Methods, sequencing methods, and devices for determining quality scores (Q-values) for base calls

Abstract

The application provides a method for determining a quality score Q value of base detection, a sequencing method and a device. The method comprises: receiving input image data for determining a quality score Q value of base detection in an Nth sequencing reaction, wherein the input image data comprises an image of a template to be detected in an Mth sequencing reaction and an image of the template to be detected in the Nth sequencing reaction, M and N are natural numbers and M is less than N; determining a predetermined prediction feature based on the input image data; determining a prediction probability value of correctly identifying a base incorporated into the template to be detected in the Nth sequencing reaction based on the predetermined prediction feature and using a trained machine learning model; and determining the quality score Q value of base detection in the Nth sequencing reaction based on the prediction probability value. The method can improve the efficiency and / or accuracy of determining the quality score Q value of base detection.

Description

Technical Field

[0001] This application relates to the field of bioinformatics, specifically to the field of nucleic acid sequencing, and more specifically to methods, sequencing methods, and apparatus for determining the quality score Q value of base detection. Background Technology

[0002] Nucleic acid sequencing is a technique used to determine the base sequence in DNA or RNA molecules. DNA and RNA are the genetic material of living organisms, each composed of four bases: adenine (A), guanine (G), thymine (T, found only in DNA), uracil (U, found only in RNA), and cytosine (C). The accuracy and order of nucleic acid sequences play a crucial role in the function and genetic information of organisms.

[0003] With technological advancements, higher-throughput, faster, and more economical nucleic acid sequencing methods have emerged, such as next-generation sequencing (NGS) technology. Sequencing-while-synthesizing refers to sequencing nucleic acids by monitoring the base sequence on the newly synthesized strand in real time during DNA or RNA synthesis. This method can acquire base sequence information during synthesis, offering advantages in real-time performance and efficiency compared to traditional methods that sequence after synthesis.

[0004] In nucleic acid sequencing, especially in next-generation sequencing (NGS), the Q-value is commonly used to measure the error rate of base detection. However, current methods for determining the Q-value still require further improvement.

[0005] Application content

[0006] This application aims to at least partially solve one of the aforementioned technical problems or at least provide a useful commercial solution. To this end, this application proposes a means for effectively determining the quality score Q of base detection.

[0007] In a first aspect of this application, a method for determining the quality score Q value of base detection is proposed. According to an embodiment of this application, the method includes: receiving input image data for determining the quality score Q value of base detection in an Nth sequencing reaction, the input image data including an image of a test template in an Mth sequencing reaction and an image of the test template in an Nth sequencing reaction, where M and N are both natural numbers and M is less than N; determining predetermined prediction features based on the input image data; determining a predicted probability value for correctly identifying bases incorporated into the test template in the Nth sequencing reaction using a trained machine learning model based on the predetermined prediction features; and determining the quality score Q value of base detection in the Nth sequencing reaction based on the predicted probability value.

[0008] According to the embodiments of this application, by using the above method, in the process of determining the quality score Q value of base detection, not only the sequencing data of the current sequencing round is considered, but also the sequencing data of other sequencing rounds, such as sequencing rounds before the current sequencing round, thereby improving the efficiency and / or accuracy of determining the quality score Q value of base detection.

[0009] In a second aspect of this application, a sequencing method is proposed, comprising: performing sequencing-while-synthesizing on a nucleic acid molecule to be tested; and determining a quality score Q value for base detection in at least one round of sequencing reactions during sequencing-while-synthesizing, according to the method described in the first aspect.

[0010] According to embodiments of this application, this method can provide an effective quality score Q value for nucleic acid sequencing, thereby identifying sequencing reads that can be used for subsequent bioinformatics analysis by analyzing the quality score Q value, improving the efficiency of bioinformatics analysis and reducing its cost. As mentioned above, in determining the quality score Q value for base detection, not only is the sequencing data from the current sequencing round considered, but also the sequencing data from other sequencing rounds, such as those prior to the current sequencing round, thereby improving the efficiency and / or accuracy of determining the quality score Q value for base detection.

[0011] In a third aspect of this application, an apparatus for determining the quality score Q value of base detection is also proposed. According to an embodiment of this application, the apparatus includes: an image data receiving unit for receiving input image data for determining the quality score Q value of base detection in the Nth round of sequencing, the input image data including an image of the template to be tested in the Mth round of sequencing and an image of the template to be tested in the Nth round of sequencing, where M and N are both natural numbers and M is less than N; a prediction feature determination unit for determining predetermined prediction features based on the input image data; a prediction unit for determining, based on the predetermined prediction features, a trained machine learning model, a prediction probability value for correctly identifying bases incorporated into the template to be tested in the Nth round of sequencing; and a Q value determination unit for determining the quality score Q value of base detection in the Nth round of sequencing based on the prediction probability value. Optionally, the template to be tested is an amplification product containing multiple identical nucleic acid molecules.

[0012] According to embodiments of this application, the method for determining the quality score Q value of base detection, as described in the first aspect, can be effectively implemented by employing this device. As previously stated, in the process of determining the quality score Q value of base detection, not only the sequencing data of the current sequencing round is considered, but also the sequencing data of other sequencing rounds, such as sequencing rounds prior to the current sequencing round, thereby improving the efficiency and / or accuracy of determining the quality score Q value of base detection.

[0013] In a fourth aspect of this application, a computing device is provided, according to an embodiment of this application, comprising: a processor and a memory; the memory for storing a computer program; and the processor for executing the computer program to implement the method as described in the first aspect.

[0014] In a fifth aspect of this application, a computer-readable storage medium, according to an embodiment of this application, includes computer instructions that, when executed by a computer, cause the computer to perform the method described in the first aspect.

[0015] In a sixth aspect of this application, a computer program product includes a computer program that, according to an embodiment of this application, implements the method described in the first aspect when executed by a processor. Attached Figure Description

[0016] The above and / or additional aspects and advantages of the embodiments of this application will become apparent and readily understood from the description of the embodiments in conjunction with the following drawings, wherein:

[0017] Figure 1 This is a flowchart illustrating a method for determining the quality score Q value of base detection according to an embodiment of this application;

[0018] Figure 2 This is a schematic diagram of the process of training a machine learning model according to an embodiment of this application;

[0019] Figure 3 This is a schematic diagram of a process for determining predictive features according to an embodiment of this application;

[0020] Figure 4 This is a ranking of the importance of model variables trained in the pre- and mid-cycles according to an embodiment of this application;

[0021] Figure 5 This is a ranking of the importance of model variables trained in the pre-, mid-, and post-cycle phases according to an embodiment of this application;

[0022] Figure 6 The curve showing the relationship between the predicted probability value and the actual Q value according to one embodiment of this application is displayed;

[0023] Figure 7 The diagram shows the fitted real Q value to the predicted probability value according to an embodiment of this application, where the horizontal axis is PredictQ, the vertical axis is RealQ, the yellow curve is the fitted curve, and the blue curve is the real curve. The fitting function is: pre_realQ = 6.35498 + P * 0.73054 + P 2 *(-0.01026)+P 3 *0.00005;

[0024] Figure 8 The comparison between the predicted Q-value and the actual Q-value according to one embodiment of this application is shown.

[0025] Figure 9 The distribution of Q-values ​​for correct and incorrect bases at false positive sites according to one embodiment of this application is shown;

[0026] Figure 10 The curve showing the relationship between the predicted probability value and the actual Q value according to one embodiment of this application is displayed;

[0027] Figure 11 The image shows the predicted probability value fitted to the true Q value according to one embodiment of this application, where the yellow curve is the fitted curve and the blue curve is the true curve. The fitting function is: pre_realQ = 4.9722 + P * 0.6759 + P 2 *(-0.01)+P 3 *0.00006 where P is the predicted probability value *100;

[0028] Figure 12 The comparison between the predicted Q-value and the actual Q-value according to one embodiment of this application is shown.

[0029] Figure 13The distribution of Q-values ​​for correct and incorrect bases at false positive sites according to one embodiment of this application is shown;

[0030] Figure 14 This illustrates a comparison of the pre-mid model and the pre-mid-post model according to one embodiment of this application on ordinary human test data; and

[0031] Figure 15 The Q values ​​for the correct and incorrect bases at false positive sites are shown in the front-middle-back model and the front-middle model according to one embodiment of this application. Detailed Implementation

[0032] The embodiments of this application are described in detail below. This application can be implemented in many different forms and is not limited to the embodiments described herein. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0033] definition

[0034] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0035] The term "base detection," also known as "base call" or "base identification," is used interchangeably in this article and refers to determining the type of base at at least one position in a nucleic acid sequence. For example, in sequencing-by-synthesis, the type of base extended in each sequencing reaction can be determined by analyzing the fluorescence signal generated in each sequencing reaction.

[0036] The term "quality score Q" used in this article, also known as "sequencing quality score," "Q value," or "base quality score," is interchangeable in its terminology. It is used to assess the error detection rate of bases and is usually expressed as the Phred quality score, calculated using the following formula:

[0037] Q = -10Log 10 (E)

[0038] Here, E represents the probability of a base detection error. Generally, as the Phred quality score (Q value) increases, the confidence level of the base detection also increases; that is, a higher Q value indicates higher confidence and a greater likelihood that the base was detected correctly. For example, the Q value can range from 0 to 50. For instance, a base with a Q value of 50 indicates a 99.999% confidence level in the correctness of the base detection, with a probability of error of 0.001%. Similarly, a base with a Q value of 30 indicates a 99.9% confidence level in the correctness of the base detection, with a probability of error of 0.1%. The Q value is crucial for applications such as SNP (single nucleotide polymorphism) analysis, variation analysis, and genome assembly. High-quality sequencing data provides more accurate results and reduces misunderstandings in subsequent analyses. More information about Q values ​​can be found at the following link: https: / / www.illumina.com.cn / science / technology / next-generation-sequencing / plan-experiments / qua lity- scores.html .

[0039] As used in this article, the term "sequencing reaction" refers to a series of reactions that occur during nucleic acid sequencing to determine the types of bases incorporated into the nucleic acid sequence. For example, in sequencing by synthesis (SBS), when a nucleotide carrying a blocking group and a fluorescent group binds to a base at a specific position in the nucleic acid sequence, the fluorescent group on the nucleotide is excited by light to generate a fluorescence signal, which is detected by a multi-channel microscopy system. Based on the detection results, the types of bases incorporated into the nucleic acid sequence and / or the base types at that position in the nucleic acid sequence can be determined. Then, the fluorescent group and the blocking group on the nucleotide are removed so that the next nucleotide carrying a blocking group and a fluorescent group can bind to the base at the next position in the nucleic acid sequence. In short, sequencing by synthesis can include the following steps:

[0040] Library construction: First, it is necessary to prepare nucleic acid samples to be sequenced. Nucleic acid samples can be, for example, DNA extracted from biological samples, and then subjected to necessary processing and preparation, such as using ultrasound to randomly break the DNA into small fragments, adding adapters, and performing PCR amplification.

[0041] Primer binding: Primers can be oligonucleotides or nucleic acid molecules that hybridize with the test template in the library. At the start of the sequencing process, the primers bind to the test template through hybridization and serve as the starting point for synthesis, initiating the synthesis of a new strand complementary to the test template.

[0042] Extension: Once the primer binds to the template, under the action of DNA polymerase and under conditions suitable for polymerase chain reaction, nucleotides carrying fluorescent and blocking groups are incorporated into the template, thereby extending the primer by one base.

[0043] Imaging: Fluorescent signals are generated by exciting fluorescent groups on nucleotides with excitation light, and the test template is imaged using a multi-channel microscopic imaging system to acquire the fluorescence signals and form an image.

[0044] Excision: After imaging, the fluorescent and blocking groups on the nucleotide are removed, freeing the 3' hydroxyl group. This 3' hydroxyl group can then be used to guide the synthesis of a new chain.

[0045] Repeat the extension-imaging-excision process multiple times until sequencing of the test template is complete.

[0046] Therefore, in this paper, the sequencing reaction can mainly include the steps of extension, imaging, and excision mentioned above.

[0047] Sequencing reactions generally involve multiple rounds to determine the sequence of multiple nucleotides / bases on a test template. In this application, each round of "determining the sequence of multiple nucleotides / bases on a test template" is referred to as "one round of sequencing." A "sequencing round" (cycle) can be defined as the determination of the base type at any specified position on the test template. Typically, a round of sequencing may include one or more base extensions (repeat). For example, if four nucleotides are added sequentially to the reaction system, and base extensions and corresponding reaction signals are collected separately, a round of sequencing includes four base extensions. Alternatively, if four nucleotides are added to the reaction system in any combination, such as in pairs or in a one-to-three combination, and base extensions and corresponding reaction signals are collected separately for each combination, a round of sequencing includes two base extensions. Yet another example is if four nucleotides are added to the reaction system simultaneously for base extensions and reaction signal collection, a round of sequencing includes one base extension.

[0048] As used in this article, the term "machine learning model" can refer to an algorithm that learns from data and makes predictions or decisions. These models can be decision trees, logistic regression, Naive Bayes, or neural networks, etc., which analyze large amounts of data to identify patterns and relationships in the data, thereby making accurate predictions for new data instances. For example, lightGBM is an efficient gradient boosting framework that uses tree models to process data and provides fast and accurate predictions in a variety of machine learning tasks.

[0049] As used in this article, the term "decision tree" refers to a tree-like model that mimics the human decision-making process, categorizing data into different classes or decisions through a series of questions. Each node represents a test of an attribute, each branch represents the result of the test, and each leaf node represents a decision or classification result. Decision trees recursively divide data in this way until a class of purity is reached or a stopping condition is met.

[0050] In this paper, the term "logistic regression" can refer to a statistical method used to predict binary classification problems. It estimates the probability of an event occurring using a logistic function (usually the sigmoid function). The logistic regression model calculates a probability value based on the input features, representing the likelihood of a particular class being present.

[0051] The term "Naive Bayes" as used in this article refers to a simple probabilistic classifier based on Bayes' theorem, which assumes that features are independent of each other. Although this assumption often does not hold true in reality, Naive Bayes classifiers still perform well in practice, especially in fields such as text classification and spam detection.

[0052] As used in this article, the term "neural network" can refer to a computational model that mimics the connections between neurons in the human brain. It consists of multiple layers, including input layers, hidden layers, and output layers. Neural networks learn complex patterns and relationships in data by adjusting the weights between layers, thereby enabling them to recognize patterns in data such as images, speech, and text.

[0053] In this paper, the term "input image data" refers to images captured during the sequencing of the test template, containing visual information about the test template in different sequencing rounds. This data is crucial for determining the base detection quality score; it records fluorescence signals during the sequencing process, which are subsequently converted into electronic data for further analysis.

[0054] In this article, the term "test template" can refer to a DNA or RNA molecule template that is being sequenced. These templates are the starting point of the sequencing process, and all sequencing reactions revolve around these templates to determine the base sequence on them.

[0055] In this paper, the term "predicted features" can refer to specific information extracted from input image data that is used to predict the quality score of base detection. These features may include base brightness, brightness ratio, etc., which are fundamental to sequencing signal analysis and are crucial for assessing sequencing quality.

[0056] In this article, the term "machine learning model" can refer to an algorithm or statistical model that learns patterns by analyzing input data and corresponding output results, and makes predictions or decisions based on these patterns. These models are able to extract knowledge from complex data patterns and can be used for various predictive tasks.

[0057] In this paper, the term "predicted probability value" refers to a value calculated by a machine learning model based on input data, representing the probability of correctly identifying a base incorporated into the template during the sequencing reaction. This value is an important indicator for evaluating sequencing accuracy.

[0058] In this paper, the terms "T base brightness," "G base brightness," "C base brightness," and "A base brightness" refer to the brightness of the fluorescence signal generated by the fluorescent groups carried by the nucleotides incorporated into the template, such as thymine (T), guanine (G), cytosine (C), and adenine (A), respectively, after excitation by light in a single sequencing reaction. These brightness values ​​are crucial for identifying and distinguishing different bases.

[0059] In this paper, the term "brightness ratio" refers to the ratio between the brightness of a specific fluorescence signal in a sequencing run, such as the ratio between the brightness of the brightest base and the second brightest base. This ratio is an indicator for evaluating the accuracy of base detection. In sequencing processes that identify incorporated base types based on fluorescence signal detection, the signal intensity / brightness at each test template location is typically represented as an array corresponding to four bases, for example, [B1, B2, B3, B4], where B1 to B4 represent signals generated at A, T, C, and G, respectively. In one run, for example, if the signal intensity at a test template location is [80, 10, 5, 5], based on the signal intensity, it can be determined that A is the base with the highest probability of being incorporated into that test template, thus determining that the extended base type corresponding to that test template location in that run is A. Here, 80 represents the maximum brightness value (or the maximum base brightness), corresponding to A being the brightest base; 10 represents the second largest brightness value (or the second largest base brightness), corresponding to T being the second brightest base, and so on. In this paper, the term "the proportion of the brightest base to the sum of the brightness of the four bases" refers to the proportion of the brightness of the brightest base (e.g., A) at a given template location to the total brightness of all bases (A, T, C, G) in a single sequencing reaction. This proportion helps assess the relative brightness of a specific base, thus affecting the accuracy of base identification.

[0060] In this article, the term "digitally encoded result" can refer to the conversion of the base detection results of a sequencing reaction, such as the types of bases detected, into a digital encoding. This encoding facilitates computer processing and analysis of sequencing data.

[0061] In this paper, the term "lightGBM" refers to a gradient boosting framework that uses tree models for fast, distributed training. This framework is particularly effective when dealing with large-scale datasets, enabling the construction of efficient and accurate prediction models.

[0062] In this paper, the terms "positive sample sequencing reaction" and "negative sample sequencing reaction" refer to sequencing reactions in which known bases were detected correctly and incorrectly, respectively, when training a machine learning model. These samples are crucial for model training, providing the foundational data for model learning.

[0063] In this paper, the term "predetermined benchmark true value" refers to the standard answer or true value used for training and evaluating machine learning models. These true values ​​serve as a benchmark for evaluating the predictive accuracy of the model.

[0064] In this paper, the term "supervised training" refers to the process of training a machine learning model using labeled data. In this training, the model adjusts its parameters by comparing predicted results with actual results to improve the accuracy of predictions.

[0065] In this paper, the term "random downsampling" refers to the process of randomly selecting a subset from the dataset. Doing so can reduce the amount of data, improve training efficiency, and help prevent model overfitting.

[0066] In this paper, the term "importance score" refers to the assessment of the importance of a feature in a machine learning model. This score indicates the magnitude of each feature's contribution to the model's predictive power.

[0067] In this paper, the term "feature combination" refers to the process of selecting and combining the optimal subset of features from an initial set of features. This process aims to improve model performance by constructing a more accurate predictive model by selecting the most informative features.

[0068] In this paper, the term "trained machine learning model" refers to a machine learning model that has learned how to map from input to output using training data. These models are able to capture complex patterns in the data and make accurate predictions based on them.

[0069] In this paper, the term "acquiring input image data for positive and negative sample sequencing reactions" refers to the need to collect known correct (positive samples) and incorrect (negative samples) sequencing data during model training. This data is crucial for training machine learning models, providing the foundation for model learning.

[0070] Phasing refers to a situation where a base that should have been incorporated into the template during, for example, cycle N, participates in the reaction later, in cycle N+1. This is likely due to insufficient enzyme activity, preventing the removal of the 3' end blocking group. If the 3' end blocking group of the base added in the previous extension reaction is not removed, the next extension reaction cannot proceed, and the captured fluorescence signal will be from the previous extension reaction, not the current one, leading to sequencing inaccuracies.

[0071] Prephasing refers to a situation where a base that should have been incorporated into the template during, for example, Cycle N, participates in the reaction prematurely during Cycle N1. This crosstalk occurs between adjacent cycles within the same channel, likely due to the instability of the 3' end blocking group in organic compounds. In the previous extension reaction, the base incorporated into the template lacked a 3' end blocking group, resulting in more than one base being incorporated. Consequently, the fluorescence signal captured in the next extension reaction is not from the current extension but from the previous one, leading to sequencing inaccuracies.

[0072] In short, both phasing and prephasing are deviations caused by chemical or enzymatic factors during sequencing, affecting the accuracy of fluorescence signals and consequently the accurate detection of bases. Both phenomena can lead to errors in sequencing results. The new algorithm, by considering the characteristics of the pre-, mid-, and post-sequencing cycles, can incorporate these effects into the model, improving the accuracy of sequencing quality assessment.

[0073] The following is a reference appendix. Figures 1-3 This application is described in detail.

[0074] This invention is based on the following discoveries of the inventors:

[0075] Traditional base Q-value assessment methods rely on statistically analyzing the error rate of four channel brightness combinations and converting it into a Q-value using a fixed formula. This method requires extensive data analysis to ensure the reliability of the results, but it has limitations, such as only considering the base brightness characteristics of the current cycle and failing to cover the effects of phasing and prephasing during sequencing. To address these limitations, the inventors of this application propose a novel algorithm. This new algorithm employs a machine learning classification algorithm, directly correlating the obtained Q-value with the traditional error rate through a fitting function, reducing the amount of data required. It also comprehensively considers the characteristics of the current cycle and its preceding and following cycles, as well as multiple features such as base brightness and call base type, to improve assessment accuracy. Furthermore, the new algorithm further enhances model performance through iterative optimization of feature selection, aiming to provide a more efficient and accurate method for assessing base sequencing quality.

[0076] In a first aspect of this application, a method for determining the quality score Q value of base detection is proposed, with reference to Figure 1 According to an embodiment of this application, the method includes:

[0077] S100: Receive input image data

[0078] In this step, in order to evaluate the quality of base detection, firstly, input image data is received to determine the quality score Q value of base detection in the Nth sequencing reaction. The input image data includes images of the template to be tested in the Mth sequencing reaction and images of the template to be tested in the Nth sequencing reaction, where M and N are natural numbers and M is less than N.

[0079] According to embodiments of this application, in this step, key image data is first received, which forms the basis for base detection quality assessment. Specifically, this data includes image information of the template to be tested in the Mth and Nth sequencing rounds. Here, the "template to be tested" can be, for example, a DNA or RNA molecule to be sequenced. M and N represent different sequencing rounds, where M is an earlier sequencing round and N is a later sequencing round, and both M and N are natural numbers, with M less than N. This image data reflects the fluorescence signals of the template to be tested in different sequencing rounds, and these signals are subsequently used to determine the base detection quality. According to embodiments of this application, images of the template to be tested from the Mth sequencing round in multiple (e.g., two or more) sequencing rounds can be used.

[0080] S200: Determine prediction features

[0081] In this step, after obtaining the input image data, predetermined prediction features are determined based on the input image data.

[0082] According to embodiments of this application, in this step, features for predicting base detection quality are extracted from the received input image data. According to embodiments of this application, these features are referred to as "predicted prediction features" and may include the brightness and brightness ratio of various fluorescence signals, as well as the digital encoding results of base detection. According to embodiments of this application, the selection of these features is based on their correlation with base detection quality; they can provide necessary information for subsequent machine learning models to predict the quality score Q value of base detection.

[0083] S300: Determine the predicted probability value

[0084] In this step, after obtaining the predetermined prediction features, a trained machine learning model can be used to determine the predicted probability value of correctly identifying the bases incorporated into the template to be tested during sequencing.

[0085] According to an embodiment of this application, in this step, a pre-trained machine learning model is used to predict the probability of correctly identifying the bases incorporated into the template to be tested in the Nth round of sequencing, based on predetermined prediction features determined in step S200. This predicted probability reflects the model's confidence level in the accuracy of base detection and is a key indicator for evaluating the quality of base detection.

[0086] S400: Determine the quality score Q value

[0087] In this step, after obtaining the predicted probability value, the quality score Q value of base detection in the Nth round of sequencing reaction can be determined based on the predicted probability value.

[0088] According to embodiments of this application, in this step, based on the predicted probability value obtained in step S300, a pre-constructed fitting formula (e.g., according to embodiments of this application, this formula may be a cubic polynomial or other mathematical model) can be used to determine the quality score Q of base detection in the Nth round of sequencing. The Q value is a standard metric used to represent the accuracy of base detection; a higher Q value indicates higher accuracy and a lower error rate. This score is ultimately used to evaluate the quality of the sequencing data and guide subsequent bioinformatics analysis.

[0089] According to embodiments of this application, the inventors have proposed two machine learning-based models aimed at improving the accuracy of base Q-value evaluation in next-generation sequencing:

[0090] 1. Pre- and Mid-Sequencing Model: This model is trained by considering the bases (mid) of the current target sequencing round and the bases (pre-) of at least one sequencing round preceding it. This approach aims to capture the interaction between adjacent sequencing rounds by analyzing the base features of two adjacent sequencing rounds to evaluate the sequencing quality of the current sequencing round.

[0091] 2. Pre-, Inter-, and Post-Sequencing Model: Compared to the pre-, intermediate, and post-sequencing model, this model further considers the base characteristics of the sequencing round following the current target sequencing round (post-sequencing round), that is, it considers the base characteristics of the three sequencing rounds simultaneously. This design helps to more comprehensively understand the dynamic changes in the sequencing process, including the impact of phasing and prephasing, thereby potentially improving the model's ability to identify sequencing errors and its prediction accuracy.

[0092] As mentioned earlier, both phasing and prephasing are deviations caused by chemical or enzymatic factors during sequencing, affecting the accuracy of fluorescence signals and consequently the accurate detection of bases. Both phenomena can lead to errors in sequencing results. The new algorithm, by considering features from the pre-, mid-, and post-sequencing cycles, can incorporate these effects into the model, improving the accuracy of sequencing quality assessment. Therefore, according to embodiments of this application, through machine learning techniques, these models can learn from sequencing data and predict sequencing quality, potentially providing more accurate results compared to traditional statistical methods.

[0093] According to embodiments of this application, the core of the algorithm lies in using machine learning technology to improve the accuracy of base Q-value evaluation in next-generation sequencing. First, useful information for the model is discovered by analyzing features such as the brightness and base combinations of the pre-, mid-, and post-sequential bases. Next, the LightGBM classification model is used, trained with the goal of correctly identifying bases incorporated into the test template. The probability value output by the model represents the likelihood of correctly identifying a base; a higher probability value indicates a higher probability of correct base identification and better base detection quality. To align with the traditional definition of Q-value, the algorithm groups the predicted data according to the probability of correctly identifying bases, calculates the probability of base detection errors for each group, and derives the true Q-value for each group. By statistically analyzing the relationship between the predicted probability value and the true Q-value, the algorithm fits a function to convert the predicted probability value into a Q-value. Finally, the model's performance is evaluated using a large amount of test data, where the relationship between the true Q-value and the probability of base detection errors follows the formula Q = -10 * log10(E). Furthermore, the model's accuracy is evaluated by setting a probability threshold. Bases with predicted probabilities exceeding this threshold are considered correctly identified as bases incorporated into the test template. The model's prediction accuracy is then calculated by comparing the predicted probability with a reference sequence. According to embodiments of this application, this method not only improves the accuracy of Q-value evaluation but also enhances the model's application value in practical sequencing.

[0094] According to the embodiments of this application, through the above steps, the technical effects of the embodiments of this application are mainly reflected in the following aspects:

[0095] First, the method according to the embodiments of this application can improve the accuracy of sequencing quality assessment. Specifically, by receiving input data containing the template images to be tested from the Mth and Nth sequencing reactions, it is ensured that the assessment process can be based on detailed fluorescence signal information, which is directly related to the detection quality of bases.

[0096] The method according to embodiments of this application can enhance the relevance and accuracy of feature selection. Specifically, according to embodiments of this application, the relevance and accuracy of feature selection are enhanced by extracting predetermined predictive features from input image data. These features include the brightness and brightness ratio of bases in different sequencing rounds, as well as the digitized encoding results, providing the most relevant and representative data for the machine learning model to predict the quality score Q value of base detection.

[0097] The method according to embodiments of this application enables accurate calculation of predicted probability values. Specifically, according to embodiments of this application, a trained machine learning model is used to predict the probability of correctly identifying a base incorporated into the template to be tested in the Nth round of sequencing, based on predetermined predictive features determined in step S200. This predicted probability value is a key indicator for evaluating the quality of base detection, reflecting the model's confidence level in the accuracy of base detection.

[0098] The method according to embodiments of this application can accurately determine the quality score Q value, thereby improving the quality of sequencing data. Specifically, according to embodiments of this application, the quality score Q value for base detection in the Nth round of sequencing is determined using a pre-constructed fitting formula based on the predicted probability value obtained in step S300. A higher Q value indicates higher accuracy in base detection and a lower error rate, thereby improving the quality of sequencing data.

[0099] The method according to the embodiments of this application can improve the reliability of bioinformatics analysis. Specifically, according to the embodiments of this application, by accurately evaluating the quality score Q value of base detection, the quality of sequencing data is ensured, thereby providing more reliable data support in genomics research, clinical diagnosis and other related fields, and enhancing the accuracy and credibility of subsequent analysis results.

[0100] The method according to embodiments of this application can optimize resource allocation and sequencing costs. Specifically, according to embodiments of this application, by automating and accelerating data processing workflows, manual intervention and repetitive work are reduced, improving work efficiency. Simultaneously, by improving sequencing quality, the number of samples requiring resequencing is reduced, thereby lowering overall sequencing costs and time consumption, enabling sequencing resources to be utilized more efficiently.

[0101] In summary, according to the embodiments of this application, this application provides an accurate and efficient method for evaluating and determining the quality score Q value of base detection in nucleic acid sequencing, which is crucial for improving the quality and reliability of sequencing data.

[0102] According to embodiments of this application, the input image data may include images of the template to be tested in the Mth sequencing reaction and images of the template to be tested in the Nth sequencing reaction, and the predetermined prediction features include at least one of the following: at least one of the brightness of T bases, G bases, C bases, and A bases in the Nth sequencing reaction; at least one of the brightness of T bases, G bases, C bases, and A bases in the Mth sequencing reaction; the brightness ratio of the brightest base to the second brightest base in the Nth sequencing reaction; the proportion of the brightest base to the sum of the brightness of the four bases in the Nth sequencing reaction; the proportion of the brightest base to the sum of the brightness of the four bases in the Mth sequencing reaction; the brightness ratio of the brightest base to the second brightest base in the Mth sequencing reaction; the digital encoding result of the combination of base detection results from the Nth and Mth sequencing reactions; the digital encoding result of the base detection result from the Nth sequencing reaction; and the digital encoding result of the base detection result from the Mth sequencing reaction. According to embodiments of this application, by employing at least one of the aforementioned predetermined predictive features, the efficiency and / or accuracy of determining the quality score Q value for base detection can be effectively improved by employing a trained machine learning model.

[0103] Specifically, in embodiments of this application, the collection of input image data can be limited to images from only two specific sequencing rounds: the template images from the Mth and Nth sequencing reactions. This limitation helps to focus the analysis on the most relevant data, thereby improving the efficiency and accuracy of the analysis. According to embodiments of this application, the selection of predetermined predictive features is crucial; these include the brightness of each base (A, T, G, C) in the Nth and Mth rounds, as well as the brightness ratio and total ratio, etc. These features provide direct information about the quality of base detection; for example, brightness indicates signal intensity, while the brightness ratio reflects the relative signal intensity between different bases. The digitized encoding results convert this information into a machine-readable format, facilitating further analysis and learning by the machine learning model. Through comprehensive analysis of these features, the machine learning model can more accurately predict the quality score Q value of base detection, thereby improving the overall reliability of the sequencing results.

[0104] According to an embodiment of this application, NM = K1, where K1 is any natural number selected from 1 to 20, optionally K1 is any natural number selected from 1 to 10, and optionally K1 is equal to 1.

[0105] In the embodiments of this application, the difference K1 between N and M represents the number of sequencing epochs from epoch M to epoch N. The value of K1 is limited to a natural number between 1 and 20, meaning that the analysis can cover cases ranging from adjacent sequencing epochs to those separated by a maximum of 20 sequencing epochs. This flexibility allows the scope of the analysis to be adjusted according to different sequencing depths and requirements. Specifically, K1 can also be further limited to a natural number between 1 and 10, which is suitable for cases requiring analysis of data from more recent sequencing epochs. In some cases, K1 can also be directly set to 1, i.e., only data from adjacent sequencing epochs are analyzed, which helps to capture the direct correlation between consecutive sequencing epochs.

[0106] According to embodiments of this application, the input image data also includes an image of the template to be tested in the Lth sequencing reaction, where L is a natural number greater than N. According to embodiments of this application, when determining the quality score Q value for base detection in the current round, further analysis of sequencing data after the current round can further improve the efficiency and / or accuracy of determining the quality score Q value for base detection using a trained machine learning model.

[0107] In embodiments of this application, the collection of input image data is not limited to rounds M and N, but may also include image data from round L, where L is a natural number greater than N. This expanded data collection helps provide more comprehensive sequencing information, especially when analyzing the base detection quality of the current sequencing round (round N). By including data from future sequencing rounds (round L), the model is able to consider sequencing results further in the future, which may help predict and correct potential errors in the current sequencing round. This approach can improve the efficiency and accuracy of machine learning models in predicting the base detection quality score Q-value because it allows the model to learn from a wider range of data and make more comprehensive predictions.

[0108] According to embodiments of this application, the predetermined prediction features include at least one of the following: at least one of the brightness of T bases, G bases, C bases, and A bases in the Nth sequencing reaction; at least one of the brightness of T bases, G bases, C bases, and A bases in the Lth sequencing reaction; at least one of the brightness of T bases, G bases, C bases, and A bases in the Mth sequencing reaction; the brightness ratio of the brightest base to the second brightest base in the Nth sequencing reaction; the proportion of the brightest base to the sum of the brightness of the four bases in the Nth sequencing reaction; the proportion of the brightest base to the sum of the brightness of the four bases in the Lth sequencing reaction; and the proportion of the brightest base to the sum of the brightness of the four bases in the Mth sequencing reaction. The characteristics of a sequencing reaction include: the proportion of the sum of the brightness of the four bases; the brightness ratio of the brightest and second brightest bases in the Lth sequencing reaction; the proportion of the brightest base in the sum of the brightness of the four bases in the Lth sequencing reaction; the brightness ratio of the brightest and second brightest bases in the Lth sequencing reaction; the combination of base detection results from the Nth, Lth, and Mth sequencing reactions; the brightness of the brightest base or the maximum base brightness in the Nth sequencing reaction; the brightness of the brightest base or the maximum base brightness in the Mth sequencing reaction; the brightness of the lowest base or the minimum base brightness in the Nth sequencing reaction; and the brightness of the lowest base or the minimum base brightness in the Mth sequencing reaction. These characteristics provide detailed information about the base brightness distribution in the sequencing reaction, which helps to assess the detection quality of specific bases.

[0109] According to an embodiment of this application, LN = K2, where K2 is any natural number selected from 1 to 20, K2 is any natural number selected from 1 to 10, and optionally, K2 equals 1. This parameter K2 represents the number of sequencing rounds from round N to round L, providing a quantitative measure of the relationship between different sequencing rounds.

[0110] According to embodiments of this application, the quality score Q is determined based on probability using a pre-constructed fitting formula. Optionally, the fitting formula is a cubic polynomial with probability as the variable. Alternatively, the quality score Q is determined based on predicted probability values ​​using a pre-constructed fitting formula. The optional fitting formula can be a cubic polynomial with predicted probability values ​​as the variable. In this way, predicted probability values ​​can be converted into quality scores Q, thus providing a quantitative measure of the accuracy of base detection. This conversion makes the calculation of Q values ​​more accurate, helping to improve the reliability of sequencing results and the accuracy of subsequent analyses.

[0111] According to embodiments of this application, the machine learning model is selected from at least one of decision trees, logistic regression, Naive Bayes, and neural networks; optionally, the machine learning model is selected from decision trees; preferably, the machine learning model is selected from lightGBM.

[0112] According to an embodiment of this application, the trained machine learning model is obtained through the following steps: acquiring input image data of positive sample sequencing reaction and negative sample sequencing reaction, wherein the base detection of positive sample is known to be a predetermined baseline true value, and the base detection of negative sample is known not to be a predetermined baseline true value; determining predetermined predictive features of positive sample sequencing reaction and negative sample sequencing reaction based on the input image data; inputting the predetermined predictive features into the machine learning model, and using the known base detection results of positive and negative samples as labels for supervised training, so as to obtain the trained machine learning model.

[0113] According to embodiments of this application, positive and negative samples are selected by randomly downsampling sequencing data from multiple sources.

[0114] According to an embodiment of this application, the predetermined predictive features are obtained by iteratively optimizing the initial predictive features by repeating the following steps at least once: using the initial predictive features, training a machine learning model and determining the importance score of the initial predictive features; and selecting a portion of the initial predictive features as an optimized feature combination based on the importance score.

[0115] According to an embodiment of this application, the template to be tested is an amplification product containing multiple identical nucleic acid molecules.

[0116] According to embodiments of this application, the template to be tested is a cluster or nanosphere.

[0117] According to embodiments of this application, the base detection in the Nth round of sequencing is the detection of the bases most likely to be incorporated into the template to be tested in the Nth round of sequencing.

[0118] According to embodiments of this application, the algorithm overcomes the limitations of traditional statistical methods by employing classification algorithms from machine learning to evaluate the quality of base sequencing. To maintain consistency with traditional base Q-values, we transform the Q-values ​​obtained from machine learning into Q-values ​​directly related to the error rate through a fitting function, thereby improving the accuracy and reliability of the evaluation. The new algorithm requires significantly less data than traditional algorithms. This is because machine learning algorithms are more efficient in feature extraction and pattern recognition, enabling accurate predictions with less data and reducing the burden of data collection and processing. This algorithm considers not only the base features of the current cycle but also the features of the previous cycle or one cycle before and after, thus incorporating the effects of phasing and prephasing during sequencing into the model, improving the model's ability to capture dynamic changes in sequencing. During feature development, this algorithm not only focuses on base brightness but also integrates multiple features such as base type combinations, the proportion of maximum base brightness, and the ratio of maximum brightness of bases before and after, enhancing the model's predictive ability and comprehensive evaluation of sequencing quality. During model training, key parameters such as num_leaves, min_data_in_leaf, subsample, lambda_l1, and lambda_l2 were adjusted to finely control the model's growth depth and prevent overfitting, thereby improving the model's generalization ability and prediction accuracy. In the model evaluation phase, in addition to using standard test data, data from false positive sites were specifically introduced for testing to verify the model's predictive performance under specific key conditions, ensuring the comprehensiveness and practicality of the model evaluation.

[0119] In a second aspect of this application, a sequencing method is proposed, comprising: performing sequencing-while-synthesizing on a nucleic acid molecule to be tested; and determining a quality score Q value for base detection in at least one round of sequencing reactions in the sequencing-while-synthesizing process, according to the method of the first aspect.

[0120] According to embodiments of this application, this method can provide an effective quality score Q value for nucleic acid sequencing, thereby determining sequencing reads that can be used for subsequent bioinformatics analysis by analyzing the quality score Q value, improving the efficiency of bioinformatics analysis and reducing its cost. As mentioned above, in the process of determining the quality score Q value for base detection, not only the sequencing data of the current sequencing round is considered, but also the sequencing data of other sequencing rounds, such as those preceding the current sequencing round, thereby improving the efficiency and / or accuracy of determining the quality score Q value for base detection.

[0121] According to embodiments of this application, the base detection in the Nth round of sequencing is the base that is most likely to be incorporated into the template to be tested in the Nth round of sequencing.

[0122] In a third aspect of this application, an apparatus for determining the quality score Q value of base detection is also proposed. According to an embodiment of this application, the apparatus includes: an image data receiving unit for receiving input image data for determining the quality score Q value of base detection in the Nth round of sequencing, the input image data including images of the template to be tested in the Mth round of sequencing and images of the template to be tested in the Nth round of sequencing, where M and N are both natural numbers and M is less than N; a prediction feature determination unit for determining predetermined prediction features based on the input image data; a prediction unit for determining, based on the predetermined prediction features, a trained machine learning model, a prediction probability value for correctly identifying bases incorporated into the template to be tested in sequencing; and a Q value determination unit for determining the quality score Q value of base detection in the Nth round of sequencing based on the prediction probability value. Optionally, the template to be tested is an amplification product containing multiple identical nucleic acid molecules.

[0123] According to embodiments of this application, the method for determining the quality score Q value of base detection, as described in the first aspect, can be effectively implemented by employing this device. As mentioned above, in the process of determining the quality score Q value of base detection, not only the sequencing data of the current sequencing round is considered, but also the sequencing data of other sequencing rounds, such as those preceding the current sequencing round, thereby improving the efficiency and / or accuracy of determining the quality score Q value of base detection.

[0124] According to embodiments of this application, the template to be tested is a cluster or nanosphere.

[0125] According to embodiments of this application, the base detection in the Nth round of sequencing is the detection of the bases most likely to be incorporated into the template to be tested in the Nth round of sequencing.

[0126] According to embodiments of this application, the input image data includes only images of the template to be tested in the Mth sequencing reaction and images of the template to be tested in the Nth sequencing reaction, and the predetermined prediction features include at least one of the following:

[0127] At least one of the T base brightness, G base brightness, C base brightness, and A base brightness in the Nth round of sequencing reaction;

[0128] At least one of the T base brightness, G base brightness, C base brightness, and A base brightness in the Mth sequencing reaction;

[0129] The ratio of the brightness of the brightest base to the second brightest base in the Nth round of sequencing reaction;

[0130] In the Nth round of sequencing, the proportion of the brightest base to the sum of the brightness of the other four bases;

[0131] In the Mth round of sequencing, the proportion of the brightest base to the sum of the brightness of the other four bases;

[0132] The ratio of the brightness of the brightest base to the second brightest base in the Mth sequencing reaction;

[0133] The digital encoding result of the combination of base detection results from the Nth and Mth sequencing reactions;

[0134] The digital encoding results of the base detection results of the Nth round of sequencing reaction;

[0135] The digital encoding results of the base detection results from the Mth round of sequencing reaction.

[0136] Optionally, NM = K1, where K1 is any natural number selected from 1 to 20, K1 is any natural number selected from 1 to 10, and K1 is equal to 1.

[0137] According to embodiments of this application, the input image data further includes an image of the template to be tested in the Lth round of sequencing reaction, where L is a natural number greater than N.

[0138] Optionally, the predetermined prediction features include at least one of the following:

[0139] At least one of the T base brightness, G base brightness, C base brightness, and A base brightness in the Nth round of sequencing reaction;

[0140] At least one of the T base brightness, G base brightness, C base brightness, and A base brightness in the Lth sequencing reaction;

[0141] At least one of the T base brightness, G base brightness, C base brightness, and A base brightness in the Mth sequencing reaction;

[0142] The ratio of the brightness of the brightest base to the second brightest base in the Nth round of sequencing reaction;

[0143] In the Nth round of sequencing, the proportion of the brightest base to the sum of the brightness of the other four bases;

[0144] In the Lth round of sequencing, the proportion of the brightest base to the sum of the brightness of the four bases;

[0145] In the Mth round of sequencing, the proportion of the brightest base to the sum of the brightness of the other four bases;

[0146] The ratio of the brightness of the brightest base to the second brightest base in the Lth sequencing reaction;

[0147] In the Lth round of sequencing, the proportion of the brightest base to the sum of the brightness of the four bases;

[0148] The ratio of the brightness of the brightest base to the second brightest base in the Lth sequencing reaction;

[0149] The combination of base detection results from the Nth, Lth, and Mth rounds of sequencing;

[0150] The brightness of the brightest base in the Nth round of sequencing;

[0151] The brightness of the brightest base in the Mth round of sequencing reaction;

[0152] The brightness of the base with the lowest brightness in the Nth round of sequencing; and

[0153] The brightness of the base with the lowest brightness in the Mth round of sequencing reaction.

[0154] Optionally, LN = K2, where K2 is any natural number selected from 1 to 20, K2 is any natural number selected from 1 to 10, and optionally, K2 equals 1.

[0155] According to embodiments of this application, the quality score Q is determined based on probability using a pre-constructed fitting formula. Optionally, the fitting formula is a cubic polynomial with probability as the variable.

[0156] Optionally, the machine learning model is selected from at least one of decision trees, logistic regression, Naive Bayes, and neural networks; optionally, the machine learning model is selected from decision trees; preferably, the machine learning model is selected from LightGBM.

[0157] According to embodiments of this application, a training unit is further included for obtaining a trained machine learning model through the following steps: acquiring input image data of positive sample sequencing reactions and negative sample sequencing reactions, wherein the base detection of positive samples is known to be a predetermined baseline true value, and the base detection of negative samples is known not to be a predetermined baseline true value; optionally, the positive sample sequencing reactions and negative sample sequencing reactions are selected by randomly downsampling sequencing data from multiple sources; determining predetermined predictive features of positive sample sequencing reactions and negative sample sequencing reactions based on the input image data; inputting the predetermined predictive features into the machine learning model, and using the known base detection results of positive and negative samples as labels for supervised training, so as to obtain a trained machine learning model.

[0158] According to an embodiment of this application, a predictive feature selection unit is further included, which is used to iteratively optimize the initial predictive features to obtain predetermined predictive features by repeating the following steps at least once: using the initial predictive features, training a machine learning model and determining the importance score of the initial predictive features; and selecting a portion of the initial predictive features as an optimized feature combination based on the importance score.

[0159] In a fourth aspect of this application, a computing device is provided, according to an embodiment of this application, comprising: a processor and a memory; the memory for storing a computer program; and the processor for executing the computer program to implement the method as described in the first aspect.

[0160] In a fifth aspect of this application, a computer-readable storage medium, according to an embodiment of this application, includes computer instructions that, when executed by a computer, cause the computer to perform the method described in the first aspect.

[0161] In a sixth aspect of this application, a computer program product includes a computer program that, according to an embodiment of this application, implements the method described in the first aspect when executed by a processor.

[0162] The following examples are provided to better illustrate the claimed invention and the embodiments described herein, but these examples should not be construed as limiting the scope of the invention. Unless otherwise specified, the specific materials listed in the following examples are for illustrative purposes only and are not intended to limit the invention. It will be apparent to those skilled in the art that various equivalents, changes, and modifications can be made without departing from the scope of the invention, and it should be understood that these equivalent embodiments will be included herein. Furthermore, all references cited in this application are incorporated herein by reference in their entirety as if fully set forth herein.

[0163] Example

[0164] 1. Data Source

[0165] Model training primarily utilizes existing human sequencing data to ensure data relevance and biological significance. The data originates from sequencing data obtained during paired-end sequencing using the LabM sequencing platform, spanning different sequencing rounds and lengths of 46, 75, and 300 cycles to ensure coverage of sequencing data of varying lengths. To maintain data diversity, sequencing data from different sources are randomly downsampled.

[0166] 2. Feature mining

[0167] The candidate features considered by the training model in the pre- and mid-sequencing rounds are summarized in Table 1 below;

[0168] Table 1

[0169]

[0170] The candidate features considered by the training model in the pre- and mid-sequencing rounds are summarized in Table 2 below;

[0171] Table 2

[0172]

[0173]

[0174] 3. Model Training

[0175] In this embodiment, the specific steps for training the model and tuning its parameters are as follows:

[0176] (1) Dataset labeling:

[0177] - Bases in the dataset are labeled into two classes based on whether they are correctly identified: correctly identified bases (positive class samples) are labeled as 1, and incorrectly identified bases (negative class samples) are labeled as 0.

[0178] (2) Dataset partitioning:

[0179] The dataset is divided into a training set and a test set in a 7:3 ratio, meaning 70% of the data is used for training and 30% for testing.

[0180] - During the partitioning process, the ratio of positive to negative samples is kept consistent with the original dataset to ensure that the model can learn fairly on samples of different classes.

[0181] (3) Model parameter initialization:

[0182] - The LightGBM model was trained with the following initialization parameters:

[0183] -`num_leaves`:200 represents the maximum number of leaf nodes in a tree.

[0184] -`min_data_in_leaf`:200 represents the minimum number of samples in a leaf node.

[0185] -`subsample`:0.7 indicates the proportion of data sampled randomly in each iteration.

[0186] -`colsample_bytree`:0.8 represents the proportion of features randomly sampled when training each tree.

[0187] -`learning_rate`:0.05, represents the learning rate of the model.

[0188] -`tree_learner`:`voting` indicates the type of tree learner; here, the voting parallel tree learner is selected.

[0189] -`metric`:`auc`, represents the evaluation metric of the model, here we choose AUC (Area Under the Curve).

[0190] -`lambda_l1`:0.5 represents the L1 regularization parameter, used to prevent the model from overfitting.

[0191] -`lambda_l2`:0.1 represents the L2 regularization parameter, which is also used to prevent the model from overfitting.

[0192] -`num_boost_round`:500 indicates the number of boost rounds, i.e., the number of iterations for model training.

[0193] (4) Ranking of variable importance:

[0194] - During model training, evaluate the contribution of each feature to the model's predictive ability, i.e., rank the variables by importance.

[0195] This helps identify which features have the greatest impact on the model's predictions, allowing for further optimization and adjustments to the model.

[0196] According to embodiments of this application, feature importance is assessed by statistically analyzing the number of times each feature is used as a split point in the model. In the LightGBM decision tree construction process, the more times a feature is used as a split point, the greater its contribution to the model's predictive ability, and therefore the more important it is considered. Specifically, the feature importance percentage is calculated by dividing the number of times that feature is split by the total number of times all features are split.

[0197] The LightGBM model employs an ensemble learning approach, improving predictive performance by constructing multiple decision trees. During the construction of each tree, each node selects a feature for classification; this process is repeated until a complete tree is formed. These trees work together, with each contributing a portion to the final prediction. After model training is complete, the feature splitting information from all trees is aggregated to calculate feature importance.

[0198] This ensemble learning approach, through the combined action of multiple trees, not only enhances the model's accuracy and generalization ability but also increases model diversity by randomly selecting subsets of data and features to construct each tree, helping to reduce the risk of overfitting. In this way, LightGBM can effectively utilize the splitting information of features, thereby comprehensively evaluating the contribution of each feature to the model.

[0199] Through these steps, the model can learn key features that distinguish between positive and negative samples while maintaining its generalization ability, and its predictive performance can be validated on the test set. Ranking the importance of variables further provides a basis for model optimization, making the model more accurate and efficient.

[0200] The results of the variable importance ranking are shown in Figure 4 and 5 In, among them, Figure 4 This demonstrates the ranking of model variable importance during the early and middle cycles of training. Figure 5 The importance ranking of model variables during the training of the early, middle, and late cycles is shown.

[0201] 4. Model Evaluation

[0202] In this embodiment, the classification performance of the LightGBM model is evaluated using the following metrics:

[0203] Positive correlation analysis: First, examine whether there is a positive correlation between the predicted probability value (P) and the real Q value (Real Q). A positive correlation means that as the predicted probability value increases, the real Q value also tends to increase, indicating that the model can identify high-quality sequencing bases well.

[0204] Fitting the mapping function: By analyzing the curves showing the changes in predicted probability values ​​and true Q values, a mapping function is fitted that can convert predicted probability values ​​into predicted Q values ​​(PredictQ). This step is to establish a bridge from the model output to traditional Q-value evaluation criteria.

[0205] Difference analysis: On the test data, compare the difference between the true Q value and the predicted Q value obtained through the mapping function. Ideally, the curves showing the difference between the two should be close to a 45° diagonal line, indicating that the difference between the predicted Q value and the true Q value is small, and the model's prediction performance is good.

[0206] Performance evaluation: The classification performance of the model can be quantitatively evaluated by observing the distribution and difference between the actual Q-value and the predicted Q-value. If the predicted Q-value closely follows the changes in the actual Q-value, then the model is considered effective in evaluating the quality of base sequencing.

[0207] This evaluation method measures the performance of machine learning models from the perspective of traditional Q-scores, while also providing a way to combine machine learning model outputs with traditional bioinformatics evaluation standards. This approach allows for more accurate evaluation and optimization of model performance in specific application scenarios.

[0208] The above results show Figures 6-15 The following sections will explain each point.

[0209] 4.1 Evaluation of Model Prediction Performance During Pre- and Mid-cycle Training

[0210] 4.1.1 Relationship curve between predicted probability value (P) and true Q value (Real Q)

[0211] During model evaluation, the relationship between the predicted probability values ​​and the actual Q-values ​​was examined. For example... Figure 6As shown, the predicted probability value and the true Q value exhibit a positive correlation. This means that as the predicted probability value increases, the true Q value increases accordingly, reflecting the accuracy of the model's predictions. The determination of the "true Q value" uses the Q-value calculation method defined by Illumina and currently widely used. Specifically, the true Q value is calculated using the following formula:

[0212] Q = -10log10(E)

[0213] Here, E represents the estimated error probability of base detection. A higher Q value indicates a lower error rate; conversely, a lower Q value may lead to a significantly higher proportion of unusable reads and a higher false-positive variant detection rate, resulting in incorrect conclusions. This method of calculating the Q value provides a standardized metric for assessing the quality of sequencing data. In model training and evaluation, this standardized Q value is used as the "true Q value" to measure the accuracy of model predictions. Model performance can be evaluated by comparing the model's predicted Q value with the true Q value calculated using the above formula. Ideally, the model's predicted Q value should be as close as possible to the true Q value to ensure the reliability and accuracy of model predictions.

[0214] 4.1.2 Fitting the predicted probability value to the true Q value

[0215] In evaluating the model's predictive performance, the inventors conducted a fitting analysis between the predicted probability values ​​and the actual Q-values. Figure 7 This diagram illustrates the relationship between predicted probability values ​​(x-axis) and true Q-values ​​(y-axis). In the figure, the yellow curve represents the true Q-value (pre_real Q or predicted Q) derived from the predicted probability value (P) output by the model, while the blue curve represents the actual true Q-value (Real Q).

[0216] The specific form of the fitting function is: pre_real Q = 6.35498 + P × 0.73054 + P 2 ×(-0.01026)+P 3 ×0.00005

[0217] Where P represents the predicted probability value.

[0218] This fitting function converts the model's predicted probability values ​​into corresponding Q-values, thus aligning them with the traditional definition of Q-values. This conversion allows for direct comparison of the model's predicted Q-values ​​with the Q-values ​​of actual sequencing data, thereby assessing the model's accuracy and reliability.

[0219] like Figure 7As shown, the yellow fitted curve should closely resemble the blue actual curve, indicating a high degree of consistency between the model's predicted Q value and the actual Q value.

[0220] 4.1.3 Comparison of predicted Q-value with actual Q-value To further verify the accuracy of the model's predicted Q-value, the inventors conducted a comparative analysis between the predicted Q-value and the actual Q-value. Figure 8 The comparison results are shown, where the horizontal axis represents the predicted Q value given by the algorithm, and the vertical axis represents the true Q value obtained by calculating the error rate.

[0221] Table 3 below details the statistical differences between the predicted Q value and the actual Q value;

[0222] Table 3

[0223]

[0224]

[0225] from Figure 8 As can be seen from the data in the table above, the mean absolute difference between the predicted Q value and the real Q value is 0.98. This low mean absolute difference indicates a high degree of consistency between the predicted and real Q values, thus demonstrating the good predictive performance of the model.

[0226] 4.1.4 Distribution of Q-values ​​for correct and incorrect bases at false positive sites

[0227] In embodiments of this invention, the inventors employed a machine learning model-based method for identifying and correcting false positive sites. False positive sites in human gene sequences refer to non-mutated sites that, due to errors during sequencing, are frequently misidentified. Once a certain threshold is reached, these sites may be incorrectly identified as mutation sites. To address this issue, the inventors identified false positive sites by comparing different datasets. Specifically, mutation sites detected by the LabM sequencing platform were first labeled as dataset A, mutation sites obtained after a series of biometric analysis processes were labeled as dataset B, and mutation sites detected by the Illumina sequencing platform were labeled as dataset C. By comparing these datasets, mutation sites present in dataset A but not in dataset B, and mutation sites present in dataset A but not in dataset C, were identified. The intersection of these two sets of sites was defined as the high-probability false positive site. Figure 9The Q-value distribution of correct and incorrect bases at these false positive sites is presented. Analysis shows that correct bases (New Q correct bases) at false positive sites generally have higher Q-values, while incorrect bases (New Q wrong bases) generally exhibit lower Q-values. This finding indicates that the model of this invention can provide more accurate Q-value estimates for false positive sites, aiding in the correction of false positives. By assigning higher Q-values ​​to correctly identified bases and lower Q-values ​​to incorrectly identified bases, the model of this invention can reduce the occurrence of false positive sites in practical applications by filtering out low-Q-value bases. This method not only improves the accuracy of sequencing data but also enhances the reliability of the model in practical applications. Through this innovative method, the inventors can more effectively identify and correct false positive sites, providing a new quality control tool for gene sequencing data analysis.

[0228] 4.2 Evaluation of Model Prediction Performance During Pre-, Mid-, and Post-Cycle Training

[0229] 4.2.1 Relationship curve between predicted probability value and actual Q value

[0230] Figure 10 This shows the positive correlation between the predicted probability value and the actual Q value. For example... Figure 10 As shown, the true Q value increases with the increase of the predicted probability value, thus reflecting the accuracy of the model's prediction.

[0231] 4.2.2 Fitting the predicted probability value to the true Q value

[0232] Figure 11 This displays the fit between the predicted probability value and the true Q value. The yellow curve represents the true Q value (pre_real Q or predicted Q) obtained from the predicted probability value (P) output by the model, while the blue curve represents the actual true Q value. The specific form of the fitting function is: pre_realQ = 4.9722 + P × 0.6759 + P 2 ×(-0.01)+P 3 ×0.00006 where P represents the predicted probability value multiplied by 100.

[0233] 4.2.3 Comparison of predicted Q-value with actual Q-value

[0234] Figure 12 The comparison between the predicted Q-values ​​and the actual Q-values ​​is shown. The data shows that the differences between the predicted Q-values ​​and the actual Q-values ​​on both the training set (Humantrain_NewQ) and the test set (Humantest_NewQ) are small, as detailed in Table 4 below.

[0235] Table 4

[0236]

[0237] These results demonstrate that the model's predicted Q-values ​​are highly consistent with the actual Q-values ​​on different datasets, proving the stability and reliability of the model's prediction performance.

[0238] 4.2.4 Distribution of Q-values ​​for correct and incorrect bases at false positive sites

[0239] Figure 13 The distribution of Q-values ​​for correct and incorrect bases at false positive sites was further illustrated. The analysis showed that correct bases (New Q correct bases) at false positive sites generally had higher Q-values, while incorrect bases (New Q wrong bases) generally exhibited lower Q-values. This finding further confirms the model's effectiveness in correcting false positive sites.

[0240] 4.3 Comparison of the pre-mid model and the pre-mid-post model

[0241] 4.3.1 Comparison on ordinary human test data

[0242] Figure 14 The comparison of the prediction performance of the pre-mid model and the pre-mid-post model on ordinary human test data is presented. By comparing the difference between the predicted Q-values ​​and the actual Q-values ​​of the two models, the performance of the two models in terms of prediction accuracy can be observed (Table 5).

[0243] Table 5

[0244]

[0245]

[0246] As can be seen from the above charts, the predictive performance of the two methods on human sequencing data is not significantly different.

[0247] 4.3.2 Comparison at false positive sites in humans Figure 15 This study demonstrates the predictive performance of the pre-, mid-, and post-test models versus the pre-, mid-test model at human false positive sites. At these sites, the pre-, mid-, and post-test model exhibits a significant advantage. Specifically, the pre-, mid-, and post-test model assigns higher Q values ​​to correctly identified bases (test pairs) and lower Q values ​​to incorrectly identified bases (test errors). This discriminative ability is crucial for improving the accuracy of sequencing data. Figure 15 The analysis leads to the following conclusions:

[0248] Correctly identified bases (tested pairs): The pre-, mid-, and post-sequencing models assign higher Q values ​​to these bases, indicating that the model is able to more accurately identify those bases that were correctly identified during the sequencing process.

[0249] Misidentified bases (test errors): In contrast, the pre-, mid-, and post-analysis models assign lower Q values ​​to these bases, which helps to identify and filter out these misidentified bases that may cause false positives in subsequent analyses.

[0250] Model Advantages: The advantage of the pre-, mid-, and post-sequencing model in distinguishing correct and incorrect bases stems from its ability to comprehensively consider the characteristics of the current cycle and the cycles before and after it. This comprehensive consideration allows the model to more fully capture the dynamic changes during the sequencing process, including the effects of phasing and prephasing.

[0251] It should be noted that, in this application, the logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0252] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0253] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0254] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0255] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0256] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application without departing from the principles and spirit of this application.

Claims

1. A method for determining the quality score Q value of base detection, characterized in that, include: Receive input image data for determining the quality score Q value of base detection in the Nth sequencing reaction. The input image data includes the image of the template to be tested in the Mth sequencing reaction and the image of the template to be tested in the Nth sequencing reaction, where M and N are both natural numbers and M is less than N. Based on the input image data, predetermined prediction features are determined; Based on the predetermined prediction features, a trained machine learning model is used to determine the predicted probability value of correctly identifying the bases incorporated into the template to be tested in the Nth round of sequencing. as well as Based on the predicted probability value, the quality score Q value of base detection in the Nth round of sequencing reaction is determined.

2. The method according to claim 1, characterized in that, The input image data includes only images of the template to be tested from the Mth sequencing round and images of the template to be tested from the Nth sequencing round, and the predetermined prediction features include at least one of the following: At least one of the T base brightness, G base brightness, C base brightness, and A base brightness of the Nth sequencing reaction; At least one of the T base brightness, G base brightness, C base brightness, and A base brightness of the Mth sequencing reaction; The ratio of the brightness of the brightest base to the second brightest base in the Nth round of sequencing reaction; In the Nth round of sequencing, the proportion of the brightest base to the sum of the brightness of the four bases; In the Mth round of sequencing, the proportion of the brightest base to the sum of the brightness of the four bases; The ratio of the brightness of the brightest base to the second brightest base in the Mth round of sequencing reaction; The digital encoding result of the combination of base detection results from the Nth and Mth sequencing reactions; The digital encoding result of the base detection result of the Nth round of sequencing reaction; The digital encoding result of the base detection result of the Mth round of sequencing reaction; Optionally, NM = K1, where K1 is any natural number selected from 1 to 20; Optionally, K1 is any natural number selected from 1 to 10; Wherein, K1 equals 1; Optionally, the input image data further includes an image of the template to be tested in the Lth sequencing reaction, where L is a natural number greater than N; Optionally, the predetermined prediction feature includes at least one of the following: At least one of the T base brightness, G base brightness, C base brightness, and A base brightness of the Nth sequencing reaction; At least one of the T base brightness, G base brightness, C base brightness, and A base brightness of the Lth sequencing reaction; At least one of the T base brightness, G base brightness, C base brightness, and A base brightness of the Mth sequencing reaction; The ratio of the brightness of the brightest base to the second brightest base in the Nth round of sequencing reaction; In the Nth round of sequencing, the proportion of the brightest base to the sum of the brightness of the four bases; In the Lth round of sequencing, the ratio of the brightest base to the sum of the brightness of the four bases; In the Mth round of sequencing, the proportion of the brightest base to the sum of the brightness of the four bases; The ratio of the brightness of the brightest base to the second brightest base in the Lth round of sequencing reaction; In the Lth round of sequencing, the ratio of the brightest base to the sum of the brightness of the four bases; The ratio of the brightness of the brightest base to the second brightest base in the Lth round of sequencing reaction; The combination of base detection results from the Nth, Lth, and Mth sequencing reactions; The brightness of the brightest base in the Nth round of sequencing reaction; The brightness of the brightest base in the Mth round of sequencing reaction; The brightness of the lowest-brightness base in the Nth round of sequencing reaction; and The brightness of the base with the lowest brightness in the Mth round of sequencing reaction; Optionally, LN = K2, where K2 is any natural number selected from 1 to 20; Optionally, K2 is any natural number selected from 1 to 10; Wherein, K2 equals 1; Optionally, the quality score Q value is determined based on the predicted probability value using a pre-constructed fitting formula; Optionally, the fitting formula is a cubic polynomial in one variable with the predicted probability value as the variable; Optionally, the machine learning model is selected from at least one of decision trees, logistic regression, Naive Bayes, and neural networks; Optionally, the machine learning model is selected from decision trees; Optionally, the machine learning model is selected from lightGBM.

3. The method according to claim 1 or 2, characterized in that, The trained machine learning model was obtained through the following steps: The input image data of positive sample sequencing reaction and negative sample sequencing reaction are obtained, wherein the base detection of the positive sample is known to be a predetermined baseline true value, and the base detection of the negative sample is known not to be a predetermined baseline true value. Based on the input image data, predetermined predictive features of the positive sample sequencing response and the negative sample sequencing response are determined; The predetermined predictive features are input into the machine learning model, and the known base detection results of the positive and negative samples are used as labels for supervised training in order to obtain the trained machine learning model. Optionally, the positive and negative samples are selected by randomly downsampling sequencing data from multiple sources; Optionally, the predetermined prediction features are obtained by iteratively optimizing the initial prediction features by repeating the following steps at least once: The machine learning model is trained using the initial prediction features, and the importance score of the initial prediction features is determined. and Based on the importance score, a subset of the initial predicted features is selected as an optimized feature combination; Optionally, the template to be tested is an amplification product containing multiple identical nucleic acid molecules; Optionally, the template to be tested is a cluster or nanospheres; Optionally, the base detection in the Nth round of sequencing is the detection of the base most likely to be incorporated into the template to be tested in the Nth round of sequencing.

4. A sequencing method, characterized in that, include: Synthesis and sequencing of nucleic acid molecules to be tested ； The method according to any one of claims 1 to 3 determines the quality score Q value of base detection in at least one round of sequencing reaction in sequencing-by-synthesis.

5. The method according to claim 4, characterized in that, The base detection in the Nth round of sequencing is the base that is most likely to be incorporated into the template to be tested in the Nth round of sequencing.

6. An apparatus for determining the quality score Q value of base detection, characterized in that, include: An image data receiving unit is used to receive input image data for determining the quality score Q value of base detection in the Nth round of sequencing reaction. The input image data includes the image of the template to be tested in the Mth round of sequencing reaction and the image of the template to be tested in the Nth round of sequencing reaction, where M and N are both natural numbers and M is less than N. The prediction feature determination unit is used to determine predetermined prediction features based on the input image data. The prediction unit is used to determine, based on the predetermined prediction features and using a trained machine learning model, the actual bases incorporated into the template to be tested in the Nth sequencing reaction that match the predicted probability value of base detection. as well as The Q-value determination unit is used to determine the quality score Q-value of base detection in the Nth round of sequencing based on the predicted probability value. Optionally, the template to be tested is an amplification product containing multiple identical nucleic acid molecules; Optionally, the template to be tested is a cluster or nanospheres; Optionally, the base detection in the Nth round of sequencing is the detection of the base most likely to be incorporated into the template to be tested in the Nth round of sequencing.

7. The device according to claim 6, characterized in that, The input image data includes only images of the template to be tested from the Mth sequencing round and images of the template to be tested from the Nth sequencing round, and the predetermined prediction features include at least one of the following: At least one of the T base brightness, G base brightness, C base brightness, and A base brightness of the Nth sequencing reaction; At least one of the T base brightness, G base brightness, C base brightness, and A base brightness of the Mth sequencing reaction; The ratio of the brightness of the brightest base to the second brightest base in the Nth round of sequencing reaction; In the Nth round of sequencing, the proportion of the brightest base to the sum of the brightness of the four bases; In the Mth round of sequencing, the proportion of the brightest base to the sum of the brightness of the four bases; The ratio of the brightness of the brightest base to the second brightest base in the Mth round of sequencing reaction; The digital encoding result of the combination of base detection results from the Nth and Mth sequencing reactions; The digital encoding result of the base detection result of the Nth round of sequencing reaction; The digital encoding result of the base detection result of the Mth round of sequencing reaction; Optionally, NM = K1, where K1 is any natural number selected from 1 to 20; Optionally, K1 is any natural number selected from 1 to 10; Wherein, K1 equals 1; Optionally, the input image data further includes an image of the template to be tested in the Lth sequencing reaction, where L is a natural number greater than N; Optionally, the predetermined prediction feature includes at least one of the following: At least one of the T base brightness, G base brightness, C base brightness, and A base brightness of the Nth sequencing reaction; At least one of the T base brightness, G base brightness, C base brightness, and A base brightness of the Lth sequencing reaction; At least one of the T base brightness, G base brightness, C base brightness, and A base brightness of the Mth sequencing reaction; The ratio of the brightness of the brightest base to the second brightest base in the Nth round of sequencing reaction; In the Nth round of sequencing, the proportion of the brightest base to the sum of the brightness of the four bases; In the Lth round of sequencing, the ratio of the brightest base to the sum of the brightness of the four bases; In the Mth round of sequencing, the proportion of the brightest base to the sum of the brightness of the four bases; The ratio of the brightness of the brightest base to the second brightest base in the Lth round of sequencing reaction; In the Lth round of sequencing, the ratio of the brightest base to the sum of the brightness of the four bases; The ratio of the brightness of the brightest base to the second brightest base in the Lth round of sequencing reaction; The combination of base detection results from the Nth, Lth, and Mth sequencing reactions; The brightness of the brightest base in the Nth round of sequencing reaction; The brightness of the brightest base in the Mth round of sequencing reaction; The brightness of the lowest-brightness base in the Nth round of sequencing reaction; and The brightness of the base with the lowest brightness in the Mth round of sequencing reaction; Optionally, LN = K2, where K2 is any natural number selected from 1 to 20; Optionally, K2 is any natural number selected from 1 to 10; Wherein, K2 equals 1; Optionally, the quality score Q value is determined based on the predicted probability value using a pre-constructed fitting formula; Optionally, the fitting formula is a cubic polynomial in one variable with the predicted probability value as the variable; Optionally, the machine learning model is selected from at least one of decision trees, logistic regression, Naive Bayes, and neural networks; Optionally, the machine learning model is selected from decision trees; Optionally, the machine learning model is selected from lightGBM; Optionally, it further includes a training unit for obtaining the trained machine learning model through the following steps: The input image data for positive sample sequencing reaction and negative sample sequencing reaction are obtained. The base detection of the positive sample is known to be a predetermined baseline true value, and the base detection of the negative sample is known not to be a predetermined baseline true value. Optionally, the positive sample sequencing reaction and negative sample sequencing reaction are selected by randomly downsampling sequencing data from multiple sources. Based on the input image data, predetermined predictive features of the positive sample sequencing response and the negative sample sequencing response are determined; The predetermined predictive features are input into the machine learning model, and the known base detection results of the positive and negative samples are used as labels for supervised training in order to obtain the trained machine learning model. Optionally, it further includes a predictive feature selection unit for iteratively optimizing the predetermined predictive features from the initial predictive features by repeating at least one round of the following steps: The machine learning model is trained using the initial prediction features, and the importance score of the initial prediction features is determined. and Based on the importance score, a subset of the initial predicted features is selected as an optimized feature combination.

8. A computing device, characterized in that, include: Processor and memory; The memory is used to store computer programs; The processor is configured to execute the computer program to implement the method as described in any one of claims 1 to 5.

9. A computer-readable storage medium, characterized in that, The storage medium includes computer instructions that, when executed by a computer, cause the computer to perform the method as described in any one of claims 1 to 5.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 3 or 4 to 5.

Images