataset, MethylBERT is able to learn diverse methylation patterns associated with various cancer types and stages.

[0143] CpG site representation: MethylBERT features an innovative CpG site embedding scheme, including four different embedding types: chromosome embedding (E c ), position embedding (E p ), methylation level embedding (E1), and gene embedding (E g ). Each embedding captures a unique aspect of CpG site information, enhancing the performance of the model.

[0144] Chromosome embedding E c: By representing the chromosome of each CpG site as an embedding, the model can learn and distinguish various chromosomal contexts, taking into account functional and structural variations between chromosomes.

[0145] Position Embedding E p : CpG sites are assigned to bins, each of length L base pairs (bps), and each bin is assigned a unique embedding. This allows the model to learn the relationship between adjacent CpG sites and capture the spatial organization of methylation patterns within the genomic landscape. In this implementation, L is set to 2000.

[0146] Methylation level embedding E1: To facilitate learning of methylation patterns, the continuous methylation level (ranging from 0 to 1) is discretized into B bins. This approach enables the model to effectively capture the subtle differences in methylation dynamics. In this case, B is set to 20. Gene embedding E g : Gene embeddings are used to associate CpG sites with their potential functional roles. For sites with known associations, the nearest gene is used; for those without, the nearest downstream gene is selected. These gene embeddings are derived from gene2vec, which learns gene-gene association information based on gene expression profiles across various tissues and conditions, thereby creating gene-gene co-expression networks. The final CpG site embedding is obtained by summing these embeddings: E = E c +E p +E1+E g .

[0147] Neural Network Architecture: The attention mechanism of the Transformer architecture exhibits quadratic computational complexity, which poses a significant challenge when processing more than 20 million CpG sites. To address this issue, the inventors used Performer, a Transformer model based on matrix factorization, to reduce the computational complexity from quadratic to linear (O(L 2 ) to O(L)), thus being able to process large-scale data efficiently. Performer uses an approximation technique called "kernelized attention" with random feature maps. Unlike the standard Transformer attention, it is expressed as Att(Q,K,V)=softmax(QK T / sqrt(d k ))V, the Performer’s attention mechanism is formulated as Att(Q,K,V)=softmax(Φ(Q)Φ(K) T / sqrt(d k))V. Here, Q, K, and V represent the query, key, and value matrices respectively, and Φ represents a feature mapping function that projects the input into a new space to facilitate an effective approximation of dot-product attention. In this study, the number of Transformer layers was set to six.

[0148] Pre-training using masked methylation level prediction: In the pre-training phase, the inventors adopted the masked language model (MLM) objective used in BERT to adapt to methylation data and generate masked methylation level prediction (MMLP). The goal of MMLP is to predict the methylation levels of some masked CpG sites given the context of their surrounding CpG sites. To achieve this, a certain percentage P m of the CpG sites in the input sequence m is masked, and the model is trained to predict their methylation levels. Following BERT, we set P i to 15%. Formally, let x i be the input CpG sequence, where i ∈ 1, …, L, and y m be the corresponding true methylation levels. During pre-training, we randomly mask P of the CpG sites and replace their methylation level embeddings with a special [MASK] token. The MMLP loss is calculated as the cross-entropy between the predicted methylation levels i and the true values y : MMLP loss where i M is the set of CpGs with masked methylation levels, and f(x

[0149] ) is the output of the MethylBERT model for the masked CpG site i. The model is optimized to minimize this loss function, encouraging it to learn biologically relevant patterns and correlations between CpG sites. m CpG site set, and N mRanging from N to 10N. Subsequently, N CpG sites are randomly selected from this set. This method ensures obtaining a representative subset of CpG sites, capturing relevant information while maintaining computational efficiency. In this study, N was set to 8192. By adopting these strategies, MethylBERT effectively addresses the computational challenges posed by large-scale DNA methylation data, enabling the model to learn meaningful CpG site representations and correlations without sacrificing performance or scalability.

[0150] Supervised learning for EOC detection: The inventors fine-tuned the pre-trained MethylBERT model for the EOC detection task. The inventors first used MethylBERT to obtain the representation of each chromosome, and then concatenated these representations to form a sample representation, which can be used for the final prediction. For a given sample, let h c represent the representation of chromosome c obtained from MethylBERT, where c ∈ {1,..., C}. We concatenated these representations to form the final sample representation z = Concat(h1, h2,..., h c ). Here, for human samples C is 23. In a training dataset of 503 EOCs and 744 healthy female samples randomly selected from the individual cohort, to predict the presence of EOC in the sample, we used a binary classification method with a fully connected layer and a sigmoid activation function to obtain the probability that the i-th sample belongs to the EOC class The model was trained to minimize the binary cross-entropy loss where y i represents the true label. By fine-tuning the MethylBERT model in a supervised manner for EOC detection, we enabled it to capture EOC-specific methylation patterns and correlations, ultimately leading to the creation of the MethylBERT-EOC diagnostic model for detecting EOC in DNA methylation data.

[0151] LASSO-EOC model

[0152] The samples in the training and validation datasets of the individual cohort were the same as those used to construct the MethylBERT-EOC diagnostic model. 493 markers screened from the pool cohort were processed by LASSO in the training dataset to distinguish EOC and healthy female samples. 500 LASSOs were performed, with 70% of the samples randomly selected each time. The markers that appeared in more than 450 LASSOs were retained and applied to the training dataset to construct an EOC diagnostic model based on logistic regression, and then this diagnostic model was tested in the validation dataset.

[0153] Prognostic model

[0154] EOC samples with complete survival information in the individual cohort were randomly divided into training and validation datasets at a ratio of 2:1. Markers were pre-screened in the pooled cohort and cell lines. The screened markers were processed by LASSO in the training dataset to distinguish the occurrence samples from other observed samples. 100 LASSO procedures were performed, with 70% of the samples randomly selected each time. Markers that appeared in more than 90 LASSO procedures were retained. Meanwhile, Unicox was also used to process the selected markers in the training dataset to distinguish the occurrence samples from other observed samples. Markers with a P-value < 0.05 in Unicox overlapped with the markers retained from LASSO. These overlapping markers were applied to the training dataset to construct an EOC prognostic model based on logistic regression, and then this prognostic model was tested in the validation dataset.

[0155] Droplet digital PCR

[0156] cfDNA samples were extracted from plasma and subjected to bisulfite conversion using EZ DNAMethylation-Lightning (ZymoResearch, Irvine, CA, USA) according to the manufacturer's instructions. Subsequent inspections and analyses were performed based on the droplet digital PCR system according to the manufacturer's instructions (Bio-Rad, Pleasanton, California, USA). FAM and HEX fluorophores were used to label the methylation probe and non-methylation probe respectively, and the sequences of the probes and primers are indicated in Table 11. For each reaction, the system and parameters were as follows:

[0157] PCR reagent mixture -

[0158]

[0159] PCR reaction conditions -

[0160] (a) 98°C for 10 minutes

[0161] (b) 98°C for 30 seconds

[0162] (c) 45.7°C for 60 seconds

[0163] Steps (b) and (c) were repeated for a total of 39 cycles

[0164] (d) 98°C for 10 minutes

[0165] (e) 4°C for 20 minutes

[0166] Cell Lines, Cell Culture, and DNA Extraction

[0167] Cell lines were obtained from the American Type Culture Collection (ATCC) and maintained according to the supplier's instructions. HO-8910, A2780, and IOSE-80 cells were grown in RPMI-1640 containing 10% FBS; SK-OV-3 cells were grown in McCoy's 5A containing 10% FBS; OVCAR-3 cells were grown in RPMI-1640 containing 20% FBS and 0.01 mg / mL insulin; Hey and Anglne cells were grown in DMEM containing 10% FBS; SW626 cells were grown in Leibovitz's L-15 containing 10% FBS. Cell DNA for targeted EPIC methylation sequencing was extracted using the QIAamp DNA Mini Kit according to the manufacturer's instructions.

[0168] Prospective EOC Screening Cohort Study Design

[0169] This prospective study was approved by the Research Ethics Committee of Zhuhai People's Hospital. We conducted a prospective EOC screening cohort study from August 2022 to July 2023 to evaluate the utility of combining the OV1 methylation marker with conventional screening methods. All participants were enrolled due to increased EOC risk, including (i) women, (ii) postmenopausal, (iii) history of breast cancer or family history of cancer, and (iv) BRCA1 / 2 gene mutations.

[0170] For the high-risk prospective cohort, 2117 subjects were screened by OV1 ddPCR and CA125 as the first-line tests, and samples predicted positive by the OV1 and CA125 combined model were further tested by TVU as the second-line test. Any positive or suspicious TVU results were then verified by abdominal MRI imaging. The verified participants were referred to a gynecologist for serial follow-up, obtaining their tissue samples and sending them to a pathologist for histological confirmation.

[0171] It will be apparent to those skilled in the art that many more modifications are possible without departing from the inventive concept herein, other than those already described. Accordingly, the subject matter of the present invention is not limited except as by the spirit of the appended claims. In addition, in interpreting the specification and claims, all terms should be interpreted in the broadest possible manner consistent with the context. In particular, the terms "comprises" and "comprising" should be interpreted as referring to elements, components, or steps in a non-exclusive manner, indicating that the recited elements, components, or steps may be present, or utilized, or combined with other elements, components, or steps not expressly recited. When the specification or claims refer to at least one of something selected from the group consisting of A, B, C... N, the text should be interpreted to mean only one element from the group, rather than A plus N or B plus N, etc.

Claims

1. A method for the auxiliary diagnosis of epithelial ovarian cancer, comprising: Isolating cell-free genetic material from an individual; And Characterizing the nucleic acid methylation of one or more genetic markers selected from the group consisting of OV1, OV2, OV6, OV3, OV4, OV5, OV6, OV6, OV6, OV6, OV7, OV8, OV9, OV10, OV11, OV12, OV13, OV14, OV15, OV16, OV17, OV18, OV19, OV20, OV21, and OV22, wherein abnormal methylation of one or more genetic markers indicates epithelial ovarian cancer.

2. The method according to claim 1, wherein the cell-free genetic material is cell-free DNA.

3. The method according to claim 1 or 2, comprising treating at least a portion of the cell-free genetic material with bisulfite.

4. The method according to any one of claims 1 to 3, comprising contacting the cell-free genetic material with nucleic acid primers and performing nucleic acid amplification to generate an amplification product, the nucleic acid primers being complementary to a portion of the genetic material adjacent to at least one of the group consisting of OV1, OV2, OV6, OV3, OV4, OV5, OV6, OV6, OV6, OV6, OV7, OV8, OV9, OV10, OV11, OV12, OV13, OV14, OV15, OV16, OV17, OV18, OV19, OV20, OV21, and OV22.

5. The method according to claim 4, comprising contacting the amplification product with a probe, the probe being complementary to at least one of the group consisting of OV1, OV2, OV6, OV3, OV4, OV5, OV6, OV6, OV6, OV6, OV7, OV8, OV9, OV10, OV11, OV12, OV13, OV14, OV15, OV16, OV17, OV18, OV19, OV20, OV21, and OV22.

6. The method according to any one of claims 1 to 4, comprising characterizing the level of CA125, wherein an elevated level of CA125 indicates epithelial ovarian cancer.

7. The method according to any one of claims 1 to 6, wherein the genetic marker is OV1.

8. A method for the auxiliary determination of the prognosis of epithelial ovarian cancer, comprising: Isolating cell-free genetic material from an individual; And Characterizing the nucleic acid methylation of one or more genetic markers selected from the group consisting of OV1, OV2, OV6, OV3, OV4, OV5, OV6, OV6, OV6, OV6, OV7, OV8, OV9, OV10, OV11, OV12, OV13, OV14, OV15, OV16, OV17, OV18, OV19, OV20, OV21, and OV22, wherein abnormal methylation of one or more genetic markers indicates a poor prognosis of epithelial ovarian cancer.

9. The method according to claim 8, wherein the cell-free genetic material is cell-free DNA.

10. The method according to claim 8 or 9, which comprises treating at least a portion of said free genetic material with bisulfite.

11. The method according to any one of claims 9 to 10, which comprises contacting said free genetic material with a nucleic acid primer and performing nucleic acid amplification to generate an amplification product, said nucleic acid primer being complementary to a portion of the genetic material adjacent to at least one selected from the group consisting of OV1, OV2, OV6, OV3, OV4, OV5, OV6, OV6, OV6, OV6, OV7, OV8, OV9, OV10, OV11, OV12, OV13, OV14, OV15, OV16, OV17, OV18, OV19, OV20, OV21, and OV22.

12. The method according to claim 11, which comprises contacting said amplification product with a probe, said probe being complementary to at least one selected from the group consisting of OV1, OV2, OV6, OV7, OV9, OV11, OV18, and OV22.

13. The method according to any one of claims 8 to 12, which comprises characterizing the level of CA125, wherein an elevated level of CA125 indicates a poor prognosis for epithelial ovarian cancer.

14. The method according to any one of claims 8 to 13, wherein said genetic marker is OV1.

15. The method according to any one of claims 1 to 14, which further comprises, prior to the step of characterizing methylation, applying an artificial intelligence algorithm to a training data set comprising individuals identified as having EOC and individuals not having EOC, and identifying a methylation pattern associated with individuals having EOC, wherein said artificial intelligence algorithm comprises associations with chromosome embedding, position embedding, methylation level embedding, and gene embedding, and wherein said methylation pattern comprises one or more genetic markers that exhibit methylation differences between individuals having EOC and individuals not having EOC.

16. The method according to claim 15, wherein the artificial intelligence algorithm includes a Transformer model based on matrix factorization, which is configured to reduce the computational complexity from quadratic to linear (O(L 2 ) to O(L)), thereby enabling efficient processing of large-scale data.

17. The method according to claim 16, wherein the Transformer model comprises a Performer, and wherein the Performer comprises an attention mechanism, the formula of which is Att(Q, K, V) = softmax(Φ(Q)Φ(K) T / sqrt(d k ))V, where Q, K, and V represent query, key, and value matrices respectively, and Φ represents a feature mapping function that projects the input into a new space, thereby improving the effective approximation of dot-product attention.

18. A composition for the diagnosis or prognosis of epithelial ovarian cancer, comprising: A first primer that is complementary to a first portion of a nucleic acid adjacent to at least one genetic marker selected from the group consisting of OV1, OV2, OV6, OV3, OV4, OV5, OV6, OV6, OV6, OV6, OV7, OV8, OV9, OV10, OV11, OV12, OV13, OV14, OV15, OV16, OV17, OV18, OV19, OV20, OV21, and OV22; and A probe that is complementary to at least one genetic marker selected from the group consisting of OV1, OV2, OV6, OV3, OV4, OV5, OV6, OV6, OV6, OV6, OV7, OV8, OV9, OV10, OV11, OV12, OV13, OV14, OV15, OV16, OV17, OV18, OV19, OV20, OV21, and OV22 and comprises a first dye or a first fluorophore.

19. The composition according to claim 18, further comprising a second primer that is complementary to a genetic marker selected from the group consisting of OV1, OV2, OV6, OV3, OV4, OV5, OV6, OV6, OV6, OV6, OV7, OV8, OV9, OV10, OV11, OV12, OV13, OV14, OV15, OV16, OV17, OV18, OV19, OV20, OV21, and OV22.

20. The composition according to claim 19, comprising a second probe that is complementary to a genetic marker selected from the group consisting of OV1, OV2, OV6, OV3, OV4, OV5, OV6, OV6, OV6, OV6, OV7, OV8, OV9, OV10, OV11, OV12, OV13, OV14, OV15, OV16, OV17, OV18, OV19, OV20, OV21, and OV22.

21. The composition according to claim 20, wherein, The second probe comprises a second dye or a second fluorophore.

Images