Biological sample processing method, device, equipment and storage medium

Abstract

The present application discloses a biological sample processing method, device, equipment and storage medium, which belongs to the field of data processing technology. The method includes: performing methylation sequencing on the nucleic acid of the biological sample to obtain the state information of the methylation site in the genome of the nucleic acid; based on the state information of the methylation site, obtaining the characterization feature of the methylation site; based on the characterization feature of the methylation site, extracting the characterization feature of the biological sample; performing cancer identification on the biological sample based on the characterization feature of the biological sample to obtain the cancer identification result of the biological sample. In the above-mentioned method, device, equipment and medium, the state information of the methylation site can reflect the characteristics of the nucleic acid of the biological sample to a large extent, and the characterization feature of the biological sample is obtained by considering the characteristics of the nucleic acid of the biological sample, which can more reliably characterize the biological sample, and performing cancer identification on the biological sample based on such characterization feature is conducive to improving the accuracy of the cancer identification result of the biological sample.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of data processing technology, and in particular to a biological sample processing method, device, equipment and storage medium. Background Art

[0002] With the development of data processing technology, more and more methods for processing biological samples (such as blood, urine, etc.) have been derived.

[0003] By processing biological samples, cancer identification results of the biological samples can be obtained, and then it can be determined whether the biological samples belong to certain cancer categories based on the cancer identification results, or the cancer risk of the biological samples can be analyzed based on the cancer identification results.

[0004] How to provide a processing method that can improve the accuracy of cancer identification results of biological samples is a problem that needs to be solved urgently. Summary of the invention

[0005] The embodiments of the present application provide a biological sample processing method, device, equipment and storage medium, which can be used to improve the accuracy of cancer identification results of biological samples.

[0006] In one aspect, an embodiment of the present application provides a method for processing a biological sample, the method comprising:

[0007] Performing methylation sequencing on the nucleic acid of the biological sample to obtain status information of the methylation sites in the genome of the nucleic acid;

[0008] Based on the state information of the methylation site, obtaining a characterization feature of the methylation site;

[0009] Extracting the characteristic features of the biological sample based on the characteristic features of the methylation sites;

[0010] Cancer identification is performed on the biological sample based on the characterizing features of the biological sample to obtain a cancer identification result of the biological sample.

[0011] On the other hand, the present invention provides another method for processing a biological sample, the method comprising:

[0012] Acquire a cancer identification tag of a first biological sample and state information of a methylation site of the first sample, wherein the state information of the methylation site of the first sample is obtained by performing methylation sequencing on a nucleic acid of the first biological sample;

[0013] Based on the state information of the methylation site of the first sample, obtaining a characterization feature of the methylation site of the first sample;

[0014] Invoking a first feature extraction model to extract a feature of the first biological sample based on the feature of the methylation site of the first sample;

[0015] Calling a first recognition model to perform cancer recognition on the first sample biological sample based on the characterization feature of the first sample biological sample to obtain a cancer recognition result of the first sample biological sample;

[0016] updating the parameters of the first feature extraction model and the parameters of the first recognition model based on the difference between the cancer recognition result of the first sample biological sample and the cancer recognition label of the first sample biological sample to obtain a trained feature extraction model and a trained recognition model;

[0017] The trained feature extraction model and the trained recognition model are used to process biological samples to obtain cancer recognition results of the biological samples.

[0018] On the other hand, an embodiment of the present application provides a biological sample processing device, the device comprising:

[0019] An acquisition module, used to perform methylation sequencing on the nucleic acid of the biological sample to obtain status information of the methylation sites in the genome of the nucleic acid;

[0020] The acquisition module is further used to acquire the characterization features of the methylation site based on the state information of the methylation site;

[0021] An extraction module, used for extracting the characteristic features of the biological sample based on the characteristic features of the methylation site;

[0022] The identification module is used to perform cancer identification on the biological sample based on the characterization features of the biological sample to obtain a cancer identification result of the biological sample.

[0023] In a possible implementation, the acquisition module is used to convert the state information of the methylation site into a first feature; convert the auxiliary characterization information of the methylation site into a second feature, wherein the auxiliary characterization information is used to assist in characterizing the methylation site based on the state information of the methylation site; and fuse the first feature and the second feature to obtain a characterization feature of the methylation site.

[0024] In a possible implementation, the auxiliary characterization information includes at least one of the first information, the second information, the third information, the fourth information or the fifth information; wherein the first information is used to indicate the position of the methylation site in the genome of the nucleic acid; the second information is used to indicate the chromosome to which the methylation site belongs; the third information is used to indicate the position of the methylation site in the chromosome to which it belongs; the fourth information is used to indicate the positional relationship between the methylation site and the functional region in the chromosome to which it belongs; and the fifth information is used to indicate the positional relationship between the methylation site and the gene region in the chromosome to which it belongs.

[0025] In a possible implementation, the extraction module is used to perform N rounds of feature transformation based on the characterization features of the methylation site, and obtain the characterization features of the biological sample based on the transformed features; wherein the i-th round of feature transformation in the N rounds of feature transformation is used to obtain the output features corresponding to the i-th round of feature transformation based on the input features corresponding to the i-th round of feature transformation; i is an integer less than or equal to N; N is an integer greater than or equal to 1; the input features corresponding to the first round of feature transformation in the N rounds of feature transformation are the characterization features of the methylation site; and the output features corresponding to the N-th round of feature transformation in the N rounds of feature transformation are the transformed features.

[0026] In one possible implementation, the extraction module is used to perform pre-processing operations on the input features corresponding to the i-th round of feature transformation to obtain the pre-processing features corresponding to the i-th round of feature transformation; transform the pre-processing features corresponding to the i-th round of feature transformation based on the attention mechanism to obtain the transformed features corresponding to the i-th round of feature transformation; perform post-processing operations on the transformed features corresponding to the i-th round of feature transformation to obtain the output features corresponding to the i-th round of feature transformation.

[0027] In a possible implementation, the extraction module is used to fuse the transformation features corresponding to the i-th round of feature transformation and the input features corresponding to the i-th round of feature transformation to obtain the fused features corresponding to the i-th round of feature transformation; normalize the fused features corresponding to the i-th round of feature transformation to obtain the normalized features corresponding to the i-th round of feature transformation; activate the normalized features corresponding to the i-th round of feature transformation to obtain the activated features corresponding to the i-th round of feature transformation; pool the activated features corresponding to the i-th round of feature transformation to obtain the output features corresponding to the i-th round of feature transformation.

[0028] In a possible implementation, the extraction module is used to call the trained feature extraction model to extract the characteristic features of the biological sample based on the characteristic features of the methylation site;

[0029] The recognition module is used to call the trained recognition model to perform cancer recognition on the biological sample based on the characterization features of the biological sample to obtain a cancer recognition result of the biological sample.

[0030] In a possible implementation, the number of the methylation sites is at least two, and the trained feature extraction model is used to perform feature extraction on sequence data;

[0031] The extraction module is used to input the characterizing features of at least two methylation sites as the sequence data into the trained feature extraction model to perform feature extraction, and use the features output by the trained feature extraction model as the characterizing features of the biological sample.

[0032] In a possible implementation, the number of the methylation sites is at least two, and the trained feature extraction model is used to perform feature extraction on the graph structure;

[0033] The extraction module is used to construct the graph structure based on the characterization features of at least two methylation sites, input the graph structure into the trained feature extraction model to perform feature extraction, and use the features output by the trained feature extraction model as the characterization features of the biological sample.

[0034] In a possible implementation, the extraction module is used to fuse the characterization features of the methylation site and the reference features of the biological sample to obtain a biological fusion feature; the reference feature is obtained based on reference biological information, and the reference biological information is obtained by processing the biological sample in a manner different from the methylation sequencing; based on the biological fusion feature, the characterization features of the biological sample are extracted.

[0035] In a possible implementation, the extraction module is used to extract a first characterization feature of the biological sample based on the characterization feature of the methylation site; extract a second characterization feature of the biological sample based on a reference feature of the biological sample; the reference feature is obtained based on reference biological information, and the reference biological information is obtained by processing the biological sample in a manner different from the methylation sequencing; and the first characterization feature and the second characterization feature are fused to obtain the characterization feature of the biological sample.

[0036] In a possible implementation, the cancer identification result of the biological sample includes a healthy category and a predicted probability of at least one cancer category; or, the cancer identification result of the biological sample includes a cancer risk value.

[0037] On the other hand, an embodiment of the present application provides another biological sample processing device, the device comprising:

[0038] an acquisition module, configured to acquire a cancer identification tag of a first biological sample and state information of a methylation site of the first sample, wherein the state information of the methylation site of the first sample is obtained by performing methylation sequencing on a nucleic acid of the first biological sample;

[0039] The acquisition module is further used to acquire the characterization features of the methylation site of the first sample based on the state information of the methylation site of the first sample;

[0040] An extraction module, configured to call a first feature extraction model to extract a feature of the first biological sample based on the feature of the methylation site of the first sample;

[0041] an identification module, configured to call a first identification model to perform cancer identification on the first sample biological sample based on the characterization feature of the first sample biological sample, and obtain a cancer identification result of the first sample biological sample;

[0042] an updating module, configured to update the parameters of the first feature extraction model and the parameters of the first recognition model based on the difference between the cancer recognition result of the first sample biological sample and the cancer recognition label of the first sample biological sample, so as to obtain a trained feature extraction model and a trained recognition model;

[0043] The trained feature extraction model and the trained recognition model are used to process biological samples to obtain cancer recognition results of the biological samples.

[0044] In a possible implementation manner, the device further includes:

[0045] A training module is used to obtain the state information of the methylation site of the second sample, where the state information of the methylation site of the second sample is obtained by performing the methylation sequencing on the nucleic acid of the second sample biological sample; based on the state information of the methylation site of the second sample, obtain the characterization features of the methylation site of the second sample; based on the characterization features of the methylation site of the second sample and the self-supervised learning task, train the initial feature extraction model to obtain the first feature extraction model.

[0046] On the other hand, a computer device is provided, comprising a processor and a memory, wherein the memory stores at least one computer program, and the at least one computer program is loaded and executed by the processor so that the computer device implements any of the above-mentioned biological sample processing methods.

[0047] On the other hand, a computer-readable storage medium is provided, wherein at least one computer program is stored in the computer-readable storage medium, and the at least one computer program is loaded and executed by a processor so that a computer implements any of the above-mentioned biological sample processing methods.

[0048] On the other hand, a computer program or a computer program product is also provided, wherein the computer program or the computer program product comprises computer instructions, and the computer instructions are loaded and executed by a processor so that a computer implements any of the above-mentioned biological sample processing methods.

[0049] The technical solution provided by the embodiments of the present application brings at least the following beneficial effects:

[0050] The technical solution provided in the embodiment of the present application is based on the characterization characteristics of the biological sample to achieve the process of performing cancer identification on the biological sample, and the characterization characteristics of the biological sample are obtained on the basis of the state information of the methylation site. Since the state information of the methylation site is obtained by performing methylation sequencing on the nucleic acid of the biological sample, the state information of the methylation site can reflect the characteristics of the nucleic acid of the biological sample to a large extent, and the characterization characteristics of the biological sample obtained on the basis of the state information of the methylation site can characterize the biological sample by considering the characteristics of the nucleic acid of the biological sample. Since the nucleic acid of the biological sample is the most basic substance of the biological sample, the characterization characteristics of the biological sample can more reliably characterize the biological sample, and performing cancer identification on the biological sample based on such characterization characteristics is conducive to improving the accuracy of the cancer identification results of the biological sample. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0052] Figure 1 is a schematic diagram of the structure of a computer system provided in an embodiment of the present application;

[0053] Figure 2 is a flow chart of a biological sample processing method provided in an embodiment of the present application;

[0054] Figure 3 is a flow chart of another biological sample processing method provided in an embodiment of the present application;

[0055] Figure 4 is a flow chart of an i-th round of feature transformation provided in an embodiment of the present application;

[0056] Figure 5 is a schematic diagram of a process of an i-th round of feature transformation provided in an embodiment of the present application;

[0057] Figure 6 is a flow chart of another biological sample processing method provided in an embodiment of the present application;

[0058] Figure 7 is a flow chart of another biological sample processing method provided in an embodiment of the present application;

[0059] Figure 8 is a flow chart of another biological sample processing method provided in an embodiment of the present application;

[0060] Fig. 9 is a flow chart of another biological sample processing method provided in an embodiment of the present application;

[0061] Fig.10 is a schematic diagram of a biological sample processing process provided in an embodiment of the present application;

[0062] Fig.11 is a schematic diagram of another biological sample processing process provided in an embodiment of the present application;

[0063] Fig.12 is a schematic diagram of an interaction process between a server and a cancer identification client provided in an embodiment of the present application;

[0064] Fig.13 is a schematic diagram of a biological sample processing device provided in an embodiment of the present application;

[0065] Fig.14 is a schematic diagram of another biological sample processing device provided in an embodiment of the present application;

[0066] Fig.15 is a schematic diagram of the structure of a terminal provided in an embodiment of the present application;

[0067] Fig.16 It is a structural diagram of a server provided in an embodiment of the present application. DETAILED DESCRIPTION

[0068] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0069] For ease of understanding, the terms involved in this application are explained.

[0070] Biological sample: A biological sample is any sample taken from an organism. An organism can be any living or non-living organism, including but not limited to humans, non-human animals, plants, bacteria, fungi, or protists. Biological samples include blood, whole blood, plasma, serum, urine, cerebrospinal fluid, feces, saliva, sweat, tears, pleural fluid, pericardial fluid, or peritoneal fluid, etc.

[0071] Nucleic acid: The general term for deoxyribonucleic acid (DNA) and ribonucleic acid (RNA). It is a biological macromolecule compound composed of many nucleotide monomers and is one of the most basic substances of life.

[0072] Genome: The complete genetic information of an organism expressed as nucleic acid sequences.

[0073] Methylation: Methylation is a form of chemical modification of nucleic acids that can change genetic expression without changing the nucleic acid sequence. During the methylation process, the hydrogen atoms on the pyrimidine ring of the cytosine base are converted to methyl groups to form 5-methylcytosine. Methylation tends to occur at the dinucleotides of cytosine and guanine at CpG sites. In other cases, methylation can occur at a portion of cytosine other than the CpG site or at another nucleotide that is not cytosine. Among them, the CpG site is a special DNA sequence whose base sequence is cytosine (C) and guanine (G), and cytosine and guanine are connected by a phosphodiester bond (p).

[0074] Methylation site: A site in the genome of a nucleic acid of a biological sample that supports the occurrence of methylation.

[0075] Methylation sequencing: a sequencing method that obtains information about the status of methylation sites in the genome of nucleic acids in biological samples. For example, methylation sequencing can be achieved by whole genome bisulfite sequencing (WGBS).

[0076] Chromosome: The genetic material that carries the genes of an organism, composed of DNA, RNA and protein.

[0077] Pre-processing operation: The operation that processes features before the transformation based on the attention mechanism.

[0078] Post-processing operations: operations that process features after the attention-based transformation.

[0079] Self-supervised learning: A machine learning method that lies between supervised learning and unsupervised learning. Unlike supervised learning, which relies on a large amount of labeled data, self-supervised learning learns the internal structure and feature representation of data by generating pseudo labels or pre-training tasks using unlabeled data. This enables self-supervised learning to effectively train high-performance models in the absence of labeled data.

[0080] Cancer, as one of the major global health threats, caused nearly 10 million deaths in 2020, and its incidence rate showed a significant upward trend with age. The occurrence of cancer is usually the result of a series of gene mutations and epigenetic changes in normal cells, and these changes often occur before clinical symptoms appear. Therefore, early screening is crucial to improving the survival rate of cancer patients. Traditional screening methods such as imaging examinations and biomarker detection have problems such as low sensitivity, lack of specificity and invasiveness, and it is urgent to develop more effective early screening technologies. The processing method of biological samples provided in the embodiments of the present application can assist in the early screening of cancer. In other words, the cancer identification results of biological samples obtained by the processing method of biological samples can be used to provide reference data for early screening of cancer.

[0081] Figure 1 The schematic diagram of the structure of the computer system provided in the embodiment of the present application is shown. The computer system may include: a computer device 11.

[0082] The computer device 11 can process the biological sample according to the method provided in the embodiment of the present application to obtain a cancer identification result of the biological sample.

[0083] The computer device 11 may be a terminal or a server, which is not limited in the embodiments of the present application. For example, the terminal may be any electronic product that can perform human-computer interaction with a user through one or more methods such as a keyboard, a touchpad, a touch screen, a remote control, voice interaction, or a handwriting device, such as a PC (Personal Computer), a mobile phone, a smart phone, a PDA (Personal Digital Assistant), a wearable device, a PPC (Pocket PC), a tablet computer, a smart car machine, a smart TV, a smart speaker, a car terminal, etc. The server may be a single server, a server cluster consisting of multiple servers, or a cloud computing service center.

[0084] See also Figure 1, the computer device 11 can obtain the biological sample 01 to be processed, perform methylation sequencing on the nucleic acid of the biological sample 01, and obtain the state information 02 of the methylation site in the genome of the nucleic acid of the biological sample 01; then, based on the state information 02 of the methylation site, obtain the characterization feature 03 of the methylation site. Next, based on the characterization feature 03 of the methylation site, extract the characterization feature 04 of the biological sample. Finally, based on the characterization feature 04 of the biological sample, perform cancer identification on the biological sample 01 to obtain the cancer identification result 05 of the biological sample 01.

[0085] In some embodiments, the process of extracting the characterization feature 04 of the biological sample can be implemented by calling the trained feature extraction model 06, and the process of obtaining the cancer identification result 05 of the biological sample 01 can be implemented by calling the trained identification model 07. In this case, the computer device 11 can also obtain the cancer identification label of the first sample biological sample; based on the cancer identification label of the first sample biological sample, the first feature extraction model 08 and the first identification model 09 are trained by supervised training to obtain the trained feature extraction model 06 and the trained identification model 07.

[0086] Those skilled in the art should understand that the above-mentioned computer device 11 is only an example, and other existing or future terminals or servers that are applicable to the present application should also be included in the protection scope of the present application and are included here by reference.

[0087] The present invention provides a method for processing a biological sample, which can be applied to the above Figure 1 The computer system shown in FIG. 1 is taken as an example of the method being applied to a computer device. Figure 2 As shown, the biological sample processing method provided in the embodiment of the present application may include the following steps 201 to 204.

[0088] In step 201, methylation sequencing is performed on the nucleic acid of the biological sample to obtain status information of methylation sites in the genome of the nucleic acid.

[0089] A biological sample refers to any sample taken from an organism, and an organism may refer to any living or non-living organism, including but not limited to humans, non-human animals, plants, bacteria, fungi or protists. A biological sample may reflect a biological state associated with an organism, and a biological sample includes nucleic acids. Exemplarily, biological samples include blood, whole blood, plasma, serum, urine, cerebrospinal fluid, feces, saliva, sweat, tears, pleural fluid, pericardial fluid or peritoneal fluid, etc. Exemplarily, the biological sample in the embodiment of the present application is a sample collected in a non-invasive manner, such as blood, to reduce the discomfort and risk of cancer identification to the organism to which the biological sample belongs. A non-invasive method refers to a sample collection method in which the instrument does not enter the body of the organism. Since the instrument does not enter the body of the organism, the risk is less and the discomfort of the organism is less.

[0090] Nucleic acid in biological samples refers to biological macromolecules composed of many nucleotide monomers in biological samples, which are one of the most basic substances of life. Nucleic acid is a type of biopolymer, an essential component of all known life forms, the most important substance among all biological molecules, and widely present in all animal and plant cells and microorganisms. Exemplarily, nucleic acid in biological samples can include deoxyribonucleic acid (DNA) and ribonucleic acid (RNA).

[0091] For example, in the case where the biological sample is blood, the nucleic acid of the biological sample may refer to cfDNA (Circulating Free DNA) in the biological sample. cfDNA refers to DNA free in the blood. During the blood collection process, a non-invasive blood sample collection kit needs to be configured to ensure that the methylation state remains stable during transportation.

[0092] Methylation sequencing is a sequencing method that obtains the status information of methylation sites in the genome of nucleic acids in biological samples. For example, methylation sequencing can be achieved by whole genome bisulfite sequencing (WGBS). Exemplarily, methylation sequencing is a high-throughput sequencing method that can obtain whole genome sequencing data. Genome refers to the complete genetic information of an organism or virus expressed in a nucleic acid sequence.

[0093] Methylation is a form of chemical modification of nucleic acids that can change genetic expression without changing the nucleic acid sequence. During the methylation process, hydrogen atoms on the pyrimidine ring of the cytosine base are converted to methyl groups to form 5-methylcytosine. In particular, methylation tends to occur at dinucleotides of cytosine and guanine in CpG sites. In other cases, methylation may occur on a cytosine that is not part of a CpG site or on another nucleotide that is not cytosine. As is well known in the art, abnormal methylation can cause different effects, which may lead to cancer. Abnormal methylation is associated with many disease processes, including cancer, and specific patterns of methylation have been determined to be associated with specific cancer symptoms.

[0094] A methylation site refers to a site in the genome of the nucleic acid of a biological sample that supports methylation, such as a CpG site. It should be noted that although the methylation site supports methylation, during the methylation sequencing process, the methylation site may or may not be methylated, which is related to the actual situation of the methylation site and the sequencing environment. It should be noted that the methylation site can be all sites in the genome of the nucleic acid of a biological sample that support methylation, or it can be a part of the sites in the genome of the nucleic acid of a biological sample that support methylation, and this embodiment of the present application does not limit this. For example, the methylation site may include 1M (million) sites in the genome of the nucleic acid of a biological sample that support methylation.

[0095] The state information of the methylation site is used to indicate the specific situation of methylation at the methylation site. For example, the specific situation of methylation at the methylation site may refer to whether the methylation site is methylated, or the probability of methylation at the methylation site. The state information of the methylation site may be expressed in various forms.

[0096] Exemplarily, the status information of the methylation site is represented by methylation, non-methylation or no detection result. If the status information of the methylation site is methylation, it means that the methylation site is methylated; if the status information of the methylation site is non-methylation, it means that the methylation site is not methylated; if the status information of the methylation site is no detection result, it means that it is impossible to determine whether the methylation site is methylated.

[0097] Exemplarily, the status information of the methylation site is represented by high methylation, low methylation or no detection result. If the status information of the methylation site is high methylation, it means that the probability of methylation at the methylation site is high; if the status information of the methylation site is low methylation, it means that the probability of methylation at the methylation site is low; if the status information of the methylation site is no detection result, it means that the probability of methylation at the methylation site cannot be determined.

[0098] Exemplarily, the status information of the methylation site is represented by a methylation ratio, which is positively correlated with the probability of methylation at the methylation site.

[0099] In an exemplary embodiment, the state information of the methylation site is determined based on the results of multiple methylation sequencing of the methylation site. The result of each methylation sequencing of the methylation site is a first result or a second result, the first result is used to indicate that the methylation site is methylated during the current methylation sequencing process, and the second result is used to indicate that the methylation site is not methylated during the current methylation sequencing process.

[0100] In the case where the state information of the methylation site is represented by methylation, non-methylation or no detection result, the method of determining the state information of the methylation site based on the results of multiple methylation sequencing of the methylation site can be: according to the results of multiple methylation sequencing, the number of first results is counted, and the ratio of the number of first results to the number of results of all methylation sequencing is calculated. If the ratio is greater than a first threshold, methylation is used as the state information of the methylation site; if the ratio is less than a second threshold, non-methylation is used as the state information of the methylation site; if the ratio is not less than the second threshold and not greater than the first threshold, no detection result is used as the state information of the methylation site. The first threshold is greater than or equal to the second threshold, for example, the first threshold and the second threshold are both 0.5; or, the first threshold is 0.7 and the second threshold is 0.3.

[0101] In the case where the state information of the methylation site is represented by high methylation, low methylation or no detection result, the method of determining the state information of the methylation site according to the results of multiple methylation sequencing of the methylation site can be: according to the results of multiple methylation sequencing, the number of first results is counted, and the ratio of the number of first results to the number of results of all methylation sequencing is calculated. If the ratio is greater than the third threshold, high methylation is used as the state information of the methylation site; if the ratio is less than the fourth threshold, low methylation is used as the state information of the methylation site; if the ratio is not less than the fourth threshold and not greater than the third threshold, no detection result is used as the state information of the methylation site. The third threshold is greater than or equal to the fourth threshold, for example, the third threshold and the fourth threshold are both 0.5; or, the third threshold is 0.8 and the fourth threshold is 0.2. The third threshold may be the same as the first threshold, or different from the first threshold; the fourth threshold may be the same as the second threshold, or different from the second threshold.

[0102] In the case where the state information of the methylation site is represented by a methylation ratio, the method of determining the state information of the methylation site based on the results of multiple methylation sequencing of the methylation site can be: according to the results of multiple methylation sequencing, the number of first results is counted, the ratio of the number of first results to the number of results of all methylation sequencing is calculated, and the value that is positively correlated with the ratio is used as the methylation ratio, and the methylation ratio is used as the state information of the methylation site. Exemplarily, the ratio is directly used as the methylation ratio, or the product of the ratio and a positive number or the ratio is used as the methylation ratio, etc.

[0103] In step 202, based on the state information of the methylation site, a characterization feature of the methylation site is obtained.

[0104] The characterization feature of the methylation site is a feature used to characterize the methylation site. The number of methylation sites is at least one, and for each methylation site, the characterization feature of the methylation site is obtained based on the state information of the methylation site.

[0105] The representation form of the features involved in the embodiments of the present application can be flexibly set as needed, and the embodiments of the present application do not limit this, as long as it is in a form that can be directly recognized and processed by a computer device. For example, the representation form of the features can be a vector or a matrix.

[0106] In one possible implementation, see Figure 3 , Figure 2 Step 202 in the method may be replaced by the following steps 2021 to 2023. That is, based on the state information of the methylation site, the process of obtaining the characterization feature of the methylation site may include the following steps 2021 to 2023.

[0107] Step 2021: Convert the state information of the methylation site into the first feature.

[0108] The first feature is an embedded feature used to characterize the state information of the methylation site. Embedding is the process of mapping discrete symbols (such as words, characters, categories, etc.) to a continuous vector space. In other words, the first feature is the feature obtained after mapping the state information of discrete methylation sites to a continuous vector space. Exemplarily, the first feature may refer to a vector of a first dimension, and by converting the state information of each methylation site into a first feature, the state information of different methylation sites can be converted into vectors of the same dimension. The first dimension can be set based on experience, or it can be flexibly adjusted according to the computing resources of the computer device, etc., and the embodiments of the present application do not limit this. For example, the first dimension may refer to 1 dimension, or it may refer to 4 dimensions, etc.

[0109] The embodiments of the present application do not limit the implementation method of converting the state information of the methylation site into the first feature. Exemplarily, the state information of the methylation site is encoded according to any data encoding method, and the vector obtained after encoding is used as the first feature. For example, the data encoding method may include but is not limited to one-hot encoding, binary encoding, hash encoding, etc. Exemplarily, based on the correspondence between the state information and the encoding vector, the encoding vector corresponding to the state information of the methylation site is searched, and the found encoding vector is used as the first feature. Exemplarily, a conversion model is called to convert the state information of the methylation site into a vector, and the vector is used as the first feature. The conversion model can be any model that can convert the state information of the methylation site into a vector.

[0110] For example, the state information of the methylation site may be methylation, non-methylation or no test result, the encoding vector corresponding to methylation is 1, the encoding vector corresponding to non-methylation is 0, and the encoding vector corresponding to no test result is -1. If the state information of the methylation site is methylation, the encoding vector 1 is used as the first feature; if the state information of the methylation site is non-methylation, the encoding vector 0 is used as the first feature; if the state information of the methylation site is no test result, the encoding vector -1 is used as the first feature.

[0111] Step 2022: converting the auxiliary characterization information of the methylation site into a second feature, wherein the auxiliary characterization information is used to assist in characterizing the methylation site based on the state information of the methylation site.

[0112] In the process of obtaining the characterization features of the methylation sites, in addition to considering the state information of the methylation sites, the auxiliary characterization information of the methylation sites is also considered to ensure the comprehensiveness of the information considered in the process of obtaining the characterization features of the methylation sites and improve the reliability of the characterization features of the methylation sites.

[0113] The embodiment of the present application does not limit the type of auxiliary characterization information of the methylation site, as long as it can assist in characterizing the methylation site.

[0114] In some embodiments, the auxiliary characterization information of the methylation site includes at least one of the first information, the second information, the third information, the fourth information or the fifth information; wherein the first information is used to indicate the position of the methylation site in the genome of the nucleic acid; the second information is used to indicate the chromosome to which the methylation site belongs; the third information is used to indicate the position of the methylation site in the chromosome to which it belongs; the fourth information is used to indicate the positional relationship between the methylation site and the functional region in the chromosome to which it belongs; and the fifth information is used to indicate the positional relationship between the methylation site and the gene region in the chromosome to which it belongs.

[0115] Since the methylation site is a site determined from the genome of the nucleic acid of the biological sample, the position of the methylation site in the genome of the nucleic acid can be determined. The embodiment of the present application does not limit the representation form of the first information, as long as it can indicate the position of the methylation site in the genome of the nucleic acid.

[0116] Exemplarily, the positions of all sites in the genome of the nucleic acid are numbered sequentially, and the number corresponding to the position of the methylation site is used as the first information. In this case, the first information is represented by the number corresponding to the position of the methylation site.

[0117] Exemplarily, the distance between the position of the methylation site and the position of the reference site in the genome is used as the first information. In this case, the first information is represented by the distance between the position of the methylation site and the position of the reference site in the genome. The reference site can be the first site, the last site, or a site at another position.

[0118] The genome of the nucleic acid of the biological sample includes the genes carried by all the chromosomes of the biological sample, that is, the genome of the nucleic acid of the biological sample can be divided into genes corresponding to each chromosome of the biological sample, and according to the position of the methylation site in the genome of the nucleic acid, it can be determined which chromosome the methylation site belongs to, so that the chromosome is used as the chromosome to which the methylation site belongs. The embodiment of the present application does not limit the representation form of the second information, as long as it can indicate the chromosome to which the methylation site belongs.

[0119] Exemplarily, all chromosomes of the biological sample are numbered respectively, and the number corresponding to the chromosome to which the methylation site belongs is used as the second information. In this case, the second information is represented by the number corresponding to the chromosome to which the methylation site belongs.

[0120] Exemplarily, the name of the chromosome to which the methylation site belongs is used as the second information. In this case, the second information is represented by the name of the chromosome to which the methylation site belongs.

[0121] After determining the chromosome to which the methylation site belongs, it is possible to further determine which site in the gene corresponding to the chromosome to which the methylation site belongs, thereby using the position of the determined site in the gene corresponding to the chromosome as the position of the methylation site in the chromosome to which the methylation site belongs. The embodiment of the present application does not limit the representation form of the third information, as long as it can indicate the position of the methylation site in the chromosome to which it belongs.

[0122] Exemplarily, the positions of all sites in the gene corresponding to the chromosome to which the methylation site belongs are numbered sequentially, and the number of the position of the methylation site in the chromosome to which it belongs is used as the third information. In this case, the third information is represented by the number of the position of the methylation site in the chromosome to which it belongs.

[0123] Exemplarily, the distance between the position of the methylation site in the chromosome to which it belongs and the position of the reference site in the gene corresponding to the chromosome is used as the third information. In this case, the third information is represented by the distance between the position of the methylation site in the chromosome to which it belongs and the position of the reference site in the gene corresponding to the chromosome. The reference site can be the first site, the last site, or a site at another position.

[0124] The positional relationship between the methylation site and the functional region in the chromosome to which it belongs is used to indicate which functional region in each functional region in the chromosome to which it belongs. The functional region in the chromosome refers to the region with different functions in the chromosome, such as, the functional region in the chromosome includes the centromere region, the chromatin arm region, the intercentromere region, and the telomeric region. The embodiment of the present application does not limit the representation of the fourth information, as long as the positional relationship between the methylation site and the functional region in the chromosome to which it belongs can be indicated. Exemplarily, the name or mark of the functional region in which the methylation site is located in each functional region in the chromosome to which it belongs is used as the fourth information. In this case, the fourth information can be represented by the name or mark of the functional region in which the methylation site is located in each functional region in the chromosome to which it belongs.

[0125] The positional relationship between the methylation site and the gene region in the chromosome to which it belongs is used to indicate which gene region in each gene region in the chromosome to which it belongs. The gene region in the chromosome refers to one or more regions in the gene corresponding to the chromosome, and the gene region in the chromosome can be one or more regions that play a specific role in genetics in the gene corresponding to the chromosome, and the gene region in the chromosome can be specified by professional and technical personnel. The embodiment of the application does not limit the representation of the fifth information, as long as the positional relationship between the methylation site and the gene region in the chromosome to which it belongs can be indicated. Exemplarily, the name or mark of the gene region where the methylation site is located in each gene region in the chromosome to which it belongs is used as the fifth information. In this case, the fifth information adopts the name or mark of the gene region where the methylation site is located in each gene region in the chromosome to which it belongs to represent.

[0126] The auxiliary characterization information of the methylation site can assist in characterizing the methylation site from aspects such as the position of the methylation site in the genome of the nucleic acid, the chromosome to which the methylation site belongs, the position of the methylation site in the chromosome to which it belongs, the positional relationship between the methylation site and the functional region in the chromosome to which it belongs, and / or the positional relationship between the methylation site and the gene region in the chromosome to which it belongs. Each of the above aspects can reflect the unique characteristics of the methylation site to a certain extent, so that the auxiliary characterization information can assist in characterizing the methylation site more accurately, which is beneficial to improve the reliability of the characterization characteristics of the methylation site obtained subsequently and improve the processing reliability of biological samples.

[0127] It should be noted that the auxiliary characterization information of the methylation site may include one or more of the first information, the second information, the third information, the fourth information or the fifth information. The specific information included in the auxiliary characterization information of the methylation site can be flexibly selected according to application requirements, and can also be flexibly adjusted according to the computing resources of the computer device, etc., which is not limited in the embodiments of the present application.

[0128] The second feature is an embedded feature of the auxiliary representation information used to represent the methylation site. That is, the second feature is a feature obtained by mapping the auxiliary representation information of the discrete methylation site to the continuous vector space.

[0129] In an exemplary embodiment, in the process of converting the auxiliary characterization information of the methylation site into the second feature, each item of information included in the auxiliary characterization information of the methylation site is converted into a second feature, in which case the number of second features is the same as the number of items of information included in the auxiliary characterization information of the methylation site. The dimensions of the second features obtained after converting different items of information may be the same or different.

[0130] The embodiments of the present application do not limit the implementation method of converting each item of information in the auxiliary characterization information of the methylation site into the second feature. Exemplarily, each item of information in the auxiliary characterization information of the methylation site is encoded according to any data encoding method, and the vector obtained after encoding is used as the second feature. For example, the data encoding method may include but is not limited to one-hot encoding, binary encoding, hash encoding, etc. Exemplarily, based on the correspondence between the information and the encoding vector, the encoding vector corresponding to each item of information in the auxiliary characterization information of the methylation site is searched, and the found encoding vector is used as the second feature. Exemplarily, a conversion model is called to convert each item of information in the auxiliary characterization information of the methylation site into a vector, and the vector is used as the second feature. The conversion model can be any model that can convert each item of information in the auxiliary characterization information of the methylation site into a vector.

[0131] In an exemplary embodiment, in the process of converting the auxiliary characterization information of the methylation site into the second feature, the various information included in the auxiliary characterization information of the methylation site is summarized into an overall information, and the overall information is converted into the second feature. In this case, the auxiliary characterization information of each methylation site corresponds to a second feature. In this case, the second feature can refer to a vector of the second dimension, and by converting the auxiliary characterization information of each methylation site into a second feature, the auxiliary characterization information of different methylation sites can be converted into vectors of the same dimension. The second dimension can be set according to experience, and can also be flexibly adjusted according to the computing resources of a computer device, etc., and the embodiment of the present application does not limit this. For example, the second dimension can refer to 1 dimension, or it can refer to 4 dimensions, etc.

[0132] The embodiment of the present application does not limit the implementation method of converting the overall information obtained by summarizing various information in the auxiliary characterization information of the methylation site into the second feature. Exemplarily, the overall information is encoded according to any data encoding method, and the vector obtained after encoding is used as the second feature. For example, the data encoding method may include but is not limited to one-hot encoding, binary encoding, hash encoding, etc. Exemplarily, according to the correspondence between the information and the encoding vector, the encoding vector corresponding to the overall information is searched, and the found encoding vector is used as the second feature. Exemplarily, the conversion model is called to convert the overall information into a vector, and the vector is used as the second feature. The conversion model can be any model that can convert the overall information into a vector.

[0133] For example, taking the auxiliary characterization information of the methylation site including the first information and the second information as an example, the first information can be converted into a second feature, and the second information can also be converted into a second feature. In the process of converting the first information into the second feature, the number corresponding to the position of the methylation site can be determined according to the first information, and the number is converted into a vector to obtain the second feature; in the process of converting the second information into the second feature, the number corresponding to the chromosome to which the methylation site belongs can be determined according to the second information, and the number is converted into a vector to obtain the second feature.

[0134] Assuming that the positions of all sites in the genome of the nucleic acid are numbered from 1 to 100, the first information can be represented by any number between 1 and 100, and any number between 1 and 100 is converted into a vector to obtain a second feature; assuming that all chromosomes of the biological sample are numbered from 1 to 23, the second information can be represented by any number between 1 and 23, and any number between 1 and 23 is converted into a vector to obtain a second feature.

[0135] Step 2023: The first feature and the second feature are integrated to obtain a characteristic feature of the methylation site.

[0136] After obtaining the first feature and the second feature, the first feature and the second feature are fused, and the fused feature is used as the characterization feature of the methylation site. Based on this, it can be ensured that the characterization feature of the methylation site can comprehensively characterize the methylation site from two perspectives: the state information of the methylation site and the auxiliary characterization information of the methylation site, thereby ensuring the reliability of the methylation site.

[0137] The present application does not limit the manner of fusing the first feature and the second feature. Exemplarily, the manner of fusing the first feature and the second feature may refer to splicing the first feature and the second feature, for example, splicing the first feature and the second feature horizontally or vertically, etc. Exemplarily, the first feature and the second feature are both represented by vectors, and splicing the first feature and the second feature horizontally refers to arranging the elements of the two vectors of the first feature and the second feature together in rows. Splicing the first feature and the second feature vertically refers to arranging the elements of the two vectors of the first feature and the second feature together in columns.

[0138] Exemplarily, when the dimensions of the first feature and the second feature are the same, the first feature and the second feature may be fused by adding elements at corresponding positions in the first feature and the second feature.

[0139] It should be noted that the number of the second features may be one or more. In the case where the number of the second features is more than one, fusing the first feature with the second feature means fusing the first feature with the more than one second features.

[0140] Through the above steps 2021 to 2023, in the process of obtaining the characterization features of the methylation site, not only the state information of the methylation site but also the auxiliary characterization information of the methylation site is considered. The auxiliary characterization information of the methylation site can assist in characterizing the methylation site on the basis of the state information of the methylation site. Based on this, in the process of obtaining the characterization features of the methylation site, more comprehensive information of the methylation site is considered, so that the characterization features of the methylation site can more comprehensively characterize the methylation site, improve the reliability of the characterization features of the methylation site, and then improve the processing reliability of the biological sample.

[0141] It should be noted that the above-described steps 2021 to 2023 are only illustrated by taking into account the state information of the methylation site and the auxiliary characterization information of the methylation site to obtain the characterization characteristics of the methylation site, and the embodiments of the present application are not limited to this. In some embodiments, the characterization characteristics of the methylation site can also be obtained by considering only the state information of the methylation site. Exemplarily, the implementation method of obtaining the characterization characteristics of the methylation site by considering only the state information of the methylation site can be: converting the state information of the methylation site into a first feature, and using the first feature as the characterization feature of the methylation site. The process of converting the state information of the methylation site into the first feature is detailed in step 2021, which will not be repeated here.

[0142] It should be further explained that the above step 202 only takes one methylation site as an example to introduce the implementation process of obtaining the characterization features of the methylation site based on the status information of the methylation site. The number of methylation sites is usually multiple. In the case where the number of methylation sites is multiple, it is necessary to obtain the characterization features of each methylation site based on the status information of each methylation site.

[0143] In step 203, based on the characteristic features of the methylation sites, characteristic features of the biological sample are extracted.

[0144] The characterization feature of a biological sample is a feature used to characterize the biological sample. In the embodiment of the present application, the characterization feature of the biological sample is extracted based on the characterization feature of the methylation site. The characterization feature of the methylation site can reflect the characteristics of the nucleic acid of the biological sample. By considering the characteristics of the nucleic acid of the biological sample to obtain the characterization feature of the biological sample, it is helpful to ensure the reliability of the characterization feature of the biological sample.

[0145] In one possible implementation, see Figure 3 , Figure 2 The step 203 in the example may be replaced by the following step 2031. That is, based on the characteristic features of the methylation sites, the process of extracting the characteristic features of the biological sample may include the following step 2031.

[0146] Step 2031: Perform N rounds of feature transformation based on the characteristic features of the methylation sites, and obtain the characteristic features of the biological sample based on the transformed features.

[0147] Among them, the i-th round of feature transformation in N rounds of feature transformation is used to obtain the output features corresponding to the i-th round of feature transformation based on the input features corresponding to the i-th round of feature transformation; i is an integer less than or equal to N; N is an integer greater than or equal to 1; the input features corresponding to the first round of feature transformation in N rounds of feature transformation are the characterization features of the methylation sites; the output features corresponding to the N-th round of feature transformation in N rounds of feature transformation are the transformed features.

[0148] Exemplarily, for the case where N is an integer greater than 1, the input features corresponding to the first round of feature transformation in N rounds of feature transformation are the characterization features of the methylation sites; the input features corresponding to the p-th round of feature transformation in N rounds of feature transformation are the output features corresponding to the p-1-th round of feature transformation in N rounds of feature transformation. Wherein, p is an integer greater than 1. In other words, the characterization features of the methylation sites are used as the input features corresponding to the first round of feature transformation to perform the first round of feature transformation, and the output features corresponding to the first round of feature transformation are obtained; the output features corresponding to the first round of feature transformation are used as the input features corresponding to the second round of feature transformation to perform the second round of feature transformation, and the output features corresponding to the second round of feature transformation are obtained; the output features corresponding to the second round of feature transformation are used as the input features corresponding to the third round of feature transformation to perform the third round of feature transformation, and the output features corresponding to the third round of feature transformation are obtained, and so on, until the output features corresponding to the Nth round of feature transformation are obtained.

[0149] Obtaining the characterization features of biological samples by performing at least one round of feature transformation based on the characterization features of methylation sites is beneficial to improving the standardization of extracting the characterization features of biological samples, improving the reliability of the characterization features of biological samples, and further improving the processing reliability of biological samples.

[0150] The transformation methods of different rounds of feature transformation in N rounds of feature transformation can be the same or different, and can be flexibly set according to application requirements, computing resources of computer equipment, etc.

[0151] The embodiment of the present application is illustrated by the example that the transformation methods of different rounds of feature transformation in N rounds of feature transformation are the same. In the case where the transformation methods of different rounds of feature transformation in N rounds of feature transformation are the same, the transformation method of the i-th round of feature transformation in N rounds of feature transformation can be set according to experience, or can be flexibly adjusted according to application requirements, and the embodiment of the present application does not limit this. Wherein, i is an integer less than or equal to N, that is, the i-th round of feature transformation is any round of feature transformation in N rounds of feature transformation.

[0152] In some embodiments, see Figure 4 The process of the i-th round of feature transformation includes the following steps 401 to 403.

[0153] Step 401: perform pre-processing operations on input features corresponding to the i-th round of feature transformation to obtain pre-processing features corresponding to the i-th round of feature transformation.

[0154] If the value of i is 1, the input feature corresponding to the i-th round of feature transformation refers to the representation feature of the methylation site. If the value of i is greater than 1, the input feature corresponding to the i-th round of feature transformation refers to the output feature corresponding to the i-1-th round of feature transformation.

[0155] Pre-processing operation refers to the operation of processing the input features corresponding to the i-th round of feature transformation before the transformation based on the attention mechanism, that is, the operation that needs to be performed before the transformation based on the attention mechanism in order to obtain the features to be transformed based on the attention mechanism. Through the pre-processing operation, the input features corresponding to the i-th round of feature transformation can be converted into features suitable for transformation based on the attention mechanism (that is, pre-processing features).

[0156] The specific implementation method of performing pre-processing operations on the input features corresponding to the i-th round of feature transformation can be set based on experience or flexibly adjusted according to application requirements, and the embodiments of the present application do not limit this.

[0157] In an exemplary embodiment, the implementation method of performing the pre-processing operation on the input features corresponding to the i-th round of feature transformation may be: normalizing the input features corresponding to the i-th round of feature transformation. By normalizing the input features corresponding to the i-th round of feature transformation, the input features corresponding to the i-th round of feature transformation may be converted to a specific scale, thereby eliminating the dimensional influence between the input features corresponding to different rounds of feature transformation.

[0158] The method used to normalize the input features corresponding to the i-th round of feature transformation can be flexibly selected based on experience, or flexibly adjusted based on application requirements, and the embodiments of the present application do not limit this. Exemplarily, the method used to normalize the input features corresponding to the i-th round of feature transformation can be root mean square normalization (Root Mean Square Normalization, RMS Norm). In this case, the process of normalizing the input features corresponding to the i-th round of feature transformation includes: calculating the root mean square value of the input features corresponding to the i-th round of feature transformation, and using the root mean square value to scale the input features corresponding to the i-th round of feature transformation.

[0159] Exemplarily, the method used to normalize the input features corresponding to the i-th round of feature transformation may be layer normalization (Layer Normalization, LayerNorm). In this case, the process of normalizing the input features corresponding to the i-th round of feature transformation includes: mapping the input features corresponding to the i-th round of feature transformation to a distribution with a mean of 0 and a standard deviation of 1.

[0160] Exemplarily, the method used to normalize the input features corresponding to the i-th round of feature transformation may be linear function normalization. In this case, the process of normalizing the input features corresponding to the i-th round of feature transformation includes: performing linear transformation on the input features corresponding to the i-th round of feature transformation, and mapping the input features corresponding to the i-th round of feature transformation to a certain range.

[0161] Of course, the method of normalizing the input features corresponding to the i-th round of feature transformation may also be other methods, which will not be introduced here one by one.

[0162] Step 402: transform the pre-processing features corresponding to the i-th round of feature transformation based on the attention mechanism to obtain the transformed features corresponding to the i-th round of feature transformation.

[0163] The attention mechanism is a method that imitates the human visual and cognitive system, allowing us to focus on relevant parts when processing input data. By introducing the attention mechanism, we can automatically learn and selectively focus on important information in the input data.

[0164] The core of the attention mechanism is to assign different weights to different parts of the input data. Its basic principles include the following steps: 1. Represent input: Represent the input data as a series of matrices, such as query matrix (Query, Q), key matrix (Key, K) and value matrix (Value, V); 2. Calculate attention weight: Calculate the attention weight by calculating the similarity between the query matrix (Q) and the key matrix (K); 3. Weighted sum: According to the calculated attention weight, the value matrix (V) is weighted and summed to obtain the final output.

[0165] Exemplarily, the implementation method of transforming the pre-processing features corresponding to the i-th round of feature transformation based on the attention mechanism to obtain the transformed features corresponding to the i-th round of feature transformation can be: based on the pre-processing features corresponding to the i-th round of feature transformation, obtain the query matrix, key matrix and value matrix; perform similarity calculation on the query matrix and the key matrix to obtain the similarity matrix; normalize the similarity matrix to obtain the attention weight; and weighted sum the attention weight and the value matrix to obtain the transformed features corresponding to the i-th round of feature transformation. The transformed features corresponding to the i-th round of feature transformation can be considered as a context-related summary of the value matrix performed according to the attention weight. Exemplarily, the pre-processing features corresponding to the i-th round of feature transformation are transformed based on the first transformation matrix to obtain the query matrix; the pre-processing features corresponding to the i-th round of feature transformation are transformed based on the second transformation matrix to obtain the key matrix; the pre-processing features corresponding to the i-th round of feature transformation are transformed based on the third transformation matrix to obtain the value matrix. The first transformation matrix is ​​a transformation matrix for obtaining the query matrix, the second transformation matrix is ​​a transformation matrix for obtaining the key matrix, and the third transformation matrix is ​​a transformation matrix for obtaining the value matrix.

[0166] It should be noted that the above introduction is only the basic principle of the attention mechanism. On the basis of the above basic principle, there can be a variety of improvements and optimizations to obtain different types of attention mechanisms. The embodiment of the present application does not limit the type of attention mechanism used. It can be selected based on experience, or it can be flexibly adjusted according to application requirements, computing resources of computer equipment, etc. Exemplarily, the type of attention mechanism may include but is not limited to a single-head attention mechanism, a multi-head attention mechanism, a sparse attention mechanism, and the like.

[0167] The single-headed attention mechanism can map a query matrix and a set of key-value pairs (that is, a key matrix and a value matrix) to an output matrix, where the output matrix is ​​the weighted sum of the value matrix, and the weight is calculated by the similarity between the query matrix and the key matrix. The multi-headed attention mechanism is an improved attention mechanism that can map a query matrix and a set of key-value pairs to an output matrix. However, unlike the single-headed attention mechanism, it projects the query matrix, key matrix, and value matrix into multiple subspaces, and then calculates the attention in each subspace, and finally concatenates the attention outputs of all subspaces to obtain the final output matrix. The multi-headed attention mechanism can be used to calculate the correlation between different parts of the input data. The sparse attention mechanism is an optimized attention mechanism that can map a query vector and a set of key-value pairs to an output matrix. However, unlike the single-headed attention mechanism and the multi-headed attention mechanism, it does not calculate the similarity between the query matrix and all key matrices, but only calculates the similarity between the query matrix and some key matrices, thereby reducing the amount of calculation and memory consumption.

[0168] In some embodiments, in the case where the attention mechanism includes a sparse attention mechanism, the sparse attention mechanism can be implemented by dilated attention, which divides the input features into segments and then sparses each segment by selecting rows at specific intervals.

[0169] Step 403: performing post-processing operations on the transformed features corresponding to the i-th round of feature transformation to obtain output features corresponding to the i-th round of feature transformation.

[0170] Post-processing operations refer to operations that process features obtained after the transformation based on the attention mechanism, that is, operations that need to be performed to obtain output features after the transformation based on the attention mechanism. Through post-processing operations, the transformed features corresponding to the i-th round of feature transformation can be converted into features suitable as inputs for the next round of feature transformation.

[0171] Each round of feature transformation obtains the output features corresponding to that round of feature transformation based on the attention mechanism. The attention mechanism enables the feature transformation process to focus on the important areas of the input features, thereby ensuring the reliability of the output features. In addition, pre-processing operations are performed before the transformation based on the attention mechanism, and post-processing operations are performed after the transformation based on the attention mechanism, which is conducive to ensuring the standardization of the feature transformation process and further improving the reliability of the output features.

[0172] The specific implementation method of performing post-processing operations on the transformed features corresponding to the i-th round of feature transformation can be set based on experience or flexibly adjusted according to application requirements, and the embodiments of the present application do not limit this.

[0173] Alternatively, see Figure 4 , step 403 can be implemented based on the following steps 4031 to 4034, that is, the process of performing post-processing operations on the transformed features corresponding to the i-th round of feature transformation can include the following steps 4031 to 4034.

[0174] Step 4031: Fuse the transformation features corresponding to the i-th round of feature transformation and the input features corresponding to the i-th round of feature transformation to obtain the fused features corresponding to the i-th round of feature transformation.

[0175] By fusing the transformation features corresponding to the i-th round of feature transformation and the input features corresponding to the i-th round of feature transformation, it is possible to determine that the basic information of the input data is not completely lost during the feature transformation process, thereby improving the stability of the feature transformation process.

[0176] Exemplarily, the method of fusing the transformation features corresponding to the i-th round of feature transformation and the input features corresponding to the i-th round of feature transformation may refer to evaluating the transformation features corresponding to the i-th round of feature transformation and the input features corresponding to the i-th round of feature transformation, such as horizontal splicing or vertical splicing.

[0177] Exemplarily, the transformation features corresponding to the i-th round of feature transformation and the input features corresponding to the i-th round of feature transformation are both matrices, and the transformation features corresponding to the i-th round of feature transformation and the input features corresponding to the i-th round of feature transformation have the same dimension. The method of fusing the transformation features corresponding to the i-th round of feature transformation and the input features corresponding to the i-th round of feature transformation can also refer to adding corresponding elements in the transformation features corresponding to the i-th round of feature transformation and the input features corresponding to the i-th round of feature transformation.

[0178] In some embodiments, the process of fusing the transformed features corresponding to the i-th round of feature transformation and the input features corresponding to the i-th round of feature transformation can also be performed as a residual connection on the transformed features corresponding to the i-th round of feature transformation and the input features corresponding to the i-th round of feature transformation.

[0179] Step 4032: Normalize the fused features corresponding to the i-th round of feature transformation to obtain the normalized features corresponding to the i-th round of feature transformation.

[0180] By normalizing the fused features corresponding to the i-th round of feature transformation, the fused features corresponding to the i-th round of feature transformation can be converted to a specific scale, thereby eliminating the dimensional influence between the fused features corresponding to different rounds of feature transformation.

[0181] The method used to normalize the fusion features corresponding to the i-th round of feature transformation can be flexibly selected based on experience, or flexibly adjusted according to application requirements, and the embodiments of the present application do not limit this. For example, the method of normalizing the fusion features corresponding to the i-th round of feature transformation may include but is not limited to root mean square normalization, layer normalization, linear function normalization, etc. The implementation methods of root mean square normalization, layer normalization, and linear function normalization have been introduced in the previous article and will not be repeated here.

[0182] Step 4033: Activate the normalized features corresponding to the i-th round of feature transformation to obtain the activated features corresponding to the i-th round of feature transformation.

[0183] Activation is used to perform nonlinear transformation on the normalized features corresponding to the i-th round of feature transformation, so as to limit the normalized features corresponding to the i-th round of feature transformation within a certain range, so as to avoid the adverse effects of too large or too small values ​​on subsequent processing.

[0184] Exemplarily, the process of activating the normalized features corresponding to the i-th round of feature transformation to obtain the activated features corresponding to the i-th round of feature transformation may include: using an activation function to activate the normalized features corresponding to the i-th round of feature transformation to obtain the activated features corresponding to the i-th round of feature transformation.

[0185] The type of activation function can be set based on experience or flexibly adjusted according to application requirements, and the embodiments of the present application do not limit this. Exemplarily, the type of activation function may include, but is not limited to, Sigmoid (S-shaped) activation function, hyperbolic tangent (tanh) activation function, ReLU (Rectified Linear Unit, corrected linear unit) activation function, Softmax (normalized exponential function) activation function, Swish activation function (a nonlinear activation function), SwiGLU activation function (a nonlinear activation function), SiLU (Sigmoid Linear Unit, S-shaped linear unit) activation function, etc.

[0186] Step 4034: Pool the activation features corresponding to the i-th round of feature transformation to obtain the output features corresponding to the i-th round of feature transformation.

[0187] Pooling is an operation for reducing the dimension of the activated features corresponding to the i-th round of feature transformation. Exemplarily, pooling may include but is not limited to maximum pooling, average pooling, and random pooling.

[0188] Exemplarily, the activation features corresponding to the i-th round of feature transformation include the activation sub-features corresponding to each methylation site, and pooling the activation features corresponding to the i-th round of feature transformation refers to pooling the activation sub-features corresponding to each methylation site. By pooling the activation sub-features corresponding to the methylation sites, the dimension of the activation sub-features corresponding to the methylation sites can be reduced.

[0189] Exemplarily, the activation features corresponding to the i-th round of feature transformation may be in the form of a matrix, and each column element in the matrix represents an activation sub-feature corresponding to a methylation site. In this case, pooling the activation features corresponding to the i-th round of feature transformation may refer to performing column pooling on the activation features corresponding to the i-th round of feature transformation. By aggregating the information of each column, the dimension of the data is reduced while retaining important information.

[0190] The post-processing operations include fusion with input features, normalization, activation and pooling. The transformation features corresponding to the i-th round of feature transformation are processed in sequence through these operations, which is conducive to performing more comprehensive and reliable further processing on the transformation features corresponding to the i-th round of feature transformation, thereby fully ensuring the reliability of the output features corresponding to the i-th round of feature transformation.

[0191] It should be noted that the method described in the above steps 401 to 403 is only an exemplary implementation method of the i-th round of feature transformation, and the embodiments of the present application are not limited to this. In some embodiments, at least one of the pre-processing operations or post-processing operations may not be performed to save computing resources. Exemplarily, the process of the i-th round of feature transformation can also be: based on the attention mechanism, the input features corresponding to the i-th round of feature transformation are transformed to obtain the output features corresponding to the i-th round of feature transformation. Exemplarily, the process of the i-th round of feature transformation can also be: pre-processing operations are performed on the input features corresponding to the i-th round of feature transformation to obtain the pre-processing features corresponding to the i-th round of feature transformation; based on the attention mechanism, the pre-processing features corresponding to the i-th round of feature transformation are transformed to obtain the output features corresponding to the i-th round of feature transformation. Exemplarily, the process of the i-th round of feature transformation can also be: based on the attention mechanism, the input features corresponding to the i-th round of feature transformation are transformed to obtain the transformed features corresponding to the i-th round of feature transformation; post-processing operations are performed on the transformed features corresponding to the i-th round of feature transformation to obtain the output features corresponding to the i-th round of feature transformation.

[0192] It should be further explained that the method described in steps 4031 to 4034 is only an exemplary implementation method of step 403, and the embodiments of the present application are not limited thereto. In some embodiments, step 403 can also be implemented by more or fewer steps. Exemplarily, the implementation process of step 403 can also be: normalizing the transformation features corresponding to the i-th round feature transformation to obtain the normalized features corresponding to the i-th round feature transformation; activating the normalized features corresponding to the i-th round feature transformation to obtain the activated features corresponding to the i-th round feature transformation; pooling the activated features corresponding to the i-th round feature transformation to obtain the output features corresponding to the i-th round feature transformation. Exemplarily, the implementation process of step 403 can also be: fusing the transformation features corresponding to the i-th round feature transformation and the input features corresponding to the i-th round feature transformation to obtain the fused features corresponding to the i-th round feature transformation; normalizing the fused features corresponding to the i-th round feature transformation to obtain the normalized features corresponding to the i-th round feature transformation; activating the normalized features corresponding to the i-th round feature transformation to obtain the output features corresponding to the i-th round feature transformation. Of course, the implementation process of step 403 can also be other processes, which will not be repeated here one by one.

[0193] In an exemplary embodiment, the process of the i-th round feature transformation can be implemented by calling a conversion sub-model, inputting the input features corresponding to the i-th round feature transformation into the conversion sub-model, and the conversion sub-model executes the above i-th round feature transformation process and outputs the output features corresponding to the i-th round feature transformation.

[0194] For example, see Figure 5 , the feature conversion sub-model includes a pre-processing module 501, an attention module 502 and a post-processing module 503. The pre-processing module 501 is used to perform pre-processing operations on the input features 51 corresponding to the i-th round of feature transformation, and output the pre-processing features 52 corresponding to the i-th round of feature transformation; the attention module 502 is used to transform the pre-processing features 52 corresponding to the i-th round of feature transformation based on the attention mechanism, and output the transformed features 53 corresponding to the i-th round of feature transformation; the post-processing module 503 is used to perform post-processing operations on the transformed features 53 corresponding to the i-th round of feature transformation, and output the output features 54 corresponding to the i-th round of feature transformation.

[0195] In some embodiments, see Figure 5 The pre-processing module 501 includes a normalization layer 5011, which is used to perform a pre-processing operation on the input features 51 corresponding to the i-th round of feature transformation by normalizing the input features 51 corresponding to the i-th round of feature transformation. The attention module 502 includes an attention layer 5021, which is used to transform the pre-processing features 52 corresponding to the i-th round of feature transformation based on a single-head attention mechanism, a multi-head attention mechanism, or a sparse attention mechanism.

[0196] The post-processing module 503 includes a residual connection layer 5031, a normalization layer 5032, an activation layer 5033 and a pooling layer 5034. The residual connection layer 5031 is used to fuse the transformation feature 53 corresponding to the i-th round feature transformation and the input feature 51 corresponding to the i-th round feature transformation, and output the fused feature 55 corresponding to the i-th round feature transformation; the normalization layer 5032 is used to normalize the fused feature 55 corresponding to the i-th round feature transformation, and output the normalized feature 56 corresponding to the i-th round feature transformation; the activation layer 5033 is used to activate the normalized feature 56 corresponding to the i-th round feature transformation, and output the activated feature 57 corresponding to the i-th round feature transformation; the pooling layer 5034 is used to pool the activated feature 57 corresponding to the i-th round feature transformation, and output the output feature 54 corresponding to the i-th round feature transformation.

[0197] After completing all N rounds of feature transformation, the output features corresponding to the Nth round of feature transformation are used as transformed features, and then the characterization features of the biological sample are obtained based on the transformed features.

[0198] The embodiment of the present application does not limit the implementation method of obtaining the characterization feature of the biological sample based on the transformed feature. Exemplarily, the implementation method of obtaining the characterization feature of the biological sample based on the transformed feature may be: using the transformed feature as the characterization feature of the biological sample. Exemplarily, the implementation method of obtaining the characterization feature of the biological sample based on the transformed feature may also be: performing an activation operation on the transformed feature to obtain the characterization feature of the biological sample.

[0199] Exemplarily, the method for obtaining the characterization features of the biological sample based on the transformed features can also be: transforming the transformed features based on the attention mechanism to obtain the characterization features of the biological sample. Exemplarily, the attention mechanism used here can be the same as or different from the attention mechanism used in each round of feature transformation. In some embodiments, an attention mechanism that can focus on global attention can be used here, for example, a single-headed attention mechanism or a multi-headed attention mechanism. This attention mechanism can increase the semantic understanding ability by calculating the correlation between all positions, thereby obtaining the characterization features of the biological sample by integrating global information.

[0200] In some embodiments, based on the characteristic features of the methylation sites, the process of extracting the characteristic features of the biological sample can be implemented by calling a trained feature extraction model. For example, see Figure 6 , step 203 can be implemented based on the following step 2032.

[0201] Step 2032: Call the trained feature extraction model to extract the characteristic features of the biological sample based on the characteristic features of the methylation sites.

[0202] Extracting the characterization features of biological samples through the trained feature extraction model is conducive to improving the standardization and reliability of the extracted characterization features of biological samples.

[0203] Exemplarily, the process of extracting the characteristic features of the biological sample by calling the trained feature extraction model based on the characteristic features of the methylation site may include: calling the trained feature extraction model to perform N rounds of feature transformation based on the characteristic features of the methylation site, and obtaining the characteristic features of the biological sample based on the transformed features. The process of each round of feature transformation can refer to the description above, and different processing processes in each round of feature transformation can be implemented by different layers in the trained feature extraction model.

[0204] The trained feature extraction model can be obtained by training the first feature extraction model. The specific training process will be introduced below and will not be introduced here.

[0205] The embodiment of the present application does not limit the type of feature extraction model after training, as long as it can perform feature extraction based on the characterization features of the methylation sites.

[0206] In some embodiments, the number of methylation sites is at least two, and the trained feature extraction model is used to perform feature extraction on the sequence data. In other words, the trained feature extraction model can be a model for performing feature extraction on the sequence data.

[0207] In the case where the trained feature extraction model is used to perform feature extraction on sequence data, the process of calling the trained feature extraction model based on the characterization features of the methylation sites to extract the characterization features of the biological sample includes: inputting the characterization features of at least two methylation sites as sequence data into the trained feature extraction model to perform feature extraction, and using the features output by the trained feature extraction model as the characterization features of the biological sample. Based on this, the trained feature extraction model can extract the characterization features of the biological sample by capturing the long-range dependency in the sequence data composed of the characterization features of the methylation sites, thereby ensuring the reliability of the characterization features of the biological sample.

[0208] The characteristics of at least two methylation sites are arranged in a certain order to obtain sequence data. For example, the characteristics of at least two methylation sites are arranged in the order of the positions of the methylation sites in the genome of the nucleic acid to obtain sequence data.

[0209] Exemplarily, in the case where the trained feature extraction model is used to perform feature extraction on sequence data, the structure of the trained feature extraction model can be any sequence processing model. The sequence processing model can be a model of a Transformer architecture. In some embodiments, the sequence processing model can refer to LongNet (long sequence processing model), which can process sequence data composed of the characteristic features of more than 1 billion methylation sites, thereby fully ensuring the comprehensiveness and reliability of the characteristic features of the biological sample finally extracted.

[0210] In some embodiments, the number of methylation sites is at least two, and the trained feature extraction model is used to perform feature extraction on the graph structure. In other words, the trained feature extraction model can be a model for performing feature extraction on the graph structure.

[0211] In the case where the trained feature extraction model is used to perform feature extraction on a graph structure, the trained feature extraction model is called to extract the feature characteristics of the biological sample based on the feature characteristics of the methylation sites, including: constructing a graph structure based on the feature characteristics of at least two methylation sites, inputting the graph structure into the trained feature extraction model to perform feature extraction, and using the features output by the trained feature extraction model as the feature characteristics of the biological sample. Based on this, the trained feature extraction model can extract the feature characteristics of the biological sample by considering the topological relationship between the methylation sites, thereby ensuring the reliability of the feature characteristics of the biological sample.

[0212] The graph structure is a structure used to characterize the topological relationship between methylation sites. Exemplarily, the process of constructing a graph structure based on the characterization features of at least two methylation sites includes: taking the methylation sites as nodes, constructing edges between nodes according to the position distance between the methylation sites, and taking the characterization features of the methylation sites as the attribute information of the nodes; and obtaining a graph structure based on the nodes, the edges between the nodes, and the attribute information of the nodes.

[0213] Optionally, the method of constructing the edge between nodes according to the position distance between the methylation sites can be: if the position distance between the two methylation sites is less than the distance threshold, then an edge is added between the two nodes corresponding to the two methylation sites. The position distance between the two methylation sites can refer to the distance between the positions of the two methylation sites in the genome of the nucleic acid.

[0214] Exemplarily, in the case where the trained feature extraction model is used to perform feature extraction on a graph structure, the structure of the trained feature extraction model can be any graph structure processing model, for example, a graph attention network (GAT), a graph convolutional network (GCN), etc. By processing methylation data through a graph structure processing model, the relationship between methylation sites can be better captured by considering the topological relationship between methylation sites, thereby improving the expression ability of the model.

[0215] The above-introduced method for extracting the characterization features of biological samples based on the characterization features of methylation sites can be considered as extracting the characterization features of biological samples directly based on the characterization features of methylation sites. In some embodiments, the method for extracting the characterization features of biological samples based on the characterization features of methylation sites can also be: extracting the characterization features of biological samples based on the characterization features of methylation sites and reference features of biological samples, the reference features are obtained based on reference biological information, and the reference biological information is obtained by processing biological samples in a manner different from methylation sequencing. In this way, on the basis of the characterization features of methylation sites, the reference features of biological samples are additionally considered to extract the characterization features of biological samples, which is conducive to enriching the information considered in the process of extracting the characterization features of biological samples, thereby improving the reliability of the extracted characterization features of biological samples and the comprehensiveness of the characterization of biological samples, thereby improving the processing reliability and accuracy of biological samples.

[0216] Exemplarily, the reference biological information may include at least one of transcriptome information of the biological sample or proteome information of the biological sample. The transcriptome information of the biological sample may be obtained by processing the biological sample using a transcriptome extraction method, and the proteome information of the biological sample may be obtained by processing the biological sample using a proteome extraction method.

[0217] The reference feature is used to characterize the reference biological information. The method of obtaining the reference feature based on the reference biological information can be flexibly selected. For example, the reference biological information can be encoded according to any data encoding method, and the vector obtained after encoding is used as the reference feature. For example, the data encoding method may include but is not limited to one-hot encoding, binary encoding, hash encoding, etc. Exemplarily, according to the correspondence between the reference biological information and the encoding vector, the encoding vector corresponding to the reference biological information can be found, and the found encoding vector can be used as the reference feature.

[0218] It should be noted that, in the case where the reference biological information includes multiple types of information, a reference feature can be obtained based on each type of information, that is, the number of reference features is the same as the number of types of information included in the reference biological information.

[0219] In an exemplary embodiment, based on the characteristic features of the methylation sites and the reference features of the biological sample, the implementation methods of extracting the characteristic features of the biological sample include but are not limited to the following two methods.

[0220] Implementation method 1: Fusion of the characterization features of the methylation sites and the reference features of the biological samples to obtain biological fusion features; based on the biological fusion features, extraction of the characterization features of the biological samples.

[0221] The method of fusing the characterization features of the methylation sites and the reference features of the biological samples can be set based on experience, or can be flexibly adjusted according to application requirements, and the embodiments of the present application are not limited to this.

[0222] Exemplarily, the manner of fusing the characteristic feature of the methylation site and the reference feature of the biological sample may refer to splicing the characteristic feature of the methylation site and the reference feature of the biological sample, such as horizontal splicing or vertical splicing.

[0223] Exemplarily, the characterization features of the methylation sites and the reference features of the biological samples are both matrices, and the characterization features of the methylation sites and the reference features of the biological samples have the same dimension. The method of fusing the characterization features of the methylation sites and the reference features of the biological samples may also refer to adding corresponding elements in the characterization features of the methylation sites and the reference features of the biological samples.

[0224] The process of extracting the characterization features of biological samples based on biological fusion features is implemented in the same principle as the process of extracting the characterization features of biological samples directly based on the characterization features of methylation sites introduced in the previous article, and will not be repeated here.

[0225] Implementation method 2: extracting a first characterization feature of a biological sample based on the characterization feature of the methylation site; extracting a second characterization feature of the biological sample based on the reference feature of the biological sample; and fusing the first characterization feature and the second characterization feature to obtain the characterization feature of the biological sample.

[0226] The first characterization feature is a feature used to characterize the biological sample extracted based on the characterization feature of the methylation site. The process of extracting the first characterization feature of the biological sample based on the characterization feature of the methylation site is the same as the process of extracting the characterization feature of the biological sample directly based on the characterization feature of the methylation site introduced in the previous text, and will not be repeated here.

[0227] The second characterization feature is a feature extracted based on the reference feature of the biological sample and used to characterize the biological sample.

[0228] In an exemplary embodiment, based on the reference feature of the biological sample, the process of extracting the second characterizing feature of the biological sample can be: performing M rounds of feature transformation based on the reference feature of the biological sample, and obtaining the second characterizing feature of the biological sample based on the transformed feature. M is an integer greater than or equal to 1, and the value of M can be the same as the value of N, or different from the value of N. The manner of performing feature transformation based on the reference feature of the biological sample and the manner of performing feature transformation based on the characterizing feature of the methylation site can be the same as or different from each other.

[0229] In an exemplary embodiment, the process of extracting the second characterizing feature of the biological sample based on the reference feature of the biological sample may be: calling a reference feature extraction model to extract the second characterizing feature of the biological sample based on the reference feature of the biological sample. The reference feature extraction model is a model for extracting features that can characterize the biological sample based on the reference feature of the biological sample. Exemplarily, the reference feature extraction model may be obtained through training.

[0230] Exemplarily, for reference features obtained based on different reference biological information, the reference feature extraction models may be the same or different.

[0231] The method of fusing the first characterization feature and the second characterization feature can be set based on experience, or can be flexibly adjusted according to application requirements, and the embodiments of the present application are not limited to this.

[0232] Exemplarily, the manner of fusing the first characterizing feature and the second characterizing feature may refer to splicing the first characterizing feature and the second characterizing feature, such as horizontal splicing or vertical splicing.

[0233] Exemplarily, the first characterization feature and the second characterization feature are both matrices, and the first characterization feature and the second characterization feature have the same dimension. The method of fusing the first characterization feature and the second characterization feature may also refer to adding corresponding elements in the first characterization feature and the second characterization feature.

[0234] Based on the characterization features of methylation sites and the reference features of biological samples, the method of extracting the characterization features of biological samples can be regarded as a method of obtaining the characterization features of biological samples by integrating multi-omics data. In the process of integrating multi-omics data, on the basis of methylation data, multi-omics data such as transcriptome and proteome are integrated. This method can capture the interaction between different omics data, provide more comprehensive biological information, improve the comprehensiveness of the characterization features of biological samples, and thus improve the accuracy of cancer identification results of biological samples. Exemplarily, both the above-mentioned implementation method 1 and implementation method 2 can be implemented by calling a multimodal deep learning model. For example, the process of implementing the above-mentioned implementation method 1 by calling a multimodal deep learning model can be regarded as a multi-instance learning method, that is, by comprehensively learning multi-omics data, the features based on which the characterization features of biological samples are extracted are obtained. The process of implementing the above-mentioned implementation method 2 by calling a multimodal deep learning model can be regarded as a tensor decomposition method, that is, the characterization features of biological samples are decomposed into multiple features for separate extraction.

[0235] In step 204, cancer identification is performed on the biological sample based on the characterizing features of the biological sample to obtain a cancer identification result of the biological sample.

[0236] The characterization features of the biological sample can characterize the biological sample. After the characterization features of the biological sample are obtained, cancer identification of the biological sample is achieved based on the characterization features of the biological sample to obtain a cancer identification result of the biological sample.

[0237] The cancer identification result of the biological sample is used to indicate the cancer screening status of the biological sample. The form of the cancer identification result of the biological sample can be flexibly set according to the needs, and the embodiment of the present application does not limit this.

[0238] In a possible implementation, the cancer identification result of the biological sample includes a healthy category and a predicted probability of at least one cancer category. The number and specific content of the cancer categories can be flexibly set according to demand. Exemplarily, at least one cancer category includes adrenal cancer, bile duct cancer, bladder cancer, bone cancer, bone marrow cancer, brain cancer, breast cancer, cervical cancer, colorectal cancer, esophageal cancer, gastric cancer, head cancer, neck cancer, hepatobiliary cancer, kidney cancer, liver cancer, lung cancer, ovarian cancer, pancreatic cancer, pelvic cancer, pleural cancer, prostate cancer, kidney cancer, skin cancer, gastric cancer, testicular cancer, thymus cancer, thyroid cancer, uterine cancer, lymphoma, melanoma, multiple myeloma or at least one of leukemia.

[0239] Exemplarily, the number of cancer categories is at least two, so that multiple cancer types can be screened simultaneously. That is, multiple cancer categories can be screened simultaneously, thereby improving screening efficiency and reducing overall costs. This "one-time test, multiple cancer screening" model can provide the possibility for large-scale screening.

[0240] Exemplarily, the sum of the predicted probability of the healthy category and the predicted probability of all cancer categories is 1 to ensure the standardization of the cancer identification results of the biological sample.

[0241] In some embodiments, for the case where the cancer identification result of a biological sample includes a healthy category and a predicted probability of at least one cancer category, the process of performing cancer identification on the biological sample based on the characterizing features of the biological sample can also be referred to as a process of classification based on the characterizing features of the biological sample.

[0242] In another possible implementation, the cancer identification result of the biological sample includes a cancer risk value. Exemplarily, the cancer risk value can be any value within a reference value range. The reference value range can be set based on experience or flexibly adjusted based on demand, and the present application embodiment does not limit this. For example, the reference value range can refer to a range of 0-1. The closer the cancer risk value is to 0, the lower the cancer risk of the biological sample; the closer the cancer risk value is to 1, the higher the cancer risk of the biological sample.

[0243] In some embodiments, when the cancer identification result of a biological sample includes a cancer risk value, the process of performing cancer identification on the biological sample based on the characterization features of the biological sample may also be referred to as a process of regression based on the characterization features of the biological sample.

[0244] The cancer identification results of biological samples can be expressed by the predicted probability of the category or the cancer risk value. The predicted probability of the category or the cancer risk value is in the form of a numerical value. This numerical form can more intuitively reflect the cancer screening situation of the biological sample.

[0245] In a possible implementation, cancer recognition is performed on a biological sample based on the characteristic features of the biological sample. The process of obtaining the cancer recognition result of the biological sample can be achieved by calling the trained recognition model, see Figure 6 , step 204 can be implemented based on the following step 2041.

[0246] Step 2041: Call the trained recognition model to perform cancer recognition on the biological sample based on the characterization features of the biological sample to obtain a cancer recognition result of the biological sample.

[0247] In other words, the characterizing features of the biological sample are input into the trained recognition model, and the trained recognition model performs cancer recognition on the biological sample based on the characterizing features of the biological sample, and outputs the cancer recognition result of the biological sample.

[0248] Performing cancer recognition on biological samples through the trained recognition model is conducive to improving the standardization and reliability of cancer recognition on biological samples, thereby improving the reliability of cancer recognition results of biological samples.

[0249] The trained recognition model can be obtained by training the first recognition model. The specific training process will be introduced below and will not be introduced here for the time being.

[0250] The embodiments of the present application do not limit the type of the trained recognition model, as long as it can perform cancer recognition on the biological sample based on the characterization features of the input biological sample and output the cancer recognition result of the biological sample. Exemplarily, the trained recognition model can be a classification model, in which case the cancer recognition result of the biological sample is a classification result, and the classification result can include the predicted probability of the health category and at least one cancer category. The trained recognition model can be a regression model, in which case the cancer recognition result of the biological sample is a regression result, and the regression result can include a cancer risk value.

[0251] In some embodiments, the number of trained recognition models can be multiple, and different trained recognition models are used to output different forms of cancer recognition results. For example, the number of trained recognition models can be three, namely, a trained binary classification model, a trained multi-classification model, and a trained regression model. After obtaining the characterization features of the biological sample, the characterization features of the biological sample can be respectively input into the trained binary classification model, the trained multi-classification model, and the trained regression model to obtain the binary classification results output by the trained binary classification model, the multi-classification results output by the trained multi-classification model, and the regression results output by the trained regression model.

[0252] Among them, binary classification results, multi-classification results and regression results are three different forms of cancer identification results. The binary classification results include the predicted probability of the health category and the predicted probability of a cancer category, which is a general category used to distinguish from the health category. The multi-classification results include the predicted probabilities of the health category and one or more specific cancer categories, and the regression results include the cancer risk value.

[0253] Exemplarily, the trained recognition model may include multiple linear layers and activation layers, which sequentially process the characterization features of the biological sample to obtain a cancer recognition result of the biological sample. In some embodiments, the recognition model may also be referred to as a prediction head.

[0254] In some embodiments, the embodiments of the present application can be implemented based on a high-performance computing server and specially developed methylation data analysis software.

[0255] In some embodiments, after obtaining the cancer identification results of biological samples, a risk assessment report can also be automatically generated to convert complex cancer identification results into an easy-to-understand clinical report, which can be presented in a doctor-side application for result interpretation and decision support.

[0256] In an embodiment of the present application, the process of performing cancer identification on a biological sample is implemented based on the characterization features of the biological sample, and the characterization features of the biological sample are obtained on the basis of the state information of the methylation site. Since the state information of the methylation site is obtained by performing methylation sequencing on the nucleic acid of the biological sample, the state information of the methylation site can reflect the characteristics of the nucleic acid of the biological sample to a large extent, and the characterization features of the biological sample obtained on the basis of the state information of the methylation site can characterize the biological sample by considering the characteristics of the nucleic acid of the biological sample. Since the nucleic acid of the biological sample is the most basic substance of the biological sample, the characterization features of the biological sample can more reliably characterize the biological sample, and performing cancer identification on the biological sample based on such characterization features is conducive to improving the accuracy of the cancer identification results of the biological sample.

[0257] The present application also provides a method for processing a biological sample, which can be applied to the above Figure 1 The computer system shown in FIG. 1 is taken as an example of the method being applied to a computer device. Figure 7 As shown, the biological sample processing method provided in the embodiment of the present application may include the following steps 701 to 705.

[0258] In step 701, a cancer identification tag of a first biological sample and status information of a methylation site of the first sample are obtained, and the status information of the methylation site of the first sample is obtained by performing methylation sequencing on the nucleic acid of the first biological sample.

[0259] The first sample biological sample refers to a biological sample with a cancer identification label, and the cancer identification label refers to a standard cancer identification result of the first sample biological sample, which can more accurately indicate the cancer screening status of the first sample biological sample. The cancer identification label of the first sample biological sample can be determined by professionals through analysis of the biological sample.

[0260] The implementation principle and method of performing methylation sequencing on the nucleic acid of the first sample biological sample to obtain the state information of the methylation site of the first sample Figure 2 The implementation principle of step 201 in the illustrated embodiment is the same and will not be described in detail here.

[0261] Exemplarily, the cancer identification tag of the first biological sample and the status information of the methylation site of the first sample are correspondingly stored in a database, and the computer device can extract the cancer identification tag of the first biological sample and the status information of the methylation site of the first sample from the database.

[0262] Exemplarily, the type of the first biological sample is Figure 2 The types of biological samples in the embodiment shown are the same, so as to ensure that the feature extraction model and recognition model obtained by training the first sample biological sample can be used for Figure 2 The biological sample in the illustrated embodiment performs cancer identification more accurately.

[0263] Exemplarily, the number of the first sample biological samples is multiple to ensure the training effect of the feature extraction model and the recognition model. Exemplarily, the number of cancer identification labels corresponding to the multiple first sample biological samples is multiple to ensure the diversity of the samples and the training effect of the feature extraction model and the recognition model.

[0264] In step 702, based on the state information of the methylation site of the first sample, a characterization feature of the methylation site of the first sample is obtained.

[0265] The characterization feature of the first sample methylation site is a feature used to characterize the first sample methylation site. The implementation principle of obtaining the characterization feature of the first sample methylation site based on the state information of the first sample methylation site is similar to Figure 2 The implementation principle of step 202 in the illustrated embodiment is the same and will not be described in detail here.

[0266] In step 703, a first feature extraction model is called to extract the feature characterization of the first sample biological sample based on the feature characterization of the methylation site of the first sample.

[0267] The characterization feature of the first sample biological sample is a feature used to characterize the first sample biological sample. The implementation principle of calling the first feature extraction model to extract the characterization feature of the first sample biological sample based on the characterization feature of the first sample methylation site is the same as Figure 2 The implementation principles of the relevant contents in step 203 in the illustrated embodiment are the same and will not be described in detail here.

[0268] In one possible implementation, see Figure 8 The process of obtaining the first feature extraction model includes the following steps 801 to 803.

[0269] Step 801: Acquire status information of methylation sites of a second sample, where the status information of methylation sites of the second sample is obtained by performing methylation sequencing on nucleic acid of a second biological sample.

[0270] Step 802: Based on the state information of the methylation site of the second sample, obtain the characterization features of the methylation site of the second sample.

[0271] Step 803: Based on the characterization features of the methylation sites of the second sample and the self-supervised learning task, an initial feature extraction model is trained to obtain a first feature extraction model.

[0272] Before using the first sample biological sample to train the first feature extraction model, the first feature extraction model is first trained through a self-supervised learning task based on the characterization features of the methylation sites of the second sample, so that the first feature extraction model has a certain ability to learn the characterization features of the methylation sites, thereby improving the convergence speed in the subsequent training of the first feature extraction model. In addition, the self-supervised learning task does not require labels, and can better utilize unlabeled data, thereby making full use of a large amount of unlabeled methylation data, and improving the representation learning ability and generalization ability of the model.

[0273] The realization principle and method of obtaining the characterization characteristics of the second sample methylation site Figure 2 The implementation principle of obtaining the characterization features of the methylation sites in the illustrated embodiments is the same and will not be described in detail here.

[0274] The initial feature extraction model is a feature extraction model to be trained. The process of training the initial feature extraction model based on the characterization features of the methylation sites of the second sample and the self-supervised learning task can be considered as a process of pre-training using unlabeled data.

[0275] The type of self-supervision task can be set based on experience or flexibly adjusted according to demand, which is not limited in the embodiments of the present application.

[0276] Exemplarily, the self-supervised task can be a task of predicting the methylation state of a masked site. In this case, based on the characterization features of the second sample methylation site and the self-supervised learning task, the process of training the initial feature extraction model includes: performing masking on the characterization features of the second sample methylation site to obtain the characterization features of the masked methylation site, and the masking is used to mask the characterization features of some methylation sites in the second sample methylation site; calling the initial feature extraction model to perform feature extraction based on the characterization features of the masked methylation site to obtain mask features; calling the prediction model to predict the characterization features of the masked methylation site based on the mask features; based on the difference between the real characterization features and the predicted characterization features of the masked methylation site, updating the parameters of the initial feature extraction model to obtain a first feature extraction model. In this case, the initial feature extraction model and the prediction model can constitute a masked autoencoder.

[0277] Exemplarily, the self-supervised task can also be a comparison learning task. In this case, based on the characterization features of the second sample methylation site and the self-supervised learning task, the process of training the initial feature extraction model includes: constructing a positive sample pair and a negative sample pair based on the characterization features of the second sample methylation site; calling the initial feature extraction model to extract the features of the two samples in the positive sample pair and the features of the two samples in the negative sample pair; with the training goal of making the features of the two samples in the positive sample pair as close as possible and the features of the two samples in the negative sample pair as far apart as possible, updating the parameters of the initial feature extraction model to obtain the first feature extraction model. Among them, the positive sample pair is composed of the characterization features of the second sample methylation site and the features that are relatively small in difference from the characterization features of the second sample methylation site; the negative sample pair is composed of the characterization features of the second sample methylation site and the features that are relatively large in difference from the characterization features of the second sample methylation site.

[0278] In step 704, the first recognition model is called to perform cancer recognition on the first sample biological sample based on the characterization features of the first sample biological sample to obtain a cancer recognition result of the first sample biological sample.

[0279] The implementation principle of calling the first feature extraction model to extract the characterization features of the first sample biological sample based on the characterization features of the first sample methylation site is as follows Figure 2 The implementation principles of the relevant contents in step 204 in the illustrated embodiment are the same and will not be described in detail here.

[0280] In step 705, based on the difference between the cancer identification result of the first sample biological sample and the cancer identification label of the first sample biological sample, the parameters of the first feature extraction model and the parameters of the first recognition model are updated to obtain a trained feature extraction model and a trained recognition model; wherein the trained feature extraction model and the trained recognition model are used to process the biological sample to obtain the cancer identification result of the biological sample.

[0281] The cancer identification result of the first sample biological sample is the result predicted after processing by the first feature extraction model and the first identification model. The cancer identification label of the first sample biological sample is a more accurate result. The difference between the cancer identification result of the first sample biological sample and the cancer identification label of the first sample biological sample can reflect the processing performance of the first feature extraction model and the identification model. For example, the greater the difference between the cancer identification result of the first sample biological sample and the cancer identification label of the first sample biological sample, the worse the processing performance of the first feature extraction model and the identification model. Therefore, based on the difference between the cancer identification result of the first sample biological sample and the cancer identification label of the first sample biological sample, the parameters of the first feature extraction model and the parameters of the first identification model are updated to improve the processing performance of the first feature extraction model and the first identification model.

[0282] In an exemplary embodiment, the process of updating the parameters of the first feature extraction model and the parameters of the first recognition model based on the difference between the cancer recognition result of the first sample biological sample and the cancer recognition label of the first sample biological sample to obtain the trained feature extraction model and the trained recognition model includes: obtaining the training loss based on the difference between the cancer recognition result of the first sample biological sample and the cancer recognition label of the first sample biological sample; updating the parameters of the first feature extraction model and the parameters of the first recognition model based on the training loss to obtain the second feature extraction model and the second recognition model; in response to the training process satisfying the training termination condition, using the second feature extraction model as the trained feature extraction model and the second recognition model as the trained recognition model.

[0283] Exemplarily, in response to the training process not satisfying the training termination condition, a new training loss is obtained based on the second feature extraction model and the second recognition model; the parameters of the second feature extraction model and the parameters of the second recognition model are updated using the new training loss to obtain a third feature extraction model and a third recognition model; in response to the training process satisfying the training termination condition, the third feature extraction model is used as the trained feature extraction model, and the third recognition model is used as the trained recognition model. In response to the training process not satisfying the training termination condition, the third feature extraction model and the third recognition model are continuously trained according to the same principle until the training process satisfies the training termination condition, and the feature extraction model and the recognition model obtained when the training termination condition is satisfied are used as the trained feature extraction model and the trained recognition model.

[0284] The training process satisfies the training termination condition, which can be set based on experience or flexibly adjusted according to the application scenario, and the embodiments of the present application do not limit this. Exemplarily, the training process satisfies the training termination condition, which can mean that the training loss is less than the loss threshold, or that the training loss converges, or that the number of training times reaches the number threshold, etc. The loss threshold and the number threshold can be set based on experience or flexibly adjusted according to the application scenario.

[0285] The training process of the feature extraction model and the recognition model provided in the embodiment of the present application optimizes the model parameters through the back propagation algorithm, so that the performance of the model on the training data is gradually improved.

[0286] After obtaining the trained feature extraction model and the trained recognition model, the biological sample is processed using the trained feature extraction model and the trained recognition model to obtain a cancer recognition result of the biological sample. The implementation process of processing the biological sample using the trained feature extraction model and the trained recognition model is shown in Figure 2 The relevant contents in the illustrated embodiment will not be described in detail here.

[0287] In an embodiment of the present application, the process of performing cancer identification on biological samples is based on a trained feature extraction model and a trained recognition model. The trained feature extraction model and the trained recognition model are obtained through supervised training, which is conducive to ensuring the reliability of cancer identification on biological samples and ensuring the accuracy of cancer identification results of biological samples.

[0288] For example, taking the biological sample as blood and the nucleic acid of the biological sample as cfDNA of the blood as an example, Fig. 9 As shown, the biological sample processing method may include the following steps 901 to 909.

[0289] Step 901: A computer device obtains a cancer identification tag of a first blood sample and status information of a methylation site of the first sample, wherein the status information of the methylation site of the first sample is obtained by performing methylation sequencing on cfDNA of the first blood sample.

[0290] Step 902: The computer device converts the state information of the methylation site of the first sample into a first sample feature; converts the auxiliary characterization information of the methylation site of the first sample into a second sample feature; and fuses the first sample feature and the second sample feature to obtain a characterization feature of the methylation site of the first sample.

[0291] The auxiliary characterization information is used to assist in characterizing the methylation site of the first sample based on the state information of the methylation site of the first sample.

[0292] Step 903: The computer device calls the first feature extraction model to extract the characterization features of the first sample blood based on the characterization features of the methylation sites of the first sample.

[0293] In some embodiments, the acquisition process of the first feature extraction model includes: acquiring state information of the methylation site of the second sample, the state information of the methylation site of the second sample is obtained by performing methylation sequencing on the cfDNA of the second sample blood; based on the state information of the methylation site of the second sample, acquiring characterization features of the methylation site of the second sample; based on the characterization features of the methylation site of the second sample and the self-supervised learning task, training the initial feature extraction model to obtain the first feature extraction model.

[0294] Step 904: The computer device calls the first recognition model to perform cancer recognition on the first blood sample based on the characterization features of the first blood sample, and obtains a cancer recognition result of the first blood sample.

[0295] Step 905: The computer device updates the parameters of the first feature extraction model and the parameters of the first recognition model based on the difference between the cancer recognition result of the first blood sample and the cancer recognition label of the first blood sample to obtain a trained feature extraction model and a trained recognition model.

[0296] Step 906: The computer device performs methylation sequencing on the cfDNA in the blood to obtain status information of the methylation sites in the genome of the cfDNA.

[0297] Step 907: The computer device converts the state information of the methylation site into a first feature; converts the auxiliary characterization information of the methylation site into a second feature; and fuses the first feature and the second feature to obtain a characterization feature of the methylation site.

[0298] The auxiliary characterization information is used to assist in characterizing the methylation site based on the state information of the methylation site.

[0299] In some embodiments, the auxiliary characterization information of the methylation site includes at least one of the first information, the second information, the third information, the fourth information or the fifth information; wherein the first information is used to indicate the position of the methylation site in the genome of cfDNA; the second information is used to indicate the chromosome to which the methylation site belongs; the third information is used to indicate the position of the methylation site in the chromosome to which it belongs; the fourth information is used to indicate the positional relationship between the methylation site and the functional region in the chromosome to which it belongs; and the fifth information is used to indicate the positional relationship between the methylation site and the gene region in the chromosome to which it belongs.

[0300] Step 908: The computer device calls the trained feature extraction model to extract the characteristic features of the blood based on the characteristic features of the methylation sites.

[0301] In some embodiments, the number of methylation sites is at least two, and the trained feature extraction model is used to perform feature extraction on the sequence data; calling the trained feature extraction model to extract the characterization features of the blood based on the characterization features of the methylation sites includes: inputting the characterization features of at least two methylation sites as sequence data into the trained feature extraction model to perform feature extraction, and using the features output by the trained feature extraction model as the characterization features of the blood.

[0302] In some embodiments, the number of methylation sites is at least two, and the trained feature extraction model is used to perform feature extraction on the graph structure; calling the trained feature extraction model to extract the characterization features of the blood based on the characterization features of the methylation sites includes: constructing a graph structure based on the characterization features of at least two methylation sites, inputting the graph structure into the trained feature extraction model to perform feature extraction, and using the features output by the trained feature extraction model as the characterization features of the blood.

[0303] In some embodiments, the process of calling a trained feature extraction model to extract characterization features of blood based on characterization features of methylation sites includes: calling a trained feature extraction model to perform N rounds of feature transformation based on the characterization features of methylation sites, and obtaining characterization features of blood based on the transformed features; wherein the i-th round of feature transformation in the N rounds of feature transformation is used to obtain output features corresponding to the i-th round of feature transformation based on input features corresponding to the i-th round of feature transformation; i is an integer less than or equal to N; N is an integer greater than or equal to 1; the input features corresponding to the first round of feature transformation in the N rounds of feature transformation are the characterization features of methylation sites; the output features corresponding to the N-th round of feature transformation in the N rounds of feature transformation are the transformed features.

[0304] In some embodiments, the i-th round of feature transformation includes: performing pre-processing operations on input features corresponding to the i-th round of feature transformation to obtain pre-processing features corresponding to the i-th round of feature transformation; transforming the pre-processing features corresponding to the i-th round of feature transformation based on the attention mechanism to obtain transformed features corresponding to the i-th round of feature transformation; performing post-processing operations on the transformed features corresponding to the i-th round of feature transformation to obtain output features corresponding to the i-th round of feature transformation.

[0305] In some embodiments, post-processing operations are performed on the transformed features corresponding to the i-th round of feature transformation to obtain output features corresponding to the i-th round of feature transformation, including: fusing the transformed features corresponding to the i-th round of feature transformation and the input features corresponding to the i-th round of feature transformation to obtain fused features corresponding to the i-th round of feature transformation, normalizing the fused features corresponding to the i-th round of feature transformation to obtain normalized features corresponding to the i-th round of feature transformation; activating the normalized features corresponding to the i-th round of feature transformation to obtain activated features corresponding to the i-th round of feature transformation; and pooling the activated features corresponding to the i-th round of feature transformation to obtain output features corresponding to the i-th round of feature transformation.

[0306] Step 909: The computer device calls the trained recognition model to perform cancer recognition on the blood based on the characterization features of the blood, and obtains a cancer recognition result of the blood.

[0307] Optionally, the cancer identification result of the blood includes a healthy category and a predicted probability of at least one cancer category; or, the cancer identification result of the blood includes a cancer risk value.

[0308] The above steps 901 to 905 are steps of the training phase of the biological sample processing model, and the biological sample processing model includes a first feature extraction model and a first recognition model. Based on the steps of the training phase, a trained biological sample processing model can be obtained.

[0309] The above steps 906 to 909 are steps in the use phase of the trained biological sample processing model. Based on the steps in the use phase, the trained biological sample processing model can be used to obtain cancer recognition results of biological samples (such as blood).

[0310] In some embodiments, when blood needs to be processed, cancer identification results of the blood are obtained by sequentially executing steps 901 to 909 .

[0311] In some embodiments, before the blood needs to be processed, steps 901 to 905 are executed to obtain a trained feature extraction model and a trained recognition model, and the trained feature extraction model and the trained recognition model are stored in a database. When the blood needs to be processed, the trained feature extraction model and the trained recognition model are extracted from the database, and then steps 906 to 909 are executed to obtain the cancer recognition results of the blood.

[0312] It should be noted that the above steps 901 to 909 are only described by taking the biological sample being blood as an example. The processing principles of other types of biological samples are the same and will not be repeated here.

[0313] In addition, the above steps 901 to 909 are only illustrated by directly extracting the characterization features of blood based on the characterization features of methylation sites. In some embodiments, the characterization features of methylation sites and the reference features of blood can also be comprehensively considered.

[0314] Exemplarily, the characterization features of the methylation sites and the reference features of the blood are fused to obtain a biological fusion feature, the reference feature is obtained based on reference biological information, and the reference biological information is obtained by processing the blood in a manner different from methylation sequencing; based on the biological fusion feature, the characterization features of the blood are extracted.

[0315] Exemplarily, based on the characterization features of methylation sites, a first characterization feature of the blood is extracted; based on the reference feature of the blood, a second characterization feature of the blood is extracted, the reference feature is obtained based on reference biological information, and the reference biological information is obtained by processing the blood in a manner different from methylation sequencing; the first characterization feature and the second characterization feature are fused to obtain the characterization feature of the blood.

[0316] Exemplarily, the biological sample processing process is implemented by calling a trained biological sample processing model, which is implemented as a deep learning large model. The biological sample processing model includes a trained feature extraction model and a trained recognition model.

[0317] In some embodiments, the biological sample processing model may include, in addition to the trained feature extraction model and the trained recognition model, a feature transformation layer disposed between the trained feature extraction model and the trained recognition model.

[0318] The trained feature extraction model is used to perform N rounds of feature transformation based on the representation features of the methylation sites to obtain output features corresponding to the Nth round of feature transformation; the feature transformation layer is used to transform the output features corresponding to the Nth round of feature transformation through global attention to obtain the representation features of the biological sample; the trained recognition model is used to perform cancer recognition based on the representation features of the biological sample to obtain the cancer recognition results of the biological sample.

[0319] The input of the trained biological sample processing model is the status information of methylation sites at the whole genome level measured based on cfDNA, and the output is the cancer identification result, such as the cancer risk value or the predicted probability of cancer category.

[0320] Exemplarily, the trained feature extraction model includes N transformation sub-models, each of which is used to perform a round of feature transformation. Each transformation sub-model includes a transformer and a pooling layer. A transformer includes a first normalization layer, an attention layer, a residual connection layer, a second normalization layer, and an activation layer.

[0321] The first normalization layer in the converter in the i-th (i is an integer less than or equal to N) conversion sub-model is used to normalize the input features corresponding to the i-th round of feature transformation to obtain the pre-processing features corresponding to the i-th round of feature transformation. The attention layer in the converter in the i-th conversion sub-model is used to transform the pre-processing features corresponding to the i-th round of feature transformation based on the attention mechanism (e.g., multi-head attention mechanism) to obtain the transformation features corresponding to the first round of feature transformation. The residual connection layer in the converter in the i-th conversion sub-model is used to fuse the transformation features corresponding to the i-th round of feature transformation and the input features corresponding to the i-th round of feature transformation to obtain the fusion features corresponding to the i-th round of feature transformation. The residual connection layer can prevent the gradient from disappearing or exploding. The second normalization layer in the converter in the i-th conversion sub-model is used to normalize the fusion features corresponding to the i-th round of feature transformation to obtain the normalized features corresponding to the i-th round of feature transformation. The activation layer in the converter in the i-th conversion sub-model is used to activate the normalized features corresponding to the i-th round of feature transformation to obtain the activation features corresponding to the i-th round of feature transformation. For example, the activation layer can use the SwiGLU activation function to replace the traditional ReLU activation function to increase the expression ability of the model. The pooling layer in the i-th conversion sub-model is used to pool the activation features corresponding to the i-th round of feature transformation to obtain the output features corresponding to the i-th round of feature transformation.

[0322] Based on this, the processing of biological samples can be as follows Fig.10shown.

[0323] Each methylation site is taken as a Token, and the state information of the methylation site is processed by TokenEmbedding. Each Token is mapped to a high-dimensional vector space to obtain the first feature (methylation state feature) used to characterize the state information of the methylation site. The first information is converted into a second feature (position feature) used to characterize the location of the methylation site in the genome; the second information is converted into a second feature (chromosome feature) used to characterize the chromosome to which the methylation site belongs. The methylation state feature, position feature, and chromosome feature are integrated to obtain the characterization feature of the biological sample.

[0324] The characterization features of the biological sample are input into the first conversion sub-model, and the characterization features of the biological sample are processed by the converter and pooling layer in the first conversion sub-model to obtain the output features corresponding to the first round of feature transformation. The output features corresponding to the first round of feature transformation are used as the input features corresponding to the second round of feature transformation, and the input features corresponding to the second round of feature transformation are processed by the converter and pooling layer in the second conversion sub-model to obtain the output features corresponding to the second round of feature transformation. This process is repeated until the output features corresponding to the Nth round of feature transformation obtained after processing by the converter and pooling layer in the Nth conversion sub-model are obtained.

[0325] After obtaining the output features corresponding to the Nth round of feature transformation, the output features corresponding to the Nth round of feature transformation are transformed based on global attention by a feature transformation layer set between the trained feature extraction model and the trained recognition model to obtain the characterization features of the biological sample. After obtaining the characterization features of the biological sample, the trained recognition model performs the final classification or regression task based on the characterization features of the biological sample to obtain the cancer recognition result of the biological sample.

[0326] For example, the feature extraction model is a long sequence language model for processing sequence data, such as Fig.11 As shown, the biological sample processing method provided in the embodiment of the present application may include the following three stages.

[0327] Stage 1: Acquisition of status information of methylation sites.

[0328] In this stage, methylation sequencing is performed on the nucleic acid of the biological sample to obtain the status information of the methylation site; then, based on the status information of the methylation site, the characterization features of the methylation site are obtained.

[0329] In the stage of obtaining the status information of methylation sites, the system uses the methylation sequencing results of nucleic acids (such as cfDNA) of biological samples, including the status information of about one million methylation sites (such as methylation / non-methylation / no detection results) as input. This stage realizes efficient data preprocessing, including data cleaning, standardization and quality control, laying the foundation for subsequent analysis.

[0330] Peripheral blood samples were collected from various cancer patients and healthy controls to ensure sample diversity and representativeness. cfDNA sulfite sequencing (WGBS) technology was used to obtain high-resolution cfDNA methylation maps. Automated sequencing quality control processes were implemented, including sequencing depth assessment, coverage analysis, and conversion efficiency calculation to ensure data reliability.

[0331] Develop efficient data cleaning algorithms to remove low-quality reads and potential sequencing errors. Standardize methylation levels to eliminate batch effects and technical biases. Design adaptive quality control thresholds to dynamically adjust filtering criteria based on sample characteristics.

[0332] Fig.11 The status information of the methylation site obtained in is only an exemplary description. Each small grid represents a site in the genome, the filled small grid represents the methylation site, and different filling methods of the filled small grids represent different status information of the methylation site.

[0333] Stage 2: Long sequence language model processing stage.

[0334] In the long sequence language model stage, the system uses advanced technologies in the field of natural language processing, innovatively treating the sequence of state information of methylation sites as a "language sequence". By designing and training a specialized Transformer architecture, the model is able to capture long-range dependencies in methylation patterns. The implementation of the self-attention mechanism in the model enables the model to highlight important methylation sites and regions, while the pre-training and fine-tuning strategies further improve the model's ability to understand methylation data.

[0335] Develop an improved model based on the long sequence processing model to adapt to long sequence methylation data input. Implement sparse attention mechanisms such as block attention and sliding window attention to reduce computational complexity and enable the model to process longer sequence inputs. Design a hierarchical model structure, including local attention layers and global attention layers, to capture methylation patterns at different scales.

[0336] An adaptive position encoding method that considers the genome structure integrates chromosome position, gene structure, and functional region information. A hybrid strategy of relative position encoding and absolute position encoding is implemented to improve the model's ability to perceive local and global methylation patterns.

[0337] Implement a dynamic multi-head attention mechanism to adaptively adjust the number and weights of attention heads according to the characteristics of the input sequence. Develop a multi-scale attention fusion module to capture both local methylation patterns and long-range dependencies. Design an attention guidance mechanism based on prior knowledge to integrate known cancer-related genomic region information.

[0338] Phase 3: Cancer identification phase.

[0339] In the cancer identification stage, the system builds a multi-label classification model based on the output of the long sequence language model to achieve early diagnosis of various cancers. This design not only reduces experimental complexity and cost, but also provides an efficient and accurate cancer identification solution.

[0340] Build a multi-label classifier based on the output of a long sequence language model to support simultaneous prediction of multiple cancer types. Implement a hierarchical classification strategy to first distinguish between cancer and non-cancer states and then subdivide specific cancer types. Develop class imbalance processing methods, such as focal loss and sample weight adjustment, to address the detection challenges of rare cancers.

[0341] Implement a dynamic threshold adjustment algorithm to adaptively set the decision threshold according to the characteristics of different cancer types. Develop a multi-stage prediction strategy that combines clinical information and other biomarkers to improve prediction accuracy.

[0342] Develop an attention-weighted visualization tool to intuitively display key methylation regions and their importance in different cancers. Implement a rule-based interpretation method to associate model predictions with known biological knowledge.

[0343] Through this innovative algorithmic framework, this application achieves high-sensitivity and high-specificity early screening of multiple cancer types. This method not only reduces experimental complexity and cost, but also provides interpretable results, providing strong support for personalized cancer risk assessment and the formulation of early intervention strategies. In addition, the modular design and flexibility of the framework make it easy to expand and adapt to new cancer types and data types, laying a solid technical foundation for future early cancer diagnosis research.

[0344] The feature extraction model and the recognition model constitute a deep learning big model. The nucleic acid of the biological sample refers to the cfDNA of the blood. The biological sample processing method proposed in this application can be regarded as a cancer early diagnosis method based on the deep learning big model technology to analyze the cfDNA methylation spectrum. By integrating high-throughput sequencing technology, innovative long sequence language models and multi-label classification algorithms, high-sensitivity and high-specificity early detection of multiple cancers is achieved. Among them, the cfDNA methylation spectrum refers to the abbreviation of the methylation state of the methylation site obtained after performing methylation sequencing on the cfDNA of the blood. Exemplarily, the methylation site refers to the methylation-related site in the whole genome obtained by sequencing.

[0345] This method constructs and trains a deep learning large model, which provides efficient and accurate cancer risk assessment and cancer category diagnosis by analyzing and utilizing the methylation status of genomic methylation sites based on cfDNA determination. This application can achieve high sensitivity and high specificity cancer recognition by utilizing the generalization ability of the large model, large-scale data pre-training / training, and the extensive life status information provided by the cfDNA methylation spectrum, thereby achieving the following effects.

[0346] 1. Improve the early diagnosis rate: The method provided in this application can provide timely diagnostic evidence in the early stages of cancer in biological samples, significantly improving the survival rate and quality of life of patients to whom the biological samples belong. By detecting the status information of methylation sites, it is possible to detect slight changes in methylation patterns, thereby identifying cancer at early stages (such as stage 0 and stage I) that cannot be detected by traditional imaging and biomarker detection, providing a critical time window for early intervention, which is expected to significantly improve patient prognosis and significantly reduce medical expenses for later treatment.

[0347] 2. Reduce misdiagnosis rate: By utilizing large-scale methylation data and advanced deep learning algorithms (such as Transformer-based long sequence learning), this application can significantly improve the accuracy of screening and reduce false positive and false negative rates. This will reduce the risk of unnecessary treatment for patients to whom the biological samples belong, while ensuring that patients who really need treatment receive timely intervention.

[0348] 3. Non-invasive testing: This method only requires testing biological samples such as blood, avoiding the invasiveness and discomfort of traditional methods. This non-invasive feature makes the screening process more convenient, and the patients to whom the biological samples belong are more accepting, which is conducive to improving the screening coverage and frequency.

[0349] 4. Rapid results: Through cfDNA sequencing and methylation status analysis, combined with a high-performance computing platform, the entire testing process is highly automated, greatly shortening the time from sampling to results, providing test results in a shorter time, significantly improving testing efficiency, and facilitating timely clinical decision-making. This rapid feedback mechanism helps reduce patients' psychological pressure and increase the rate of early intervention.

[0350] 5. Simultaneous screening of multiple cancer types: This application can screen multiple cancer types at the same time, improve screening efficiency and reduce overall costs. This "one test, multiple cancer screening" model provides the possibility for large-scale population screening.

[0351] 6. Personalized risk assessment: By analyzing individual-specific methylation patterns, this application can provide more accurate personalized cancer risk assessment. This helps to develop personalized screening and prevention strategies and achieve precision medicine. In addition, advanced model interpretation technology can be used to provide intuitive result interpretation, enhance doctors and patients' understanding and trust in the predicted results, and improve the transparency of clinical decision-making.

[0352] 7. Cost-effectiveness: Although the initial investment may be high, considering the ability to screen multiple cancers and the treatment cost savings brought about by early diagnosis, this application has significant cost-effectiveness advantages in the long run.

[0353] 8. Excellent performance in technical adaptability: The model architecture has good adaptability and scalability, and can quickly adapt to newly discovered cancer-related methylation markers. When new cancer types are added, through transfer learning, the model can quickly reach a performance level comparable to the original cancer types.

[0354] Exemplarily, the computer device that executes the biological sample processing method provided in the embodiment of the present application can be a server. The server is used to provide a background server for the cancer recognition client installed in the terminal of the medical staff. The cancer recognition client is a client that can provide the medical staff with the cancer recognition results of the biological sample.

[0355] like Fig.12 As shown, the interaction process between the server and the cancer identification client may include the following steps 1201 to 1207 .

[0356] Step 1201: In response to the input operation of the identification information of the biological sample, the cancer identification client sends a biological sample processing request to the server, where the biological sample processing request carries the identification information of the biological sample.

[0357] The identification information of the biological sample is used to identify the biological sample. Exemplarily, the identification information of the biological sample is the number of the biological sample.

[0358] Step 1202: The server receives a request for processing a biological sample.

[0359] Step 1203: In response to the biological sample processing request, the server performs methylation sequencing on the nucleic acid of the biological sample through a methylation sequencing platform to obtain status information of the methylation site; based on the status information of the methylation site, obtains the characterization features of the methylation site.

[0360] Step 1204: The server calls the trained feature extraction model to extract the characterization features of the biological sample based on the characterization features of the methylation sites; calls the trained recognition model to perform cancer recognition on the biological sample based on the characterization features of the biological sample to obtain a cancer recognition result of the biological sample.

[0361] Step 1205: The server sends the cancer identification result of the biological sample to the cancer identification client.

[0362] Step 1206: The cancer identification client receives the cancer identification result of the biological sample.

[0363] Step 1207: The cancer identification client displays the cancer identification result of the biological sample.

[0364] See also Fig.13 , an embodiment of the present application provides a biological sample processing device, the device comprising:

[0365] An acquisition module 1301 is used to perform methylation sequencing on the nucleic acid of the biological sample to obtain status information of methylation sites in the genome of the nucleic acid;

[0366] The acquisition module 1301 is further used to acquire the characterization features of the methylation site based on the state information of the methylation site;

[0367] An extraction module 1302 is used to extract the characteristic features of the biological sample based on the characteristic features of the methylation sites;

[0368] The identification module 1303 is used to perform cancer identification on the biological sample based on the characteristic features of the biological sample to obtain a cancer identification result of the biological sample.

[0369] In one possible implementation, the acquisition module 1301 is used to convert the state information of the methylation site into a first feature; convert the auxiliary characterization information of the methylation site into a second feature, the auxiliary characterization information is used to assist in characterizing the methylation site based on the state information of the methylation site; and fuse the first feature and the second feature to obtain a characterization feature of the methylation site.

[0370] In a possible implementation, the auxiliary characterization information includes at least one of the first information, the second information, the third information, the fourth information or the fifth information; wherein the first information is used to indicate the position of the methylation site in the genome of the nucleic acid; the second information is used to indicate the chromosome to which the methylation site belongs; the third information is used to indicate the position of the methylation site in the chromosome to which it belongs; the fourth information is used to indicate the positional relationship between the methylation site and the functional region in the chromosome to which it belongs; and the fifth information is used to indicate the positional relationship between the methylation site and the gene region in the chromosome to which it belongs.

[0371] In one possible implementation, the extraction module 1302 is used to perform N rounds of feature transformation based on the characterization features of the methylation sites, and obtain the characterization features of the biological sample based on the transformed features; wherein the i-th round of feature transformation in the N rounds of feature transformation is used to obtain the output features corresponding to the i-th round of feature transformation based on the input features corresponding to the i-th round of feature transformation; i is an integer less than or equal to N; N is an integer greater than or equal to 1; the input features corresponding to the first round of feature transformation in the N rounds of feature transformation are the characterization features of the methylation sites; and the output features corresponding to the N-th round of feature transformation in the N rounds of feature transformation are the transformed features.

[0372] In one possible implementation, the extraction module 1302 is used to perform pre-processing operations on the input features corresponding to the i-th round of feature transformation to obtain pre-processing features corresponding to the i-th round of feature transformation; transform the pre-processing features corresponding to the i-th round of feature transformation based on the attention mechanism to obtain transformed features corresponding to the i-th round of feature transformation; perform post-processing operations on the transformed features corresponding to the i-th round of feature transformation to obtain output features corresponding to the i-th round of feature transformation.

[0373] In one possible implementation, the extraction module 1302 is used to fuse the transformation features corresponding to the i-th round of feature transformation and the input features corresponding to the i-th round of feature transformation to obtain the fused features corresponding to the i-th round of feature transformation; normalize the fused features corresponding to the i-th round of feature transformation to obtain the normalized features corresponding to the i-th round of feature transformation; activate the normalized features corresponding to the i-th round of feature transformation to obtain the activated features corresponding to the i-th round of feature transformation; pool the activated features corresponding to the i-th round of feature transformation to obtain the output features corresponding to the i-th round of feature transformation.

[0374] In a possible implementation, the extraction module 1302 is used to call the trained feature extraction model to extract the feature characteristics of the biological sample based on the feature characteristics of the methylation site;

[0375] The identification module 1303 is used to call the trained identification model to perform cancer identification on the biological sample based on the characterization features of the biological sample to obtain a cancer identification result of the biological sample.

[0376] In a possible implementation, the number of methylation sites is at least two, and the trained feature extraction model is used to perform feature extraction on the sequence data;

[0377] The extraction module 1302 is used to input the characteristic features of at least two methylation sites as sequence data into the trained feature extraction model to perform feature extraction, and use the features output by the trained feature extraction model as the characteristic features of the biological sample.

[0378] In a possible implementation, the number of methylation sites is at least two, and the trained feature extraction model is used to perform feature extraction on the graph structure;

[0379] The extraction module 1302 is used to construct a graph structure based on the characterization features of at least two methylation sites, input the graph structure into the trained feature extraction model to perform feature extraction, and use the features output by the trained feature extraction model as the characterization features of the biological sample.

[0380] In one possible implementation, the extraction module 1302 is used to fuse the characterization features of the methylation sites and the reference features of the biological samples to obtain biological fusion features; the reference features are obtained based on reference biological information, and the reference biological information is obtained by processing the biological samples in a manner different from methylation sequencing; based on the biological fusion features, the characterization features of the biological samples are extracted.

[0381] In one possible implementation, the extraction module 1302 is used to extract a first characterization feature of the biological sample based on the characterization feature of the methylation site; extract a second characterization feature of the biological sample based on the reference feature of the biological sample; the reference feature is obtained based on reference biological information, and the reference biological information is obtained by processing the biological sample in a manner different from methylation sequencing; and the first characterization feature and the second characterization feature are fused to obtain the characterization feature of the biological sample.

[0382] In a possible implementation, the cancer identification result of the biological sample includes a healthy category and a predicted probability of at least one cancer category; or, the cancer identification result of the biological sample includes a cancer risk value.

[0383] See also Fig.14 The present application also provides a biological sample processing device, which includes:

[0384] An acquisition module 1401 is used to acquire a cancer identification tag of a first biological sample and state information of a methylation site of the first sample, where the state information of the methylation site of the first sample is obtained by performing methylation sequencing on the nucleic acid of the first biological sample;

[0385] The acquisition module 1401 is further used to acquire the characterization features of the methylation sites of the first sample based on the state information of the methylation sites of the first sample;

[0386] An extraction module 1402 is used to call a first feature extraction model to extract a feature of the first biological sample based on the feature of the methylation site of the first sample;

[0387] The identification module 1403 is used to call the first identification model to perform cancer identification on the first sample biological sample based on the characterization feature of the first sample biological sample, and obtain a cancer identification result of the first sample biological sample;

[0388] An updating module 1404 is used to update the parameters of the first feature extraction model and the parameters of the first recognition model based on the difference between the cancer recognition result of the first sample biological sample and the cancer recognition label of the first sample biological sample to obtain a trained feature extraction model and a trained recognition model;

[0389] The trained feature extraction model and the trained recognition model are used to process biological samples to obtain cancer recognition results of the biological samples.

[0390] In a possible implementation, the device further includes:

[0391] A training module is used to obtain the state information of the methylation site of the second sample, where the state information of the methylation site of the second sample is obtained by performing methylation sequencing on the nucleic acid of the second sample biological sample; based on the state information of the methylation site of the second sample, the characterization features of the methylation site of the second sample are obtained; based on the characterization features of the methylation site of the second sample and the self-supervised learning task, an initial feature extraction model is trained to obtain a first feature extraction model.

[0392] It should be noted that the device provided in the above embodiment only uses the division of the above functional modules as an example to implement its functions. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiment belong to the same concept, and their specific implementation process and beneficial effects are detailed in the method embodiment, which will not be repeated here.

[0393] In an exemplary embodiment, a computer device is also provided, the computer device includes a processor and a memory, and at least one computer program is stored in the memory. The at least one computer program is loaded and executed by one or more processors, so that the computer device implements any of the above-mentioned biological sample processing methods. The computer device can be a server or a terminal. Next, the structures of the terminal and the server are introduced respectively.

[0394] Fig.15 1 is a schematic diagram of a terminal structure provided by an embodiment of the present application. The terminal may be: a PC, a mobile phone, a smart phone, a PDA, a wearable device, a PPC, a tablet computer, a smart car machine, a smart TV, a smart speaker, or a vehicle-mounted terminal. The terminal may also be referred to as a user device, a portable terminal, a laptop terminal, a desktop terminal, or other names.

[0395] Typically, the terminal includes: a processor 1501 and a memory 1502 .

[0396] The processor 1501 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 1501 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 1501 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 1501 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 1501 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0397] The memory 1502 may include one or more computer-readable storage media, which may be non-transitory. The memory 1502 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 1502 is used to store at least one instruction, which is used to be executed by the processor 1501 so that the terminal implements the biological sample processing method provided in the method embodiment of the present application.

[0398] In some embodiments, the terminal may further optionally include: a peripheral device interface 1503 and at least one peripheral device. The processor 1501, the memory 1502 and the peripheral device interface 1503 may be connected via a bus or a signal line. Each peripheral device may be connected to the peripheral device interface 1503 via a bus, a signal line or a circuit board. Specifically, the peripheral device includes: at least one of a radio frequency circuit 1504, a display screen 1505, a camera assembly 1506, an audio circuit 1507 and a power supply 1508.

[0399] The peripheral device interface 1503 may be used to connect at least one peripheral device related to I / O (Input / Output) to the processor 1501 and the memory 1502. In some embodiments, the processor 1501, the memory 1502, and the peripheral device interface 1503 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 1501, the memory 1502, and the peripheral device interface 1503 may be implemented on a separate chip or circuit board, which is not limited in this embodiment.

[0400] The radio frequency circuit 1504 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The radio frequency circuit 1504 communicates with the communication network and other communication devices through electromagnetic signals. The radio frequency circuit 1504 converts the electrical signal into an electromagnetic signal for transmission, or converts the received electromagnetic signal into an electrical signal. Optionally, the radio frequency circuit 1504 includes: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, and the like. The radio frequency circuit 1504 can communicate with other terminals through at least one wireless communication protocol. The wireless communication protocol includes, but is not limited to: a metropolitan area network, various generations of mobile communication networks (2G, 3G, 4G and 5G), a wireless local area network and / or a WiFi (Wireless Fidelity) network. In some embodiments, the radio frequency circuit 1504 may also include circuits related to NFC (Near Field Communication), which is not limited in this application.

[0401] The display screen 1505 is used to display a UI (User Interface). The UI may include graphics, text, icons, videos, and any combination thereof. When the display screen 1505 is a touch display screen, the display screen 1505 also has the ability to collect touch signals on the surface or above the surface of the display screen 1505. The touch signal can be input to the processor 1501 as a control signal for processing. At this time, the display screen 1505 can also be used to provide virtual buttons and / or virtual keyboards, also known as soft buttons and / or soft keyboards. In some embodiments, the display screen 1505 can be one, which is set on the front panel of the terminal; in other embodiments, the display screen 1505 can be at least two, which are respectively set on different surfaces of the terminal or are folded; in other embodiments, the display screen 1505 can be a flexible display screen, which is set on the curved surface or folded surface of the terminal. Even, the display screen 1505 can also be set as a non-rectangular irregular shape, that is, a special-shaped screen. The display screen 1505 can be made of materials such as LCD (Liquid Crystal Display), OLED (Organic Light-Emitting Diode), etc.

[0402] The camera assembly 1506 is used to capture images or videos. Optionally, the camera assembly 1506 includes a front camera and a rear camera. Typically, the front camera is disposed on the front panel of the terminal, and the rear camera is disposed on the back of the terminal. In some embodiments, there are at least two rear cameras, which are any one of a main camera, a depth of field camera, a wide-angle camera, and a telephoto camera, so as to realize the fusion of the main camera and the depth of field camera to realize the background blur function, the fusion of the main camera and the wide-angle camera to realize the panoramic shooting and VR (Virtual Reality) shooting function or other fusion shooting functions. In some embodiments, the camera assembly 1506 may also include a flash. The flash may be a monochrome temperature flash or a dual-color temperature flash. A dual-color temperature flash refers to a combination of a warm light flash and a cold light flash, which can be used for light compensation at different color temperatures.

[0403] The audio circuit 1507 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, and convert the sound waves into electrical signals and input them into the processor 1501 for processing, or input them into the RF circuit 1504 to achieve voice communication. For the purpose of stereo acquisition or noise reduction, there may be multiple microphones, which are respectively arranged at different parts of the terminal. The microphone may also be an array microphone or an omnidirectional acquisition microphone. The speaker is used to convert the electrical signal from the processor 1501 or the RF circuit 1504 into sound waves. The speaker may be a traditional film speaker or a piezoelectric ceramic speaker. When the speaker is a piezoelectric ceramic speaker, it can not only convert the electrical signal into sound waves audible to humans, but also convert the electrical signal into sound waves inaudible to humans for purposes such as distance measurement. In some embodiments, the audio circuit 1507 may also include a headphone jack.

[0404] The power supply 1508 is used to supply power to various components in the terminal. The power supply 1508 may be an alternating current, a direct current, a disposable battery, or a rechargeable battery. When the power supply 1508 includes a rechargeable battery, the rechargeable battery may support wired charging or wireless charging. The rechargeable battery may also be used to support fast charging technology.

[0405] In some embodiments, the terminal further includes one or more sensors 1509 , including but not limited to: an acceleration sensor 1510 , a gyroscope sensor 1511 , a pressure sensor 1512 , an optical sensor 1513 , and a proximity sensor 1514 .

[0406] The acceleration sensor 1510 can detect the magnitude of acceleration on the three coordinate axes of the coordinate system established by the terminal. For example, the acceleration sensor 1510 can be used to detect the components of gravity acceleration on the three coordinate axes. The processor 1501 can control the display screen 1505 to display the user interface in a horizontal view or a vertical view according to the gravity acceleration signal collected by the acceleration sensor 1510. The acceleration sensor 1510 can also be used for collecting game or user motion data.

[0407] The gyro sensor 1511 can detect the body direction and rotation angle of the terminal, and the gyro sensor 1511 can cooperate with the acceleration sensor 1510 to collect the user's 3D actions on the terminal. The processor 1501 can implement the following functions based on the data collected by the gyro sensor 1511: motion sensing (such as changing the UI according to the user's tilt operation), image stabilization during shooting, game control, and inertial navigation.

[0408] The pressure sensor 1512 may be provided at the side frame of the terminal and / or the lower layer of the display screen 1505. When the pressure sensor 1512 is provided at the side frame of the terminal, it may detect the user's holding signal of the terminal, and the processor 1501 may perform left and right hand recognition or shortcut operation according to the holding signal collected by the pressure sensor 1512. When the pressure sensor 1512 is provided at the lower layer of the display screen 1505, the processor 1501 may control the operability control on the UI interface according to the user's pressure operation on the display screen 1505. The operability control includes at least one of a button control, a scroll bar control, an icon control, and a menu control.

[0409] The optical sensor 1513 is used to collect the ambient light intensity. In one embodiment, the processor 1501 can control the display brightness of the display screen 1505 according to the ambient light intensity collected by the optical sensor 1513. Specifically, when the ambient light intensity is high, the display brightness of the display screen 1505 is increased; when the ambient light intensity is low, the display brightness of the display screen 1505 is reduced. In another embodiment, the processor 1501 can also dynamically adjust the shooting parameters of the camera component 1506 according to the ambient light intensity collected by the optical sensor 1513.

[0410] The proximity sensor 1514, also known as a distance sensor, is usually arranged on the front panel of the terminal. The proximity sensor 1514 is used to collect the distance between the user and the front of the terminal. In one embodiment, when the proximity sensor 1514 detects that the distance between the user and the front of the terminal is gradually decreasing, the processor 1501 controls the display screen 1505 to switch from the screen-on state to the screen-off state; when the proximity sensor 1514 detects that the distance between the user and the front of the terminal is gradually increasing, the processor 1501 controls the display screen 1505 to switch from the screen-off state to the screen-on state.

[0411] Those skilled in the art will understand that Fig.15 The structure shown in the figure does not constitute a limitation on the terminal, and may include more or less components than those shown in the figure, or combine certain components, or adopt a different component arrangement.

[0412] Fig.16This is a schematic diagram of the structure of a server provided in an embodiment of the present application. The server may have relatively large differences due to different configurations or performances, and may include one or more processors (Central Processing Units, CPU) 1601 and one or more memories 1602, wherein the one or more memories 1602 store at least one computer program, and the at least one computer program is loaded and executed by the one or more processors 1601, so that the server implements the biological sample processing method provided in the above-mentioned various method embodiments. Of course, the server may also have components such as a wired or wireless network interface, a keyboard, and an input and output interface for input and output, and the server may also include other components for implementing device functions, which will not be repeated here.

[0413] In an exemplary embodiment, a computer-readable storage medium is also provided, in which at least one computer program is stored. The at least one computer program is loaded and executed by a processor of a computer device to enable the computer to implement any of the above-mentioned biological sample processing methods.

[0414] In one possible implementation, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc (CD-ROM), a magnetic tape, a floppy disk, an optical data storage device, etc.

[0415] In an exemplary embodiment, a computer program or a computer program product is also provided. The computer program or the computer program product includes computer instructions. The computer instructions are loaded and executed by a processor to enable a computer to implement any of the above-mentioned biological sample processing methods.

[0416] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.) and signals involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions. For example, the biological samples involved in this application were obtained with full authorization.

[0417] The terms "first", "second", etc. in the application are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described here can be implemented in an order other than those illustrated or described here. The embodiments described in the above exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are only examples of devices and methods consistent with some aspects of the present application.

[0418] It should be understood that the "plurality" mentioned in this article refers to two or more. "And / or" describes the association relationship of the associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the associated objects are in an "or" relationship.

[0419] A person of ordinary skill in the art will understand that all or part of the steps to implement the above embodiments can be completed by hardware, or can be completed by instructing related hardware through a program. Optionally, the program is stored in a computer-readable storage medium. Optionally, the above-mentioned storage medium is a read-only memory, a disk or an optical disk, etc.

[0420] The above description is only an exemplary embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for processing a biological sample, characterized in that: The method comprises: Performing methylation sequencing on the nucleic acid of the biological sample to obtain status information of the methylation site in the genome of the nucleic acid, wherein the status information of the methylation site is used to indicate the specific situation of methylation of the methylation site; Converting the state information of the methylation site into a first feature; The auxiliary characterization information of the methylation site is converted into a second feature, wherein the auxiliary characterization information is used to assist in characterizing the methylation site based on the state information of the methylation site; the auxiliary characterization information includes at least one of the first information, the second information, the third information, the fourth information or the fifth information; the first information is used to indicate the position of the methylation site in the genome of the nucleic acid; the second information is used to indicate the chromosome to which the methylation site belongs; the third information is used to indicate the position of the methylation site in the chromosome to which it belongs; the fourth information is used to indicate the positional relationship between the methylation site and the functional region in the chromosome to which it belongs; and the fifth information is used to indicate the positional relationship between the methylation site and the gene region in the chromosome to which it belongs; The first feature and the second feature are combined to obtain a characteristic feature of the methylation site; Calling the trained feature extraction model to extract the characteristic features of the biological sample based on the characteristic features of the methylation site; The trained recognition model is called to perform cancer recognition on the biological sample based on the characterizing features of the biological sample to obtain a cancer recognition result of the biological sample.

2. The method according to claim 1, characterized in that: The calling of the trained feature extraction model to extract the feature characteristics of the biological sample based on the feature characteristics of the methylation site includes: Calling the trained feature extraction model to perform N rounds of feature transformation based on the characterization features of the methylation sites, and obtaining the characterization features of the biological sample based on the transformed features; Among them, the i-th round of feature transformation in the N rounds of feature transformation is used to obtain the output features corresponding to the i-th round of feature transformation based on the input features corresponding to the i-th round of feature transformation; i is an integer less than or equal to N; N is an integer greater than or equal to 1; the input features corresponding to the first round of feature transformation in the N rounds of feature transformation are the characterization features of the methylation sites; the output features corresponding to the N-th round of feature transformation in the N rounds of feature transformation are the transformed features.

3. The method according to claim 2, characterized in that The i-th round of feature transformation includes: Performing a pre-processing operation on the input features corresponding to the i-th round of feature transformation to obtain pre-processing features corresponding to the i-th round of feature transformation; Transforming the pre-processing features corresponding to the i-th round of feature transformation based on the attention mechanism to obtain the transformation features corresponding to the i-th round of feature transformation; A post-processing operation is performed on the transformed features corresponding to the i-th round of feature transformation to obtain output features corresponding to the i-th round of feature transformation.

4. The method according to claim 3, characterized in that: The performing a post-processing operation on the transformed features corresponding to the i-th round of feature transformation to obtain output features corresponding to the i-th round of feature transformation includes: Fusing the transformation features corresponding to the i-th round of feature transformation and the input features corresponding to the i-th round of feature transformation to obtain fused features corresponding to the i-th round of feature transformation; Normalizing the fused features corresponding to the i-th round of feature transformation to obtain normalized features corresponding to the i-th round of feature transformation; Activate the normalized features corresponding to the i-th round of feature transformation to obtain the activated features corresponding to the i-th round of feature transformation; The activated features corresponding to the i-th round of feature transformation are pooled to obtain the output features corresponding to the i-th round of feature transformation.

5. The method according to claim 1, characterized in that The number of the methylation sites is at least two, and the trained feature extraction model is used to perform feature extraction on sequence data; The calling of the trained feature extraction model to extract the feature characteristics of the biological sample based on the feature characteristics of the methylation site includes: The characterizing features of at least two methylation sites are input as the sequence data into the trained feature extraction model to perform feature extraction, and the features output by the trained feature extraction model are used as the characterizing features of the biological sample.

6. The method according to claim 1, characterized in that The number of the methylation sites is at least two, and the trained feature extraction model is used to perform feature extraction on the graph structure; The calling of the trained feature extraction model to extract the feature characteristics of the biological sample based on the feature characteristics of the methylation site includes: The graph structure is constructed based on the characterization features of at least two methylation sites, the graph structure is input into the trained feature extraction model to perform feature extraction, and the features output by the trained feature extraction model are used as the characterization features of the biological sample.

7. The method according to claim 1, characterized in that The calling of the trained feature extraction model to extract the feature characteristics of the biological sample based on the feature characteristics of the methylation site includes: Fusion of the characteristic feature of the methylation site and the reference feature of the biological sample to obtain a biological fusion feature; the reference feature is obtained based on reference biological information, and the reference biological information is obtained by processing the biological sample in a manner different from the methylation sequencing; The trained feature extraction model is called to extract the characterization features of the biological sample based on the biological fusion features.

8. The method according to claim 1, characterized in that: The calling of the trained feature extraction model to extract the feature characteristics of the biological sample based on the feature characteristics of the methylation site includes: Calling the trained feature extraction model to extract a first characterization feature of the biological sample based on the characterization feature of the methylation site; Extracting a second characterizing feature of the biological sample based on a reference feature of the biological sample; the reference feature is obtained based on reference biological information obtained by processing the biological sample in a manner different from the methylation sequencing; The first characterizing feature and the second characterizing feature are fused to obtain the characterizing feature of the biological sample.

9. The method according to any one of claims 1 to 8, characterized in that: The cancer identification result of the biological sample includes a healthy category and a predicted probability of at least one cancer category; or, The cancer identification result of the biological sample includes a cancer risk value.

10. A method for processing a biological sample, characterized in that: The method comprises: Acquire a cancer identification tag of a first biological sample and status information of a methylation site of the first sample, wherein the status information of the methylation site of the first sample is obtained by performing methylation sequencing on the nucleic acid of the first biological sample, and the status information of the methylation site of the first sample is used to indicate a specific situation in which methylation occurs at the methylation site of the first sample; Converting the state information of the methylation site of the first sample into a first sample feature; Converting auxiliary characterization information of the first sample methylation site into a second sample feature, the auxiliary characterization information being used to assist in characterizing the first sample methylation site based on the state information of the first sample methylation site; the auxiliary characterization information comprising at least one of the first information, the second information, the third information, the fourth information or the fifth information; the first information being used to indicate the position of the first sample methylation site in the genome of the nucleic acid of the first sample biological sample; the second information being used to indicate the chromosome to which the first sample methylation site belongs; the third information being used to indicate the position of the first sample methylation site in the chromosome to which it belongs; the fourth information being used to indicate the positional relationship between the first sample methylation site and the functional region in the chromosome to which it belongs; and the fifth information being used to indicate the positional relationship between the first sample methylation site and the gene region in the chromosome to which it belongs; The first sample feature and the second sample feature are combined to obtain a characteristic feature of the methylation site of the first sample; Invoking a first feature extraction model to extract a feature of the first biological sample based on the feature of the methylation site of the first sample; Calling a first recognition model to perform cancer recognition on the first sample biological sample based on the characterization feature of the first sample biological sample to obtain a cancer recognition result of the first sample biological sample; updating the parameters of the first feature extraction model and the parameters of the first recognition model based on the difference between the cancer recognition result of the first sample biological sample and the cancer recognition label of the first sample biological sample to obtain a trained feature extraction model and a trained recognition model; The trained feature extraction model and the trained recognition model are used to process biological samples to obtain cancer recognition results of the biological samples.

11. The method according to claim 10, characterized in that The method further comprises: Acquiring state information of the methylation site of the second sample, where the state information of the methylation site of the second sample is obtained by performing the methylation sequencing on the nucleic acid of the second biological sample; Based on the state information of the methylation site of the second sample, obtaining a characterization feature of the methylation site of the second sample; Based on the characterizing features of the methylation sites of the second sample and the self-supervised learning task, an initial feature extraction model is trained to obtain the first feature extraction model.

12. A biological sample processing device, characterized in that: The device comprises: An acquisition module, used to perform methylation sequencing on the nucleic acid of the biological sample to obtain status information of the methylation site in the genome of the nucleic acid, wherein the status information of the methylation site is used to indicate the specific situation of methylation of the methylation site; The acquisition module is further used to convert the state information of the methylation site into a first feature; convert the auxiliary characterization information of the methylation site into a second feature, wherein the auxiliary characterization information is used to auxiliary characterize the methylation site based on the state information of the methylation site; and fuse the first feature and the second feature to obtain the characterization feature of the methylation site; the auxiliary characterization information includes at least one of the first information, the second information, the third information, the fourth information or the fifth information; the first information is used to indicate the position of the methylation site in the genome of the nucleic acid; the second information is used to indicate the chromosome to which the methylation site belongs; the third information is used to indicate the position of the methylation site in the chromosome to which it belongs; the fourth information is used to indicate the positional relationship between the methylation site and the functional region in the chromosome to which it belongs; and the fifth information is used to indicate the positional relationship between the methylation site and the gene region in the chromosome to which it belongs; An extraction module, used to call the trained feature extraction model to extract the feature characteristics of the biological sample based on the feature characteristics of the methylation site; The recognition module is used to call the trained recognition model to perform cancer recognition on the biological sample based on the characterization features of the biological sample to obtain a cancer recognition result of the biological sample.

13. The device according to claim 12, characterized in that The extraction module is used to call the trained feature extraction model to perform N rounds of feature transformation based on the characterization features of the methylation site, and obtain the characterization features of the biological sample based on the transformed features; wherein the i-th round of feature transformation in the N rounds of feature transformation is used to obtain the output features corresponding to the i-th round of feature transformation based on the input features corresponding to the i-th round of feature transformation; i is an integer less than or equal to N; N is an integer greater than or equal to 1; the input features corresponding to the first round of feature transformation in the N rounds of feature transformation are the characterization features of the methylation site; the output features corresponding to the N-th round of feature transformation in the N rounds of feature transformation are the transformed features.

14. The device according to claim 13, characterized in that The extraction module is used to perform pre-processing operations on the input features corresponding to the i-th round of feature transformation to obtain the pre-processing features corresponding to the i-th round of feature transformation; transform the pre-processing features corresponding to the i-th round of feature transformation based on the attention mechanism to obtain the transformation features corresponding to the i-th round of feature transformation; perform post-processing operations on the transformation features corresponding to the i-th round of feature transformation to obtain the output features corresponding to the i-th round of feature transformation.

15. The device according to claim 14, characterized in that The extraction module is used to fuse the transformation features corresponding to the i-th round of feature transformation and the input features corresponding to the i-th round of feature transformation to obtain the fused features corresponding to the i-th round of feature transformation; normalize the fused features corresponding to the i-th round of feature transformation to obtain the normalized features corresponding to the i-th round of feature transformation; activate the normalized features corresponding to the i-th round of feature transformation to obtain the activated features corresponding to the i-th round of feature transformation; pool the activated features corresponding to the i-th round of feature transformation to obtain the output features corresponding to the i-th round of feature transformation.

16. The device according to claim 12, characterized in that The number of the methylation sites is at least two, and the trained feature extraction model is used to perform feature extraction on sequence data; The extraction module is used to input the characterizing features of at least two methylation sites as the sequence data into the trained feature extraction model to perform feature extraction, and use the features output by the trained feature extraction model as the characterizing features of the biological sample.

17. The device according to claim 12, characterized in that The number of the methylation sites is at least two, and the trained feature extraction model is used to perform feature extraction on the graph structure; The extraction module is used to construct the graph structure based on the characterization features of at least two methylation sites, input the graph structure into the trained feature extraction model to perform feature extraction, and use the features output by the trained feature extraction model as the characterization features of the biological sample.

18. The device according to claim 12, characterized in that The extraction module is used to fuse the characterization feature of the methylation site and the reference feature of the biological sample to obtain a biological fusion feature; The reference feature is obtained based on reference biological information, and the reference biological information is obtained by processing the biological sample in a manner different from the methylation sequencing; the trained feature extraction model is called to extract the characterization feature of the biological sample based on the biological fusion feature.

19. The device according to claim 12, characterized in that The extraction module is used to call the trained feature extraction model to extract the first characterization feature of the biological sample based on the characterization feature of the methylation site; Extracting a second characterizing feature of the biological sample based on the reference feature of the biological sample; The reference feature is obtained based on reference biological information, and the reference biological information is obtained by processing the biological sample in a manner different from the methylation sequencing; the first characterization feature and the second characterization feature are fused to obtain the characterization feature of the biological sample.

20. The device according to any one of claims 12 to 19, characterized in that The cancer identification result of the biological sample includes a healthy category and a predicted probability of at least one cancer category; or, the cancer identification result of the biological sample includes a cancer risk value.

21. A biological sample processing device, characterized in that: The device comprises: an acquisition module, configured to acquire a cancer identification tag of a first sample biological sample and status information of a first sample methylation site, wherein the status information of the first sample methylation site is obtained by performing methylation sequencing on the nucleic acid of the first sample biological sample, and the status information of the first sample methylation site is used to indicate a specific situation in which methylation occurs at the first sample methylation site; The acquisition module is further used to convert the state information of the first sample methylation site into a first sample feature; convert the auxiliary characterization information of the first sample methylation site into a second sample feature, the auxiliary characterization information is used to assist in characterizing the first sample methylation site based on the state information of the first sample methylation site; the first sample feature and the second sample feature are fused to obtain the characterization feature of the first sample methylation site; the auxiliary characterization information includes at least one of the first information, the second information, the third information, the fourth information or the fifth information; the first information is used to indicate the position of the first sample methylation site in the genome of the nucleic acid of the first sample biological sample; the second information is used to indicate the chromosome to which the first sample methylation site belongs; the third information is used to indicate the position of the first sample methylation site in the chromosome to which it belongs; the fourth information is used to indicate the positional relationship between the first sample methylation site and the functional region in the chromosome to which it belongs; the fifth information is used to indicate the positional relationship between the first sample methylation site and the gene region in the chromosome to which it belongs; An extraction module, configured to call a first feature extraction model to extract a feature of the first biological sample based on the feature of the methylation site of the first sample; an identification module, configured to call a first identification model to perform cancer identification on the first sample biological sample based on the characterization feature of the first sample biological sample, and obtain a cancer identification result of the first sample biological sample; an updating module, configured to update the parameters of the first feature extraction model and the parameters of the first recognition model based on the difference between the cancer recognition result of the first sample biological sample and the cancer recognition label of the first sample biological sample, so as to obtain a trained feature extraction model and a trained recognition model; The trained feature extraction model and the trained recognition model are used to process biological samples to obtain cancer recognition results of the biological samples.

22. The device according to claim 21, characterized in that The device also includes: A training module is used to obtain the state information of the methylation site of the second sample, where the state information of the methylation site of the second sample is obtained by performing the methylation sequencing on the nucleic acid of the second sample biological sample; based on the state information of the methylation site of the second sample, obtain the characterization features of the methylation site of the second sample; based on the characterization features of the methylation site of the second sample and the self-supervised learning task, train the initial feature extraction model to obtain the first feature extraction model.

23. A computer device, characterized in that: The computer device includes a processor and a memory, wherein at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor so that the computer device implements the biological sample processing method as described in any one of claims 1 to 11.

24. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one computer program, and the at least one computer program is loaded and executed by a processor so that a computer implements the biological sample processing method according to any one of claims 1 to 11.

25. A computer program product, characterized in that The computer program product comprises computer instructions, which are loaded and executed by a processor so as to enable a computer to implement the biological sample processing method according to any one of claims 1 to 11.

Images