Methods and apparatuses for methylation detection of ribonucleic acid, devices, media, computer program products

Abstract

The present disclosure relates to a method and device, equipment, medium and computer program product for methylation detection of ribonucleic acid, the method comprising: obtaining a target motif in a ribonucleic acid (RNA) sequence to be detected; performing at least one nanopore direct RNA sequencing on the target motif to obtain a target sequencing signal of each sequencing; determining an index result corresponding to each target sequencing signal based on the target sequencing signal; performing word embedding on the index result of each target sequencing signal to obtain a corresponding input matrix and inputting each input matrix into a methylation detection model for calculation to obtain a detection result of each target sequencing signal; and determining a final detection result representing a probability of methylation of the target motif based on the detection results of all target sequencing signals. The present disclosure realizes high-precision methylation detection by combining signal index results and comparison index results and using word embedding technology to fuse context information.

Description

Technical Field

[0001] This disclosure relates to the field of gene detection technology, and in particular to a method, apparatus, equipment, medium, and computer program product for detecting methylation of ribonucleic acid. Background Technology

[0002] Ribose nucleic acid (RNA) modification is an important research area in epitranscriptomics, and its detection technology has undergone a revolution from traditional biochemical methods to single-molecule sequencing. N6-methyladenosine (m6A), as one of the most abundant chemical modifications in mRNA, is widely involved in RNA metabolic regulation, but its accurate detection has long been limited by the insufficient single-base resolution and experimental bias of traditional immunoprecipitation methods (such as MeRIP-seq and miCLIP). Although newly developed m6A-SAC-Seq and GLORI methods have achieved single-nucleotide-level quantification through enzymatic labeling or chemical deamination strategies, the complex library preparation process still restricts its application.

[0003] In recent years, breakthroughs in nanopore direct RNA sequencing (DRS) technology have brought new opportunities for in-situ detection. This technology, by analyzing changes in electrical signals caused by modified bases, has led to the development of statistical methods such as DRUMMER that do not require pre-training and machine learning models such as m6anet. However, the former relies on control samples, increasing experimental costs, while the latter, although eliminating the need for controls, has not yet broken through the bottleneck in recognition accuracy. Both of these technical approaches face the challenge of optimizing signal analysis algorithms, and there is an urgent need to develop more accurate and efficient analytical frameworks. Summary of the Invention

[0004] In view of this, this disclosure proposes a method, apparatus, equipment, medium, and computer program product for the detection of methylation of ribonucleic acid, which can achieve high-precision methylation detection.

[0005] According to one aspect of this disclosure, a method for detecting methylation of ribonucleic acid (RNA) is provided, comprising: acquiring a target motif from a RNA sequence to be detected; performing at least one nanopore direct RNA sequencing on the target motif to obtain a target sequencing signal from each sequencing; determining an indicator result corresponding to each target sequencing signal based on each target sequencing signal, wherein each indicator result includes a signal indicator result and an alignment indicator result, wherein the signal indicator result represents the statistical characteristics of the corresponding target sequencing signal, and the alignment indicator result represents the anomalies in the base sequence identified based on the corresponding target sequencing signal; performing word embedding on the indicator results of each target sequencing signal to obtain a corresponding input matrix, and inputting each input matrix into a methylation detection model to calculate a detection result for each target sequencing signal, wherein each detection result represents the probability of methylation of the target motif predicted based on the corresponding target sequencing signal; and determining a final detection result based on the detection results of all target sequencing signals, wherein the final detection result represents the probability of methylation of the target motif. In this way, by acquiring the sequencing signal of the target motif and analyzing it from two dimensions, namely signal index and alignment index, the impact of methylation modification on the sequencing signal can be captured comprehensively and accurately, thereby improving the accuracy of detection. Furthermore, by using the methylation detection model to calculate the input matrix, the influence of multiple factors on the methylation detection results can be comprehensively considered, further improving the accuracy of detection and reducing the possibility of misjudgment.

[0006] In one possible implementation, determining the index results corresponding to each of the target sequencing signals based on each target sequencing signal includes: determining a base reference sequence that matches the target motif, the base reference sequence being used to indicate the actual base fragment of the target motif; performing base identification on each of the target sequencing signals to obtain a base identification sequence corresponding to each target sequencing signal, each base identification sequence being used to indicate the identified base fragment of the target motif; comparing the identified base fragment corresponding to each of the target sequencing signals with the actual base fragment to obtain the alignment index results of each of the target sequencing signals; and determining the corresponding signal index results based on each of the target sequencing signals; wherein each alignment index result includes at least one of the quality score, insertion information, deletion information, and matching information of the identified base fragment, and the signal index results include at least one of the signal mean, signal median, signal standard deviation, and signal length of the identified base fragment. In this way, by simultaneously considering alignment metrics (such as quality score, insertion information, missing information, and matching information) and signal metrics (such as signal mean, signal median, signal standard deviation, and signal length), the methylation status of the target motif can be comprehensively evaluated from multiple perspectives. This multi-dimensional analysis method avoids misjudgments that may arise from a single metric, thus improving the accuracy and reliability of detection. The alignment metrics provided in this disclosure can accurately identify the differences between the base recognition sequence and the actual base fragment, thereby capturing the impact of methylation modification on the base sequence; the signal metrics, starting from the statistical characteristics of the sequencing signal, further analyze the impact of methylation modification on signal intensity and stability, which helps to more accurately determine the methylation status.

[0007] In one possible implementation, the identified base fragments corresponding to each target sequencing signal are compared with the actual base fragments to obtain the alignment index results of each target sequencing signal, including at least one of the following: if it is determined that the number of bases in the identified base fragment is greater than the number of bases in the actual base fragment, then the insertion information of the identified base fragment is determined to have an insertion; if it is determined that the number of bases in the identified base fragment is less than the number of bases in the actual base fragment, then the missing information of the identified base fragment is determined to have a missing information; if it is determined that the bases at the same position in the identified base fragment and the actual base fragment are different, then the matching information of the identified base fragment is determined to have a mismatch; based on the comparison results of the bases at the same position in the identified base fragment and the actual base fragment, the quality score of the identified base fragment is determined. In this way, by using multiple alignment metrics such as insertion, deletion, matching, and quality scores, the differences between the base sequence in the target sequencing signal and the real base fragment can be comprehensively and accurately identified. This multi-dimensional alignment method avoids anomalies that may be missed by a single metric, ensuring the comprehensiveness and accuracy of the detection results. Insertion, deletion, and mismatch information can accurately capture minute changes in the base sequence, and even differences in a single base can be accurately identified. The introduction of quality scores provides a quantitative measure of the accuracy of base identification. By comparing the identified base fragment with the base fragment at the same position in the real base fragment, the reliability of each base identification can be accurately assessed. This quantitative assessment method helps to distinguish the influence of real methylation modification and sequencing errors or noise, thereby improving the reliability of detection. By clearly identifying anomalies such as insertion, deletion, and mismatch, problems in the base sequence can be quickly located. This not only helps to detect the presence or absence of methylation modification, but also provides detailed reference information for subsequent analysis.

[0008] In one possible implementation, the methylation detection model includes a first feature extraction module, a long short-term memory network module, a second feature extraction module, a gated recurrent unit, and a linear layer. The first feature extraction module extracts a first feature from the input matrix to obtain a first feature. The long short-term memory network module extracts temporal dynamic information from the first feature and integrates it to obtain a second feature. The second feature extraction module extracts a second feature from the second feature to obtain a third feature. The gated recurrent unit extracts long-term dependencies from the third feature and integrates them to obtain a fourth feature. The linear layer obtains the detection result based on the fourth feature. Thus, through deep learning architectures (LSTM, GRU, etc.), the methylation detection model can automatically learn complex patterns and rules in the data, effectively reducing human error and improving the accuracy and reliability of methylation detection.

[0009] In one possible implementation, the method further includes: acquiring a training sample set, which includes multiple samples and corresponding sample labels, wherein the sample labels indicate whether the corresponding sample is methylated or not; training an initial methylation detection model using the training sample set until the loss function value of the current methylation detection model meets a preset condition, thereby obtaining a trained methylation detection model. The loss function value includes a first function value and a second function value. The first function value is calculated based on the label information of each sample pair, the Euclidean distance between the two samples in each sample pair, and a preset boundary coefficient. The second function value is calculated based on the sample label of each sample and the predicted probability of methylation for each sample. Thus, by combining the first and second function values ​​to train the methylation detection model, the methylation detection model can not only learn the relationship between sample labels and methylation prediction probabilities but also consider the similarity and differences between samples. This dual optimization method enables the methylation detection model to more accurately capture the features of methylation, thereby improving the detection accuracy.

[0010] In one possible implementation, determining the final detection result based on the detection results of all target sequencing signals includes: processing the detection results of each target sequencing signal to obtain a corresponding processed probability; determining the final probability based on the product of all processed probabilities; and using the final probability as the final detection result. In this way, by processing and integrating the detection results of all target sequencing signals, information from multiple sequencing runs can be comprehensively considered, avoiding misjudgments that may be caused by random errors or noise in a single sequencing run, thereby improving the accuracy and reliability of methylation detection results.

[0011] According to another aspect of this disclosure, a device for detecting methylation of ribonucleic acid (RNA) is provided, comprising: an acquisition module for acquiring a target motif in a RNA sequence to be detected, performing at least one nanopore direct RNA sequencing on the target motif to obtain a target sequencing signal from each sequencing; an index determination module for determining an index result corresponding to each target sequencing signal based on each target sequencing signal, wherein each index result includes a signal index result and an alignment index result, wherein the signal index result represents the statistical characteristics of the corresponding target sequencing signal, and the alignment index result represents the anomalies in the base sequence identified based on the corresponding target sequencing signal; a model prediction module for performing word embedding on the index results of each target sequencing signal to obtain a corresponding input matrix, and inputting each input matrix into a methylation detection model to calculate the detection result of each target sequencing signal, wherein each detection result represents the probability of methylation of the target motif predicted based on the corresponding target sequencing signal; and a result determination module for determining a final detection result based on the detection results of all target sequencing signals, wherein the final detection result represents the probability of methylation of the target motif. In this way, by acquiring the sequencing signal of the target motif and analyzing it from two dimensions, namely signal index and alignment index, the impact of methylation modification on the sequencing signal can be captured comprehensively and accurately, thereby improving the accuracy of detection. Furthermore, by using the methylation detection model to calculate the input matrix, the influence of multiple factors on the methylation detection results can be comprehensively considered, further improving the accuracy of detection and reducing the possibility of misjudgment.

[0012] In one possible implementation, determining the index results corresponding to each of the target sequencing signals based on each target sequencing signal includes: determining a base reference sequence that matches the target motif, the base reference sequence being used to indicate the actual base fragment of the target motif; performing base identification on each of the target sequencing signals to obtain a base identification sequence corresponding to each target sequencing signal, each base identification sequence being used to indicate the identified base fragment of the target motif; comparing the identified base fragment corresponding to each of the target sequencing signals with the actual base fragment to obtain the alignment index results of each of the target sequencing signals; and determining the corresponding signal index results based on each of the target sequencing signals; wherein each alignment index result includes at least one of the quality score, insertion information, deletion information, and matching information of the identified base fragment, and the signal index results include at least one of the signal mean, signal median, signal standard deviation, and signal length of the identified base fragment. In this way, by simultaneously considering alignment metrics (such as quality score, insertion information, missing information, and matching information) and signal metrics (such as signal mean, signal median, signal standard deviation, and signal length), the methylation status of the target motif can be comprehensively evaluated from multiple perspectives. This multi-dimensional analysis method avoids misjudgments that may arise from a single metric, thus improving the accuracy and reliability of detection. The alignment metrics provided in this disclosure can accurately identify the differences between the base recognition sequence and the actual base fragment, thereby capturing the impact of methylation modification on the base sequence; the signal metrics, starting from the statistical characteristics of the sequencing signal, further analyze the impact of methylation modification on signal intensity and stability, which helps to more accurately determine the methylation status.

[0013] In one possible implementation, the identified base fragments corresponding to each target sequencing signal are compared with the actual base fragments to obtain the alignment index results of each target sequencing signal, including at least one of the following: if it is determined that the number of bases in the identified base fragment is greater than the number of bases in the actual base fragment, then the insertion information of the identified base fragment is determined to have an insertion; if it is determined that the number of bases in the identified base fragment is less than the number of bases in the actual base fragment, then the missing information of the identified base fragment is determined to have a missing information; if it is determined that the bases at the same position in the identified base fragment and the actual base fragment are different, then the matching information of the identified base fragment is determined to have a mismatch; based on the comparison results of the bases at the same position in the identified base fragment and the actual base fragment, the quality score of the identified base fragment is determined. In this way, by using multiple alignment metrics such as insertion, deletion, matching, and quality scores, the differences between the base sequence in the target sequencing signal and the real base fragment can be comprehensively and accurately identified. This multi-dimensional alignment method avoids anomalies that may be missed by a single metric, ensuring the comprehensiveness and accuracy of the detection results. Insertion, deletion, and mismatch information can accurately capture minute changes in the base sequence, and even differences in a single base can be accurately identified. The introduction of quality scores provides a quantitative measure of the accuracy of base identification. By comparing the identified base fragment with the base fragment at the same position in the real base fragment, the reliability of each base identification can be accurately assessed. This quantitative assessment method helps to distinguish the influence of real methylation modification and sequencing errors or noise, thereby improving the reliability of detection. By clearly identifying anomalies such as insertion, deletion, and mismatch, problems in the base sequence can be quickly located. This not only helps to detect the presence or absence of methylation modification, but also provides detailed reference information for subsequent analysis.

[0014] In one possible implementation, the methylation detection model includes a first feature extraction module, a long short-term memory network module, a second feature extraction module, a gated recurrent unit, and a linear layer. The first feature extraction module extracts a first feature from the input matrix to obtain a first feature. The long short-term memory network module extracts temporal dynamic information from the first feature and integrates it to obtain a second feature. The second feature extraction module extracts a second feature from the second feature to obtain a third feature. The gated recurrent unit extracts long-term dependencies from the third feature and integrates them to obtain a fourth feature. The linear layer obtains the detection result based on the fourth feature. Thus, through deep learning architectures (LSTM, GRU, etc.), the methylation detection model can automatically learn complex patterns and rules in the data, effectively reducing human error and improving the accuracy and reliability of methylation detection.

[0015] In one possible implementation, the device further includes a training module for acquiring a training sample set, which includes multiple samples and corresponding sample labels. The sample labels indicate whether a sample is methylated or not. The initial methylation detection model is trained using the training sample set until the loss function value of the current methylation detection model meets a preset condition, resulting in a trained methylation detection model. The loss function value includes a first function value and a second function value. The first function value is calculated based on the label information of each sample pair, the Euclidean distance between the two samples in each sample pair, and a preset boundary coefficient. The second function value is calculated based on the sample label of each sample and the predicted probability of methylation for each sample. Thus, by combining the first and second function values ​​to train the methylation detection model, the model can not only learn the relationship between sample labels and methylation prediction probabilities but also consider the similarity and differences between samples. This dual optimization method enables the methylation detection model to more accurately capture methylation features, thereby improving detection accuracy.

[0016] In one possible implementation, determining the final detection result based on the detection results of all target sequencing signals includes: processing the detection results of each target sequencing signal to obtain a corresponding processed probability; determining the final probability based on the product of all processed probabilities; and using the final probability as the final detection result. In this way, by processing and integrating the detection results of all target sequencing signals, information from multiple sequencing runs can be comprehensively considered, avoiding misjudgments that may be caused by random errors or noise in a single sequencing run, thereby improving the accuracy and reliability of methylation detection results.

[0017] According to another aspect of this disclosure, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-described method.

[0018] According to another aspect of this disclosure, a non-volatile computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the above-described method.

[0019] According to another aspect of this disclosure, a computer program product is provided, including a computer program or a non-volatile computer-readable storage medium carrying the computer program, wherein the computer program, when executed by a processor, implements the steps of the above-described method.

[0020] This disclosure obtains the target motif from the RNA sequence to be detected, performs at least one nanopore direct RNA sequencing on the target motif, and obtains the target sequencing signal from each sequencing. Based on each target sequencing signal, it determines the corresponding index result, which includes signal index result and alignment index result. The signal index result represents the statistical characteristics of the corresponding target sequencing signal, and the alignment index result represents the anomalies in the base sequence identified based on the corresponding target sequencing signal. It performs word embedding on the index results of each target sequencing signal to obtain the corresponding input matrix, and inputs each input matrix into a methylation detection model to calculate the detection result of each target sequencing signal. Each detection result represents the probability of methylation of the target motif predicted based on the corresponding target sequencing signal. Based on the detection results of all target sequencing signals, it determines the final detection result, which represents the probability of methylation of the target motif. In this way, by combining the signal index result and the alignment index result, and fusing contextual information through word embedding technology, high-precision methylation detection is achieved.

[0021] Other features and aspects of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0022] The accompanying drawings, which are included in and form part of this specification, illustrate exemplary embodiments, features, and aspects of this disclosure together with the specification and serve to explain the principles of this disclosure.

[0023] Figures 1 to 2 This diagram illustrates a method for detecting methylation of ribonucleic acid provided in an embodiment of this disclosure.

[0024] Figure 3 This diagram illustrates the detection performance of the methylation detection model and other models provided in the embodiments of this disclosure.

[0025] Figure 4 A block diagram of a methylation detection apparatus for ribonucleic acid provided in an embodiment of this disclosure is shown. Detailed Implementation

[0026] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0027] As used herein, the terms “comprising,” “including,” “having,” or variations thereof are open-ended and include one or more of the stated features, integrals, elements, steps, components, or functions, but do not exclude the presence or addition of one or more other features, integrals, elements, steps, components, functions, or groups thereof.

[0028] When an element is referred to as “connected,” “coupled,” “responding,” or a variation thereof relative to another element, it may be directly connected, coupled, or responding to another element, or there may be an intermediate element present.

[0029] Although the terms first, second, third, etc., may be used herein to describe various elements / operations, these elements / operations should not be limited by these terms. These terms are only used to distinguish one element / operation from another. Therefore, without departing from the teachings of the inventive concept, a first element / operation in some embodiments may be referred to as a second element / operation in other embodiments.

[0030] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0031] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.

[0032] To facilitate understanding of the technical solutions provided by the embodiments of this disclosure by those skilled in the art, the technical environment for implementing the technical solutions will be described below.

[0033] Currently, over 170 chemical modifications have been identified in RNA. As one of the most common RNA modifications, m6A is present in approximately 25% of messenger ribonucleic acid (mRNA). In eukaryotic RNA, m6A is primarily found in the DRACH or RRACH motif. m6A affects the structure and stability of mRNA and plays a crucial role in various biological processes, such as mRNA translation, heterogeneous nuclear ribonucleic acid (pre-mRNA) splicing, and microRNA processing.

[0034] Currently, methods relying on immunoprecipitation or chemoselective modification, such as MeRIP-seq, miCLIP, and m6ACE-Seq, are the gold standard for detecting RNA m6A. However, these methods have limitations, such as lack of single-nucleotide resolution, isoform ambiguity, and experimental bias. Recently, a research team proposed m6A-SAC-Seq using the dimethyltransferase MjDim1 to label m6A. Another research group proposed GLORI, a method for quantifying m6A by using glyoxal and nitrite to mediate the deamination of unmethylated adenosine while maintaining the integrity of m6A. Both m6A-SAC-Seq and GLORI can quantify m6A modification in mammalian transcriptomes at single-nucleotide resolution. However, both methods still require complex experimental steps during library preparation.

[0035] Nanopore direct RNA sequencing (DRS) technology provides a novel approach for detecting RNA m6A. When RNA molecules pass through a nanopore, different nucleotides in the RNA generate different electrical signals, and modifications within these nucleotides also affect these signals. Recently, several statistical and machine learning-based methods have been developed to identify m6A modifications from nanopore DRS data.

[0036] Statistical methods, including DRUMMER, ELIGOOS, Yanocomp, Nanocompore, and xPore, identify m6A sites by comparing differences between two groups of samples: one group contains samples with the m6A sites to be identified, and the other group contains unmodified control samples. This method does not require pre-training. However, the control samples required by these methods incur additional costs and are sometimes difficult to generate.

[0037] Machine learning-based methods include EpiNano, MINES, nanom6A, m6anet, and DENA. These methods utilize datasets containing both m6A and non-m6A site markers to identify m6A sites from nanopore DRS readings through supervised training, and can perform detection without any control samples. However, the recognition accuracy of these methods still has limitations.

[0038] To address the aforementioned technical problems, this disclosure provides a deep learning-based method for methylation detection of RNA, capable of determining the final detection result of the RNA sequence to be detected, i.e., the probability of methylation of each motif in the RNA sequence, thereby achieving high-precision methylation detection. Now, in conjunction with... Figures 1 to 3 The methylation detection method provided in the embodiments of this disclosure is illustrated below. Figure 1 As shown, the methylation detection method may include the following steps S101 to S104.

[0039] Step S101: Obtain the target motif from the RNA sequence to be detected, and perform at least one nanopore direct RNA sequencing on the target motif to obtain the target sequencing signal from each sequencing.

[0040] Nanopore direct RNA sequencing is a single-molecule sequencing method based on nanopore technology. Its core principle is that when a single strand of RNA passes through a nanopore under voltage, different bases or modifications cause changes in current. These current signals (i.e., sequencing signals) are recorded and used to identify the RNA's base sequence. Based on the obtained RNA sequence, at least one motif can be identified. Therefore, performing nanopore direct RNA sequencing on the obtained RNA sequence—that is, performing nanopore direct RNA sequencing on the target motif—can yield the target sequencing signal corresponding to the motif. The target sequencing signals obtained from different sequencing runs may not be exactly the same.

[0041] After obtaining the RNA sequence to be detected, a base reference sequence matching the RNA sequence can be determined. This base reference sequence refers to the actual base sequence of the corresponding RNA sequence. Since the RNA sequence may include at least one motif, and the number of motifs in the RNA sequence and the position of each motif in the RNA sequence can be known in advance, that is, if the number and position of the motifs included in the RNA sequence are determined, the actual base fragments corresponding to each motif can be determined from the base reference sequence. The specific determination method can be determined according to relevant technologies, and this disclosure does not limit it.

[0042] Based on the obtained RNA sequence to be detected, at least one motif included in the RNA sequence can be identified. Each motif can be considered a target motif, or only the motif of interest can be selected as the target motif; this disclosure does not limit this approach. For each target motif, performing this methylation detection method can yield the probability of methylation occurring for that target motif.

[0043] Step S102: Determine the index results corresponding to each target sequencing signal based on each target sequencing signal. Each index result includes signal index results and alignment index results. The signal index results represent the statistical characteristics of the corresponding target sequencing signal, and the alignment index results represent the abnormalities in the base sequence identified based on the corresponding target sequencing signal.

[0044] Step S102, determining the index results corresponding to each target sequencing signal based on each target sequencing signal, may include: determining a base reference sequence matching the target motif, wherein the base reference sequence is used to indicate the actual base fragment of the target motif, for example, the actual base fragment of the target motif may include 5 bases, and the number of bases included in the target motif can be flexibly set according to actual detection needs, for example, it may be set to include 7 bases; performing base identification on each target sequencing signal to obtain the base identification sequence corresponding to each target sequencing signal, specifically, a suitable base identification tool can be selected to perform base identification on the target sequencing signal according to the actual situation, and this tool may be, but is not limited to, a nanopore sequencer. Each of the base recognition sequences is used to indicate the recognition base fragment of the target motif. Similarly, each recognition base fragment may also include multiple bases, but these recognition base fragments may differ from the real base fragments. For example, a target motif may include the five bases TAACT, but the recognition base fragment corresponding to this target motif is AAACT, where T represents thymine, A represents adenine, and C represents cytosine. The recognition base fragments corresponding to each target sequencing signal are compared with the real base fragments to obtain the alignment index results of each target sequencing signal. The corresponding signal index results are determined based on each target sequencing signal. The embodiments of this disclosure do not specifically limit the order of determining the alignment index results and the signal index results.

[0045] The signal index results include at least one of the following: signal mean, signal median, signal standard deviation, and signal length of the identified base fragment of the target motif. Preferably, the signal index results may include the signal mean, signal median, signal standard deviation, and signal length of each base in the identified base fragment of the target motif. The signal mean of a base refers to the average value of all signals corresponding to that base in the target sequencing signal. The signal median of a base refers to the median of all signals corresponding to that base in the target sequencing signal. The signal standard deviation of a base refers to the standard deviation of all signals corresponding to that base in the target sequencing signal. The signal length of a base refers to the length of all signals corresponding to that base in the target sequencing signal. Taking the signal mean as an example, the mapping relationship between bases and target sequencing signals can be obtained using Tombo, a tool used for depth analysis of nanopore sequencing data. That is, which signals in the target sequencing signal a specific base corresponds to. After determining the specific signals corresponding to a base, the signal mean of that base can be calculated based on these specific signals.

[0046] Specifically, the alignment index results for each target sequencing signal are obtained by comparing the identified base fragment corresponding to each target sequencing signal with the actual base fragment, including at least one of the following: if it is determined that the number of bases in the identified base fragment is greater than the number of bases in the actual base fragment, then the insertion information of the identified base fragment is determined to have an insertion; if it is determined that the number of bases in the identified base fragment is less than the number of bases in the actual base fragment, then the missing information of the identified base fragment is determined to have a missing information; if it is determined that the bases at the same position in the identified base fragment and the actual base fragment are different, then the matching information of the identified base fragment is determined to have a mismatch; and the quality score of the identified base fragment is determined based on the comparison results of the bases at the same position in the identified base fragment and the actual base fragment.

[0047] The alignment index results can indicate anomalies in the base sequence obtained from the target sequencing signal, such as deletions. The alignment index results include at least one of the following: quality score, insertion information, deletion information, and matching information of the identified base fragment of the target motif. Preferably, the alignment index results may include the quality score, insertion information, deletion information, and matching information of the identified base fragment of the target motif. The quality score refers to the confidence score in base identification, which is an indicator for quantitatively evaluating the accuracy of identification of each base in the target motif using nanopore direct RNA sequencing technology. The insertion information indicates whether a new base exists in the identified base fragment of the target motif. If, compared to the actual base fragment of the target motif, the identified base fragment of the target motif contains an additional base, then the insertion information of the identified base fragment of the target motif indicates the presence of an insertion; otherwise, no insertion exists. Missing information indicates whether the identified base fragment of the target motif is missing any bases. If the identified base fragment of the target motif is missing one base compared to the actual base fragment, then the missing information indicates a missing base; otherwise, no missing base is found. Matching information indicates whether the identified base fragment of the target motif contains bases that do not match those in the actual base fragment. For example, if the target motif includes the five bases TAACT, but the corresponding identified base fragment is AAACT, this indicates a mismatch, specifically a mismatch at the first position.

[0048] Step S103: Word embedding is performed on the index results of each target sequencing signal to obtain the corresponding input matrix, and each input matrix is ​​input into the methylation detection model to calculate the detection result of each target sequencing signal. Each detection result represents the probability of methylation of the target motif predicted based on the corresponding target sequencing signal.

[0049] For example, word embedding of the index results of a target sequencing signal for a target motif yields a corresponding input matrix. This input matrix represents the relevant information of all bases in the target motif. Specifically, the input matrix can include 128 dimensions of information, of which 120 dimensions represent the identified base fragments of the target motif, 4 dimensions represent the alignment index results of the target motif, and the remaining 4 dimensions represent the signal index results of the target motif. By inputting this input matrix of the target sequencing signal into a methylation detection model for predicting the probability of methylation, the predicted probability of methylation of the target motif based on the target sequencing signal can be obtained. Specifically, this can be the probability of methylation at a target site within the target motif. The specific location of the target site is selected based on actual conditions, and this embodiment does not limit this. By calculating the input matrix corresponding to each target sequencing signal of the target motif using the methylation detection model, the detection result corresponding to each target sequencing signal can be obtained.

[0050] The methylation detection model may include a first feature extraction module, a long short-term memory network module, a second feature extraction module, a gated recurrent unit, and a linear layer.

[0051] The first feature extraction module can be used to extract first features from the input matrix. Specifically, the first feature extraction involves the model representation after global modeling. It can capture the correlation between local sequence features and global context by combining self-attention mechanisms and convolution operations. The first feature extraction module can employ, but is not limited to, the Conformer module. The Conformer module can include self-attention modules based on self-attention mechanisms, convolutional modules, and feedforward networks. The self-attention module is one of the core parts of the Conformer module, mainly responsible for capturing global dependencies in the sequence. It dynamically adjusts the weights of information by calculating the relationship between each position in the sequence and other positions, thereby extracting global features. The convolutional module is mainly used to extract local features in the sequence. It captures local patterns and structural information in the sequence through convolutional operations. The sequence can be used to indicate the index results corresponding to the target sequencing signal. Specifically, the methylation detection model predicts the probability of methylation of the target motif by calculating the input matrix of the sequence, i.e., the corresponding target sequencing signal. The convolutional module can use depthwise separable convolution to improve computational efficiency. The feedforward network can be mainly used to perform further nonlinear transformations on the features extracted by the convolutional module to enhance the model's expressive power.

[0052] The Long Short-Term Memory (LSTM) module is used to extract temporal dynamic information from the first feature and integrate it to obtain the second feature. The LSTM module may include a first LSTM submodule and a second LSTM submodule. The first LSTM submodule is used to forward process the first feature output by the first feature extraction module to obtain a first processing result, and the second LSTM submodule is used to reverse process the first feature output by the first feature extraction module to obtain a second processing result. Specifically, the LSTM module integrates (or "merges") the first and second processing results at each time step to obtain the final output, i.e., the second feature. The forward processing in the first LSTM submodule can be considered as an ascending order, which typically refers to the sequence of bases obtained through base identification. The reverse processing in the second LSTM submodule can be considered as an inverted order, i.e., the reverse sequence of the ascending order. Taking the sequence of bases obtained through base identification as AAACT as an example, the forward processing in the LSTM module refers to the extraction and integration of temporal dynamic information along the AAACT direction, and the reverse processing refers to the extraction and integration of temporal dynamic information along the TCAAA direction. The processing of the first LSTM submodule and the second LSTM submodule can be performed simultaneously or sequentially.

[0053] The second feature extraction module can be used to extract a third feature from the second feature. This second feature extraction captures the correlation between local sequence features and global context information by combining self-attention mechanisms and convolutional operations, while preventing gradient vanishing. The second feature extraction module can employ, but is not limited to, an improved Conformer module. Besides self-attention modules, convolutional modules, and feedforward networks (not elaborated further, see previous text), the improved Conformer module adds residual connections to the standard Conformer module, thus preventing gradient vanishing.

[0054] The gated recurrent unit (GRU) can be used to extract long-term dependencies from the third feature and integrate them to obtain the fourth feature. The GRU may include a first gated recurrent subunit and a second gated recurrent subunit. The first gated recurrent subunit can be used to forward process the third feature output by the second feature extraction module to obtain the third processing result. The second gated recurrent subunit can be used to reverse process the third feature to obtain the fourth processing result. Specifically, at each time step, the gated recurrent unit integrates (or "assembles") the first and second processing results as the output of the current time step, and only extracts the output of the last time step as the fourth feature for probability prediction of whether it is a methylation site. The forward direction in the forward processing corresponding to the first gated recurrent subunit can be considered as the forward order, which usually refers to the order of bases obtained sequentially through base recognition. The reverse direction in the reverse processing corresponding to the second gated recurrent subunit can be considered as the reverse order, i.e., the opposite order of the forward order. Taking the sequence of bases obtained through base identification as AAACT as an example, the forward processing in the gated recurrent unit refers to the extraction and integration of long-term dependencies along the AAACT direction, while the reverse processing refers to the extraction and integration of long-term dependencies along the TCAAA direction.

[0055] The linear layer can be used to obtain the detection result based on the fourth feature, that is, to reduce the dimension of the fourth feature to 1 dimension, and to activate the transformation of the last 1-dimensional feature using the sigmoid function to obtain the probability of methylation of the target motif in the range of [0,1], that is, the detection result of methylation detection.

[0056] For example, such as Figure 2 As shown, the methylation detection model of this embodiment can be structured as follows: 4 Conformer modules, 2 bidirectional LSTM modules, 4 improved Conformer modules, 2 bidirectional GRU modules / units, and a fully connected layer (also called a "linear layer"). The Conformer modules and improved Conformer modules can be used for sequence information interaction and context modeling, while the bidirectional LSTM and bidirectional GRU can be used to extract feature vectors that can characterize the entire text.

[0057] This methylation detection method may further include a training process for a methylation detection model: acquiring a training sample set, which includes multiple samples and corresponding sample tags. Each sample includes a sample sequencing signal obtained by performing nanopore direct RNA sequencing on the sample RNA sequence. Specifically, the sample RNA sequence can be subjected to at least one nanopore direct RNA sequencing to obtain the sample sequencing signal for each sequencing. The meaning and acquisition process of the sample sequencing signal are similar to the meaning and acquisition process of the target sequencing signal, as detailed above, and will not be repeated here. The sample RNA sequence may include multiple different RNA sequences, and each RNA sequence may include at least one sample motif. Thus, the sample sequencing signal in the training sample set may include sample sequencing signals of multiple sample motifs. The sample tag indicates whether the corresponding sample is methylated or not, i.e., whether the corresponding sample motif is methylated. The initial methylation detection model is trained using the training sample set until the loss function value of the current methylation detection model meets a preset condition (e.g., the loss function value is minimized), thus obtaining the trained methylation detection model. The loss function value includes a first function value and a second function value. The first function value is calculated based on the label information of each sample pair, the Euclidean distance between the two samples in each sample pair, and a preset boundary coefficient. The second function value is calculated based on the sample label of each sample and the predicted probability of methylation corresponding to each sample.

[0058] For example, the methylation detection model can be trained using both contrastive learning and multi-instance learning.

[0059] The computational principle of contrastive learning can be expressed by the following equation 1:

[0060]

[0061] In Equation 1, Loss Contrastive Let y represent the first function value, N represent the number of sample pairs in the training sample set (the number of sample pairs is 1 / 2 of the total number of samples), and y represent the first function value. i This represents the label information for sample pairs, indicating whether the sample pairs are similar. This label information is calculated by the methylation detection model; d i This represents the Euclidean distance between sample pairs; m is a predefined boundary used to control the minimum distance between dissimilar samples, which can be set to 0.1. During the training of the methylation detection model, two samples (i.e., a sample pair) can be input simultaneously into the currently trained methylation detection model. If the methylation detection model calculates that these two samples belong to the same category, for example, both belong to the target category (or both are positive samples), then the y calculated by the methylation detection model... iThis can be represented by 1; if the methylation detection model calculates that the two samples belong to different categories, for example, one sample is a positive sample and the other is a negative sample, then the y calculated by the methylation detection model is... i It can be represented by 0, where a positive sample is a sample belonging to the target category, which represents the category with methylation, and a negative sample is a sample belonging to a non-target category, which represents the category without methylation. A positive sample can be a sample sequencing signal of a sample motif with methylation, and a negative sample can be a sample sequencing signal of a sample motif without methylation.

[0062] Multi-instance learning achieves convergence training through binary cross-entropy loss, and its calculation principle can be expressed by the following equation 2:

[0063]

[0064] In Equation 2, Loss BCE q represents the value of the second function, M represents the number of samples in the training sample set; i Indicates sample label; p i This represents the probability that samples in the training sample set are methylated, as predicted by the methylation detection model currently in training.

[0065] Multi-instance learning involves ensembling the predicted probabilities of sequencing signals corresponding to each read length of a sample motif to obtain the predicted probabilities of methylation sites at the transcriptome level. The calculation principle can be expressed by Equation 3 below:

[0066]

[0067] In Equation 3, P t This represents the integration probability of methylation at the same site in sequencing signals of different read lengths from the sample motif, where n represents the number of sequencing signals of different read lengths from the sample motif. This represents the predicted probability that the sequencing signal of the k-th read will be methylated at the same site. It can be calculated using the sigmod function shown in Equation 4 below:

[0068]

[0069] In Equation 4, x represents the dimensionality reduction result of the fourth feature output by the gated recurrent unit of the methylation detection model in the current training. Equation 4 can be used to obtain the probability of methylation of the target site in the target motif within the range of [0,1].

[0070] Step S104: Determine the final detection result based on the detection results of all target sequencing signals. The final detection result represents the probability that the target motif is methylated.

[0071] Step S104, determining the final detection result based on the detection results of all target sequencing signals, may include: processing the detection results of each target sequencing signal to obtain a corresponding processed probability; determining the final probability based on the product of all processed probabilities; and using the final probability as the final detection result. For example, the final detection result can be determined using the following formula 5:

[0072]

[0073] In Equation 5, P final Let j represent the final probability, and j represent the number of target sequencing signals corresponding to the target motif. This represents the detection result of the q-th target sequencing signal. This represents the probability after processing.

[0074] This disclosure compares the performance of the provided methylation detection model with other m6A modification recognition models built using deep learning neural network methods, such as RedNano, m6Anet, and DENA. The results are shown in Table 1. Figure 3 It can be seen that this methylation detection model performs well in terms of accuracy, recall, false positive rate (FPR), and true positive rate (TPR), or sensitivity, for three types of RNA molecules produced by artificial synthesis technology, Arabidopsis, and humans.

[0075] Table 1. Performance comparison of different RNA m6A methylation site recognition methods

[0076]

[0077] This disclosure also provides a methylation detection device for ribonucleic acid (RNA), comprising: an acquisition module for acquiring a target motif in a RNA sequence to be detected, performing at least one nanopore direct RNA sequencing on the target motif to obtain a target sequencing signal from each sequencing; an index determination module for determining an index result corresponding to each target sequencing signal based on each target sequencing signal, wherein each index result includes a signal index result and an alignment index result, wherein the signal index result represents the statistical characteristics of the corresponding target sequencing signal, and the alignment index result represents the anomalies in the base sequence identified based on the corresponding target sequencing signal; a model prediction module for performing word embedding on the index results of each target sequencing signal to obtain a corresponding input matrix, and inputting each input matrix into a methylation detection model to calculate the detection result of each target sequencing signal, wherein each detection result represents the probability of methylation of the target motif predicted based on the corresponding target sequencing signal; and a result determination module for determining a final detection result based on the detection results of all target sequencing signals, wherein the final detection result represents the probability of methylation of the target motif.

[0078] In one possible implementation, determining the index results corresponding to each of the target sequencing signals based on each target sequencing signal includes: determining a base reference sequence that matches the target motif, the base reference sequence being used to indicate the actual base fragment of the target motif; performing base identification on each of the target sequencing signals to obtain a base identification sequence corresponding to each target sequencing signal, each base identification sequence being used to indicate the identified base fragment of the target motif; comparing the identified base fragment corresponding to each of the target sequencing signals with the actual base fragment to obtain the alignment index results of each of the target sequencing signals; and determining the corresponding signal index results based on each of the target sequencing signals; wherein each alignment index result includes at least one of the quality score, insertion information, deletion information, and matching information of the identified base fragment, and the signal index results include at least one of the signal mean, signal median, signal standard deviation, and signal length of the identified base fragment.

[0079] In one possible implementation, the identified base fragments corresponding to each target sequencing signal are compared with the actual base fragments to obtain the alignment index results of each target sequencing signal, including at least one of the following: if it is determined that the number of bases in the identified base fragment is greater than the number of bases in the actual base fragment, then the insertion information of the identified base fragment is determined to have an insertion; if it is determined that the number of bases in the identified base fragment is less than the number of bases in the actual base fragment, then the missing information of the identified base fragment is determined to have a missing information; if it is determined that the bases at the same position in the identified base fragment and the actual base fragment are different, then the matching information of the identified base fragment is determined to have a mismatch; based on the comparison results of the bases at the same position in the identified base fragment and the actual base fragment, the quality score of the identified base fragment is determined.

[0080] In one possible implementation, the methylation detection model includes a first feature extraction module, a long short-term memory network module, a second feature extraction module, a gated recurrent unit, and a linear layer. The first feature extraction module is used to extract a first feature from the input matrix to obtain a first feature. The long short-term memory network module is used to extract the temporal dynamic information from the first feature and integrate it to obtain a second feature. The second feature extraction module is used to extract a second feature from the second feature to obtain a third feature. The gated recurrent unit is used to extract the long-term dependencies from the third feature and integrate them to obtain a fourth feature. The linear layer is used to obtain the detection result based on the fourth feature.

[0081] In one possible implementation, the device further includes a training module for acquiring a training sample set, which includes multiple samples and corresponding sample labels, wherein the sample labels indicate whether the corresponding sample is methylated or not. The initial methylation detection model is trained using the training sample set until the loss function value of the current methylation detection model meets a preset condition, resulting in a trained methylation detection model. The loss function value includes a first function value and a second function value. The first function value is calculated based on the label information of each sample pair, the Euclidean distance between the two samples in each sample pair, and a preset boundary coefficient. The second function value is calculated based on the sample label of each sample and the predicted probability of methylation occurring for each sample.

[0082] In one possible implementation, determining the final detection result based on the detection results of all target sequencing signals includes: processing the detection results of each target sequencing signal to obtain a corresponding processed probability; determining the final probability based on the product of all processed probabilities; and using the final probability as the final detection result.

[0083] In some embodiments, the functions or modules of the apparatus provided in this disclosure can be used to perform the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0084] This disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0085] This disclosure also provides a non-volatile computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described method.

[0086] This disclosure also provides a computer program product, including a computer program or a non-volatile computer-readable storage medium carrying the computer program, wherein the computer program, when executed by a processor, implements the steps of the above method.

[0087] Figure 4 A block diagram of a methylation detection apparatus for ribonucleic acid provided in an embodiment of this disclosure is shown. For example, apparatus 1900 may be provided as a server or terminal device. (Refer to...) Figure 4 The apparatus 1900 includes a processing component 1922, which further includes one or more processors, and memory resources represented by memory 1932 for storing instructions, such as application programs, that can be executed by the processing component 1922. The application programs stored in memory 1932 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 1922 is configured to execute instructions to perform the methods described above.

[0088] Device 1900 may also include a power supply component 1926 configured to perform power management of device 1900, a wired or wireless network interface 1950 configured to connect device 1900 to a network, and an input / output interface 1958 (I / O interface). Device 1900 can operate on an operating system, such as Windows Server, stored in memory 1932. TM macOS X TM Unix TM Linux TM FreeBSD TM Or similar.

[0089] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions that can be executed by a processing component 1922 of the device 1900 to perform the above-described method.

[0090] Computer-readable storage media can be tangible devices capable of holding and storing programs / instructions used by instruction execution devices. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0091] The computer program (or computer-readable program instructions) described herein can be downloaded from a computer-readable storage medium to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage medium in the respective computing / processing device.

[0092] The computer program (or computer program instructions) used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuits, such as programmable logic circuits, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), are personalized by utilizing state information of computer-readable program instructions. These electronic circuits can execute computer-readable program instructions to implement various aspects of this disclosure.

[0093] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0094] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0095] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0096] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0097] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for detecting methylation of ribonucleic acid, characterized in that, include: Obtain the target motif from the RNA sequence to be detected, and perform at least one nanopore direct RNA sequencing on the target motif to obtain the target sequencing signal from each sequencing. Based on each of the target sequencing signals, the corresponding index results are determined. Each index result includes signal index results and alignment index results. The signal index results represent the statistical characteristics of the corresponding target sequencing signal, and the alignment index results represent the abnormalities in the base sequences identified based on the corresponding target sequencing signals. Word embedding is performed on the index results of each target sequencing signal to obtain the corresponding input matrix. Each input matrix is ​​then input into the methylation detection model to calculate the detection result of each target sequencing signal. Each detection result represents the probability of methylation of the target motif predicted based on the corresponding target sequencing signal. The final detection result is determined based on the detection results of all target sequencing signals, and the final detection result represents the probability that the target motif is methylated.

2. The method according to claim 1, characterized in that, Based on the target sequencing signals, the corresponding index results for each target sequencing signal are determined, including: A base reference sequence matching the target motif is determined, the base reference sequence being used to indicate the actual base fragment of the target motif; The target sequencing signals are subjected to base identification of the target motif to obtain the base identification sequence corresponding to each target sequencing signal. Each base identification sequence is used to indicate the identification base fragment of the target motif. The identification base fragments corresponding to each target sequencing signal are compared with the real base fragments to obtain the alignment index results of each target sequencing signal, and the corresponding signal index results are determined based on each target sequencing signal. The comparison index results include at least one of the quality score, insertion information, missing information, and matching information of the identified base fragment, and the signal index results include at least one of the signal mean, signal median, signal standard deviation, and signal length of the identified base fragment.

3. The method according to claim 2, characterized in that, The identified base fragments corresponding to each target sequencing signal are compared with the actual base fragments to obtain the alignment index results for each target sequencing signal, including at least one of the following: If it is determined that the number of bases in the identified base fragment is greater than the number of the actual base fragment, then the insertion information of the identified base fragment is determined to be that an insertion exists; If it is determined that the number of bases in the identified base fragment is less than the number of the actual base fragment, then the missing information of the identified base fragment is determined to be missing. If it is determined that the bases at the same position in the identified base fragment and the real base fragment are different, then the matching information of the identified base fragment is determined to be a mismatch. The quality fraction of the identified base fragment is determined based on the comparison results of the bases at the same position in the identified base fragment and the actual base fragment.

4. The method according to any one of claims 1 to 3, characterized in that, The methylation detection model includes a first feature extraction module, a long short-term memory network module, a second feature extraction module, a gated recurrent unit, and a linear layer; The first feature extraction module is used to perform first feature extraction on the input matrix to obtain the first feature; The long short-term memory network module is used to extract the time dynamic information in the first feature and integrate it to obtain the second feature; The second feature extraction module is used to extract a second feature from the second feature to obtain a third feature; The gated loop unit is used to extract long-term dependencies from the third feature and integrate them to obtain the fourth feature; The linear layer is used to obtain the detection result based on the fourth feature.

5. The method according to any one of claims 1 to 3, characterized in that, The method further includes: Obtain a training sample set, which includes multiple samples and corresponding sample labels, wherein the sample labels indicate whether the corresponding sample is methylated or not. The initial methylation detection model is trained using the training sample set until the loss function value of the current methylation detection model meets the preset conditions, thus obtaining the trained methylation detection model. The loss function value includes a first function value and a second function value. The first function value is calculated based on the label information of each sample pair, the Euclidean distance between the two samples in each sample pair, and the preset boundary coefficient. The second function value is calculated based on the sample label of each sample and the predicted probability of methylation occurring for each sample.

6. The method according to any one of claims 1 to 3, characterized in that, The final detection results are determined based on the detection results of all target sequencing signals, including: The detection results of the sequencing signals of each target are processed to obtain the corresponding processed probabilities; The final probability is determined by multiplying all processed probabilities, and the final probability is used as the final detection result.

7. A device for detecting methylation of ribonucleic acid, characterized in that, include: The acquisition module is used to acquire the target motif in the RNA sequence to be detected, and to perform at least one nanopore direct RNA sequencing on the target motif to obtain the target sequencing signal from each sequencing. The indicator determination module is used to determine the indicator results corresponding to each of the target sequencing signals based on each of the target sequencing signals. Each indicator result includes signal indicator results and alignment indicator results. The signal indicator results represent the statistical characteristics of the corresponding target sequencing signal, and the alignment indicator results represent the abnormalities in the base sequences identified based on the corresponding target sequencing signals. The model prediction module is used to perform word embedding on the index results of each target sequencing signal to obtain the corresponding input matrix, and input each input matrix into the methylation detection model to calculate the detection result of each target sequencing signal. Each detection result represents the probability of methylation of the target motif predicted based on the corresponding target sequencing signal. The result determination module is used to determine the final detection result based on the detection results of all target sequencing signals, wherein the final detection result represents the probability that the target motif has been methylated.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 6.

9. A non-volatile computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product comprising a computer program, or a non-volatile computer-readable storage medium carrying a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images