Deep learning-based gene sequencing base quality assessment method, product, device, and medium

Abstract

The present application provides a deep learning-based gene sequencing base quality assessment method, a product, a device, and a medium. The method comprises: taking fluorescence image data to be analyzed comprising base signal acquisition units as an input of a base recognition model trained on the basis of a deep learning model, and obtaining input data of an output layer of the base recognition model, wherein the input data comprises multi-channel feature data of a plurality of channels respectively corresponding to a plurality of base categories; on the basis of the input data, performing normalization processing on the multi-channel feature data respectively corresponding to the plurality of base signal acquisition units to obtain corresponding standardized data, and establishing a quality assessment table on the basis of the standardized data of the plurality of base signal acquisition units and the statistics of whether the base categories are correct; and in the gene sequencing process, for each base signal acquisition unit, predicting a corresponding base category on the basis of the multi-channel feature data, and on the basis of the quality assessment table, determining a quality assessment value corresponding to the base category.

Description

Gene sequencing base quality assessment methods, products, equipment and media based on deep learning

[0001] The present invention claims priority to Chinese patent application No. 202410160733.5, filed with the Patent Office of China on February 5, 2024, entitled “Methods, products, equipment and media for gene sequencing base quality assessment based on deep learning”, the entire contents of which are incorporated herein by reference. Technical Field

[0002] The present application relates to the field of gene sequencing technology, and in particular to a gene sequencing base quality assessment method based on deep learning, a computer program product, a gene sequencing device, and a computer-readable storage medium. Background Art

[0003] Gene sequencing refers to the process of identifying and analyzing genetic samples at the microscopic level through certain technical means.

[0004] A second-generation sequencer is a widely used instrument for genome sequencing, capable of quickly and accurately determining DNA sequences. The entire sequencing process primarily involves acquiring a fluorescent image of the sample through an optical system, performing a basecalling process, and obtaining and evaluating the sequencing results. The basecalling process (base recognition algorithm) uses the brightness information of the sample fluorescence image generated by laser light (sequencing signals corresponding to different base classes) to determine whether the sample belongs to one of the four base classes: A, C, G, or T. After basecalling, the sample information is converted from a digital image into sequence information, representing the sample's DNA sequence, for subsequent analysis and evaluation.

[0005] The accuracy and quality of base sequence sequencing by base recognition algorithms are crucial. Due to its powerful fitting capabilities, deep learning has been widely used in many studies to apply deep learning to base recognition in second-generation sequencers. In their research, the inventors of this application analyzed the currently known methods of applying deep learning to gene sequencing as follows: 1) inputting an image into a deep learning model, and then outputting base information; 2) inputting uncorrected fluorescence brightness information (fluorescence brightness extracted from an image, without removing optical interference, biochemical interference, and other interference factors) into a deep learning model, and then outputting base information; 3) inputting corrected fluorescence brightness information (fluorescence brightness removed from optical, biochemical, and other factors) into a deep learning model, and then outputting base information. Deep learning solutions can greatly improve the accuracy of base recognition.

[0006] However, due to the uninterpretability of the processing of deep learning models, and the fact that base quality (accuracy of base categories) in gene sequencing is important information and is of great significance to the subsequent bioinformatics analysis process, the uninterpretable gene sequencing results will greatly hinder the promotion and application of deep learning models in the field of gene sequencing. Summary of the Invention

[0007] The present application provides a deep learning-based gene sequencing base quality assessment method, computer program product, gene sequencing device and computer-readable storage medium that can provide a quality assessment system for deep learning gene sequencing results and has higher accuracy in gene sequencing quality assessment.

[0008] In a first aspect of the embodiments of the present application, a method for evaluating base quality in gene sequencing based on deep learning is provided, comprising:

[0009] Using the fluorescence image data to be measured including the base signal acquisition unit as input to a base recognition model trained based on a deep learning model, and obtaining input data for an output layer of the base recognition model; the input data includes multi-channel feature data of multiple channels corresponding to multiple base categories;

[0010] According to the input data, based on the multi-channel feature data corresponding to the plurality of base signal acquisition units, corresponding standardized data is obtained by normalization processing, and a quality assessment table is established based on the standardized data of the plurality of base signal acquisition units and statistics on whether the base categories are correct;

[0011] In the gene sequencing process, for each base signal acquisition unit, the corresponding base category is predicted according to the multi-channel feature data, and the quality assessment value corresponding to the base category is determined based on the quality assessment table.

[0012] In a second aspect, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the steps of the deep learning-based gene sequencing base quality assessment method as described in any embodiment of the present application.

[0013] In a third aspect, a gene sequencing device is provided, comprising a processor and a memory connected to the processor, wherein the memory stores a computer program executable by the processor, and when the computer program is executed by the processor, the steps of the deep learning-based gene sequencing base quality assessment method as described in any embodiment of the present application are implemented.

[0014] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the deep learning-based gene sequencing base quality assessment method described in any embodiment of the present application are implemented.

[0015] In the above embodiment, by obtaining the input data of the output layer of the base recognition model, the multi-channel feature data of multiple channels input to the output layer is normalized to obtain corresponding standardized data, and a quality assessment table is established based on the standardized data of multiple base signal acquisition units and the statistics of whether the base category is correct. In the subsequent gene sequencing process, the quality assessment value corresponding to the base category can be determined based on the quality assessment table for the predicted result of the base category of each base signal acquisition unit. In this way, the data used to construct the quality assessment table is the original data before being processed by the output layer of the base recognition model. After normalization, the quality assessment table is established based on the correct situation of the base category. This can improve the accuracy of the evaluation of the quality of the base category prediction. In the subsequent gene sequencing process, the base category of each base signal acquisition unit is predicted by the base recognition model to obtain a prediction result. At the same time, the quality assessment value of the current prediction result can be judged by comparing it with the quality assessment table. The quality assessment value reflects the reliability of the current base category prediction result, making the processing process of the deep learning-based base recognition model more visual and interpretable.

[0016] In the above embodiments, the computer program product, gene sequencing device, and computer-readable storage medium belong to the same concept as the corresponding gene sequencing base quality assessment method embodiment based on deep learning, and thus have the same technical effects as the corresponding gene sequencing base quality assessment method embodiment based on deep learning, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] FIG1 is a schematic diagram of a process for evaluating base quality in gene sequencing based on deep learning in one embodiment;

[0018] FIG2 is a schematic diagram of the principle of the output layer of the base recognition model in one embodiment;

[0019] FIG3 is a schematic diagram showing the principle of a method for assessing base quality in gene sequencing based on deep learning in one embodiment;

[0020] FIG4 is a schematic diagram of a flow chart of a method for evaluating gene sequencing base quality based on deep learning in an optional specific example;

[0021] FIG5 is a schematic diagram of the structure of a gene sequencing base quality assessment device based on deep learning in one embodiment;

[0022] FIG6 is a schematic diagram of the structure of a gene sequencing device in one embodiment. DETAILED DESCRIPTION

[0023] The technical solution of the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] In order to make the purpose, technical solutions and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limiting this application. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0025] In the following description, the expression "some embodiments" is involved, which describes a subset of all possible embodiments. It should be noted that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict.

[0026] In the following description, the terms "first, second, and third" are merely used to distinguish similar objects and do not represent a specific ordering of the objects. It is understandable that "first, second, and third" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0027] In the field of gene sequencing technology, based on the powerful fitting ability of deep learning models, more and more studies are applying deep learning to base recognition in gene sequencers, using various deep learning models to train base recognition models to predict base categories. In order to solve the application obstacles brought about by the unexplainable nature of the processing process of deep learning models, the inventors of this application have summarized and analyzed the currently known methods of applying deep learning to gene sequencing, and mainly divided the input of base classification of base recognition models into the following three categories: fluorescence images collected in real time during the sequencing process, fluorescence brightness information extracted from fluorescence images (fluorescence brightness information of uncorrected fluorescence images), and fluorescence brightness information extracted after preprocessing the fluorescence images (fluorescence brightness information of corrected fluorescence images). These different types of inputs to the base recognition model are essentially based on the fluorescence brightness information of the base signal acquisition unit contained in the fluorescence image collected in real time during the sequencing process. The base recognition model identifies the base category of each base signal acquisition unit in different sequencing cycles through feature extraction and classification prediction. In the embodiment of the present application, the different types of inputs to the base recognition model: fluorescence image, fluorescence brightness information of uncorrected fluorescence image, and fluorescence brightness information of corrected fluorescence image are all classified within the range of the fluorescence image data to be tested that includes the base signal acquisition unit.

[0028] On this basis, the inventors of this application conducted the following research and analysis during the study of how to make the processing of the base recognition model explainable: the main functions of the output layer of the base recognition model are: first, to map the output value to between 0 and 1; second, to ensure that the sum of the output values ​​of multiple channels corresponding to the base category is 1; and third, to facilitate the derivative calculation during the gradient descent process. In this way, the exponential operation introduced by the output layer may distort the relative relationship between the data of multiple channels, and the exponential operation will amplify the distance between the output values ​​of multiple channels.

[0029] For ease of understanding, let's use a softmax output layer as an example. The softmax layer predicts base classification by calculating the relative relationships between the four base classes: A, T, C, and G. This relative value eliminates absolute values, resulting in a loss of key information and inaccurate base quality assessment using the output of the base recognition model. For example, consider two base clusters, cluster A and cluster B. For various reasons, cluster A exhibits an inadequate response, resulting in low brightness across all channels. Before the softmax layer, the four values ​​for cluster A are: 0.054, -2.44, -2.44, -2.44. After the softmax layer, the corresponding values ​​for these four channels are: 0.802, 0.065, 0.065, 0.065. Cluster B exhibits normal brightness. Before the softmax layer, the four values ​​for cluster B are: 2, -0.5, -0.5, -0.5. After the softmax layer, the corresponding values ​​for these four channels are: 0.802, 0.065, 0.065, 0.065. From this, we can see that the base classification prediction outputs of clusterA and clusterB by the base recognition model are essentially identical. If the base quality scores are evaluated based on the base recognition model output, the quality assessments of clusterA and clusterB will also be similar. However, because clusterA's response is insufficient, the signal is weak, the signal-to-noise ratio is low, and it is easily affected by other interference factors, the accuracy of clusterA is inevitably lower than that of clusterB. As a result, using the output of the softmax layer of the base recognition model to evaluate base quality is inaccurate. Similarly, other methods that use the output of the base recognition model's output layer as a basis for base quality assessment also suffer from similar issues to some extent.

[0030] Based on this, the inventors of this application proposed a technical idea of ​​constructing a quality value evaluation system based on the input data of the output layer of the base recognition model. The feature data representing the base category extracted by the base recognition model before the output layer is used as the basis for base quality evaluation, and a quality evaluation table is established to guide the quality evaluation value of the base category in the subsequent gene sequencing process. This not only makes the processing process of the subsequent gene sequencing process visual and more interpretable, but also improves the accuracy of the base quality evaluation.

[0031] Please refer to FIG1 , which shows a method for assessing base quality of gene sequencing based on deep learning according to an embodiment of the present application, comprising the following steps:

[0032] S101, using the fluorescent image data to be tested including the base signal acquisition unit as the input of the base recognition model trained based on the deep learning model, and obtaining the input data of the output layer of the base recognition model; the input data includes multi-channel feature data of multiple channels corresponding to multiple base categories.

[0033] The gene sequencing process is to amplify the DNA molecules to be tested through specific amplification technology, amplify each DNA fragment (single-stranded library molecule) to form a base cluster, use the base cluster detection results to construct the template points of the base cluster on the sequencing chip, and then perform base recognition and other operations based on the base cluster template points to obtain the base sequence. It can be understood that the deep learning-based gene sequencing base quality assessment method provided in the embodiment of the present application uses a deep learning model training strategy to improve the base recognition accuracy. The input of the base recognition model is based on the fluorescence image obtained after the amplification of the single-stranded library molecules on the sequencing chip to perform base cluster positioning detection and base category identification. Here, each base cluster refers to a base signal acquisition unit, so it is not limited to which amplification technology is used for the single-stranded library molecules. That is, the base type recognition of the base signal acquisition unit of the sequencing chip in other massively parallel sequencing technologies is also applicable. For example, the base signal acquisition unit can refer to the base cluster obtained by the bridge amplification technology in the Illumina sequencing technology, and also includes the nanospheres obtained by the rolling circle amplification technology (RCA), etc. This application does not impose any restrictions on this.

[0034] It should be noted that the fluorescence image data to be tested that includes the base signal acquisition unit can be directly a fluorescence image, or it can be the fluorescence brightness information corresponding to the base signal acquisition unit obtained based on the fluorescence image extraction. These fluorescence brightness information come directly or indirectly from the fluorescence image, and thus also fall within the scope of the fluorescence image data to be tested.

[0035] The output layer of a base calling model is the final layer of a neural network model (also known as a deep learning model), responsible for processing the input and outputting the results. Typically, a neural network model includes an input layer (the first layer) and a hidden layer connected between the input and output layers. The input layer interacts with the outside world to obtain input data and converts it into a format that the neural network can process, such as a data vector. The hidden layer processes the input data and converts it into a higher-level feature representation. The design of the output layer needs to be tailored to the specific task. The output is usually the neural network's prediction or classification of the input data. For example, in classification tasks, the output layer typically uses a softmax function to calculate the probability distribution of each class. In regression tasks, the output layer typically has only one neuron, whose output value represents the predicted value for the input data. The error between the output of the output layer and the true value is the neural network's loss function, and the goal of deep learning model training is to minimize this loss function. Therefore, the input data for the output layer of a base calling model refers to the feature data obtained by the hidden layer through processing and conversion of the fluorescent image data to be detected, including the base signal acquisition unit, and is input to the output layer as the basis for class probability prediction.

[0036] S103, according to the input data, based on the normalization processing of the multi-channel feature data corresponding to the multiple base signal acquisition units, corresponding standardized data are obtained, and a quality assessment table is established based on the statistics of the standardized data of the multiple base signal acquisition units and whether the base categories are correct.

[0037] Multiple channels correspond to multiple base categories. For example, if the base categories are A, T, C, and G, there are four channels. For each base signal acquisition unit, the probability of the current base signal acquisition unit belonging to the four base categories A, T, C, and G is determined based on the input data of the output layer, that is, based on the feature data output by the four channels corresponding to the four base categories A, T, C, and G, respectively.

[0038] It should be noted that input data refers to the data before being input into the output layer of the neural network model. The corresponding standardized data is obtained by normalizing the multi-channel feature data corresponding to the multiple base signal acquisition units. Here, it means decoupling the normalization of the input data from the normalization in the output layer of the neural network model, and using a preset normalization method independent of the output layer of the neural network model to process the input data to obtain standardized data. Therefore, the normalization processing of the input data is no longer limited to the method in which the output layer of the neural network model needs to perform normalization processing on the four channel feature data of each single base signal acquisition unit, but can be performed by simultaneously obtaining the multi-channel feature data corresponding to a certain number of base signal acquisition units to avoid the normalization results of the multi-channel feature data of the base signal acquisition units with insufficient reactions affecting the accuracy of the overall standardized data.

[0039] Among them, the characteristic data of the four channels corresponding to each base signal acquisition unit is a multi-channel characteristic data. Normalization processing is performed on each multi-channel characteristic data to obtain the corresponding standardized data. First, the standardized data for each base signal acquisition unit is obtained based on the channel characteristic data after normalization processing. Secondly, the characteristic data of these channels are also used as the basis for determining the base category to which the current base signal acquisition unit belongs. The quality assessment table is established by analyzing the distribution of the normalized data of multiple base signal acquisition units after normalization processing, and combining the results of whether the base categories of these base signal acquisition units are correct. In this way, the quality assessment table is constructed by utilizing the distribution of the standardized data of a certain amount of base signal acquisition units and the accuracy under different distribution conditions.

[0040] S105 , in the gene sequencing process, for each base signal acquisition unit, predict the corresponding base category according to the multi-channel feature data, and determine the quality evaluation value corresponding to the base category based on the quality evaluation table.

[0041] After establishing a quality assessment table based on the statistics of a certain number of base signal acquisition units, in the gene sequencing process, the quality assessment value of each base signal acquisition unit can be determined accordingly according to the quality assessment table by determining the distribution of the standardized data of each base signal acquisition unit.

[0042] In the above embodiment, by obtaining the input data of the output layer of the base recognition model, the multi-channel feature data of multiple channels input to the output layer is normalized to obtain corresponding standardized data, and then a quality assessment table is established based on the standardized data of multiple base signal acquisition units and the statistics of whether the base category is correct. In the subsequent gene sequencing process, for the prediction result of the base category of each base signal acquisition unit, the quality assessment value corresponding to the base category can be determined based on the quality assessment table. In this way, the data used to construct the quality assessment table is the original data before being processed by the output layer of the base recognition model. After normalization, the quality assessment table is established in combination with the correctness of the base category to improve the accuracy of the base category prediction quality assessment. In the subsequent gene sequencing process, the base category of each base signal acquisition unit is predicted by the base recognition model to obtain a prediction result. At the same time, the quality assessment value of the current prediction result can be judged by comparing it with the quality assessment table. The quality assessment value reflects the reliability of the current base category prediction result, making the processing process of the deep learning-based base recognition model more visual and interpretable.

[0043] In some embodiments, step S103 includes:

[0044] According to the input data, corresponding standardized data is obtained by normalizing the multi-channel feature data corresponding to the plurality of base signal acquisition units;

[0045] Based on whether the standardized data of the multiple base signal acquisition units correspond to the correct quality assessment intervals and base categories, the base category accuracy rates corresponding to different quality assessment intervals are counted, and a quality assessment table containing the correspondence between quality assessment intervals and base category accuracy rates is established.

[0046] The multi-channel feature data corresponding to each base signal acquisition unit is normalized to generate corresponding standardized data, which can be considered as probability parameters representing the probability that the current base signal acquisition unit belongs to different base categories. When establishing a quality index table, multiple quality assessment intervals can be defined. By statistically analyzing the quality assessment intervals to which the standardized data of a certain amount of base signal acquisition units corresponds and the accuracy of the base category predicted based on the multi-channel feature data, the base category accuracy corresponding to each quality assessment interval can be obtained.

[0047] In the above embodiment, the quality assessment table is set to include the correspondence between quality assessment intervals and base category accuracy, and the number of quality assessment intervals can be used to form a corresponding number of assessment levels, which not only simplifies the logic of establishing the quality assessment table, but also facilitates obtaining more intuitive and easy-to-visualize quality assessment results.

[0048] In some embodiments, the standardized data based on the plurality of base signal acquisition units respectively correspond to whether the quality assessment intervals and base categories to which they belong are correct, and the statistical calculation of the base category accuracy rates corresponding to different quality assessment intervals includes:

[0049] Determine a plurality of numerical intervals based on the value range from 0 to 1 and divided into a plurality of numerical intervals according to a preset interval, and divide the quality assessment intervals based on the combination of the numerical intervals;

[0050] For each of the standardized data corresponding to the base signal acquisition unit, determining the quality assessment interval to which the standardized data corresponds, and determining whether the predicted base category of the base signal acquisition unit is correct;

[0051] According to the statistical results of the multiple base signal acquisition units, the base category accuracy rates corresponding to the different quality assessment intervals are obtained.

[0052] A numerical interval refers to a numerical interval that can correspond to the size of the standardized data, and a combination of numerical intervals refers to a combination of multiple standardized data falling into different numerical intervals. In this embodiment, the preset interval is 0.1, and 0 to 1 is divided into ten numerical intervals (Q1, Q2, Q3, Q4, Q5, Q6, Q7, Q8, Q9, Q10) with an interval of 0.1. Based on the combination of the ten numerical intervals (such as Q1 Q2, Q1 Q3...), a 10×10 quality assessment interval is obtained. For each base signal acquisition unit, the input data of the output layer is the feature data of the corresponding four channels. The feature data of the four channels are normalized to obtain a set of standardized data (NO_A, NO_C, NO_G, NO_T) corresponding to the four channels. According to the two numerical intervals (such as Q2 and Q10) in which the maximum value (NO_Max) and the minimum value (NO_Min) in a set of standardized data fall, the current base signal acquisition unit is classified into the corresponding quality assessment interval. Then, the accuracy of the base signal acquisition unit in each quality assessment interval is used as the guidance accuracy of the category to obtain the corresponding 10×10 quality assessment table.

[0053] In the above embodiment, the quality assessment intervals in the quality assessment table are set based on the distribution of the numerical intervals to which the standardized data corresponding to the base signal acquisition units belong, and the accuracy of the base category prediction results in the quality assessment intervals corresponding to these base signal acquisition units are counted, so that a quality assessment table for base category quality guidance that can cover more possible data range changes can be established, so that a more accurate quality assessment value can be obtained according to the range of the characteristic data of each base signal acquisition unit.

[0054] In some embodiments, in the base recognition model, the output layer is a softmax layer, and the base recognition model includes an input layer, a hidden layer and the output layer; wherein the input data includes the multi-channel feature data output by the hidden layer based on feature extraction of the fluorescent image data to be tested, and the determined logarithmic probability corresponding to each base category to which the base signal acquisition unit in the fluorescent image data to be tested belongs.

[0055] Please refer to Figure 2. The output layer is a softmax layer. The input data of the output layer are the raw scores of the four channels output by the base recognition model, indicating that the base signal acquisition unit belongs to the four base categories. The output layer calculates based on the raw scores of the four channels and converts them into probability distribution outputs indicating that the base signal acquisition unit belongs to the four base categories. Specifically, the hidden layer processes and converts the feature data obtained by feature extraction from the input of the base recognition model to determine the logarithmic probability (Z_A, Z_C, Z_G, Z_T) that the currently extracted feature data represents the corresponding base signal acquisition unit belonging to the four base categories. Based on the input logarithmic probability data, the softmax layer outputs the probability values ​​(Y_A, Y_C, Y_G, Y_T) of the base signal acquisition unit belonging to the four base categories.

[0056] In the above embodiments, the softmax layer is usually used as the output layer in the base recognition model. The input data of the softmax layer is used to establish a quality evaluation system that serves as a guide for the subsequent quality evaluation of the prediction results of the base category, which is conducive to improving the accuracy of the quality evaluation. Compared with the application method of using the softmax output value to establish the quality evaluation system, it can also effectively reduce the consumption of computing resources. The correspondence between the output value of the output layer of the base recognition model and the accuracy is a relatively complex nonlinear mapping equation. In the application of gene sequencing, each base sequence is subjected to such a transformation, which consumes a lot of speed and computing resources.

[0057] In some embodiments, in step S103, the step of obtaining corresponding standardized data based on the input data and the multi-channel feature data corresponding to each of the plurality of base signal acquisition units through normalization processing includes:

[0058] According to the base recognition model, a standard fluorescence image is used as the fluorescence image data to be tested as the input to obtain the input data of the output layer of the base recognition model. Based on the multi-channel feature data corresponding to the multiple base signal acquisition units in the input data, the corresponding normalization parameters are determined through statistical analysis using a preset normalization method. Based on the normalization parameters, the multi-channel feature data of each base signal acquisition unit are processed respectively to obtain standardized data corresponding to each channel.

[0059] Among them, the input of the base recognition model is a fluorescent image. A standard fluorescent image refers to a fluorescent image of the base categories corresponding to each base signal acquisition unit that is known to be contained therein. In an optional example, the standard fluorescent image can be determined by calculating the position of each base signal acquisition unit in the fluorescent image and the base category of the base signal acquisition unit at the corresponding position using a traditional algorithm. To establish a quality assessment table, a certain amount of predicted results of the base categories of base signal acquisition units are required as statistical objects. Here, a certain amount of base signal acquisition units can be base signal acquisition units from the same standard fluorescent image, or from multiple standard fluorescent images. In the gene sequencing process, the identification of a base category for a base signal acquisition unit can be called a cycle, and an input data of the output layer corresponds to a four-channel feature data composed of the four channel outputs in a cycle. The value of the standardized data is related to the type of normalization parameter used to perform normalization processing on the multi-channel feature data, that is, it corresponds to the normalization method adopted. In this embodiment, the normalization parameter used to obtain the standardized data may refer to a set of preset normalization parameter values ​​of a specified type obtained by statistically analyzing the base signal acquisition unit contained in the standard fluorescence image.

[0060] In the above embodiment, an implementation method is provided for establishing a quality assessment table by using a base recognition model to predict the base category of the base signal acquisition unit contained in a standard fluorescence image. The standard fluorescence image can be repeatedly used for different base recognition models to provide guidance for establishing a quality assessment table, which is more illustrative of the accuracy of the base quality assessment of the base recognition model.

[0061] In some embodiments, in step S103, the step of obtaining corresponding standardized data based on the input data and the multi-channel feature data corresponding to each of the plurality of base signal acquisition units through normalization processing includes:

[0062] According to the base recognition model, the real-time fluorescence image to be tested obtained by the current sequencing is used as input to obtain the input data of the output layer of the base recognition model. Based on the multi-channel feature data corresponding to the multiple base signal acquisition units in the input data, the corresponding normalization parameters are determined through statistical analysis using a preset normalization method. Based on the normalization parameters, the multi-channel feature data of each base signal acquisition unit are processed respectively to obtain standardized data corresponding to each channel.

[0063] The input of the base recognition model is a fluorescence image. In the process of establishing a quality evaluation table, the input of the base recognition model is the real-time fluorescence image to be tested obtained by the current sequencing. To establish a quality evaluation table, the prediction results of the base categories of a certain number of base signal acquisition units are required as statistical objects. Here, the certain number of base signal acquisition units can be base signal acquisition units from the same real-time fluorescence image to be tested, or can be from multiple real-time fluorescence images to be tested. In the implementation scheme of establishing a quality evaluation table using the data of the base signal acquisition units contained in the real-time fluorescence image to be tested, a traditional algorithm can be used to calculate and determine the position of each base signal acquisition unit in the real-time fluorescence image to be tested and the base category of the base signal acquisition unit at the corresponding position. The base category result obtained by the traditional algorithm is used to judge whether the prediction result of the base category output by the base recognition model is correct, so that the accuracy of the prediction results of the base category of a certain number of base signal acquisition units can be used to establish the quality evaluation table.

[0064] In the above embodiment, an implementation method is provided for establishing a quality assessment table by using a base recognition model to predict the base category of the base signal acquisition unit contained in the real-time fluorescent image to be tested. The real-time fluorescent image to be tested can include the characteristics of the current sequencing device and the current sequencing environment, and the quality assessment table established by the fluorescent image is obtained accordingly. The guidance on the sequencing results of the base category of the base signal acquisition unit in the fluorescent image collected in the current sequencing process can be more targeted.

[0065] Specifically, for each sequencing process, the base recognition model uses the real-time fluorescence images to be tested, collected in the previous cycle or cycles, to establish a quality assessment table. On the one hand, when the base recognition model performs base classification processing on these real-time fluorescence images to be tested, the input data of the output layer of the base recognition model is normalized to obtain corresponding standardized data. On the other hand, a traditional algorithm is also used to determine the position of each base signal acquisition unit in these real-time fluorescence images to be tested and the base category of the base signal acquisition unit at the corresponding position, so as to determine whether the base category prediction results of the base signal acquisition units obtained by the base recognition model from the base category recognition processing of these real-time fluorescence images to be tested are correct. In this way, after the quality assessment table is established based on the processing of the base signal acquisition units from these real-time fluorescence images to be tested, the quality assessment value of each base signal acquisition unit can be determined based on the quality assessment table in subsequent gene sequencing processes. The quality assessment value can be used to characterize the base category recognition processing process of the base recognition model, thereby achieving the purpose of objectively evaluating the accuracy of the processing results of the base recognition model to a certain extent.

[0066] In some embodiments, the determining of corresponding normalization parameters based on the multi-channel feature data corresponding to the plurality of base signal acquisition units in the input data by statistical analysis using a preset normalization method includes one of the following:

[0067] Performing statistical analysis based on the multi-channel feature data corresponding to the plurality of base signal acquisition units in the input data using maximum value normalization to obtain a normalized maximum value and a normalized minimum value corresponding to each channel;

[0068] Performing statistical analysis based on the multi-channel feature data corresponding to the plurality of base signal acquisition units in the input data by using mean-variance normalization to obtain a normalized mean and a normalized variance;

[0069] Based on the multi-channel feature data corresponding to the multiple base signal acquisition units in the input data, quantile normalization is used for statistical analysis, and the corresponding normalized score maximum value is obtained by the preset ratio of the normalized maximum value corresponding to each channel, and the corresponding normalized score minimum value is obtained by the preset ratio of the normalized minimum value.

[0070] In the process of establishing the quality assessment table, for the input data of the output layer of the base recognition model, the normalization processing method includes but is not limited to maximum value normalization, mean variance normalization and quantile normalization. In maximum value normalization, for the four-channel feature data corresponding to the four channels corresponding to the multiple base signal acquisition units, a preset normalization method is used to statistically analyze and obtain the corresponding normalization parameters. After the four-channel feature data are normalized based on the normalization parameters, the standardized data corresponding to each of the four channels are obtained. The quality assessment interval is determined according to the data interval corresponding to the maximum and minimum values ​​in the standardized data corresponding to each of the four channels. In mean variance normalization, for the four-channel feature data corresponding to the four channels of each base signal acquisition unit, the normalization method is used to obtain the mean and variance. After the four-channel feature data are normalized based on the mean and variance, the corresponding standardized data are obtained. The quality assessment interval is determined according to the data interval where the standardized data is located. In quantile normalization, for the maximum and minimum values ​​in the standardized data corresponding to each of the four channels, the maximum and minimum values ​​are replaced accordingly according to a certain ratio of the maximum and minimum values, that is, the maximum and minimum normalized scores are obtained as normalization parameters. After normalizing the four-channel feature data according to the maximum and minimum normalized scores, the corresponding standardized data are obtained. The quality assessment interval is determined according to the data interval corresponding to the standardized data.

[0071] In the above embodiments, a variety of optional normalization processing methods are provided. In actual application, one of them can be selected to implement the deep learning-based gene sequencing base quality assessment method described in the embodiment of the present application. Alternatively, multiple normalization methods can be used to execute and compare them separately, and then one of them can be selected to implement the deep learning-based gene sequencing base quality assessment method described in the embodiment of the present application, which is more convenient to meet different needs in different application scenarios.

[0072] In some embodiments, step S105 includes:

[0073] In the gene sequencing process, for each base signal acquisition unit, corresponding standardized data is obtained based on the normalization processing of the multi-channel feature data, the base category of the base signal acquisition unit is predicted based on the standardized data corresponding to the multiple channels, and the corresponding quality assessment interval in the quality assessment table is determined based on the standardized data, and the quality assessment value corresponding to the base category is determined based on the base category accuracy corresponding to the quality assessment interval.

[0074] Among them, the quality assessment table can be established in a one-to-one correspondence with the base recognition model, that is, for the same base recognition model, only one quality assessment table needs to be established; or, the quality assessment table can also be established in a one-to-one correspondence with each gene sequencing process of the base recognition model, that is, the same base recognition model re-establishes the corresponding quality assessment table before performing a gene sequencing. Please refer to Figure 3. Establishing a quality assessment table for the trained base recognition model includes: ①, inputting the fluorescence image data to be tested containing the base signal acquisition unit (such as a standard fluorescence image, a real-time fluorescence image to be tested collected in the previous cycle or the previous cycles of a gene sequencing process) into the base recognition model; ②, obtaining the input data of the output layer of the base recognition model, each input data correspondingly refers to the multi-channel feature data output by multiple channels that represent that a base signal acquisition unit belongs to different base categories, that is, the original score output by the base recognition model; ③, the input data of the output layer ⑤, the quality assessment interval to which the current base signal acquisition unit belongs is determined based on the standardized data, and whether the prediction result of the base category of the current base signal acquisition unit is correct is counted; ⑥, a quality assessment table is established by counting the classification and accuracy of a certain number of base signal acquisition units corresponding to the quality assessment interval. In the gene sequencing process, using the quality assessment table to determine the quality assessment value of the base category includes: ⑦, in the gene sequencing process, real-time acquisition of the fluorescence image data to be tested is input into the base recognition model; ⑧, obtaining the input data of the output layer of the base recognition model, each input data corresponding to the multi-channel feature data output by multiple channels representing a base signal acquisition unit belonging to different base categories, that is, the original score output by the base recognition model; ⑨, preprocessing the input data of the output layer, that is, normalizing the multi-channel feature data according to the normalization parameters determined by the previous statistical analysis to obtain the standardized data corresponding to each channel; ⑩, according to the quality assessment table, determining the quality assessment interval in which the standardized data of the current base signal acquisition unit is located, and using the accuracy corresponding to the quality assessment interval as the quality assessment value of the currently obtained base category prediction result.

[0075] In the above embodiment, in the gene sequencing process, for the prediction of the base category of the base signal acquisition unit in the real-time collected fluorescent image data to be tested, the base category of the base signal acquisition unit can be predicted based on the standardized data after the multi-channel feature data is normalized. The prediction of the base category and the determination of the quality assessment value can share the result of normalization processing of the input data of the output layer of the base recognition model, thereby realizing the synchronization of base category prediction and base quality assessment, reducing the amount of calculation and improving the accuracy of base quality assessment.

[0076] In some embodiments, step S105 includes:

[0077] In the gene sequencing process, for each base signal acquisition unit, the base category of the base signal acquisition unit is predicted according to the output result of the output layer, and the multi-channel feature data of the multiple channels in the input data of the output layer are normalized to obtain corresponding standardized data. Based on the standardized data, the corresponding quality assessment interval in the quality assessment table is determined, and the quality assessment value corresponding to the base category is determined according to the base category accuracy corresponding to the quality assessment interval.

[0078] The process of establishing a quality assessment table for the trained base recognition model and using the quality assessment table to determine the quality assessment value of the base category in the gene sequencing process can be as described above. In this embodiment, in the gene sequencing process, the prediction of the base category by the base signal acquisition unit in the real-time fluorescent image data to be tested can be based on the base category prediction results of the output layer of the base recognition model. The quality assessment value of the base category prediction results obtained by the output layer can be determined based on the standardized data after normalization of the multi-channel feature data.

[0079] In the above embodiment, the base recognition model uses the results of the output layer to determine the base category, and uses the results of the non-output layer as the basis for base quality assessment. On the basis of retaining the original complete base recognition function of the base recognition model, a base quality assessment system is added to evaluate the accuracy of the current base category prediction results.

[0080] In order to provide a more comprehensive understanding of the deep learning-based gene sequencing base quality assessment method provided in the embodiments of the present application, please refer to FIG4 . A specific example is provided below to illustrate the deep learning-based gene sequencing base quality assessment method. The method includes:

[0081] S11, using the fluorescence image data to be measured including the base signal acquisition unit as the input of the base recognition model, extracting the input data of the output layer of the base recognition model, and performing data preprocessing.

[0082] Taking the output layer of the base recognition model as a softmax layer as an example, the input data of the output layer is logits (log probability). The raw scores output by the base recognition model as shown in Figure 2 are normalized for each channel in a certain way. Normalization methods include but are not limited to: quantile normalization, maximum normalization, mean variance normalization, etc. After selecting the normalization method, normalization of each channel according to the selected normalization method can be: Option 1: preset normalization parameters, and obtain the normalization parameters of the four channels by counting the base signal acquisition units of a large batch; Option 2: count the normalization parameters of the four channels obtained by the base signal acquisition units of the current batch.

[0083] S12, perform statistical analysis on the data preprocessing results and establish a quality assessment table.

[0084] After the raw scores output by the base call model are preprocessed in step S11, the data is divided into N categories according to a certain rule. The accuracy rate of the data in each category is then calculated. Finally, this accuracy rate is used as the guide accuracy rate for the data in that category to obtain a base quality assessment table Table Q. The certain rule includes but is not limited to the following scheme: Assume that the raw scores output by the base call model after the preprocessing in step S11 are normalized scores: NO_A, NO_C, NO_G, NO_T. The maximum value (NO_Max) and minimum value (NO_Min) of the normalized scores of the four channels of each base signal acquisition unit (cluster) are counted, and then 0 to 1 is divided into 10 intervals (Q1, Q2, Q3, Q4, Q5, Q6, Q7, Q8, Q9, Q10) with an interval of 0.1. Then, the NO_Max and NO_Min of each cluster are judged to fall into which interval, and the clusters are divided into 100 categories. Then, the accuracy of the category prediction of the clusters in these 100 categories is counted as the guidance accuracy of the cluster, thereby obtaining a 10×10 base quality assessment table.

[0085] S13, in gene sequencing, uses a quality assessment table to evaluate the quality of the base category prediction results of the base recognition model.

[0086] In gene sequencing, the real-time collected fluorescence image data is used as the input of the base recognition model. The raw score output by the base recognition model is preprocessed in step S11, and the category X to which the cluster belongs is determined according to a certain rule. Then, the base quality assessment table TableQ generated in step S12 is indexed to obtain the guidance quality value of the cluster.

[0087] The deep learning-based gene sequencing base quality assessment method provided in the embodiments of the present application uses the original score of the base recognition model as the basis for evaluating the base quality value, and adopts a statistical method to obtain a base quality assessment system. In this way, on the one hand, by using the results of the non-output layer as the basis for base quality assessment, the base quality assessment is made more accurate; on the other hand, by combining a non-fitting method with a statistical method to obtain a quality value assessment system, the amount of calculation brought about by the base quality assessment can be reduced.

[0088] On the other hand, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the deep learning-based gene sequencing base quality assessment method described in any embodiment of the present application.

[0089] In the computer program product, an optional implementation form of the program module architecture of the computer program that implements each step of the method is a deep learning-based gene sequencing base quality assessment device. Referring to Figure 5, the deep learning-based gene sequencing base quality assessment device includes: an acquisition module 11, configured to use the fluorescence image data to be measured, including the base signal acquisition unit, as input to a base recognition model trained based on a deep learning model to obtain input data for the output layer of the base recognition model; the input data includes multi-channel feature data of multiple channels corresponding to multiple base categories; a quality system construction module 12, configured to obtain corresponding standardized data based on the input data, based on the multi-channel feature data corresponding to the multiple base signal acquisition units, and to establish a quality assessment table based on the standardized data of the multiple base signal acquisition units and statistics on whether the base categories are correct; and an assessment module 13, configured to predict the corresponding base category for each base signal acquisition unit according to the multi-channel feature data in the gene sequencing process, and determine the quality assessment value corresponding to the base category based on the quality assessment table.

[0090] Optionally, the quality system construction module 12 is used to obtain corresponding standardized data based on the input data and the normalization processing of the multi-channel feature data corresponding to the multiple base signal acquisition units; based on whether the quality assessment intervals and base categories corresponding to the standardized data of the multiple base signal acquisition units are correct, the accuracy rates of the base categories corresponding to the different quality assessment intervals are counted, and a quality assessment table containing the correspondence between the quality assessment intervals and the base category accuracy rates is established.

[0091] Optionally, the quality system construction module 12 is also used to determine multiple numerical intervals based on 0 to 1 divided at preset intervals, and divide the quality assessment intervals based on the combinations between the numerical intervals; for the standardized data corresponding to each of the base signal acquisition units, determine the quality assessment interval to which the standardized data corresponds, and determine whether the predicted base category of the base signal acquisition unit is correct; based on the statistical results of the multiple base signal acquisition units, obtain the base category accuracy corresponding to the different quality assessment intervals.

[0092] Optionally, the output layer is a softmax layer, and the base recognition model includes an input layer, a hidden layer and the output layer; wherein the input data includes the multi-channel feature data output by the hidden layer based on feature extraction of the fluorescent image data to be tested, and the determined logarithmic probability corresponding to each base category to which the base signal acquisition unit in the fluorescent image data to be tested belongs.

[0093] Optionally, the quality system construction module 12 is also used to input the input data of the output layer of the base recognition model using a standard fluorescent image as the fluorescent image data to be tested according to the base recognition model, and to process the multi-channel feature data corresponding to the multiple base signal acquisition units in the input data using a preset normalization method to obtain standardized data corresponding to each channel.

[0094] Optionally, the quality system construction module 12 is also used to obtain input data of the output layer of the base recognition model based on the real-time fluorescence image to be tested obtained by current sequencing according to the base recognition model, and to process the multi-channel feature data corresponding to the multiple base signal acquisition units in the input data through a preset normalization method to obtain standardized data corresponding to each channel.

[0095] Optionally, optionally, the quality system construction module 12 is also used to process the multi-channel feature data corresponding to the multiple base signal acquisition units in the input data through maximum value normalization to obtain the normalized maximum value and normalized minimum value corresponding to each channel; process the multi-channel feature data corresponding to the multiple base signal acquisition units in the input data through mean variance normalization to obtain the normalized mean and normalized variance; process the multi-channel feature data corresponding to the multiple base signal acquisition units in the input data through quantile normalization to obtain the corresponding normalized score maximum value and the corresponding normalized score minimum value according to the preset ratio of the normalized maximum value corresponding to each channel and the preset ratio of the normalized minimum value.

[0096] Optionally, the evaluation module 13 is also used to, in the gene sequencing process, for each of the base signal acquisition units, obtain corresponding standardized data based on the normalization processing of the multi-channel feature data, predict the base category of the base signal acquisition unit based on the standardized data corresponding to the multiple channels, and determine the corresponding quality assessment interval in the quality assessment table based on the standardized data, and determine the quality assessment value corresponding to the base category according to the base category accuracy corresponding to the quality assessment interval.

[0097] Optionally, the evaluation module 13 is also used to predict the base category of each base signal acquisition unit according to the output result of the output layer in the gene sequencing process, and obtain corresponding standardized data based on the multi-channel feature data of the multiple channels in the input data of the output layer through normalization processing, determine the corresponding quality evaluation interval in the quality evaluation table based on the standardized data, and determine the quality evaluation value corresponding to the base category according to the base category accuracy corresponding to the quality evaluation interval.

[0098] It should be noted that the deep learning-based gene sequencing base quality assessment device provided in the above embodiment only uses the division of the above-mentioned program modules as an example to illustrate the process of performing gene sequencing and base quality assessment simultaneously. In actual applications, the above-mentioned processing can be assigned to different program modules as needed, that is, the internal structure of the device can be divided into different program modules to complete all or part of the method steps described above. In addition, the deep learning-based gene sequencing base quality assessment device provided in the above embodiment and the deep learning-based gene sequencing base quality assessment method embodiment are based on the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0099] On the other hand, the present application also provides a gene sequencing device. Please refer to Figure 6, which is an optional hardware structure diagram of a gene sequencing device. The gene sequencing device includes a processor 212 and a memory 211 connected to the processor 212. The memory 211 stores a computer program for implementing the gene sequencing base quality assessment method based on deep learning provided in any embodiment of the present application, so that when the corresponding computer program is executed by the processor, the steps of the gene sequencing base quality assessment method based on deep learning provided in any embodiment of the present application are implemented. The gene sequencing device loaded with the corresponding computer program has the same technical effect as the corresponding method embodiment. To avoid repetition, it will not be repeated here.

[0100] On the other hand, the embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, each process of the above-mentioned embodiment of the method for assessing base quality of gene sequencing based on deep learning is implemented, and the same technical effect is achieved. To avoid repetition, it is not described here. Wherein, the computer-readable storage medium is, for example, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0101] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or apparatus comprising the element.

[0102] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), including a number of instructions for a terminal (which can be a mobile phone, computer, server, spatial genomics sequencing platform, gene sequencer, or network equipment, etc.) to execute the methods described in each embodiment of the present invention.

[0103] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A gene sequencing base quality assessment method based on deep learning, characterized in that: include: Using the fluorescence image data to be measured including the base signal acquisition unit as input to a base recognition model trained based on a deep learning model, and obtaining input data for an output layer of the base recognition model; the input data includes multi-channel feature data of multiple channels corresponding to multiple base categories; According to the input data, based on the multi-channel feature data corresponding to the plurality of base signal acquisition units, corresponding standardized data is obtained by normalization processing, and a quality assessment table is established based on the standardized data of the plurality of base signal acquisition units and statistics on whether the base categories are correct; In the gene sequencing process, for each base signal acquisition unit, the corresponding base category is predicted according to the multi-channel feature data, and the quality assessment value corresponding to the base category is determined based on the quality assessment table.

2. The method for evaluating gene sequencing base quality based on deep learning according to claim 1, wherein: The method further comprises: performing normalization processing on the multi-channel feature data corresponding to each of the plurality of base signal acquisition units to obtain corresponding standardized data according to the input data; and establishing a quality assessment table based on statistics of the standardized data of the plurality of base signal acquisition units and the accuracy of the base categories, including: According to the input data, corresponding standardized data is obtained by normalizing the multi-channel feature data corresponding to the plurality of base signal acquisition units; Based on whether the standardized data of the multiple base signal acquisition units correspond to the correct quality assessment intervals and base categories, the base category accuracy rates corresponding to different quality assessment intervals are counted, and a quality assessment table containing the correspondence between quality assessment intervals and base category accuracy rates is established.

3. The method for evaluating gene sequencing base quality based on deep learning according to claim 2, wherein: The step of obtaining corresponding standardized data based on the input data and the multi-channel feature data corresponding to each of the plurality of base signal acquisition units through normalization processing includes: According to the input data, the multi-channel feature data corresponding to the multiple base signal acquisition units are processed by a preset normalization method independent of the output layer to obtain corresponding standardized data.

4. The method for evaluating gene sequencing base quality based on deep learning according to claim 2, wherein: The standardized data based on the plurality of base signal acquisition units respectively correspond to whether the quality assessment intervals and base categories to which they belong are correct, and the statistical calculation of the base category accuracy rates corresponding to different quality assessment intervals includes: Determine a plurality of numerical intervals based on the value range from 0 to 1 and divided into a plurality of numerical intervals according to a preset interval, and divide the quality assessment intervals based on the combination of the numerical intervals; For each of the standardized data corresponding to the base signal acquisition unit, determining the quality assessment interval to which the standardized data corresponds, and determining whether the predicted base category of the base signal acquisition unit is correct; According to the statistical results of the multiple base signal acquisition units, the base category accuracy rates corresponding to the different quality assessment intervals are obtained.

5. The method for evaluating gene sequencing base quality based on deep learning according to claim 4, wherein: The preset interval is 0.1, and the multiple numerical intervals are determined based on the preset intervals from 0 to 1, and the quality assessment intervals are divided based on the combination of the numerical intervals, including: The range from 0 to 1 is divided into ten numerical intervals with an interval of 0.1, and a 10×10 quality evaluation interval is obtained based on the combination of the ten numerical intervals.

6. The method for evaluating gene sequencing base quality based on deep learning according to claim 4, wherein: The output layer is a softmax layer, and the base recognition model includes an input layer, a hidden layer and the output layer; wherein the input data includes multi-channel feature data output by the hidden layer based on feature extraction of the fluorescent image data to be tested, and the determined logarithmic probability corresponding to each base category to which the base signal acquisition unit in the fluorescent image data to be tested belongs.

7. The method for evaluating gene sequencing base quality based on deep learning according to claim 2, wherein: The step of obtaining corresponding standardized data based on the input data and the multi-channel feature data corresponding to each of the plurality of base signal acquisition units through normalization processing includes: According to the base recognition model, a standard fluorescence image is used as the fluorescence image data to be tested as the input to obtain the input data of the output layer of the base recognition model. Based on the multi-channel feature data corresponding to the multiple base signal acquisition units in the input data, the corresponding normalization parameters are determined through statistical analysis using a preset normalization method. Based on the normalization parameters, the multi-channel feature data of each base signal acquisition unit are processed respectively to obtain standardized data corresponding to each channel.

8. The method for evaluating gene sequencing base quality based on deep learning according to claim 2, wherein: The step of obtaining corresponding standardized data based on the input data and the multi-channel feature data corresponding to each of the plurality of base signal acquisition units through normalization processing includes: According to the base recognition model, the real-time fluorescence image to be tested obtained by the current sequencing is used as input to obtain the input data of the output layer of the base recognition model. Based on the multi-channel feature data corresponding to the multiple base signal acquisition units in the input data, the corresponding normalization parameters are determined through statistical analysis using a preset normalization method. Based on the normalization parameters, the multi-channel feature data of each base signal acquisition unit are processed respectively to obtain standardized data corresponding to each channel.

9. The method for evaluating gene sequencing base quality based on deep learning according to claim 7 or 8, wherein: The determining of corresponding normalization parameters based on the multi-channel feature data respectively corresponding to the plurality of base signal acquisition units in the input data by statistical analysis using a preset normalization method includes one of the following: Performing statistical analysis based on the multi-channel feature data corresponding to the plurality of base signal acquisition units in the input data using maximum value normalization to obtain a normalized maximum value and a normalized minimum value corresponding to each channel; Performing statistical analysis based on the multi-channel feature data corresponding to the plurality of base signal acquisition units in the input data by using mean-variance normalization to obtain a normalized mean and a normalized variance; Based on the multi-channel feature data corresponding to the multiple base signal acquisition units in the input data, quantile normalization is used for statistical analysis, and the corresponding normalized score maximum value is obtained by the preset ratio of the normalized maximum value corresponding to each channel, and the corresponding normalized score minimum value is obtained by the preset ratio of the normalized minimum value.

10. The method for evaluating gene sequencing base quality based on deep learning according to claim 1, wherein: In the gene sequencing process, for each base signal acquisition unit, predicting the corresponding base category according to the multi-channel feature data, and determining the quality assessment value corresponding to the base category based on the quality assessment table, including: In the gene sequencing process, for each base signal acquisition unit, corresponding standardized data is obtained based on the normalization processing of the multi-channel feature data, the base category of the base signal acquisition unit is predicted based on the standardized data corresponding to the multiple channels, and the corresponding quality assessment interval in the quality assessment table is determined based on the standardized data, and the quality assessment value corresponding to the base category is determined based on the base category accuracy corresponding to the quality assessment interval.

11. The method for evaluating gene sequencing base quality based on deep learning according to claim 1, wherein: In the gene sequencing process, for each base signal acquisition unit, predicting the corresponding base category according to the multi-channel feature data, and determining the quality assessment value corresponding to the base category based on the quality assessment table, including: In the gene sequencing process, for each base signal acquisition unit, the base category of the base signal acquisition unit is predicted according to the output result of the output layer, and the multi-channel feature data of the multiple channels in the input data of the output layer are normalized to obtain corresponding standardized data. Based on the standardized data, the corresponding quality assessment interval in the quality assessment table is determined, and the quality assessment value corresponding to the base category is determined according to the base category accuracy corresponding to the quality assessment interval.

12. The method for evaluating gene sequencing base quality based on deep learning according to claim 1, wherein: The multiple base categories include four base categories of A, T, C, and G, and the multi-channel feature data includes feature data corresponding to four channels corresponding to the four base categories of A, T, C, and G, respectively.

13. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 12 are implemented.

14. A gene sequencing device comprising a memory, a processor, and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 12.

15. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which implements the steps of the method according to any one of claims 1 to 12 when executed by a processor.

Images