A method and system for eliminating false positives from label jumping contamination in NGS sequencing

By using Dual-indexing splitting process and feedforward neural network model in NGS sequencing, the false positive sample problem caused by label jump contamination is solved, the accuracy and efficiency of detection is improved, and diagnostic reports are reduced review workload.

CN118711657BActive Publication Date: 2024-11-12HANGZHOU MATRIDX BIOTECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411206931.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-30
Publication Date
2024-11-12
Estimated Expiration
2044-08-30

AI Technical Summary

Technical Problem

In NGS sequencing, label jump contamination leads to the generation of false positive samples, affecting the accuracy of pathogen detection and diagnostic reports. The existing technology is difficult to effectively avoid and deal with, resulting in high-cost review workload.

Method used

By evaluating the hopping pressure of each sample, using the data characteristics and mathematical model of the Dual-indexing splitting process, the prediction results of weakly positive samples were screened out, and the pollution pressure value was predicted by combining the feedforward neural network model, false positive samples were excluded, and diagnosis was improved. Report efficiency.

Benefits of technology

It effectively reduces the number of false positive samples, reduces the review workload of diagnostic reports, improves the accuracy and efficiency of testing, and saves public medical resources.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118711657B_ABST
    Figure CN118711657B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of bioinformatics, and in particular to a method and system for excluding false positives for label jumping contamination in NGS sequencing. The main steps of the method are: obtaining the number of labeled pathogenic detection sequences and the number of labeled intermediate sequences of each sample, and determining the strong positive sample and source value; calculating the predicted contamination pressure of the benchmark strong positive sample on all other weak positive samples, and obtaining the predicted contamination pressure value of each weak positive sample by testing the prediction model; then summing all the predicted contamination pressure values ​​of any weak positive sample to obtain the actual Hopping expected value of the weak positive sample; sequentially comparing the actual Hopping expected value of each weak positive sample with the number of labeled pathogenic detection sequences of the weak positive sample, and outputting the predicted result of the weak positive sample. The method can evaluate the Hopping pressure of the sample on other samples, give the positive / negative prediction results of each false positive sample, and help the diagnostic report personnel improve the reporting efficiency.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of bioinformatics, and in particular to a method and system for eliminating false positives caused by label jumping contamination in NGS sequencing. Background Art

[0002] NGS sequencer, or Next Generation Sequencing, is also known as a high-throughput sequencer. It is a sequencing device that can read the sequence of hundreds of thousands to millions of nucleic acid molecules in parallel at one time. During the sequencing process of the NGS sequencer, a phenomenon called hopping (label hopping) occurs, which is usually called "index hopping" or "barcode hopping". When performing NGS sequencing, in order to improve efficiency, libraries containing different indexes (labels) are usually mixed together and sequenced at the same time (library refers to a collection of short nucleic acid fragments obtained after a series of processing and preparation of DNA or RNA samples before sequencing). During the construction process, each library will be assigned a unique index sequence (i.e., index tag, used to distinguish different samples) so that the sequencing data can be split back into the corresponding library through the index tag after sequencing.

[0003] However, in some cases, the index tag of one sample may appear incorrectly in the sequencing data of another sample. This phenomenon is called Hopping, commonly known as "index hopping" or "barcode hopping". Hopping can cause the ends of reads (nucleic acid fragments) with pathogenic nucleic acid sequences in some positive samples to be connected to the index tags of other negative samples in mixed testing, making these negative samples identified as weak positive samples in statistical analysis.

[0004] The above problems will have serious consequences in the field of pathogen detection. Usually, the throughput of mNGS (non-targeted sequencing) is as high as 10 million sequences, and the throughput of tNGS (targeted NGS) is also more than 100,000. The existing Hopping rate greatly affects the technology's ability to distinguish weak positive samples, which can easily lead to misjudgments and incorrect diagnosis and treatment. In an NGS sequencing, there are usually a majority of negative samples and a relatively small number of positive samples. The above Hopping phenomenon will cause multiple negative samples to be identified as weak (false) positive samples. Diagnostic report personnel need to conduct a second review of all detected weak positive samples.

[0005] In the existing technology, high-quality reagents and strict aseptic operation are usually used to reduce the transfer and mismatch of index tags during the mixing of different libraries, so as to reduce the incidence of Hopping phenomenon. However, based on the current NGS technology, Hopping phenomenon can be reduced but cannot be avoided. In the field of NGS pathogen diagnosis, even if there are only a small number of incorrectly assigned sequences, the large number of weak positive samples generated still need to be further verified manually. In large-scale testing, the high cost of rechecking weak positive samples has brought great challenges to public medical defense. Summary of the invention

[0006] The purpose of the present invention is to provide a method and system for eliminating false positives for label jumping contamination in NGS sequencing. The method can evaluate the Hopping pressure of each sample from other samples, and provide positive / negative prediction results for each false positive sample by filtering out a large amount of Hopping background. The prediction results can help diagnostic reporting personnel improve reporting efficiency.

[0007] In order to achieve the above purpose, the technical solution adopted by this method is as follows:

[0008] A method for eliminating false positives due to label jumping contamination in NGS sequencing, characterized in that the method comprises:

[0009] S1. Perform NGS mixed detection on all samples to be tested to obtain the number of labeled pathogenic detection sequences and the number of labeled intermediate sequences of each sample; define the sample with the number of labeled pathogenic detection sequences greater than or equal to the strong positive judgment threshold as a strong positive sample, and define this number as the source of the strong positive sample;

[0010] S2. Select a benchmark strong positive sample and obtain the predicted contamination pressure of the benchmark strong positive sample on all other weak positive samples: package the number of labeled intermediate sequences of any weak positive sample and the source value of the benchmark strong positive sample as a predicted input set of the weak positive sample; bring the predicted input set into the test prediction model to obtain the predicted contamination pressure value of the weak positive sample; repeat the input until the predicted contamination pressure values ​​of all weak positive samples on the benchmark strong positive sample are obtained;

[0011] S3, obtaining different predicted contamination pressure values ​​of all weak positive samples for each benchmark strong positive sample; summing up all predicted contamination pressure values ​​of any weak positive sample to obtain the actual Hopping expected value of the weak positive sample;

[0012] S4. Compare the actual Hopping expected value of each weak positive sample with the number of labeled pathogenic detection sequences of the weak positive sample in turn. If the former is greater than or equal to the latter, the prediction result of the weak positive sample is output as negative; if the former is less than the latter, the prediction result of the weak positive sample is output as positive.

[0013] Therefore, in actual NGS sequencing, all samples to be tested are sequenced on the same detection chip, and libraries need to be prepared one by one before testing. The long-chain nucleic acid molecules in each sample need to be first broken into a large number of short sequence fragment reads by enzymes, and then an index tag for specific identity recognition is added to both ends of the short sequence fragment reads (F end and R end) to form a labeled sequence, which is then assembled into a library, and then all libraries are mixed for sequencing.

[0014] Due to the Hopping phenomenon, the index tags at both ends of different samples may be incorrectly exchanged, so that the reads with pathogenic nucleic acid sequences are connected to the index tags of negative samples at both ends. When detected later, it will appear that pathogenic nucleic acid sequences are detected in multiple samples. However, since the Hopping rate is usually between 0.1% and 2%, the number of short sequence fragment reads with pathogenic nucleic acid sequences in the above samples generated based on the Hopping phenomenon is small, usually less than 1000 (strong positive judgment threshold); similarly, for true positive samples, the number of short sequence fragment reads with pathogenic nucleic acid sequences detected will obviously or even far exceed the strong positive judgment threshold, and the sample can be judged as a strong positive sample at this time. The above-mentioned weakly positive samples may be true positive samples or negative samples (which are contaminated by Hopping of strong positive samples participating in mixed tests at the same time). Diagnostic report personnel often need to re-examine all the weakly positive samples mentioned above, which is a huge workload.

[0015] This method is implemented on the basis of multiple weakly positive samples obtained after NGS sequencing.

[0016] Use A i To represent the index labels at both ends of different weak positive samples, use Sa i To represent the reads of different weak positive samples; use B j To represent the index tags at both ends of different strong positive samples, use Sb jto represent the reads of different strong positive samples (in one NGS mixed test, there may be multiple different strong positive samples, such as samples containing a large number of tuberculosis pathogenic nucleic acid sequences, samples containing a large number of Staphylococcus aureus pathogenic nucleic acid sequences, samples containing a large number of Klebsiella pneumoniae pathogenic nucleic acid sequences, and samples containing a large number of Acinetobacter baumannii pathogenic nucleic acid sequences, etc.), where i represents different weak positive samples and j represents different strong positive samples.

[0017] Based on the data characteristics of the dual-indexing splitting process, in an NGS test, by reading the data of the NGS sequencing chip, without the Hopping phenomenon, the following 10 types of reads connected with index tags can be directly obtained: (B1: Sb1: B1), (A1: Sa1: A1), (A2: Sa2: A2), (A3: Sa3: A3), (A4: Sa4: A4), (A5: Sa5: A5), (A6: Sa6: A6), (A7: Sa7: A7), (A8: Sa8: A8) and (A9: Sa9: A9).

[0018] For example, after a body fluid sample of a pulmonary tuberculosis patient is made into a library, all reads in the library connected to the index tag are represented by (B1: Sb1: B1), and if Sb1 in the middle of (B1: Sb1: B1) contains a tuberculosis pathogenic nucleic acid sequence, it is represented by (B1: Sb1*: B1) (not all Sb1 in a strongly positive sample will carry a pathogenic nucleic acid sequence). In the remaining 9 sample libraries mixed with the pulmonary tuberculosis patient, reads are represented by Sa1~Sa9 respectively.

[0019] In actual situations, due to the existence of Hopping, abnormal exchange of index labels on one end or both ends may lead to (A i :Sa i : A i )、(B j :Sb j :B j )、(A i :Sb j : A i )、(B j :Sa i :B j )、(A i :Sb j :B j )、(B j :Sb j : A i )、(A i :Sa i :Bj ) or (B j :Sa i : A i ) The first two types are original short sequence fragment reads without hopping, the second and third types are completely erroneous short sequence fragment reads with hopping at both ends, and the last four types are partially erroneous short sequence fragment reads with hopping at one end.

[0020] Furthermore, there will be (B j :Sb j *: B j )、(A i :Sb j *: A i )、(A i :Sb j *: B j ) and (B j :Sb j *: A i The pathogenic detection sequence with label is (A i :Sb j *: A i ) or (B j :Sb j *: B j ), among all the weak positive samples, the former has a small number of detections, usually below 1000 (the threshold for strong positive determination), so the weak positive sample is very likely to be affected by Sb j *True negative samples contaminated by Hopping; the latter are detected in larger numbers, usually significantly higher than 1000 (strong positive judgment threshold).

[0021] Based on the data characteristics of the dual-indexing splitting process, a mathematical model for Hopping is established to summarize and define the quantity of the above-mentioned short sequence fragment reads that may appear, specifically:

[0022] A i :Sb j *: B j ) and (A i :Sa i :B j ) is defined as (A i xB j )

[0023] B j :Sb j *: A i ) and (B j :Sa i : A i) is defined as (B j xA i )

[0024] A i :Sa i : A i ) and (A i :Sb j *: A i ) is defined as (A i xA i )

[0025] B j :Sb j *: B j ) and (B j :Sa i :B j ) is defined as (B j xB j )

[0026] Furthermore, (A i :x:B j ) and (A i :x:A i ) is defined as (A i xX) one of the following;

[0027] Furthermore, (B j :x:A i ) and (A i :x:A i ) is defined as (XxA i )

[0028] Furthermore, (B j :x:A i ) and (B j :x:B j ) is defined as (B j xX) one of the following;

[0029] Furthermore, (A i :x:B j ) and (B j :x:B j ) is defined as (XxB j )

[0030] Where x represents any reads, which can be Sa i Can also be Sb j *; X represents any index tag, which can be A i Can also be B j.

[0031] Since Hopping is random, there will be (A i xB j ) and (B j xA i ) This type of direct intermediate; Since different index labels between different weak positive / strong positive samples may also jump, so further induction, there is also (A i xX)、(XxA i )、(B j xX) and (XxB j ) In the existing traditional analysis process, researchers often do not pay attention to the number of the above intermediates. In this method, the data of the intermediates that were originally discarded are extracted, and the data are defined, classified and counted, and input into the model as input parameters to finally obtain the prediction results.

[0032] Specifically, through NGS detection, a detection chip can be obtained, the above (A i xX)、(XxA i )、(B j xX)、(XxB j )、(A i xB j ) and (B j xA i ) their respective quantities.

[0033] Next, a benchmark strong positive sample is selected, for example, j=1, Sb1* represents the short sequence fragment reads with the pathogenic nucleic acid sequence of Mycobacterium tuberculosis. After NGS detection, the number of labeled pathogenic detection sequences of the benchmark strong positive sample can be obtained, that is, the number of (B1: Sb1*: B1), which is defined as the source of the strong positive sample.

[0034] Afterwards, for the weak positive sample i=1, for the selected benchmark strong positive sample, through NGS detection, the number of 6 labeled intermediate sequences of the weak positive sample can be obtained, namely (A1xX), (XxA1), (B1xX), (XxB1), (A1xB1) and (B1xA1); the 7 values ​​of the number of the above 6 intermediate sequences and the number of sources of the selected benchmark strong positive sample are packaged as the prediction input set of the weak positive sample, and the prediction input set is brought into the obtained test prediction model to obtain the predicted contamination pressure value of the weak positive sample. This predicted contamination pressure value reflects the Hopping pressure of the benchmark strong positive sample on the weak positive sample i=1, which is the number of (A1:Sb1*:A1) that the weak positive sample i=1 will produce when it is affected by the Hopping of this benchmark strong positive sample.

[0035] Among them, the above-mentioned predicted contamination pressure value, that is, the number of (A1: Sb1*: A1) is obtained through a test prediction model. The model can select the feedforward neural network model MLP in the prior art and use the TanH (hyperbolic tangent function) activation function to implement it. The model is trained by introducing a training data set in advance, and the test prediction model after training is obtained. When obtaining the predicted contamination pressure value of any weak positive sample for any benchmark strong positive sample, the prediction input set formed by the above-mentioned 7 values ​​of the number of 6 labeled intermediate sequences and the number of sources of the benchmark strong positive sample is input into the model to obtain the prediction result.

[0036] In addition, there may be multiple different microbial pathogen sequences in a mixed sequencing, such as multiple samples of Mycobacterium tuberculosis, Staphylococcus aureus, and Klebsiella pneumoniae. This method predicts all strong positive samples of a specific pathogenic microorganism each time. For example, for Mycobacterium tuberculosis, assuming that there are 3 strong positive samples of Mycobacterium tuberculosis in a sequencing, the above 3 strong positive samples are used as benchmark strong positive samples in turn, and the Hopping pressure for each weak positive sample is calculated respectively. For example, the corresponding three libraries (B1: Sb1*: B1), (B2: Sb2*: B2) and (B3: Sb3*: B3) will produce Hopping pressure for any weak positive sample, so it is necessary to repeat the above model prediction process to obtain the three different predicted contamination pressure values ​​of all weak positive samples for each benchmark strong positive sample (the above three strong positive samples of Mycobacterium tuberculosis), and then sum up the above three different predicted contamination pressure values ​​to obtain the actual Hopping expected value of a weak positive sample. This actual Hopping expected value reflects the total contamination pressure of all strong positive samples of a specific pathogenic microorganism on the weak positive sample in the same sequencing, that is, the number of false positive samples predicted by this method (A1: Sb j *:A1), where j=1, 2, 3.

[0037] Next, the actual Hopping expected value of each weak positive sample is compared with the number of labeled pathogenic detection sequences of the weak positive sample. For example, for sample i=1, that is, the comparison (A1:Sb j *: A1) predicted value and NGS detection value. If the former is greater than or equal to the latter, the predicted result of the weak positive sample is output as predicted negative; if the former is less than the latter, the predicted result of the weak positive sample is output as predicted positive.

[0038] Finally, through the above method, by predicting a large number of false positive samples as negative samples, and with the help of the model's own high prediction accuracy, it can help the diagnosis report personnel exclude a large number of cases where the true negative is contaminated and shows weak positive due to the Hopping phenomenon. The diagnosis report personnel only need to conduct a second review of the samples that actually show false positives and the predicted results are also positive, which greatly saves the review workload.

[0039] In addition, for other strongly positive samples of multiple different pathogenic microorganisms that may exist in the same NGS sequencing, predictions need to be made for the pathogenic microorganisms separately, that is, each prediction is only for all strongly positive samples of one of the pathogenic microorganisms.

[0040] As a preferred embodiment of the present invention, the test prediction model is a feedforward neural network model MLP, and the feedforward neural network model MLP comprises an input layer, a hidden layer and an output layer.

[0041] MLP is a feedforward neural network consisting of multiple neuron layers, including an input layer, one or more hidden layers, and an output layer. The neurons in each layer are fully connected to all neurons in the next layer, and information is transmitted through weights and biases. MLP contains at least three layers of network structure, namely the input layer, hidden layer (there can be multiple layers), and output layer. The input layer is responsible for receiving raw data; the hidden layer is responsible for extracting intermediate features of the data, which is the most critical part of MLP. Complex nonlinear relationships can be learned through the combination of multiple hidden layers; the output layer is responsible for outputting the final prediction results. When given a set of input data, MLP processes it by forward propagation. Each neuron calculates the weighted sum of its input and weight, transforms it through a nonlinear activation function, and then passes the output to the next layer of neurons. This process continues until the output layer, producing the final prediction result.

[0042] As a preferred embodiment of the present invention, in the feedforward neural network model MLP, the hidden layer uses 3 TanH activation functions and 1 linear fitting function to extract and analyze features of input data.

[0043] Activation functions play a vital role in MLP. They introduce nonlinear factors to neurons, allowing MLP to learn and represent complex nonlinear functions. Common activation functions include Sigmoid, ReLU (Rectified Linear Unit), Tanh, etc. In this method, the hidden layer of the feedforward neural network model MLP uses three TanH activation functions and one linear fitting function. The TanH activation function is a hyperbolic tangent function, TanH(x)=(ex – ex) / (ex +ex), where e is the natural logarithm, taking the first 5 decimal places, that is, e=2.72818; x is the input value. The Tanh function is nonlinear and can introduce nonlinear factors, so that the neural network can learn the seven input values ​​of the number of 6 labeled intermediate sequences of weak positive samples, namely (A1xX), (XxA1), (B1xX), (XxB1), (A1xB1) and (B1xA1) and the number of sources of the above selected benchmark strong positive samples. The linear fitting function can learn and analyze linear factors. One linear fitting function performs further analysis and calculation based on the output values ​​of the above three TanHs, which satisfies the analysis and prediction of the above input values ​​by this method.

[0044] As a preferred embodiment of the present invention, the value range of the strong positive determination threshold is [1000, 10000].

[0045] According to the early simulation calculations and experimental historical data of this method, [1000, 10000] is selected as the value range of the strong positive judgment threshold. The specific value is affected by factors such as the sample type involved in NGS sequencing and the type of pathogenic nucleic acid sequence, and can be preferably selected from values ​​such as 1000, 3000, 5000, etc.

[0046] As a preferred embodiment of the present invention, step S3 further includes a Cutoff threshold acquisition step: respectively adding the actual Hopping expected value of each weak positive sample to the range adjustment parameter k to obtain the Cutoff threshold.

[0047] Since the predicted negative / positive output by this method is a predicted value, according to data research, after arranging all weak positive samples in turn according to the size of the actual Hopping expected value, it is found that the number of labeled pathogenic detection sequences of each weak positive sample shows an overall fluctuating distribution trend. In order to maximize the false positive filtering rate of this method while meeting the prediction accuracy, the range adjustment parameter k is introduced, and the range adjustment parameter k is added to the actual Hopping expected value of each weak positive sample to obtain the Cutoff threshold of each weak positive sample. The Cutoff threshold is more meaningful for filtering than the original actual Hopping expected value.

[0048] As a preferred embodiment of the present invention, in step S4, the Cutoff threshold of each weak positive sample is compared with the number of labeled pathogenic detection sequences of the weak positive sample in turn. If the former is greater than or equal to the latter, the prediction result of the weak positive sample is output as a predicted negative; if the former is less than the latter, the prediction result of the weak positive sample is output as a predicted positive.

[0049] The Cutoff threshold is used to compare with the number of labeled pathogenic detection sequences corresponding to it. If the former is greater than or equal to the latter, the prediction result of the weak positive sample is output as predicted negative; if the former is less than the latter, the prediction result of the weak positive sample is output as predicted positive. In this way, most of the distribution points fall below the Cutoff threshold, which increases the false positive filtering rate.

[0050] As a preferred embodiment of the present invention, the value range of the range adjustment parameter k is [3, 7].

[0051] The value range of the range adjustment parameter k can be determined by experimental methods or adaptively adjusted according to the experience of the inspector, so that the Cutoff threshold has a higher screening rate as much as possible while ensuring the prediction accuracy.

[0052] As a preferred embodiment of the present invention, the value of the range adjustment parameter k is 5.

[0053] When the range adjustment parameter k is set to 5, the sensitivity and specificity values ​​are the highest.

[0054] A false positive exclusion system for label jumping contamination in NGS sequencing, characterized in that the system comprises:

[0055] The sample differentiation and data acquisition module is used to perform NGS mixed detection on all samples to be tested, and obtain the number of labeled pathogenic detection sequences and the number of labeled intermediate sequences of each sample; the sample with the number of labeled pathogenic detection sequences greater than or equal to the strong positive judgment threshold is defined as a strong positive sample, and this number is defined as the source of the strong positive sample;

[0056] A predicted contamination pressure value acquisition module is used to package the number of labeled intermediate sequences of any weak positive sample and the source value of the benchmark strong positive sample into a predicted input set of the weak positive sample; bring the predicted input set into the test prediction model to obtain the predicted contamination pressure value of the weak positive sample; repeat the input until the predicted contamination pressure values ​​of all weak positive samples for the benchmark strong positive sample are obtained;

[0057] The actual Hopping expected value calculation module is used to obtain different predicted contamination pressure values ​​of all weak positive samples for each benchmark strong positive sample; sum up all predicted contamination pressure values ​​of any weak positive sample to obtain the actual Hopping expected value of the weak positive sample;

[0058] The prediction result output module is used to compare the actual Hopping expected value of each weak positive sample with the number of labeled pathogenic detection sequences of the weak positive sample in turn. If the former is greater than or equal to the latter, the prediction result of the weak positive sample is output as negative; if the former is less than the latter, the prediction result of the weak positive sample is output as positive.

[0059] In summary, the present invention has the following beneficial effects:

[0060] 1. By predicting a large number of false positive samples as negative samples and leveraging the model's own high prediction accuracy, it can help diagnostic report personnel exclude a large number of cases where true negative samples may be contaminated and show weak positives due to the Hopping phenomenon. Diagnostic report personnel only need to conduct a second review of samples that actually show false positives and are also predicted to be positive, which greatly saves the review workload.

[0061] 2. Introduce the range adjustment parameter k, and add the range adjustment parameter k to the actual Hopping expected value of each weak positive sample to obtain the Cutoff threshold of each weak positive sample. The Cutoff threshold is more meaningful for filtering than the original actual Hopping expected value. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 This is a flowchart of the false positive exclusion method for index jumping contamination in NGS sequencing;

[0063] Figure 2 This is a system block diagram of the false positive exclusion system for index jumping contamination in NGS sequencing;

[0064] Figure 3 This is a structural diagram of the inspection prediction model of Example 1. DETAILED DESCRIPTION

[0065] The preferred embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Anyone can implement the present disclosure in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to make the present disclosure more thorough and complete, and to fully convey the scope of the present disclosure to those skilled in the art.

[0066] In this embodiment 1, according to Figure 1 The method shown in the figure is used to exclude false positives. All samples to be tested are 48 samples in one chip that are actually tested after NGS testing, of which 7 are negative samples (No. 02 to No. 08), 40 are weakly positive samples (No. 01; No. 09 to No. 36; No. 38 to No. 48) and 1 is strongly positive (No. 37), as shown in Table 1. It is known that sample No. 37 is strongly positive for TB. The specific judgment process is that the number of labeled pathogenic detection sequences of sample No. 37 is 594667. Assuming that the value of the strong positive judgment threshold corresponding to the detection chip is 1000, 594667>1000, so sample No. 37 is defined as a strong positive sample, and the source value is 594667. For negative samples 02 to 08, the number of labeled pathogenic detection sequences is 0, so they are defined as negative samples; for the 40 samples 01, 09 to 38, and 38 to 48, the number of labeled pathogenic detection sequences is not 0, and is less than the strong positive judgment threshold of 1000, so samples 01, 09 to 36, and 38 to 48 are weakly positive samples. Then, the above 40 weakly positive samples are predicted, and the decision on whether to retest in the next step is made based on the negative / positive prediction results.

[0067] Table 1 Weakly positive sample detection table

[0068] Sample No. Number of labeled pathogenic sequences Review and verification results Model prediction results Cutoff Threshold source value <![CDATA[(A i xX) quantity]]> <![CDATA[(XxA i ) Quantity]]> <![CDATA[(B j xX) quantity]]> <![CDATA[(XxB j ) Quantity]]> <![CDATA[(A i xB j ) Quantity]]> <![CDATA[(B j xA i ) Quantity]]> 01 6 Clinical review negative Predict negative 8.16 594667 1638912 1632601 1242707 1230403 122 945 09 1 Clinical review negative Predict negative 2.81 594667 1482726 1482395 1242707 1230403 80 343 10 4 Clinical review negative Predict negative 5.21 594667 1128114 1127971 1242707 1230403 58 284 11 6 Clinical review negative Predict negative 3.11 594667 1215208 1214146 1242707 1230403 65 358 12 2 Clinical review negative Predict negative 2.12 594667 1346379 1344317 1242707 1230403 67 417 13 5 Clinical review negative Predict negative 3.24 594667 1461374 1462178 1242707 1230403 63 357 14 2 Clinical review negative Predict negative 1.82 594667 1563027 1562107 1242707 1230403 87 413 15 1 Clinical review negative Predict negative 2.82 594667 1373553 1372975 1242707 1230403 76 357 16 3 Clinical review negative Predict negative 5.65 594667 1422746 1423951 1242707 1230403 181 596 17 5 Clinical review negative Predict negative 5.93 594667 1246970 1246317 1242707 1230403 62 256 18 4 Clinical review negative Predict negative 5.20 594667 1826688 1824390 1242707 1230403 49 327 19 1 Clinical review negative Predict negative 1.81 594667 1161484 1160425 1242707 1230403 58 430 20 4 Clinical review negative Predict negative 1.09 594667 1272939 1270705 1242707 1230403 53 562 21 4 Clinical review negative Predict negative 3.85 594667 1649734 1649509 1242707 1230403 73 306 22 3 Clinical review negative Predict negative 4.92 594667 1499770 1498950 1242707 1230403 63 293 23 3 Clinical review negative Predict negative 3.44 594667 1476273 1473503 1242707 1230403 74 324 24 4 Clinical review negative Predict negative 2.07 594667 1225841 1222494 1242707 1230403 72 552 25 3 Clinical review negative Predict negative 3.33 594667 1093037 1094633 1242707 1230403 178 399 26 5 Clinical review negative Predict negative 4.59 594667 1337617 1338045 1242707 1230403 75 275 27 6 Clinical review negative Predict negative 3.35 594667 1308225 1306160 1242707 1230403 52 379 28 2 Clinical review negative Predict negative 4.97 594667 1395475 1392944 1242707 1230403 97 509 29 9 Clinical review negative Predict negative 7.95 594667 1492893 1495243 1242707 1230403 181 603 30 3 Clinical review negative Predict negative 3.84 594667 1714970 1717386 1242707 1230403 135 324 31 5 Clinical review negative Predict negative 2.48 594667 1777398 1776678 1242707 1230403 92 315 32 18 Clinical review negative Predict negative 16.55 594667 1600793 1600137 1242707 1230403 173 492 33 8 Clinical review negative Predict positive 2.74 594667 1446801 1445706 1242707 1230403 117 917 34 2 Clinical review negative Predict negative 7.48 594667 1060636 1060854 1242707 1230403 56 209 35 8 Clinical review negative Predict negative 7.99 594667 4776311 4772337 1242707 1230403 217 889 36 2 Clinical review negative Predict negative 1.40 594667 1827039 1825282 1242707 1230403 105 368 38 2 Clinical review negative Predict negative 2.03 594667 1605718 1603485 1242707 1230403 95 378 39 27 Clinical review positive Predict positive 11.35 594667 1416431 1417527 1242707 1230403 154 532 40 3 Clinical review negative Predict negative 0.50 594667 1051759 1052374 1242707 1230403 93 345 41 4 Clinical review negative Predict negative 6.44 594667 1471515 1474067 1242707 1230403 87 177 42 5 Clinical review negative Predict negative 4.53 594667 1681514 1683496 1242707 1230403 86 248 43 1 Clinical review negative Predict negative 8.72 594667 1231134 1231335 1242707 1230403 52 192 44 1 Clinical review negative Predict negative 1.22 594667 1333539 1333481 1242707 1230403 204 598 45 2 Clinical review negative Predict negative 2.34 594667 1613816 1614708 1242707 1230403 104 371 46 3 Clinical review negative Predict negative 2.41 594667 1020908 1032578 1242707 1230403 306 234 47 3 Clinical review negative Predict negative 2.80 594667 1471449 1468901 1242707 1230403 69 364 48 5 Clinical review negative Predict negative 4.41 594667 773111 775232 1242707 1230403 198 295

[0069] The benchmark strong positive sample No. 37 has been selected, and the predicted contamination pressure of sample No. 37 on all other weak positive samples is obtained. According to the results of the detection chip, the number of labeled intermediate sequences corresponding to each weak positive sample No. 01, No. 09 to No. 36, and No. 38 to No. 48 is obtained, that is, (A i xX)、(XxA i )、(B j xX)、(XxB j )、(A i xB j ) and (B j xA i ) respectively; the values ​​of i are 01, 09, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47 and 48 respectively; the value of j is 37. As shown in Table 1, for example, for sample 01, the results of the detection chip show that the number of its labeled pathogenic detection sequences is 6, and the number of its labeled intermediate sequences is (A 01 xX) is 1638912, (XxA 01 ) is 1632601, (B 37 xX) is 1242707, (XxB 37 ) is 1230403, (A 01 xB 37 ) is 122, (B 37 xA 01 ) is 945. And it is known that the source value is 594667 labeled pathogenic detection sequences of the benchmark strong positive sample No. 37, so the above 6 intermediate sequence quantity values ​​and source values ​​of the weak positive sample No. 01, a total of 7 data, are packaged as the prediction input set of the weak positive sample No. 01.

[0070] In this embodiment, Figure 3 As shown in the figure, the test prediction model is a feedforward neural network model MLP, which includes an input layer, a hidden layer and an output layer. The hidden layer uses three TanH activation functions and one linear fitting function to extract and analyze the input data. The three TanH activation functions are the hyperbolic tangent function TanH(x)=(ex–ex) / (ex +ex), where e is the natural logarithm, taking the first 5 decimal places, that is, e=2.72818. The neural model is obtained by calculating the simulated sample data, and its prediction formula is:

[0071] H1_1_1 = TanH{10481.2990590665+(-0.0160097822355114)*(A i xB j )+0.0000003440660243193 *(A i xX)+(-0.00380336143440503 )*(B j xA i ) + (-0.00716596268814964 )*(B j xX)+ 0.00541562739510937 *source +0.0000003698754670991 *(XxA i )+( -0.00389606387574936 )*(XxB j )};

[0072] H1_2_1 = TanH{(-1102.18297297801) +( -0.00266805966119159)*(A i xB j )+0.0000000692870223697 *(A i xX)+(-0.000901520957513312)*(B j xA i )+0.00151754333270918 *(B j xX)+(-0.000767339189925724)* source + (-0.0000000296501853509) * (XxA i ) + (-0.000265654497086211 )*(XxB j )};

[0073] H1_3_1 = TanH(507.64919739075+0.0204329985096802*(A i xB j )+ (-0.0000022236231360201)*(A i xX)+0.00441048472618185*(B j xA i )+ (-0.00584724608508914)*(B jxX) + 0.00183036662074535 *source + (-0.0000022103841716681) * (XxA i ) + 0.00460896265525032 * (XxB j )};

[0074] Then use the linear fitting function to obtain the predicted pollution pressure value:

[0075] Lm(1) = predicted pollution pressure value = 9.98338721822971 + (-22.8380508568465) * (H1_1_1) + 41.8875507178956 * (H1_2_1) -11.1075699029386 * (H1_3_1).

[0076] The predicted input set of the above-mentioned weak positive sample No. 01 is brought into the above-mentioned test prediction model to obtain the predicted contamination pressure value of the weak positive sample No. 01; the predicted contamination pressure value of the weak positive sample No. 01 is 3.16; according to the above method, the predicted contamination pressure values ​​of the remaining 39 weak positive samples are obtained in turn.

[0077] Since there is only one strong positive sample, sample No. 37, in this embodiment, the actual Hopping expected value of each weak positive sample is equal to its respective predicted contamination pressure value, that is, the actual Hopping expected value of weak positive sample No. 01 is 3.16.

[0078] Next, the range adjustment parameter k is introduced. In this embodiment, the range adjustment parameter k is set to 5, and the cutoff threshold of the weak positive sample No. 01 is 3.16+5=8.16. The cutoff thresholds of 40 weak positive samples are obtained in sequence.

[0079] Afterwards, the actual Hopping expected values ​​of each of the 40 weak positive samples are compared with the number of labeled pathogenic detection sequences of the weak positive samples. If the former is greater than or equal to the latter, the prediction result of the weak positive sample is output as negative prediction; if the former is less than the latter, the prediction result of the weak positive sample is output as positive prediction. As shown in Table 1, among the 40 weak positive samples, the number of labeled pathogenic detection sequences of 38 samples is less than their actual Hopping expected values, so the prediction result is negative, and the number of labeled pathogenic detection sequences of 2 samples (No. 33 and No. 39) is greater than or equal to their actual Hopping expected values, and the prediction result is positive.

[0080] Finally, the diagnostic report personnel only need to review samples No. 33 and No. 39, which greatly saves the review workload.

[0081] In this embodiment, according to the actual clinical review results, sample No. 33 was predicted to be positive, and the review result was negative; sample No. 39 was predicted to be positive, and the review result was positive. From the perspective of prediction accuracy, 38 predictions were negative, all were correct, and the negative prediction accuracy was 100%.

[0082] In another embodiment, Figure 2 As shown, a false positive exclusion system for label jumping contamination in NGS sequencing is also disclosed, the system comprising:

[0083] The sample differentiation and data acquisition module is used to perform NGS mixed detection on all samples to be tested, and obtain the number of labeled pathogenic detection sequences and the number of labeled intermediate sequences of each sample; the sample with the number of labeled pathogenic detection sequences greater than or equal to the strong positive judgment threshold is defined as a strong positive sample, and this number is defined as the source of the strong positive sample;

[0084] A predicted contamination pressure value acquisition module is used to package the number of labeled intermediate sequences of any weak positive sample and the source value of the benchmark strong positive sample into a predicted input set of the weak positive sample; bring the predicted input set into the test prediction model to obtain the predicted contamination pressure value of the weak positive sample; repeat the input until the predicted contamination pressure values ​​of all weak positive samples for the benchmark strong positive sample are obtained;

[0085] The actual Hopping expected value calculation module is used to obtain different predicted contamination pressure values ​​of all weak positive samples for each benchmark strong positive sample; sum up all predicted contamination pressure values ​​of any weak positive sample to obtain the actual Hopping expected value of the weak positive sample;

[0086] The prediction result output module is used to compare the actual Hopping expected value of each weak positive sample with the number of labeled pathogenic detection sequences of the weak positive sample in turn. If the former is greater than or equal to the latter, the prediction result of the weak positive sample is output as negative; if the former is less than the latter, the prediction result of the weak positive sample is output as positive.

[0087] The sample differentiation and data acquisition module, the predicted contamination pressure value acquisition module, the actual Hopping expected value calculation module and the prediction result output module are interconnected. The predicted contamination pressure value acquisition module receives data from the sample differentiation and data acquisition module, brings it into the test prediction model for prediction, and obtains the predicted contamination pressure value of each weak-positive sample; the actual Hopping expected value calculation module calculates the actual Hopping expected value of each weak-positive sample and sends the actual Hopping expected value to the prediction result output module, and the prediction result output module outputs the prediction result of the weak-positive sample according to the actual Hopping expected value and the number of labeled pathogenic detection sequences of each weak-positive sample.

[0088] The system can help diagnostic report personnel exclude a large number of cases where true negatives are contaminated and appear weakly positive due to the Hopping phenomenon. Diagnostic report personnel only need to conduct a second review of samples that actually appear as false positives and are also predicted to be positive, which greatly saves the review workload.

[0089] A number of embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method for eliminating false positives from label jumping contamination in NGS sequencing, characterized in that: include: S1. Perform NGS mixed detection on all samples to be tested to obtain the number of labeled pathogenic detection sequences and the number of labeled intermediate sequences of each sample; The samples whose number of labeled pathogenic detection sequences is greater than or equal to the strong positive judgment threshold are defined as strong positive samples, and this number is defined as the source of the strong positive sample; S2. Select a benchmark strong positive sample and obtain the predicted contamination pressure of the benchmark strong positive sample on all other weak positive samples: package the number of labeled intermediate sequences of any weak positive sample and the source value of the benchmark strong positive sample as a predicted input set of the weak positive sample; bring the predicted input set into the test prediction model to obtain the predicted contamination pressure value of the weak positive sample; repeat the input until the predicted contamination pressure values ​​of all weak positive samples on the benchmark strong positive sample are obtained; S3, obtaining different predicted contamination pressure values ​​of all weak positive samples for each benchmark strong positive sample; summing up all predicted contamination pressure values ​​of any weak positive sample to obtain the actual Hopping expected value of the weak positive sample; S4. Compare the actual Hopping expected value of each weak positive sample with the number of labeled pathogenic detection sequences of the weak positive sample in turn. If the former is greater than or equal to the latter, the prediction result of the weak positive sample is output as negative; if the former is less than the latter, the prediction result of the weak positive sample is output as positive.

2. The method for eliminating false positives from label jumping contamination in NGS sequencing according to claim 1, characterized in that: The test prediction model is a feedforward neural network model MLP, and the feedforward neural network model MLP includes an input layer, a hidden layer and an output layer.

3. The method for eliminating false positives for label jumping contamination in NGS sequencing according to claim 2, characterized in that: In the feedforward neural network model MLP, the hidden layer uses three TanH activation functions and one linear fitting function to extract and analyze features of input data.

4. The method for eliminating false positives from label jumping contamination in NGS sequencing according to claim 1, characterized in that: The value range of the strong positive determination threshold is [1000, 10000].

5. The method for eliminating false positives for label jumping contamination in NGS sequencing according to claim 3, characterized in that: Step S3 also includes a Cutoff threshold acquisition step: the actual Hopping expected value of each weak positive sample is added to the range adjustment parameter k to obtain the Cutoff threshold.

6. The method for eliminating false positives from label jumping contamination in NGS sequencing according to claim 5, characterized in that: In step S4, the Cutoff threshold of each weak positive sample is compared with the number of labeled pathogenic detection sequences of the weak positive sample in turn. If the former is greater than or equal to the latter, the prediction result of the weak positive sample is output as predicted negative; if the former is less than the latter, the prediction result of the weak positive sample is output as predicted positive.

7. The method for eliminating false positives from label jumping contamination in NGS sequencing according to claim 6, characterized in that: The value range of the range adjustment parameter k is [3, 7].

8. The method for eliminating false positives from label jumping contamination in NGS sequencing according to claim 7, characterized in that: The value of the range adjustment parameter k is 5.

9. A false positive exclusion system for label jumping contamination in NGS sequencing, characterized in that: The system includes: The sample differentiation and data acquisition module is used to perform NGS mixed detection on all samples to be tested, and obtain the number of labeled pathogenic detection sequences and the number of labeled intermediate sequences of each sample; the sample with the number of labeled pathogenic detection sequences greater than or equal to the strong positive judgment threshold is defined as a strong positive sample, and this number is defined as the source of the strong positive sample; The module for obtaining the predicted contamination pressure value is used to package the number of labeled intermediate sequences of any weak positive sample and the source value of the benchmark strong positive sample into a predicted input set of the weak positive sample; bring the predicted input set into the test prediction model to obtain the predicted contamination pressure value of the weak positive sample; repeat the input until the predicted contamination pressure values ​​of all weak positive samples for the benchmark strong positive sample are obtained; The actual Hopping expected value calculation module is used to obtain different predicted contamination pressure values ​​of all weak positive samples for each benchmark strong positive sample; sum up all predicted contamination pressure values ​​of any weak positive sample to obtain the actual Hopping expected value of the weak positive sample; The prediction result output module is used to compare the actual Hopping expected value of each weak positive sample with the number of labeled pathogenic detection sequences of the weak positive sample in turn. If the former is greater than or equal to the latter, the prediction result of the weak positive sample is output as negative; if the former is less than the latter, the prediction result of the weak positive sample is output as positive.