Method for analyzing disease, method and apparatus for training model, and storage medium

Abstract

A method for analyzing a disease, a method and apparatus for training a model, and a storage medium. The method for training the disease analysis model comprises: dividing a reference genome into a plurality of genomic intervals, and determining the genomic interval to which each sequencing read in free DNA sequencing data belongs; counting the number of short fragments and the number of long fragments in each genomic interval on the basis of the length of each sequencing read, and selecting at least a portion of the plurality of genomic intervals as differential genomic intervals, wherein the ratio of the number of short fragments to the number of long fragments in the differential genomic intervals is greater than the ratio of the number of short fragments to the number of long fragments in the non-differential genomic intervals; acquiring at least partial read alignments of each sequencing read within each differential genomic interval, and encoding the acquired read alignments to form a complete data set; and constructing a neural network model, and training and evaluating the neural network model by using the complete data set.

Description

Disease analysis method, model training method and device, storage medium Technical Field

[0001] The embodiments of the present disclosure relate to, but are not limited to, the field of biotechnology, and in particular to a disease analysis method, a model training method and device, and a storage medium. Background Art

[0002] Machine learning models usually require a large sample size for training, but clinically there are relatively few samples of cancer patients. Therefore, when conducting disease analysis, there is often a problem of being unable to build a model due to the small number of samples or lack of data to effectively evaluate model performance.

[0003] Summary of the Invention

[0004] The following is a summary of the subject matter described in detail herein. This summary is not intended to limit the scope of the claims.

[0005] The present disclosure provides a method for training a disease analysis model, including:

[0006] Divide the reference genome into multiple genomic intervals and determine the genomic interval to which each sequencing sequence in the free DNA sequencing data belongs;

[0007] Counting the number of short fragments and the number of long fragments in each genomic interval according to the length of each sequencing sequence, and selecting at least some of the genomic intervals among the multiple genomic intervals as differential genomic intervals, wherein the ratio of the number of short fragments to the number of long fragments in the differential genomic interval is greater than the ratio of the number of short fragments to the number of long fragments in the non-differential genomic interval;

[0008] Obtaining at least a partial sequence arrangement of each sequencing sequence within each of the differential genomic intervals, encoding the obtained sequence arrangement to form a total data set;

[0009] A neural network model is constructed, and the neural network model is trained and evaluated using the total data set.

[0010] An embodiment of the present disclosure also provides a training device for a disease analysis model, comprising a memory; and a processor connected to the memory, wherein the memory is used to store instructions, and the processor is configured to execute the steps of the training method for the disease analysis model described in any embodiment of the present disclosure based on the instructions stored in the memory.

[0011] An embodiment of the present disclosure also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the training method of the disease analysis model described in any embodiment of the present disclosure.

[0012] The present disclosure also provides a training device for a disease analysis model, comprising: a first segmentation module, a difference interval determination module, a first encoding module, and a model training module, wherein:

[0013] The first partitioning module is configured to partition the reference genome into a plurality of genomic intervals and determine the genomic interval to which each sequencing sequence in the cell-free DNA sequencing data belongs;

[0014] The difference interval determination module is configured to count the number of short fragments and the number of long fragments in each genomic interval according to the length of each sequencing sequence, and select at least some of the genomic intervals among the multiple genomic intervals as difference genomic intervals, wherein the ratio of the number of short fragments to the number of long fragments in the difference genomic interval is greater than the ratio of the number of short fragments to the number of long fragments in the non-differential genomic interval;

[0015] The first encoding module is configured to obtain at least a partial sequence arrangement of each sequencing sequence within each of the differential genomic intervals, encode the obtained sequence arrangement, and form a total data set;

[0016] The model training module is configured to construct a neural network model and use the total data set to train and evaluate the neural network model.

[0017] The present disclosure also provides a disease analysis method, including:

[0018] Determine the differential genomic intervals to which multiple sequencing sequences in cell-free DNA sequencing data belong;

[0019] Obtaining at least a partial sequence arrangement of each sequencing sequence within each differential genomic interval, encoding the obtained sequence arrangement to form a test data set, wherein the test data set includes a plurality of test data;

[0020] Inputting the plurality of test data into a trained neural network model to obtain a plurality of model prediction values;

[0021] Data processing is performed on the multiple model prediction values ​​to obtain disease analysis results.

[0022] An embodiment of the present disclosure also provides a disease analysis device, comprising a memory; and a processor connected to the memory, wherein the memory is used to store instructions, and the processor is configured to execute the steps of the disease analysis method described in any embodiment of the present disclosure based on the instructions stored in the memory.

[0023] An embodiment of the present disclosure further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the disease analysis method described in any embodiment of the present disclosure.

[0024] The present disclosure also provides a disease analysis device, comprising: a second segmentation module, a second encoding module, and a prediction module, wherein:

[0025] The second partitioning module is configured to determine the differential genomic intervals to which multiple sequencing sequences in the cell-free DNA sequencing data belong;

[0026] The second encoding module is configured to obtain at least a partial sequence arrangement of each sequencing sequence in each differential genomic interval, encode the obtained sequence arrangement, and form a test data set, wherein the test data set includes a plurality of test data;

[0027] The prediction module is configured to input the multiple test data into a trained neural network model to obtain multiple model prediction values; and perform data processing on the multiple model prediction values ​​to obtain disease analysis results.

[0028] Other aspects will become apparent upon reading and understanding the drawings and detailed description.

[0029] Summary of the Figures

[0030] The accompanying drawings are intended to provide a further understanding of the technical solutions of the present disclosure and constitute a part of the specification. Together with the embodiments of the present disclosure, they are used to explain the technical solutions of the present disclosure and do not constitute a limitation of the technical solutions of the present disclosure. The shapes and sizes of the components in the drawings do not reflect the actual scale and are intended only to illustrate the contents of the present disclosure.

[0031] FIG1 is a flow chart of a method for training a disease analysis model according to an exemplary embodiment of the present disclosure;

[0032] FIG2 is a schematic diagram of a reference genome divided into multiple genomic intervals provided by an exemplary embodiment of the present disclosure;

[0033] FIG3 is a schematic diagram showing the situation in which each genomic interval in FIG2 contains short fragments and long fragments;

[0034] FIG4 is a schematic diagram of a method for encoding an acquired sequence arrangement provided by an exemplary embodiment of the present disclosure;

[0035] FIG5 is a schematic structural diagram of a neural network model provided by an exemplary embodiment of the present disclosure;

[0036] FIG6 is a flow chart of a disease analysis method provided by an exemplary embodiment of the present disclosure;

[0037] FIG7 is a schematic structural diagram of a training device for a disease analysis model provided by an exemplary embodiment of the present disclosure;

[0038] FIG8 is a schematic structural diagram of a disease analysis device provided by an exemplary embodiment of the present disclosure;

[0039] FIG9 is a schematic structural diagram of another disease analysis model training device provided by an exemplary embodiment of the present disclosure;

[0040] FIG10 is a schematic structural diagram of another disease analysis device provided by an exemplary embodiment of the present disclosure.

[0041] Details

[0042] To make the objectives, technical solutions and advantages of the present disclosure more clearly understood, the embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings. It should be noted that, unless there is a conflict, the embodiments and features in the embodiments of the present disclosure can be combined with each other in any manner.

[0043] Unless otherwise defined, the technical or scientific terms used in the embodiments of the present disclosure should have the ordinary meaning understood by people with ordinary skills in the field to which the present disclosure belongs. The words "first", "second" and similar words used in the embodiments of the present disclosure do not indicate any order, quantity or importance, but are only used to distinguish different components. The words "include" or "comprising" and similar words mean that the elements or objects preceding the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects.

[0044] Circulating free DNA (cfDNA) refers to highly fragmented DNA that circulates in the human blood and exists outside of cells. Clinically, analysis of cfDNA fragmentation data can be used for non-invasive disease diagnosis, disease monitoring, and treatment response assessment. The study of cfDNA fragmentation is crucial for understanding disease biology and developing relevant biomarkers. Current research suggests that early cancer screening models can be developed based on information on cfDNA methylation, fragmentation, and mutations.

[0045] As shown in FIG1 , an embodiment of the present disclosure provides a method for training a disease analysis model, including:

[0046] Step 101: Divide the reference genome into multiple genomic intervals, and determine the genomic interval to which each sequencing sequence (read) in the free DNA sequencing data belongs;

[0047] Step 102: Count the number of short fragments and the number of long fragments in each genomic interval based on the length of each sequencing sequence, and select some genomic intervals from the multiple genomic intervals as differential genomic intervals, where the ratio of the number of short fragments to the number of long fragments in the differential genomic interval is greater than the ratio of the number of short fragments to the number of long fragments in the non-differential genomic interval;

[0048] Step 103: Obtain at least a partial sequence arrangement of each sequencing sequence within each differential genomic interval, and encode the obtained sequence arrangement to form a total data set;

[0049] Step 104: Build a neural network model, and use the total data set to train and evaluate the neural network model.

[0050] Reads are short sequence fragments obtained in high-throughput sequencing. Sequencing the entire genome will generate millions of reads.

[0051] The training method for the disease analysis model provided in the embodiments of the present disclosure determines the differential genomic interval based on the ratio of the number of short fragments to the number of long fragments, and encodes at least part of the sequence arrangement of each sequencing sequence in the differential genomic interval to form a total data set. The total data set is used to train and evaluate the neural network model, which greatly increases the number of samples and solves the clinical problem of being unable to build a model or lacking the amount of data to effectively evaluate model performance due to too few samples.

[0052] In the disclosed embodiment, the disease analysis model can be used for at least one of the following: probability analysis of a patient being ill, probability analysis of a patient's expected treatment being effective, probability analysis of residual lesions, and the like.

[0053] The training method of the disease analysis model provided in the embodiments of the present disclosure is suitable for predicting the stage of tumor occurrence and development, such as early tumor screening, treatment efficacy evaluation, minimal residual disease (MRD) detection and other clinical applications.

[0054] Early cancer screening involves using a series of diagnostic tests to promptly detect potential malignancies, enabling early treatment and disease control. Treatment efficacy assessment involves using scientific methods to evaluate and determine the effectiveness of treatment, as well as the safety and reliability of treatment. MRD is primarily used to determine whether cancer cells, known as minimal residual disease, remain in patients who have received treatment. This article will use treatment efficacy assessment as an example to illustrate this.

[0055] In some exemplary embodiments, the reference genome is the hg19 version or the hg38 version.

[0056] Both hg19 (Human Genome Version 19) and hg38 (Human Genome Version 38) are versions of the human genome. They represent the human genome reference sequence released at different points in time and are widely used standards in scientific research. This disclosure uses hg19 as an example for illustration.

[0057] In some exemplary embodiments, a window with a window length a is slid across the reference genome with a step size b to divide the reference genome into a plurality of genomic intervals, where a is greater than or equal to b.

[0058] When a is greater than b, there is a certain proportion of overlap between each two adjacent genomic intervals; when a=b, there is no overlap between each two adjacent genomic intervals.

[0059] For example, when the reference genome is divided into multiple genomic intervals, the window length a and the step length b can both be set to 5Mb. However, the embodiments of the present disclosure do not limit this, and the sizes of the window length a and the step length b can be set as needed.

[0060] The reference genome was partitioned by obtaining the autosome 1 to 22 interval information from the hg19 fasta index file hg19.fai. The window length w and step size s were both set to 5 Mb, resulting in a total of 589 genomic intervals, as shown in Figure 2. In the figure, one bin represents one genomic interval, with n = 589. The partitioned interval file contains three columns: chromosome number, start position, and end position, from left to right.

[0061] In some exemplary embodiments, determining the genomic interval to which each sequencing sequence in the cell-free DNA sequencing data belongs includes:

[0062] Perform data quality control and cleaning on cell-free DNA sequencing data to obtain data to be compared;

[0063] Compare the data to be compared with the reference sequence of each genomic interval to obtain the chromosome number and start and end position information of each sequencing sequence;

[0064] Determine the intersection ratio of each sequencing sequence with one or more genomic intervals based on the chromosome number and start and end position information of each sequencing sequence;

[0065] The genomic interval to which each sequencing sequence belongs is determined based on the intersection ratio between each sequencing sequence and one or more genomic intervals.

[0066] In the disclosed embodiments, using treatment efficacy evaluation as an example, during the training phase, the cell-free DNA sequencing data may include sequencing data from multiple patients who responded to treatment and those who did not respond. Before sequence alignment, the cell-free DNA sequencing data undergoes data quality control and cleaning to ensure data quality. Subsequently, the QC- and cleaned sequencing data is aligned to a reference genome, generating a bam file.

[0067] The third column RNAME in the bam file is the chromosome number. Compare the sizes of the fourth column POS and the eighth column MPOS in the bam format file. Use the smaller value as the read start position. Add the absolute value of the ninth column ISIZE to the read start position to obtain the read end position. Then, retrieve the chromosome number, start position, end position, and read length of all double-end matched reads in sequence.

[0068] The start and end information of the reads under the same chromosome are intersected with the previously divided genomic interval information, and the genomic interval to which each read belongs is determined based on the intersection ratio of each read with one or more genomic intervals.

[0069] In some exemplary embodiments, determining the genomic interval to which each read belongs based on the intersection ratio of each read with one or more genomic intervals includes:

[0070] When the intersection ratio of a read and a genomic interval is greater than 50%, the read is determined to fall within the genomic interval.

[0071] In the disclosed embodiment, when the intersection ratio is greater than 50% of the read length, the read is considered to fall within the corresponding genomic interval, so the distribution of reads in all genomic intervals can be obtained.

[0072] In some exemplary embodiments, the method further comprises:

[0073] Set the short fragment length range and long fragment length range;

[0074] According to the length of each sequencing sequence, it is determined whether each sequencing sequence is a short fragment, a long fragment or a useless fragment.

[0075] In the embodiment of the present disclosure, the short fragment length range and the long fragment length range can be set as needed.

[0076] In some exemplary embodiments, the length of the short fragment may range from n1 bp to n2 bp, and the length of the long fragment may range from (n2+1) bp to n3 bp, wherein n1 <n2<n3。

[0077] In some exemplary embodiments, n1≧100, 150≦n2≦160, and n3≦250.

[0078] For example, the length of short fragments can range from 100bp to 150bp, and the length of long fragments can range from 151bp to 220bp. Sequencing sequences shorter than 100bp or longer than 220bp are considered useless fragments. That is, read sequences with a length of 100bp to 150bp are regarded as short fragments, and read sequences with a length of 151bp to 220bp are regarded as long fragments. The number of short fragments and the number of long fragments in each genomic interval are counted, as shown in Figure 3. Based on the number of long and short fragments of reads distributed in each interval, the ratio of the number of short fragments to the number of long fragments can be calculated, and this ratio can be used as a statistic for subsequent determination of differential genomic intervals.

[0079] In some exemplary embodiments, selecting at least some of the genomic intervals in the plurality of genomic intervals as differential genomic intervals comprises:

[0080] Calculate the fold change of each genomic interval (fold change = log2 (number of short fragments / number of long fragments);

[0081] Sort the multiple genomic intervals by their multiple difference scores from large to small, and set the top N genomic intervals as N differential genomic intervals, where N is a natural number greater than or equal to 1.

[0082] For example, N = 100, however, this is not a limitation of the present disclosure. Taking N = 100 as an example, the log2 (number of short fragments / number of long fragments) of each genomic interval is calculated as the Fold Change value, and the top 100 genomic intervals with the largest Fold Change values ​​are taken as the differential genomic intervals.

[0083] This embodiment defines the difference factor as the log2-transformed value of the ratio of the number of short fragments to the number of long fragments, making the data more evenly distributed and facilitating subsequent research. In other exemplary embodiments, the difference factor can also be defined as the inverse of the ratio of the number of short fragments to the number of long fragments, etc., which is not limited in this embodiment.

[0084] In some exemplary embodiments, obtaining at least a partial sequence arrangement of each sequencing sequence within each differential genomic interval includes: obtaining the arrangement of the first 4 bases of the 5' end of each sequencing sequence within each differential genomic interval.

[0085] Count the reads that fall into the differential genomic interval and take the first 4 bp base arrangement of the 5' end of each read. Since there are 4 possible bases at each 1 bp position: A, T, C, G, the maximum possible number of base arrangements is 4 4 =256.

[0086] In some exemplary embodiments, encoding the acquired sequence arrangement comprises:

[0087] The first 4 bases at the 5' end are encoded as a 4*4 matrix, with each base corresponding to a row of the 4*4 matrix.

[0088] As shown in Figure 4, using one-hot encoding, the four bases A, T, C, and G are converted into four lists: [1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0], and [0, 0, 0, 1], respectively. Since the 5' end sequence arrangement of each read obtained is the arrangement of the first four bases at the 5' end of each read, a 4*4 one-hot encoding matrix can be formed after combination from top to bottom. Each matrix is ​​regarded as an independent sample, and the corresponding label is the patient's post-treatment response number (this embodiment takes the treatment efficacy evaluation as an example, and the evaluation results include: treatment effectiveness or ineffectiveness, represented by 1 or 0 respectively).

[0089] Since each patient's cfDNA sequencing data contains millions of reads, the matrix corresponding to each read can be used as a sample. Because each read is labeled, the amount of sample data can be greatly increased for modeling.

[0090] The performance of machine learning models is highly dependent on sample size. Clinically, due to the small number of patient samples and dispersed data resources, the sample size in existing machine learning model construction for clinical research is generally small, resulting in poor performance and generalization of the model. The disclosed embodiment treats each read from all differential genomic intervals of a patient (an existing sample) as a sample. A single patient contains tens of thousands of reads, and each read can be assigned a label for the patient to whom it belongs. This greatly expands the sample size and increases the reliability of model construction.

[0091] In some exemplary embodiments, the neural network model includes one or more convolutional pooling modules and one or more fully connected layers;

[0092] Each convolution-pooling module includes a convolution layer and a pooling layer. The convolution layer is configured to extract feature information from the input data; the pooling layer is configured to compress the input data.

[0093] The fully connected layer is configured to map the feature information output by the convolutional pooling module into the output space to obtain the prediction result.

[0094] Due to the large sample size, a neural network approach can be used to build the model. When building the model, as shown in Figure 5, a 4*4 matrix is ​​linearized using one-dimensional convolution. One-dimensional pooling is then used to retain key features. Finally, the model is trained by passing it through several fully connected layers, using an activation function to return a probability value between 0 and 1. A loss function is defined for the classification task for backpropagation. Depending on the specific situation, the one-dimensional convolution, one-dimensional maximum pooling, and fully connected network layers can be repeated multiple times. The average or median of the probability values ​​predicted by the model is used as the final probability.

[0095] In some exemplary embodiments, the neural network model is trained and evaluated using the total dataset, including:

[0096] Split the total dataset into training and testing sets;

[0097] The training set is used to train the neural network model, and the test set is used to evaluate the neural network model.

[0098] The matrix data for each patient were combined to form a total dataset, which included both matrix data for effective treatment and matrix data for ineffective treatment. Stratified sampling was used to split the total dataset into a training set and a test set in an 8:2 ratio.

[0099] Taking the construction of a neural network based on two convolution pooling and two full connections as an example, the number of input channels of the one-dimensional convolution layer 1 is set to 4, the number of output channels is set to 8, and the convolution kernel size is 2; the number of input channels of the one-dimensional convolution layer 2 is set to 8, the number of output channels is set to 16, and the convolution kernel size is 2; the convolution kernel size of the one-dimensional maximum pooling layer is set to 2; the number of input features of the fully connected layer 1 is set to 16, and the number of output features is set to 8; the number of input features of the fully connected layer 2 is set to 8, and the number of output features is set to 1; the sigmoid function is selected as the activation function, the binary cross-entropy loss (BCELoss) is used as the loss function, the stochastic gradient descent (SGD) function or the adaptive moment estimation algorithm (Adam) function is used as the optimizer, the early stopping parameter is set to 10, the model is trained using the training set, and the model is evaluated using the test set. The average or median value of the probability values ​​predicted by the model is used as the final probability. The final probability range is between 0 and 1, and the probability threshold is set to 0.5. If the final probability is greater than 0.5, it means that the patient's expected treatment is effective. The closer the final probability value is to 1, the greater the chance that the patient's expected treatment will be effective.

[0100] In some exemplary embodiments, the training method of the disease analysis model provided by the embodiments of the present disclosure includes the following steps: dividing the human genome interval into equal distances to form several genome intervals; inputting cfDNA sequencing data, and for each genome interval, counting the number of short fragments and long fragments, and calculating the short fragment / long fragment ratio to obtain the differential genome interval; converting the 4bp base arrangement at the beginning of the 5' end of the read distributed in the differential genome interval into a one-hot encoding matrix, taking each matrix as a sample, and using the label information of the patient corresponding to the matrix as the label of the sample, and using the neural network method to construct the model.

[0101] The training method for the disease analysis model provided in the embodiments of the present disclosure greatly increases the number of samples by treating each read of the patient's cfDNA sequencing data as a sample and the corresponding patient marker as the sample marker, thereby making up for the problem of insufficient sample number in clinical model construction; in addition, by comprehensively considering the long and short fragment information and 5' end base arrangement information of cfDNA fragment omics, the long and short fragment information is used to obtain differential genomic intervals, and the read 5' end sequence information of the differential genomic interval is used to construct the model, thereby increasing the reliability of model construction.

[0102] As shown in FIG6 , the present disclosure also provides a disease analysis method, including:

[0103] Step 601: determining the differential genomic intervals to which multiple sequencing sequences in the cell-free DNA sequencing data belong;

[0104] Step 602: Obtain at least a partial sequence arrangement of each sequencing sequence within each differential genomic interval, encode the obtained sequence arrangement, and form a test data set, wherein the test data set includes a plurality of test data;

[0105] Step 603: Input a plurality of test data into the trained neural network model to obtain a plurality of model prediction values;

[0106] Step 604: perform data processing on multiple model prediction values ​​to obtain disease analysis results.

[0107] In the disclosed embodiment, in step 601, the differential genomic interval is the differential genomic interval determined during the training phase of the neural network model. When the trained neural network model is used for disease analysis, it is not necessary to redetermine the differential genomic interval. Taking the evaluation of treatment efficacy as an example, assuming that disease analysis is required for patient sample A, in step 601, the cell-free DNA sequencing data of patient sample A is first obtained, and the differential genomic interval determined during the training phase is obtained. The sequencing data of patient sample A is aligned to the reference genome, and it is determined which reads of patient sample A belong to the differential genomic interval, and the reads belonging to the differential genomic interval are selected.

[0108] In some exemplary embodiments, obtaining at least a partial sequence arrangement of each sequencing sequence within each differential genomic interval includes: obtaining the arrangement of the first 4 bases of the 5' end of each sequencing sequence within each differential genomic interval.

[0109] In some exemplary embodiments, encoding the acquired sequence arrangement comprises:

[0110] The first 4 bases at the 5' end are encoded as a 4*4 matrix, with each base corresponding to a row of the 4*4 matrix.

[0111] In some exemplary embodiments, the disease analysis result is at least one of the following: the probability that the patient has the disease, the probability that the patient is expected to be effectively treated, and the probability that there is residual disease.

[0112] In some exemplary embodiments, data processing is performed on multiple model prediction values ​​to obtain disease analysis results, including:

[0113] Calculate the mean or median of multiple model predictions;

[0114] The disease analysis results are obtained based on the calculated mean or median value.

[0115] Taking the evaluation of treatment efficacy as an example, the result of disease analysis can be that the treatment is effective or ineffective.

[0116] The training method of the neural network model in the embodiment of the present disclosure can be referred to as described above and will not be repeated here.

[0117] As shown in FIG7 , the embodiment of the present disclosure further provides a training device for a disease analysis model, comprising: a first segmentation module 701 , a difference interval determination module 702 , a first encoding module 703 , and a model training module 704 , wherein:

[0118] A first partitioning module 701 is configured to partition the reference genome into a plurality of genomic intervals and determine the genomic interval to which each sequencing sequence in the cell-free DNA sequencing data belongs;

[0119] The differential interval determination module 702 is configured to count the number of short fragments and the number of long fragments in each genomic interval according to the length of each sequencing sequence, and select at least some of the genomic intervals from the plurality of genomic intervals as differential genomic intervals, wherein the ratio of the number of short fragments to the number of long fragments in the differential genomic interval is greater than the ratio of the number of short fragments to the number of long fragments in the non-differential genomic interval;

[0120] A first encoding module 703 is configured to obtain at least a partial sequence arrangement of each sequencing sequence within each of the differential genomic intervals, encode the obtained sequence arrangement, and form a total data set;

[0121] The model training module 704 is configured to build a neural network model and use the total data set to train and evaluate the neural network model.

[0122] In some exemplary embodiments, the differential interval determining module 702 selects at least some of the genomic intervals from the plurality of genomic intervals as differential genomic intervals, comprising:

[0123] Calculate the fold difference of each genomic interval, fold difference = log2 (number of short fragments / number of long fragments);

[0124] Sort the multiple genomic intervals by their multiple difference scores from large to small, and set the top N genomic intervals as N differential genomic intervals, where N is a natural number greater than or equal to 1.

[0125] In some exemplary embodiments, the first encoding module 703 obtains at least a partial sequence arrangement of each sequencing sequence within each differential genomic interval, including: obtaining the arrangement of the first 4 bases of the 5' end of each sequencing sequence within each differential genomic interval.

[0126] In some exemplary embodiments, the first encoding module 703 encodes the acquired sequence arrangement, including:

[0127] The first 4 bases at the 5' end are encoded as a 4*4 matrix, with each base corresponding to a row of the 4*4 matrix.

[0128] In some exemplary embodiments, the neural network model includes one or more convolutional pooling modules and one or more fully connected layers;

[0129] Each convolution-pooling module includes a convolution layer and a pooling layer. The convolution layer is configured to extract feature information from the input data; the pooling layer is configured to compress the input data.

[0130] The fully connected layer is configured to map the feature information output by the convolutional pooling module into the output space to obtain the prediction result.

[0131] In some exemplary embodiments, the first partitioning module 701 performs a sliding operation on the reference genome with a window of a length a and a step size b to partition the reference genome into a plurality of genomic intervals, where a is greater than or equal to b.

[0132] In some exemplary embodiments, the first partitioning module 701 determines the genomic interval to which each sequencing sequence in the cell-free DNA sequencing data belongs, including:

[0133] Perform data quality control and cleaning on cell-free DNA sequencing data to obtain data to be compared;

[0134] Compare the data to be compared with the reference sequence of each genomic interval to obtain the chromosome number and start and end position information of each sequencing sequence;

[0135] Determine the intersection ratio of each sequencing sequence with one or more genomic intervals based on the chromosome number and start and end position information of each sequencing sequence;

[0136] The genomic interval to which each sequencing sequence belongs is determined based on the intersection ratio between each sequencing sequence and one or more genomic intervals.

[0137] In some exemplary embodiments, the first partitioning module 701 determines the genomic interval to which each sequencing sequence belongs based on the intersection ratio between each sequencing sequence and one or more genomic intervals, including:

[0138] When the intersection ratio between a sequencing sequence and a genomic interval is greater than 50%, the sequencing sequence is determined to fall within the genomic interval.

[0139] In some exemplary embodiments, the model training module 704 uses the total dataset to train and evaluate the neural network model, including:

[0140] Split the total dataset into training and testing sets;

[0141] The training set is used to train the neural network model, and the test set is used to evaluate the neural network model.

[0142] As shown in FIG8 , the embodiment of the present disclosure further provides a disease analysis device, comprising: a second division module 801 , a second encoding module 802 , and a prediction module 803 , wherein:

[0143] The second partitioning module 801 is configured to determine the differential genomic intervals to which multiple sequencing sequences in the cell-free DNA sequencing data belong;

[0144] The second encoding module 802 is configured to obtain at least a partial sequence arrangement of each sequencing sequence within each differential genomic interval, encode the obtained sequence arrangement, and form a test data set, where the test data set includes a plurality of test data;

[0145] The prediction module 803 is configured to input a plurality of test data into the trained neural network model to obtain a plurality of model prediction values; perform data processing on the plurality of model prediction values ​​to obtain disease analysis results.

[0146] In the disclosed embodiment, the second partitioning module 801 determines the differential genomic intervals to which multiple sequencing sequences in the cell-free DNA sequencing data belong, including: aligning the cell-free DNA sequencing data to the reference genome, determining which reads belong to the differential genomic intervals, and selecting the reads belonging to the differential genomic intervals.

[0147] In the embodiment of the present disclosure, the method by which the second encoding module 802 encodes the acquired sequence arrangement is similar to the method by which the first encoding module 703 encodes the acquired sequence arrangement. The specific implementation can be referred to above and will not be repeated here.

[0148] In some exemplary embodiments, the disease analysis result is at least one of the following: the probability that the patient has the disease, the probability that the patient is expected to be effectively treated, and the probability that there is residual disease.

[0149] In some exemplary embodiments, the prediction module 803 performs data processing on the multiple model prediction values ​​to obtain disease analysis results, including:

[0150] Calculate the mean or median of multiple model predictions;

[0151] The disease analysis results are obtained based on the calculated mean or median value.

[0152] An embodiment of the present disclosure also provides a training device for a disease analysis model, comprising a memory; and a processor connected to the memory, wherein the memory is used to store instructions, and the processor is configured to execute the steps of the training method for the disease analysis model as described in any embodiment of the present disclosure based on the instructions stored in the memory.

[0153] As shown in Figure 9, in one example, the training device of the disease analysis model may include: a first processor 910, a first memory 920, a first bus system 930 and a first transceiver 940, wherein the first processor 910, the first memory 920 and the first transceiver 940 are connected through the first bus system 930, the first memory 920 is used to store instructions, and the first processor 910 is used to execute the instructions stored in the first memory 920 to control the first transceiver 940 to send and receive signals. Specifically, the first transceiver 940 can obtain free DNA sequencing data and a reference genome under the control of the first processor 910. The first processor 910 divides the reference genome into multiple genomic intervals and determines the genomic interval to which each sequencing sequence in the free DNA sequencing data belongs; according to the length of each sequencing sequence, the number of short fragments and the number of long fragments in each genomic interval are counted, and at least part of the genomic intervals in the multiple genomic intervals are selected as differential genomic intervals, and the ratio of the number of short fragments to the number of long fragments in the differential genomic interval is greater than the ratio of the number of short fragments to the number of long fragments in the non-differential genomic interval; obtain at least a partial sequence arrangement of each sequencing sequence in each of the differential genomic intervals, encode the obtained sequence arrangement to form a total data set; construct a neural network model, and use the total data set to train and evaluate the neural network model.

[0154] It should be understood that the first processor 910 may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0155] The first memory 920 may include a read-only memory and a random access memory, and provides instructions and data to the first processor 910. A portion of the first memory 920 may also include a non-volatile random access memory. For example, the memory 720 may also store device type information.

[0156] In addition to the data bus, the first bus system 930 may also include a power bus, a control bus, a status signal bus, etc. However, for the sake of clarity, various buses are labeled as the first bus system 930 in FIG.

[0157] During the implementation process, the processing performed by the processing device can be completed by the hardware integrated logic circuit in the first processor 910 or by instructions in the form of software. That is, the method steps of the embodiment of the present disclosure can be embodied as being executed by a hardware processor, or being executed by a combination of hardware and software modules in the processor. The software module can be located in a storage medium such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the first memory 920, and the first processor 910 reads the information in the first memory 920 and completes the steps of the above method in combination with its hardware. To avoid repetition, it will not be described in detail here.

[0158] The present disclosure also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the disease analysis model training method described in any of the embodiments of the present disclosure. The method for driving the disease analysis model training by executing executable instructions is substantially the same as the disease analysis model training method provided in the aforementioned embodiments of the present disclosure and is not further described here.

[0159] In some possible embodiments, various aspects of the disease analysis model training method provided by the present disclosure can also be implemented in the form of a program product, which includes program code. When the program product is run on a computer device, the program code is used to enable the computer device to execute the steps of the disease analysis model training method according to various exemplary embodiments of the present disclosure described above in this specification. For example, the computer device can execute the disease analysis model training method recorded in the embodiments of the present disclosure.

[0160] An embodiment of the present disclosure also provides a disease analysis device, comprising a memory; and a processor connected to the memory, wherein the memory is used to store instructions, and the processor is configured to execute the steps of the disease analysis method described in any embodiment of the present disclosure based on the instructions stored in the memory.

[0161] As shown in FIG10 , in one example, a disease analysis apparatus may include: a second processor 1010, a second memory 1020, a second bus system 1030, and a second transceiver 1040. The second processor 1010, the second memory 1020, and the second transceiver 1040 are connected via the second bus system 1030. The second memory 1020 is configured to store instructions, and the second processor 1010 is configured to execute the instructions stored in the second memory 1020 to control the second transceiver 1040 to transmit and receive signals. Specifically, under the control of the second processor 1010, the second transceiver 1040 may acquire cell-free DNA sequencing data and differential genomes. The second processor 1010 may determine the differential genome intervals to which multiple sequencing sequences in the cell-free DNA sequencing data belong; acquire at least a partial sequence arrangement of each sequencing sequence within each differential genome interval; encode the acquired sequence arrangement to form a test data set, wherein the test data set includes multiple test data; input the multiple test data into a trained neural network model to obtain multiple model prediction values; and perform data processing on the multiple model prediction values ​​to obtain a disease analysis result.

[0162] It should be understood that the second processor 1010 may be a central processing unit (CPU), or may be another general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc.

[0163] The second memory 1020 may include a read-only memory and a random access memory, and provides instructions and data to the second processor 1010. A portion of the second memory 1020 may also include a non-volatile random access memory. For example, the second memory 1020 may also store device type information.

[0164] In addition to the data bus, the second bus system 1030 may also include a power bus, a control bus, a status signal bus, etc. However, for the sake of clarity, in FIG10 , various buses are labeled as the second bus system 1030 .

[0165] During the implementation process, the processing performed by the processing device can be completed by the hardware integrated logic circuit in the second processor 1010 or by instructions in the form of software. That is, the method steps of the embodiment of the present disclosure can be embodied as being executed by a hardware processor, or being executed by a combination of hardware and software modules in the processor. The software module can be located in a storage medium such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the second memory 1020, and the second processor 1010 reads the information in the second memory 1020 and completes the steps of the above method in combination with its hardware. To avoid repetition, it will not be described in detail here.

[0166] The present disclosure also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the disease analysis method described in any of the embodiments of the present disclosure. The method for driving disease analysis by executing executable instructions is substantially the same as the disease analysis method described in the aforementioned embodiments of the present disclosure and is not further described here.

[0167] In some possible embodiments, various aspects of the disease analysis method provided by the present disclosure can also be implemented in the form of a program product, which includes program code. When the program product is run on a computer device, the program code is used to enable the computer device to execute the steps of the disease analysis method according to various exemplary embodiments of the present disclosure described above in this specification. For example, the computer device can execute the disease analysis method recorded in the embodiments of the present disclosure.

[0168] The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection having one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0169] It will be appreciated by those skilled in the art that all or some of the steps, systems, and functional modules / units in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. In hardware implementations, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed by several physical components in cooperation. Some or all components may be implemented as software executed by a processor, such as a digital signal processor or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or temporary medium). As is well known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable, and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is well known to those skilled in the art that communication media generally embodies computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0170] It should be noted that the above-described embodiments or implementations are merely illustrative and not restrictive. Therefore, the present disclosure is not limited to what is specifically shown and described herein. Various modifications, substitutions, or omissions may be made to the forms and details of the implementations without departing from the scope of the present disclosure.

Claims

1. A method for training a disease analysis model, comprising: Divide the reference genome into multiple genomic intervals and determine the genomic interval to which each sequencing sequence in the free DNA sequencing data belongs; Counting the number of short fragments and the number of long fragments in each genomic interval according to the length of each sequencing sequence, and selecting at least some of the genomic intervals among the multiple genomic intervals as differential genomic intervals, wherein the ratio of the number of short fragments to the number of long fragments in the differential genomic interval is greater than the ratio of the number of short fragments to the number of long fragments in the non-differential genomic interval; Obtaining at least a partial sequence arrangement of each sequencing sequence within each of the differential genomic intervals, encoding the obtained sequence arrangement to form a total data set; A neural network model is constructed, and the neural network model is trained and evaluated using the total data set.

2. The method according to claim 1, wherein The selecting at least some of the genomic intervals among the plurality of genomic intervals as differential genomic intervals comprises: Calculate the difference fold of each genomic interval, where the difference fold = log2 (number of short fragments / number of long fragments); The difference multiple genomic intervals are sorted from large to small, and the genomic intervals ranked in the top N positions are set as N said difference genomic intervals, where N is a natural number greater than or equal to 1.

3. The method according to claim 1, wherein The obtaining of at least a partial sequence arrangement of each sequencing sequence within each of the differential genomic intervals includes obtaining the arrangement of the first four bases at the 5' end of each sequencing sequence within each of the differential genomic intervals.

4. The method according to claim 3, wherein: The encoding of the acquired sequence arrangement includes: The first 4 bases at the 5' end are arranged and encoded into a 4*4 matrix, with each base corresponding to a row of the 4*4 matrix.

5. The method according to claim 1, wherein The neural network model includes one or more convolutional pooling modules and one or more fully connected layers; Each of the convolution and pooling modules includes a convolution layer and a pooling layer, wherein the convolution layer is configured to extract feature information from the input data; and the pooling layer is configured to compress the input data; The fully connected layer is configured to map the feature information output by the convolutional pooling module into the output space to obtain a prediction result.

6. The method according to claim 1, wherein The reference genome is divided into a plurality of genomic intervals by sliding a window with a window length a and a step length b on the reference genome, where a is greater than or equal to b.

7. The method according to claim 1, wherein Determining the genomic interval to which each sequencing sequence in the cell-free DNA sequencing data belongs includes: Performing data quality control and cleaning on the cell-free DNA sequencing data to obtain data to be compared; Compare the data to be compared with the reference sequence of each genomic interval to obtain the sequence of each sequencing sequence. Chromosome number and start and end position information; Determine the intersection ratio of each sequencing sequence with one or more of the genomic intervals based on the chromosome number and start and end position information of each sequencing sequence; The genomic interval to which each sequencing sequence belongs is determined based on the intersection ratio between each sequencing sequence and one or more of the genomic intervals.

8. The method according to claim 7, wherein: Determining the genomic interval to which each sequencing sequence belongs based on the intersection ratio between each sequencing sequence and one or more genomic intervals includes: When the intersection ratio between a sequencing sequence and a genomic interval is greater than 50%, the sequencing sequence is determined to fall within the genomic interval.

9. The method according to claim 1, wherein The method of using the total data set to train and evaluate the neural network model includes: Splitting the total dataset into a training set and a test set; The neural network model is trained using the training set, and the neural network model is evaluated using the test set.

10. A training device for a disease analysis model, comprising a memory; and a processor connected to the memory, wherein the memory is used to store instructions, and the processor is configured to execute the steps of the training method for the disease analysis model as described in any one of claims 1 to 9 based on the instructions stored in the memory.

11. A computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the method for training a disease analysis model according to any one of claims 1 to 9 is implemented.

12. A disease analysis method comprising: Determine the differential genomic intervals to which multiple sequencing sequences in cell-free DNA sequencing data belong; Obtaining at least a partial sequence arrangement of each sequencing sequence within each differential genomic interval, encoding the obtained sequence arrangement to form a test data set, wherein the test data set includes a plurality of test data; Inputting the plurality of test data into a trained neural network model to obtain a plurality of model prediction values; Data processing is performed on the multiple model prediction values ​​to obtain disease analysis results.

13. The method according to claim 12, wherein: The disease analysis result is at least one of the following: the probability that the patient is ill, the probability that the patient's expected treatment will be effective, and the probability that there are residual lesions.

14. The method according to claim 12, wherein: The data processing of the multiple model prediction values ​​to obtain disease analysis results includes: Calculating the average or median of the multiple model prediction values; The disease analysis result is obtained based on the calculated mean value or median value.

15. A disease analysis device comprising a memory; and a processor connected to the memory, the memory being used to store instructions, the processor being configured to execute the steps of the disease analysis method according to any one of claims 12 to 14 based on the instructions stored in the memory. 16 . A computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the disease analysis method according to claim 12 is implemented.

17. A training device for a disease analysis model, comprising: A first division module, a difference interval determination module, a first encoding module and a model training module, wherein: The first partitioning module is configured to partition the reference genome into a plurality of genomic intervals and determine the genomic interval to which each sequencing sequence in the cell-free DNA sequencing data belongs; The difference interval determination module is configured to count the number of short fragments and the number of long fragments in each genomic interval according to the length of each sequencing sequence, and select at least some of the genomic intervals among the multiple genomic intervals as difference genomic intervals, wherein the ratio of the number of short fragments to the number of long fragments in the difference genomic interval is greater than the ratio of the number of short fragments to the number of long fragments in the non-differential genomic interval; The first encoding module is configured to obtain at least a partial sequence arrangement of each sequencing sequence within each of the differential genomic intervals, encode the obtained sequence arrangement, and form a total data set; The model training module is configured to construct a neural network model and use the total data set to train and evaluate the neural network model.

18. A disease analysis device comprising: A second partitioning module, a second encoding module and a prediction module, wherein: The second partitioning module is configured to determine the differential genomic intervals to which multiple sequencing sequences in the cell-free DNA sequencing data belong; The second encoding module is configured to obtain at least a partial sequence arrangement of each sequencing sequence in each differential genomic interval, encode the obtained sequence arrangement, and form a test data set, wherein the test data set includes a plurality of test data; The prediction module is configured to input the multiple test data into a trained neural network model to obtain multiple model prediction values; and perform data processing on the multiple model prediction values ​​to obtain disease analysis results.

Images