A multicolor digital nucleic acid amplification and detection method combining deep learning algorithms

Abstract

The present invention discloses a multicolor digital nucleic acid amplification detection method combined with a deep learning algorithm, and belongs to the field of multiple nucleic acid detection technology. The detection method comprises: first, adding a sample to be tested into a LAMP reaction system containing a hydrogel monomer to form a hydrogel system, the reaction system comprises specific amplification primers and quenching probes for different target molecules, wherein a fluorescent group is marked on one inner primer, and the quenching probe is partially complementary to the inner primer and carries a corresponding quenching group; after the isothermal amplification reaction, the fluorescence point images of the hydrogel system under different fluorescence channels are recorded, the images are merged, input into a deep learning model, and the counting results of multiple molecules are output. The present invention realizes the rapid detection of multiple template molecules with the help of different molecular probes; an asymmetric quenching probe is used to reduce fluorescence background interference; and a deep learning algorithm is used to accurately identify nucleic acid amplification points. The present invention provides an accurate method for the identification and absolute quantification of multicolor molecular detection.

Description

Technical Field

[0001] The present invention relates to the field of multiple nucleic acid detection technology, and in particular to a multicolor digital nucleic acid amplification detection method combined with a deep learning algorithm. Background Art

[0002] Molecular diagnosis is one of the important frontiers of rapid detection in the contemporary era. As an important nucleic acid analysis tool, PCR has gradually developed from qualitative detection to single-molecule quantification, which can realize digital and precise analysis in the true sense. In addition to PCR, there are many isothermal nucleic acid amplification technologies (such as LAMP) that can achieve large-scale amplification of templates at a constant temperature, getting rid of the strict temperature control requirements of the PCR process. The so-called digital nucleic acid amplification (i.e. digital PCR or digital LAMP) is to distribute the diluted sample to a large number of independent micro-reaction units, and analyze it based on the end-point fluorescence reading. It is not affected by the reaction amplification efficiency, has strong anti-interference ability and the ability to detect multiple gene targets simultaneously. Multiplex PCR refers to a detection technology that can amplify multiple target genes in the same system. When digital PCR is combined with multiplex PCR, it can realize the qualitative detection of multiple target genes to the digital quantification of multiple targets, that is, multiplex digital PCR. According to the labeling of different fluorescent and quenching group DNA probes, the preparation of different types of micro-chamber chips or micro-droplets, and the use of different cycles of nucleic acid amplification, the performance of multiple digital PCR detection can be greatly improved, and the parallel detection of multiple indicators can be achieved, more abundant detection information can be obtained, and the detection cost can be significantly reduced. Therefore, it is of great significance to develop a multiple molecular rapid detection system and a corresponding automatic intelligent recognition model for molecular detection results.

[0003] Compared with the digital quantification achieved by the microfluidic system, the hydrogel digital LAMP can achieve the relative separation of nucleic acids in the internal nanopores without complicated operations. The hydrogel polymer is mixed with the LAMP reagent and the sample to be tested, which can be polymerized to form a gel in a short time and fix the sample in the porous channel. If the target DNA / RNA exists in the sample, fluorescent amplification points will be generated in the porous channel. The template molecules are fixed by the three-dimensional nano-confinement effect of the gel to form a micro-reaction unit, which does not require the micro-nano processing and other methods in conventional digital detection to complete the distribution of nucleic acid samples, providing a simple and convenient platform for on-site detection.

[0004] However, current hydrogel amplification is a single-plex detection, that is, it can only detect a single target and cannot detect multiple targets at the same time. Hydrogel multiple nucleic acid digital amplification based on multicolor imaging has not yet been reported. The confined area advantage of the nanopores inside the hydrogel makes it have great application potential in the field of detecting multiple foodborne pathogens.

[0005] Existing multiple molecular detection technologies often use TaqMan probes of various colors to distinguish different targets. The two ends of this nucleic acid probe are labeled with a fluorescent group and a quencher group respectively, so the complete probe chain cannot show a fluorescent signal. During template amplification, the probe chain will be cut, and the fluorescent group will be dispersed into the solution to emit corresponding fluorescence. However, this method cannot achieve multi-color digital detection in hydrogels, because the free fluorescent group will be dispersed throughout the system, and the target cannot be confined to single-molecule quantification. At the same time, the TaqMan probe needs to be carefully designed and has a high binding efficiency with the template chain to ensure that the probe chain is cut and releases fluorescence during the amplification process. Therefore, it is necessary to develop a new multi-color technology to achieve multi-color digital nucleic acid detection in hydrogels.

[0006] At present, for the result counting of multi-color digital detection, the multi-color PCR digital detection completed by microspheres, microdroplets or microchambers with relatively uniform sizes can simply realize the recognition and counting of images. However, the fluorescent amplification points formed by the hydrogel LAMP system are of different sizes, and the hydrogel matrix occasionally produces bubbles after the hydrogel is heated as a flexible material. At the same time, the nucleic acid products of multiple molecular amplifications in the three-dimensional network structure will approach and connect with each other during the diffusion process. It is time-consuming and laborious to identify and count with the naked eye. Commercial software cannot automatically identify the detection results of multiple template molecules, and cannot accurately count the fluorescent points of the detection target.

[0007] Therefore, developing a multiplex digital nucleic acid detection method that utilizes a hydrogel system to simultaneously detect multiple nucleic acid molecules and can accurately analyze multicolor signals is a technical problem that needs to be solved by those skilled in the art. Summary of the invention

[0008] The purpose of the present invention is to provide a multi-color digital nucleic acid amplification detection method for non-diagnostic purposes, which utilizes a hydrogel system to simultaneously detect multiple nucleic acid molecules and combines machine learning methods to achieve accurate analysis of multi-color signals.

[0009] To achieve the above object, the present invention adopts the following technical solution:

[0010] The present invention provides a multicolor digital nucleic acid amplification detection method combined with a deep learning algorithm, comprising the following steps:

[0011] (1) adding a sample to be tested to a loop-mediated isothermal amplification reaction system solution, wherein the reaction system comprises a specific amplification primer pair for different target molecules and a quenching probe, wherein the primer pair comprises an inner primer, wherein one inner primer is labeled with a fluorescent group, and the inner primers for different target molecules are labeled with different spectrally distinguishable fluorescent groups, wherein the quenching probe is complementary to a partial sequence base of the inner primer and has a dissolution temperature lower than the loop-mediated isothermal amplification reaction temperature, and the quenching probe carries a quenching group for quenching the luminescence of the fluorescent group on the inner primer complementary thereto, and then adding a hydrogel monomer, cross-linking to form a hydrogel, and obtaining a hydrogel reaction system of the sample to be tested;

[0012] (2) placing the hydrogel reaction system under constant temperature conditions to perform isothermal amplification reaction;

[0013] (3) After the reaction is completed, the fluorescence spot images of the hydrogel system under different fluorescence channels are recorded using fluorescence imaging technology, and the multi-color molecular fluorescence spot images are obtained by merging them. The merged images are input into the deep learning model to output the counting results of the multi-color molecules;

[0014] The method for constructing the deep learning model includes:

[0015] S1, according to steps (1)-(3), the sample to be tested is replaced with a standard solution containing different target molecules to obtain a multicolor molecular hydrogel reaction system for isothermal amplification reaction, and a merged image is obtained after fluorescence imaging;

[0016] S2, annotate the fluorescent spots in the merged image to obtain a data set, and then train the YOLOv8 model on the data set to obtain a deep learning model.

[0017] The present invention utilizes the unique molecular sieve characteristics and three-dimensional network structure of hydrogel, replaces the traditional microfluidic technology for nucleic acid amplification compartmentalization, adopts specific probes for multi-color molecular detection, and avoids the use of non-specific nucleic acid dyes. The nanoporous matrix of the hydrogel fixes multiple template molecules in the pores, and nucleic acid amplification reaction reagents, primers, quenching probes, etc. can diffuse freely in the hydrogel matrix. Since the dissolution temperature of the quenching probe and the inner primer is lower than the reaction temperature of the loop-mediated isothermal amplification, under the temperature conditions of the loop-mediated isothermal amplification, the inner primer labeled with the fluorescent group continuously participates in the LAMP reaction process, enters the nucleic acid product, generates a bright fluorescent signal, and the amplified product is confined in the nanopore to form a fluorescent spot. After the reaction is completed, the fluorescence of the excess inner primer is quenched by the hybridization of the complementary probe under room temperature conditions, which reduces the fluorescence interference of the background of the amplification point and realizes the molecular amplification of the multi-color hydrogel LAMP.

[0018] The target molecule may be foodborne pathogenic microorganism DNA, and the pathogenic microorganism includes bacteria, fungi, viruses, and may be but not limited to Escherichia coli, Salmonella or Listeria.

[0019] The sample to be tested can be first subjected to nucleic acid extraction treatment and then added to the LAMP reaction system; it can also be directly added to the LAMP reaction system containing lysozyme or reverse transcriptase, wherein the lysozyme can directly lyse the bacteria in the sample to be tested and the reverse transcriptase can reverse transcribe RNA into cDNA.

[0020] In step (1), a corresponding specific amplification primer pair and a quenching probe are designed according to each target molecule. Specifically, fluorescent groups with different luminescent wavelengths are modified on each inner primer, and then the corresponding probe sequence and quenching group are designed based on complementary bases.

[0021] For different target molecules, different spectrally distinguishable fluorescent groups are marked on the inner primers, and spectrally distinguishable amplification products are obtained by nucleic acid amplification. Further, the fluorescent group is any one of 6-FAM, AMCA, PacificBlue, Atto 425, HEX, VIC, TET, ROX, JOE, Cy3, and Cy5.

[0022] Furthermore, the inner primers include an upstream inner primer (FIP) and a downstream inner primer (BIP), and a fluorescent group is labeled at the 5' end of the upstream inner primer; the quenching probe is complementary to the sequence of the inner primer near the 5' end, the complementary part accounts for 50-60% of the length of the inner primer sequence and the melting temperature is 8-12°C lower than the LAMP amplification temperature, and the quenching group used to quench the fluorescent group is modified at the 3' end of the quenching probe.

[0023] The present invention uses an asymmetric quenching probe of the fluorescent amplification signal to achieve specific multi-color nucleic acid molecule detection. The fluorescent group is labeled at the 5' end of the inner primer, and a fluorescent quencher is designed at the 3' end of the complementary sequence to form a higher signal-to-noise ratio and clearly distinguish multiple molecular detection signals. The melting temperature of the two complementary sequences needs to be about 10°C lower than the LAMP amplification temperature so as not to interfere with the normal hydrogel LAMP multi-color molecular detection process.

[0024] Specifically, the present invention uses FAM and ROX fluorescent groups as green and red channel markers for fluorescent amplification points, respectively, without the need for additional components, and the nucleic acid groups inside the hydrogel can directly emit flocculated fluorescent signals. To further improve the signal-to-noise ratio, TAMRA and BHQ2 are used to quench the excess free green and red fluorescence that is not incorporated into the amplification product.

[0025] The present invention combines molecular probe technology to construct a hydrogel multicolor molecular detection system according to the number of target molecular species. The loop-mediated isothermal amplification reaction system includes: isothermal amplification buffer, magnesium ions, dNTPs, Bst 2.0 DNA polymerase, primer mixture, quenching probe and hydrogel monomer. When constructing the reaction system, the amount of each raw material added can be adjusted according to actual conditions.

[0026] Furthermore, the primers for each target molecule in the reaction system included 1.6 μM FIB and BIP, 0.2 μM F3 and B3, and 0.8 μM LF and LB.

[0027] Furthermore, when the fluorescent group is 6-FAM and the quenching group is TAMRA, the molar ratio of the upstream inner primer to the quenching probe in the reaction system is 5:4; when the fluorescent group is ROX and the quenching group is BHQ2, the molar ratio of the upstream inner primer to the quenching probe in the reaction system is 5:1.

[0028] Furthermore, the present invention uses polyethylene glycol-based hydrogel, wherein the hydrogel monomers include four-arm ethylene glycol acrylate (4arm-PEG-ACRL) and thiol-polyethylene glycol-thiol (SH-PEG-SH). The above two monomers are mixed in an aqueous solution and gelled at room temperature for 3-5 minutes.

[0029] Furthermore, the molecular weight of four-arm ethylene glycol acrylate is 10000 MW, the molecular weight of mercapto-polyethylene glycol-mercapto is 3400 MW, and the molar ratio of the two is 1:2.

[0030] In step (2), the hydrogel system is placed in a sealed space to prevent water evaporation for isothermal amplification reaction. The sealed space is used to prevent water evaporation in the hydrogel, thereby ensuring the accuracy of the test results. Specifically, the mixed system solution is added dropwise to a transparent sealed chamber, allowed to stand to form a gel, and sealed with a film. The sealed system composed of the sealed chamber and the film can prevent water evaporation.

[0031] Furthermore, the isothermal amplification reaction was performed under the conditions of heating at 65-70° C. for 20-30 min.

[0032] In step (3), under excitation light sources of different wavelengths, the multicolor detection system presents amplification points with different fluorescence, and the fluorescence signal images of each channel are taken separately, and the images are merged using software to obtain multiple amplification point signals in the same image.

[0033] Furthermore, under the excitation light source corresponding to the fluorescent group, the fluorescence signal image is collected, and then the images under different fluorescence channels are merged using Image J software.

[0034] The PEG hydrogel of the present invention is a three-dimensional mesh structure, and the template nucleic acid molecules of multicolor detection are confined to the internal nanoporous matrix, and a fluorescent nucleic acid amplification group is formed after amplification. However, the imaging result is shot in a two-dimensional plane space, and there may be a situation where a variety of fluorescent points are connected. Manual counting cannot meet the needs of fast and simple counting, and Image J software cannot accurately count the multicolor PEG-LAMP molecular detection results. Therefore, the present invention uses a deep learning algorithm to train a model, accurately identify amplification points from fluorescent impurities and bubbles, and distinguish overlapping fluorescent points to achieve accurate counting of multicolor digital detection.

[0035] Deep learning is a machine learning method based on neural networks. It extracts features from raw data for detection, classification or regression. With its powerful automatic feature learning capabilities, it can solve problems intelligently and quickly and effectively. The present invention can effectively distinguish bubbles and fluorescent impurities and accurately count connected amplification points by training the best neural network model with the aid of artificial feature extraction.

[0036] Specifically, the present invention adopts the YOLOv8 algorithm to construct a recognition counting model. During machine learning, the labeled hydrogel image is used as input, and the number of fluorescent nucleic acid spots and / or the image of the labeled fluorescent spots is used as output.

[0037] In order to increase the performance, speed and accuracy of the model, the collected image data was processed. Specifically, the Roboflow platform was used to annotate the amplification points in the merged image to obtain a data set, and then the data set was increased by 3 times, which can better extract the characteristics of the fluorescent amplification points and generalize the model. The model performance was evaluated according to the mAP value, and the best model hyperparameters and configurations were selected. Specifically, the hyperparameters and configurations of the model training were: Adam optimizer, 16Batch size, 0.001lr0 and 0.01lrf.

[0038] The deep learning model trained by the present invention can effectively distinguish amplification bubbles, automatically identify and accurately count multiple nucleic acid molecules with high spatial proximity from a two-dimensional image, and detect each image within a time range of 3 to 8 seconds.

[0039] The present invention has the following beneficial effects:

[0040] (1) The present invention provides a multi-color digital nucleic acid amplification detection method for deep learning recognition, which realizes the rapid detection of multiple template molecules with the help of different molecular probes. The hydrogel nanoporous matrix fixes multiple template molecules in the pores to realize in situ amplification. The inner primers labeled with fluorescent groups continuously participate in the LAMP reaction process, enter the nucleic acid product, and generate bright fluorescent signals, while the fluorescence of the excess inner primers is quenched by hybridization of asymmetric complementary probes, reducing the fluorescence background interference of the amplification point, forming a higher signal-to-noise ratio, and realizing the specific molecular amplification of multi-color PEG-LAMP.

[0041] (2) The present invention combines a deep learning algorithm to construct a model for automatic recognition and counting of multi-color PEG-LAMP molecular detection images. The model can be used to clearly distinguish nucleic acid amplification points from fluorescent impurities and bubbles, and accurately count multiple nucleic acid molecules with high spatial proximity. The counting accuracy is higher than that of the commonly used Image J software, providing an accurate method for the recognition and absolute quantitative counting of multi-color molecular detection.

[0042] (3) The multicolor digital nucleic acid amplification detection method combined with deep learning provided by the present invention can perform direct and non-interference multiple amplification in the actual sample mixed fruit and vegetable juice. The operation process is very simple, the amplification time is fast, the multicolor molecular detection accuracy is high, and the deep learning model has a fast automatic recognition speed. It has good commercial application value in the field of multicolor foodborne pathogen molecular detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 Endpoint fluorescence plot (a), amplification spot count (b), and signal-to-noise ratio (c) of molecular probe concentration optimization for E. coli.

[0044] Figure 2 Endpoint fluorescence plot (a) and amplification spot count (b) of the molecular probe time optimization for E. coli.

[0045] Figure 3 Endpoint fluorescence plots (a), amplification spot counts (b), and signal-to-noise ratios (c) for optimization of molecular probe concentrations for Listeria monocytogenes.

[0046] Figure 4 Endpoint fluorescence plot (a) and amplification spot count (b) of the time-optimized molecular probe for Listeria monocytogenes.

[0047] Figure 5 Mean average precision of the deep learning single-count model (a) and comparison of amplification point counts (b).

[0048] Figure 6This is a comparison of the counting operation steps using a deep learning model and using Image J software. Step ① of the 1-step deep learning recognition is the automatic output count of the input image; steps ①, ②, and ③ of the 3-step commercial software recognition are color selection setting, manual threshold adjustment, and analysis count, respectively.

[0049] Figure 7 This is a fluorescence image of multi-color PEG-LAMP molecule detection.

[0050] Figure 8 Multicolor molecule detection fluorescence images for deep learning models.

[0051] Fig. 9 End-point fluorescence amplification graph (a) and deep learning counting graph (b) of multicolor molecular detection of mixed fruit and vegetable juice. DETAILED DESCRIPTION

[0052] The present invention is further described below using specific examples. The following examples are only used to illustrate the present invention, but the embodiments of the present invention are not limited to the following examples. According to the technical solutions and concepts described below, various other corresponding changes and deformations are made, which all belong to the protection scope of the claims of the present invention.

[0053] Unless otherwise stated, the various reagents, reaction conditions, detection methods, etc. used in the following examples are all conventional reagents, reaction conditions and detection methods used in the art.

[0054] The present invention utilizes the unique three-dimensional mesh structure of PEG hydrogel to compartmentalize nucleic acid amplification, and uses specific probes for multi-color molecular detection. The nanoporous PEG matrix fixes multiple template molecules in the pores, and the inner primers labeled with fluorescent groups continuously participate in the LAMP reaction process, enter the nucleic acid product, and generate bright fluorescent signals. The fluorescence of the redundant inner primers is quenched by hybridization of complementary probes, which reduces the fluorescence interference of the background of the amplification point, realizes the molecular amplification of multi-color PEG-LAMP, and can present the position of nucleic acid molecules in situ. At the same time, the collected PEG-LAMP fluorescence amplification images are used as data sets in combination with the deep learning algorithm, and the artificial annotation recognition is trained, and the optimal model hyperparameters and configurations are selected after the model performance is evaluated. The optimal single and multiple molecular detection models obtained are respectively applied to the tasks of single and multiple molecular counting of the collected hydrogel fluorescence amplification point images, and the nucleic acid molecules of multiple foodborne pathogens with high spatial proximity are automatically identified.

[0055] The sequences of primers and probes for E. coli LAMP amplification are as follows, from left to right from 5' to 3':

[0056] EC-F3: GCCATCTCCTGATGACGC;

[0057] EC-B3: ATTTACCGCAGCCAGACG;

[0058] EC-FAM-FIP:

[0059] 6-FAM-CATTTTGCAGCTGTACGCTCGCAGCCCATCATGAATGTTGCT;

[0060] EC-BIP:

[0061] CTGGGGCGAGGTCGTGGTATTCCGACAAACACCACGAATT;

[0062] EC-LF: CTTTGTAACAACCTGTCATCGACA;

[0063] EC-LB: ATCAATCTCGATATCCATGAAGGTG;

[0064] EC-quencher probe: GAGCGTACAGCTGCAAAATG-TAMRA.

[0065] The sequences of primers and probes for Listeria monocytogenes LAMP amplification are as follows:

[0066] LM-F3: AAGCTGCTTTTGATGCTG;

[0067] LM-B3:TCGATTAAAAGTAGCGCCTT;

[0068] LM-ROX-FIP:

[0069] ROX-CGGCTTTGAAGGAAGAATTTTTGATCGTAAGCGGAAAATC TGTCT;

[0070] LM-BIP:

[0071] TACGGAGGTTCCGCAAAAGATTTTCAAAATATCGCGTAAGTCT; LM-LB: TGAAGTTCAAATCATCGACGGC;

[0072] LM-quencher probe: GATCAAAAATTCTTCCTTCAAAGCCG-BHQ2.

[0073] All LAMP primers and probe sequences were synthesized by Shanghai Bioengineering Co., Ltd. LAMP amplification reagents were purchased from New England Biolab. 4arm-PEG-ACRL (molecular weight 10,000) and SH-PEG-SH (molecular weight 3,400) were purchased from Xi'an Kaixin Biological Co., Ltd. The reaction chamber was purchased from BIO-RAD.

[0074] Example 1

[0075] This embodiment provides a multi-color digital nucleic acid detection method combined with deep learning recognition, which is as follows:

[0076] 1. Design molecular probes for a variety of foodborne pathogens. Modify the fluorescent groups on their respective inner primers, and then design the corresponding probe sequences and quenching groups based on complementary bases. Use FAM and ROX fluorescent groups as green and red channel markers for fluorescent amplification points, and use TAMRA and BHQ2 to quench excess free green and red fluorescent groups, respectively.

[0077] 2. Attach the reaction chamber to the glass slide, mix the amplification system according to the proportion, transfer the amplification system to the reaction chamber, seal the chamber, and let it stand at room temperature to crosslink into gel. Prepare PEG hydrogel LAMP single and multiple detection systems.

[0078] Various PEG-LAMP molecular detection systems are as follows:

[0079] The PEG-LAMP chip Escherichia coli monochrome detection system consists of the following components: 1.6μM EC-FIP, 1.6μM E.C-BIP, 0.2μM EC-F3, 0.2μM EC-B3, 0.8μM EC-LF, 0.8μM EC-LB, 1.28μM EC-quenching probe, 1.4mM dNTPs, 640U / mL Bst 2.0 polymerase, 1×LAMP buffer, 6mM MgSO4, 1.6mg 4arm-PEG-ACRL (MW:10000), 1.1mg SH-PEG-SH (MW:3400), 2μL of DNA template, and then add ddH2O to the reaction system to 25μL.

[0080] The PEG-LAMP chip single-color detection system for Listeria monocytogenes consists of the following components: 1.6μM LM-FIP, 1.6μM LM-BIP, 0.2μM LM-F3, 0.2μM LM-B3, 0.8μM LM-LF, 0.8μM LM-LB, 0.32μM L.M-quenching probe, 1.4mM dNTPs, 640U / mL Bst 2.0 polymerase, 1×LAMP buffer, 6mM MgSO4, 1.6mg4arm-PEG-ACRL (MW:10000), 1.1mg SH-PEG-SH (MW:3400), 2μL of DNA template, and then ddH2O is added to the reaction system to 25μL.

[0081] The PEG-LAMP chip multicolor detection system was composed of the following components: 1.6 μM EC-FIP, 1.6 μM EC-BIP, 0.2 μM EC-F3, 0.2 μM EC-B3, 0.8 μM EC-LF, 0.8 μM EC-LB, 1.28 μM EC quenching probe, 1.6 μM L.M-FIP, 1.6 μM LM-BIP, 0.2 μM LM-F3, 0.2 μM LM-B3, 0.8 μM LM-LF, 0.8 μM LM-LB, 0.32 μM LM quenching probe, 1.4 mM dNTPs, 640 U / mL Bst 2.0 polymerase, 1× LAMP buffer, 6 mM MgSO4, 3.2 mg 4arm-PEG-ACRL (MW: 10000), 2.2 mg SH-PEG-SH (MW: 3400), 2 μL of Escherichia coli nucleic acid genome, 2 μL of Listeria monocytogenes nucleic acid genome, and then add ddH2O to the reaction system to 50 μL.

[0082] 3. Transfer the cross-linked hydrogel molecular detection system to a 65°C heating plate for 20-30 minutes to perform LAMP amplification reaction.

[0083] 4. After the amplification is completed, the reaction chamber is placed under a fluorescence microscope. Under fluorescent light sources of different wavelengths, the amplification signals of molecular probes in different fluorescent channels of FAM and ROX are excited. The PEG-LAMP chip multicolor detection system presents amplification points with different fluorescence. The endpoint pictures of each channel are taken separately and merged using Image J software to obtain the fluorescence signals of various foodborne pathogens on the same hydrogel system.

[0084] 5. Combine the YOLOv8m algorithm structure to train the neural network model, optimize various hyperparameters and configurations, mainly use mAP values ​​to evaluate the model performance, and select the best hyperparameter configuration to achieve accurate single and multiple PEG-LAMP molecule counting.

[0085] Example 2

[0086] This example compares the ratios and reaction times of different molecular probes, as follows:

[0087] 1. The molecular probe design rules and sequences are the same as those described in Example 1.

[0088] 2. The preparation process of the PEG-LAMP chip is the same as that described in Example 1. Except for the quenching probe, the other reagent components are the same as those described in Example 1. The concentration of the quenching probe added to the system is changed, and the number of fluorescence amplification points and the fluorescence background signal-to-noise ratio are evaluated to select the appropriate quenching probe concentration, and amplification is performed at different times at this concentration.

[0089] 3. The heating amplification conditions of the hydrogel molecule detection system are consistent with the steps in Example 1.

[0090] 4. The hydrogel photography was carried out in accordance with the steps of Example 1.

[0091] 5. Result analysis:

[0092] like Figure 1 As shown in a, under a series of E. coli quenching probe concentrations, the PEG-LAMP chip showed fluorescence amplification spots. Without the need for additional components, the E. coli amplification group inside the hydrogel can directly emit flocculated fluorescence signals. However, when the quenching probe concentration is low, the background fluorescence is high, indicating that the free FAM-labeled primer will cause a high fluorescence background signal. Figure 1 The fluorescence amplification points and signal-to-noise ratio at each quenching probe concentration are shown in bc. The 1.28 μM E. coli quenching probe concentration was selected for the subsequent preparation of the PEG-LAMP chip system.

[0093] like Figure 2 As shown in the figure, after the amplification points appeared, the number of points increased with the extension of reaction time, and no obvious new amplification points appeared after 20 minutes, indicating that 20 minutes is the best time for E. coli amplification saturation.

[0094] like Figure 3As shown in a, under a series of concentrations of Listeria monocytogenes quenching probe, the PEG-LAMP chip showed fluorescence amplification spots. Without the need for additional components, the Listeria monocytogenes amplification group inside the hydrogel can directly emit flocculated fluorescence signals. However, when the concentration of the quenching probe is low, the background fluorescence is high, indicating that the free ROX-labeled primer will cause a signal with a high fluorescence background. Figure 3 The fluorescence amplification points and signal-to-noise ratios at various quenching probe concentrations are shown in bc, and a 0.16 μM Listeria monocytogenes quenching probe concentration was selected for subsequent PEG-LAMP chip system preparation.

[0095] like Figure 4 As shown, after the amplification points appeared, the number of points increased with the extension of reaction time, and no obvious new amplification points appeared after 15 minutes, indicating that 15 minutes is the best time for the saturation of Listeria monocytogenes amplification.

[0096] Example 3

[0097] This example combines the deep learning algorithm to detect and identify the E. coli single-color PEG-LAMP molecule, as follows:

[0098] 1. The molecular probe design rules and sequences are the same as those described in Example 1.

[0099] 2. The preparation process of the PEG-LAMP chip and the reagent components are the same as those described in Example 1.

[0100] 3. The heating amplification conditions of the hydrogel molecule detection system are consistent with the steps in Example 1.

[0101] 4. The hydrogel photography was carried out in accordance with the steps of Example 1.

[0102] 5. Build a deep learning model. The data collection image for deep learning is a monochrome PEG-LAMP molecular detection image. The amplification points are annotated using the Roboflow platform. The data set is then enhanced and the model is generalized. The collected data set is then used to train the YOLOv8m model, with the annotated hydrogel image as input and the number of nucleic acid fluorescence points as output. The model is trained using the Adam optimizer to achieve the task of identifying and counting hydrogel fluorescence amplification points. Evaluate the model performance based on the mAP value and select the best model hyperparameters and configuration. Using the trained model, the analysis task of monochrome PEG-LAMP molecular detection fluorescence images can be performed.

[0103] like Figure 5As shown in a, according to the highest mAP50 value, configuration 3 is selected as the best hyperparameter. The model hyperparameter configuration for single-color PEG-LAMP molecule detection is Adam optimizer, 16Batch size, 0.001lr0 and 0.01lrf.

[0104] 6. Use the above deep learning model to count multiple PEG-LAMP hydrogel amplification images and compare the counting results with manual recognition and commercial software counting.

[0105] like Figure 5 As shown in Figure b, in the analysis of gradient concentration fluorescence amplification points, the deep learning model and manual recognition have high consistency, but the commercial Image J software will have counting differences when the number of amplification points increases, and it is impossible to distinguish points that are close to each other, indicating that the molecular detection image counting combined with the deep learning algorithm has extremely high accuracy.

[0106] like Figure 6 As shown in the figure, the deep learning model recognition method only requires one step, and the output result can be obtained by inputting the picture. However, the commercial Image J software requires three steps: setting the color threshold, manual adjustment, and analysis counting to achieve the amplification point counting result of an input picture. Counting using commercial Image J software relies more on manual operation and cannot complete automatic counting. The deep learning model can realize automatic counting of multiple pictures at the same time, and the detection time for each picture is about 3 to 8 seconds.

[0107] Example 4

[0108] This example explores multicolor PEG-LAMP molecular detection, as follows:

[0109] 1. The molecular probe design rules and sequences are the same as those described in Example 1.

[0110] 2. The preparation process of the PEG-LAMP chip and the reagent components are the same as those described in Example 1.

[0111] 3. The heating amplification conditions of the hydrogel molecular detection system are consistent with the steps in Example 1, and the heating time is 20 min-30 min.

[0112] 4. The hydrogel photography was carried out in accordance with the steps of Example 1.

[0113] like Figure 7As shown in a, the multicolor PEG-LAMP molecular detection image shows multiple fluorescence channels. The Escherichia coli nucleic acid product labeled with FAM and TAMRA probe system excites green fluorescence in the green channel, and the Listeria monocytogenes nucleic acid product labeled with ROX and BHQ2 probe system excites red fluorescence in the red channel. Image J software merges multiple fluorescence channels to obtain a multicolor PEG-LAMP molecular detection image.

[0114] Example 5

[0115] This example explores the deep learning and commercial software comparison of multi-color PEG-LAMP molecular detection and counting, as follows:

[0116] 1. The molecular probe design rules and sequences are the same as those described in Example 1.

[0117] 2. The preparation process of the PEG-LAMP chip and the reagent components are the same as those described in Example 1.

[0118] 3. The heating amplification conditions of the hydrogel molecule detection system are consistent with the steps in Example 1.

[0119] 4. The hydrogel photography was carried out in accordance with the steps of Example 1.

[0120] 5. The data collection diagram for deep learning is a multi-color PEG-LAMP molecular detection image. The Roboflow platform is used to annotate the amplification points and distinguish different fluorescent colors. The data set is enhanced three times to generalize the model. The collected data set is then used to train the YOLOv8m model, with multi-color annotated hydrogel images as input and the number of different color fluorescent nucleic acid spots and / or images annotated with different color fluorescent spots as output. The Adam optimizer is used to train the model to achieve the task of identifying and counting hydrogel fluorescent amplification points. Using the trained model, the analysis task of multi-color PEG-LAMP molecular detection fluorescent images can be performed.

[0121] like Figure 8 As shown in the figure, the multicolor PEG-LAMP molecular detection model of the deep learning algorithm can accurately identify the nucleic acid amplification points of Escherichia coli and Listeria monocytogenes, distinguish the fluorescent points from the fluorescent bubbles, and distinguish the amplification products with spatial overlap, indicating that the multicolor detection model can identify and count multiple template molecules with high spatial proximity, and realize multicolor PEG-LAMP molecular detection. However, the commercial Image J software sets the selected threshold of the target object and cannot distinguish the colors of the multicolor amplification points of different probes, and cannot realize the counting of multicolor PEG-LAMP molecular fluorescence detection.

[0122] Example 6

[0123] This embodiment combines the deep learning algorithm to perform multi-color PEG-LAMP molecule detection and identification on mixed fruit and vegetable juice, as follows:

[0124] 1. The molecular probe design rules and sequences are the same as those described in Example 1.

[0125] 2. The preparation process of the PEG-LAMP chip and the reagent components of the multicolor PEG-LAMP molecular detection system are the same as those described in Example 1 except for the addition of mixed fruit and vegetable juice.

[0126] 3. The heating amplification conditions of the hydrogel molecule detection system are consistent with the steps in Example 1.

[0127] 4. The hydrogel photography was carried out in accordance with the steps of Example 1.

[0128] 5. Model training and testing are consistent with the steps in Example 1.

[0129] 6. Result analysis:

[0130] like Fig. 9 As shown in a, in the mixed fruit and vegetable juice sample, the multicolor PEG-LAMP molecular detection will not be affected, and has a good anti-interference effect. Multiple nucleic acid molecules can be amplified to produce corresponding fluorescent spots.

[0131] like Fig. 9 As shown in Figure b, the deep learning multi-color detection model can accurately identify and analyze multi-color PEG-LAMP molecular detection images, realize automatic counting of amplification signals of multiple nucleic acid molecules, and has the potential for excellent on-site absolute quantification of nucleic acid molecules in actual samples.

[0132] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A multicolor digital nucleic acid amplification detection method combined with a deep learning algorithm for non-diagnostic purposes, characterized in that: The following steps are involved: (1) adding the sample to be tested to a loop-mediated isothermal amplification reaction system solution, wherein the reaction system includes a specific amplification primer pair for Escherichia coli and Listeria monocytogenes and a quenching probe, the 5' end of the upstream inner primer is labeled with a fluorescent group, and a quenching group for quenching the fluorescent group is modified at the 3' end of the quenching probe, and then adding a hydrogel monomer to form a hydrogel through cross-linking to obtain a hydrogel reaction system of the sample to be tested; The hydrogel is obtained by mixing four-arm ethylene glycol acrylate and mercapto-polyethylene glycol-mercapto in a molar ratio of 1:2 in an aqueous solution for 3-5 minutes and gelling at room temperature; the molecular weight of the four-arm ethylene glycol acrylate is 10000 MW, and the molecular weight of the mercapto-polyethylene glycol-mercapto is 3400 MW; The specific amplification primer pair and quenching probe sequence of Escherichia coli are as follows, and the sequence is from 5' end to 3' end from left to right: EC-F3: GCCATCTCCTGATGACGC; EC-B3: ATTTACCGCAGCCAGACG; EC-FAM-FIP: 6-FAM-CATTTTGCAGCTGTACGCTCGCAGCCCATCATGAATGTTGCT; EC-BIP: CTGGGGCGAGGTCGTGGTATTCCGACAAACACCACGAATT; EC-LF: CTTTGTAACAACCTGTCATCGACA; EC-LB: ATCAATCTCGATATCCATGAAGGTG; EC-quenching probe: GAGCGTACAGCTGCAAAATG-TAMRA; The specific amplification primer pair and quenching probe for Listeria monocytogenes are as follows, with the sequence from left to right from 5' to 3': LM-F3: AAGCTGCTTTTGATGCTG; LM-B3:TCGATTAAAAGTAGCGCCTT; LM-ROX-FIP: ROX-CGGCTTTGAAGGAAGAATTTTTGATCGTAAGCGGAAAATCTGTCT; LM-BIP: TACGGAGGTTCCGCAAAAGATTTTCAAAATATCGCGTAAGTCT; LM-LB: TGAAGTTCAAATCATCGACGGC; LM-quenching probe: GATCAAAAATTCTTCCTTCAAAGCCG-BHQ2; (2) placing the hydrogel reaction system under constant temperature conditions to perform isothermal amplification reaction; (3) After the reaction is completed, the fluorescence spot images of the hydrogel system under different fluorescence channels are recorded using fluorescence imaging technology, and the multi-color molecular fluorescence spot images are obtained by merging them. The merged images are input into the deep learning model to output the counting results of the multi-color molecules; The method for constructing the deep learning model comprises the following steps: S1. According to steps (1)-(3), the sample to be tested is replaced with a standard solution containing different target molecules to obtain a multicolor molecular hydrogel reaction system for isothermal amplification reaction, and a merged image is obtained after fluorescence imaging; S2, marking the fluorescent spots in the merged image to obtain a data set, and then training the YOLOv8 model on the data set to obtain a deep learning model; In step S2, the hyperparameters and configurations of the model training are: Adam optimizer, 16 Batch size, 0.001lr0 and 0.01 lrf; In step S2, the Roboflow platform is used to annotate the amplification points in the merged image to obtain a data set, which is increased by 3 times to generalize the model.

2. The detection method according to claim 1, characterized in that The molar ratio of the upstream inner primer to the quenching probe of Escherichia coli in the reaction system is 5:4; the molar ratio of the upstream inner primer to the quenching probe of Listeria monocytogenes in the reaction system is 5:

1.

3. The detection method according to claim 1, characterized in that In step (2), the isothermal amplification reaction is carried out at 65-70°C for 20-30 min.

4. The detection method according to claim 1, characterized in that In step (3), under the excitation light source corresponding to the fluorescent group, the fluorescence signal image is collected, and then the images under different fluorescence channels are merged using Image J software.

Images