Universal disease prediction device based on microbial marker and isothermal amplification

Abstract

The invention provides a universal disease prediction device based on microbial markers and isothermal amplification, and belongs to the technical field of medical data processing. A universal disease prediction device based on microbial markers and isothermal amplification comprises a microbial marker amplification primer design module used for designing amplification primers for disease-related microbial markers; the isothermal amplification module is used for carrying out isothermal amplification reaction on the disease-related microbial marker by utilizing the amplification primer; the color development acquisition module is used for acquiring color development information of isothermal amplification reactions of different microbial markers; the disease-color development model module is used for inputting the color development information into a disease prediction model to judge a disease risk result; and the calculation report module is used for feeding back the illness risk result to an operator. The content of disease-related microbial markers is reflected through an isothermal amplification result, the color of a reaction system is converted into data, and a disease risk prediction model is established, so that rapid and simple on-site diagnosis is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical data processing, and particularly relates to a general disease prediction device based on microbial markers and isothermal amplification. Background Art

[0002] The human microbiome is a microbial community jointly shaped by environmental and host-related factors. Changes in the composition and function of the microbiome are associated with a series of diseases, such as cardiometabolic disorders, hypertension, kidney function, neuropsychiatric disorders, cancer, autoimmune diseases, lung diseases, etc. Some of these associations can be explained by the gut-X axis (gut-liver axis, gut-brain axis, gut-lung axis, etc.) mechanism, and there are still many associations that have not been fully elucidated waiting for further exploration in the field of microbiome science. Disease prediction models based on the comparison of the microbiomes of healthy people and patients are emerging rapidly, and some of these models have been commercially applied.

[0003] Currently, common disease prediction models based on microbial markers usually rely on high-throughput sequencing technologies, such as 16s amplicon sequencing and metagenomic sequencing. Microbiome methods based on high-throughput sequencing of feces and saliva, without the need for blood collection or biopsy, are powerful methods for non-invasive diagnosis. High-throughput sequencing can accurately measure the abundance of the microbiome, but the disadvantage is that it depends on the laboratory environment and cannot meet the requirements of POCT for on-site diagnosis. The main feature of POCT is that reliable test results can be obtained quickly, thereby improving the clinical medical effect of patients. This requires miniaturization of experimental instruments, simplification of operation methods, and instant reporting of results. Therefore, certain technical means are needed to replace laboratory operations such as high-throughput sequencing and detect the abundance of microbial markers in a more easily achievable detection method to achieve on-site diagnosis.

[0004] Isothermal amplification has low hardware requirements and is an optional solution for bedside diagnosis. Similar to high-throughput sequencing, the number of Reads representing a specific microorganism reflects its relative abundance, and the end-point concentration of the polymerase chain reaction of a biomarker also characterizes its relative abundance in the sample. The former uses the relative abundance of biomarkers based on the number of Reads to construct various disease models and effectively evaluate disease risks; similarly, the concentration characteristics at the end-point of the polymerase chain reaction, such as the color, turbidity, fluorescence, etc. of the reaction system, can be used to construct disease models and also achieve the purpose of disease risk assessment.

[0005] The design of isothermal amplification primers for biomarkers is one of the difficulties in achieving this goal. Different from previous isothermal amplification (such as LAMP) for pathogen detection, disease-related biomarkers are usually distributed at various biological classification levels of phylum, class, order, family, genus, species, and strain (or serotype). Biomarkers above the species level usually contain multiple strain types, and other classification levels may contain dozens or even hundreds of microbial strain types. Therefore, one challenge is how to design primers suitable for the isothermal amplification system for biomarkers at different classification levels. Summary of the Invention

[0006] In view of this, the purpose of the present invention is to provide a general disease prediction device based on microbial biomarkers and isothermal amplification, which specifically characterizes the content of multi-microbial biomarkers related to diseases through isothermal amplification, constructs characteristic data through the color of the reaction system, and establishes a disease risk prediction model, so as to achieve rapid and simple on-site diagnosis.

[0007] The present invention provides a general disease prediction device based on microbial biomarkers and isothermal amplification, including the following modules:

[0008] A microbial biomarker amplification primer design module, which is used to design amplification primers for disease-related microbial biomarkers;

[0009] An isothermal amplification module, which is used to perform isothermal amplification reactions on different disease-related microbial biomarkers using the amplification primers;

[0010] A color development collection module, which is used to collect the color development information of isothermal amplification reactions of different microbial biomarkers;

[0011] A disease-color development model module, which is used to train a disease prediction model based on the color development information;

[0012] And a calculation and reporting module, which is used to feedback the disease risk result to the operator.

[0013] Preferably, the disease-related microbial biomarkers include at least one of the following levels of microorganisms: phylum, class, order, family, genus, species, and strain.

[0014] Preferably, the method for designing amplification primers for disease-related microbial biomarkers includes the following steps:

[0015] Download the genomic data of model microorganisms with the full sequence length in the bacterial domain from a public database, screen out all the complete genomic data under the target microbial marker category to form a target set, and the remaining genomic data forms an exclusion set. Use a Kmer statistical tool to obtain the Kmer information of all genomes in the target set and the exclusion set, select the Kmer that only exists in the target set and does not exist in the exclusion set to assemble into Contig, and sequentially paste the Contig onto the genomes of the exclusion set. Design primers for isothermal amplification using the Contig with more than 2 different sites per 15 bp as a template.

[0016] Preferably, the isothermal amplification includes at least one of the following techniques: LAMP loop-mediated isothermal amplification technique, RPA recombinase polymerase amplification technique, RCA rolling circle amplification technique, CPA cross-priming amplification technique, SDA strand displacement amplification technique, and HDA helicase-dependent amplification technique.

[0017] Preferably, in the isothermal amplification module, there are also negative control reaction wells and positive control reaction wells;

[0018] The reaction system of the positive control reaction well is 25 μL, specifically including 12.5 μL of WarmStart LAMP 2× premix, 1 μL of 40 pmol / μL FIPc, 1 μL of 40 pmol / μL BIPc, 0.5 μL of 10 pmol / μL F3c, 0.5 μL of 10 pmol / μL B3c, 2 μL of 20 pmol / μL LoopB, 1 μL of EH template, and 6.5 μL of ddH2O;

[0019] The nucleotide sequence of FIPc is as shown in SEQ ID NO:91;

[0020] The nucleotide sequence of BIPc is as shown in SEQ ID NO:92;

[0021] The nucleotide sequence of F3c is as shown in SEQ ID NO:93;

[0022] The nucleotide sequence of B3c is as shown in SEQ ID NO:94;

[0023] The nucleotide sequence of LoopB is as shown in SEQ ID NO:95;

[0024] The nucleotide sequence of the EH template is as shown in SEQ ID NO:96;

[0025] The reaction system of the negative control reaction well is to replace the EH template with ddH2O on the basis of the reaction system of the positive control reaction well.

[0026] Preferably, in the color development and acquisition module, after shooting with a color acquisition device, through the steps of denoising, cropping the ROI, extracting the B channel, calibration, and normalization, it is converted into a grayscale data matrix representing the sample to be tested.

[0027] Preferably, the disease prediction model module is a disease prediction model trained based on a random forest model. The learning curve is plotted to evaluate the fitting situation of the disease prediction model, the receiver operating characteristic curve is used to evaluate the effect of the disease prediction model, and the holdout set is used to evaluate the accuracy of the disease prediction model.

[0028] Preferably, for the random forest model training, the grid search method is used to adjust the training parameters of the random forest model, and the random forest model is trained with ten-fold cross-validation.

[0029] Preferably, the diseases include liver cancer and / or colorectal cancer;

[0030] When the disease is liver cancer, the amplification primers for disease-related microbial markers are nucleotide sequences such as SEQ ID NO: 1 to SEQ ID NO: 36;

[0031] When the disease is colorectal cancer, the amplification primers for disease-related microbial markers are nucleotide sequences such as SEQ ID NO: 37 to SEQ ID NO: 90.

[0032] Preferably, it further includes a sample processing module for collecting, nucleic acid separating, and loading the sample to be tested.

[0033] The present invention provides a universal disease prediction device based on microbial markers and isothermal amplification, including the following modules: a microbial marker amplification primer design module for designing amplification primers for disease-related microbial markers; an isothermal amplification module for performing isothermal amplification reactions on different disease-related microbial markers using the amplification primers; a color development and acquisition module for collecting the color development information of different microbial marker isothermal amplification reactions; a disease-color model module for inputting the color development information into a disease prediction model to judge the disease risk result; and a calculation and reporting module for feeding back the disease risk result to the operator. The present invention specifically characterizes the contents of multiple disease-related microbial markers through isothermal amplification, constructs characteristic data through the color of the reaction system, and establishes a disease risk prediction model, thereby realizing rapid and simple on-site diagnosis. The present invention provides a multi-microbial marker LAMP amplification that can be carried out in a non-laboratory environment to meet the requirements of on-site diagnosis, and a disease model is established based on the color information of the reaction system for rapid assessment of disease risk.

[0034] Furthermore, the present invention specifically defines a method for designing amplification primers for disease-related microbial markers. The primer design for a single microorganism is relatively simple, but it is more difficult for microorganisms at the phylum, class, order, family, and genus levels. Usually, to predict genes and annotate using the whole genome sequence, it is necessary to find gene families unique to the microorganisms of this category and design primers on these gene family genes. As the microbial category increases (from genus to phylum), the computational amount and design difficulty increase. This is because the probability of having homologous sequences between microorganisms with a more distant evolutionary relationship is smaller, and the probability of obtaining universal primers is also smaller. The present invention does not consider gene information but starts from the Kmer information of the genome sequence to find specific amplification primers for microorganisms at the phylum, class, order, family, and genus levels, achieving the amplification of multi-level microbial markers, thereby establishing a general-purpose disease prediction device. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 Schematic diagram of a general-purpose disease prediction device based on multi-level microbial markers and isothermal amplification;

[0036] Figure 2 Schematic diagram of the color processing scheme;

[0037] Figure 3 Color atlas of LAMP reactors of fecal samples from liver cancer patients and normal people with primers for liver cancer markers; NC: negative control; PC: positive control; Biomarker A: Prevotella 9; B: Ruminococcus; C: Bacteroides; D: Lachnospira; E: Streptococcu; F: Blautia; G: Escherichia-Shigella.;

[0038] Figure 4 Receiver operating characteristic curve of isothermal amplification random forest modeling of microbial markers in liver cancer samples;

[0039] Figure 5 Confusion matrix of the liver cancer random forest model;

[0040] Figure 6 Color atlas of LAMP reactors of fecal samples from colorectal cancer patients and normal people with primers for liver cancer markers; NC: negative control; PC: positive control; A: Escherichia-Shigella 9; B: Veillonella; C: Porphyromonas; D: Peptostreptococcus; E: Bacteroides; F: Blautia; G: Faecalibacterium; H: Clostridium; I: Alistipes; J: Prevotella;

[0041] Figure 7 Isothermal amplification random forest modeling receiver operating characteristic curve for microbial markers in colorectal cancer samples;

[0042] Figure 8 Is the confusion matrix of the random forest model for colorectal cancer. Detailed implementation manner

[0043] The present invention provides a general disease prediction device based on microbial markers and isothermal amplification, comprising the following modules:

[0044] Microbial marker amplification primer design module, used to design amplification primers for disease-related microbial markers;

[0045] Isothermal amplification module, used to perform isothermal amplification reactions on different disease-related microbial markers using the amplification primers;

[0046] Color development acquisition module, used to collect color development information of isothermal amplification reactions of different microbial markers;

[0047] Disease-color development model module, used to train a disease prediction model based on the color development information;

[0048] And a calculation and reporting module, used to feedback the disease risk result to the operator.

[0049] The general disease prediction device provided by the present invention includes a microbial marker amplification primer design module, used to design amplification primers for disease-related microbial markers.

[0050] The present invention does not make special restrictions on the types of the diseases, and any disease types that can be diagnosed by reported microbial markers are acceptable. In the present invention, the diseases preferably include tumors, cancers, cardiovascular and cerebrovascular diseases, diabetes, digestive tract-related diseases, etc. In the embodiments of the present invention, in order to illustrate the prediction results of the general disease prediction device developed by the present invention, liver cancer and colorectal cancer are taken as examples for illustration, but it should not be construed as a limitation to the protection scope of the present invention.

[0051] In the present invention, the sources of the disease-related microbial markers can be determined by retrieving public databases or literature reports, or microbial markers can be obtained by comparing the differential microorganisms in the disease group and the normal group. The disease-related microbial markers refer to microbial classifications that show a strong positive / negative association with the disease and can be used for disease prediction. They can be either a single microorganism (species, strain, serotype) or a category of microorganisms, such as multiple microorganisms under phylum, class, order, family, and genus. The number of the disease-related microbial markers includes N species, and N is an integer greater than 1. The microbial marker amplification primer design module designs a pair of specific primers for each microbial marker.

[0052] In the present invention, the method for designing amplification primers for disease-related microbial markers preferably comprises the following steps: downloading genomic data of model microorganisms with all complete sequence lengths under the bacterial domain from a public database, screening out all complete genomic data belonging to the target microbial marker category to form a target set, and the remaining genomic data to form an exclusion set, obtaining Kmer information of all genomes in the target set and the exclusion set using a Kmer statistical tool, selecting Kmers that only exist in the target set and do not exist in the exclusion set to assemble into Contigs, successively pasting the Contigs onto the genomes of the exclusion set, and designing primers for isothermal amplification using Contigs with more than 2 different sites per 15 bp as templates. The public database includes the NCBI database, the EMBL database, or other websites related to disease-related microbial genomes. The target microbial marker category means that when the target microorganism is a phylum-level microorganism, all microorganisms belonging to the phylum are selected; when the target microorganism is a genus category, all microorganisms at the species level within the genus of the target microorganism are selected; and so on. The Kmer information = 99 - 199. The Kmer statistical tool is preferably kmc, the short sequence assembly tool is preferably skesa, the sequence pasting tool is preferably MiniMap2, and the primer design tool is preferably Primer3. The specific detection tool for Contigs with more than 2 different sites per 15 bp is preferably RE-PCR and MFEprimer, and the LMAP primer design tool is preferably LAVA. The isothermal amplification preferably comprises at least one of the following techniques: LAMP loop-mediated isothermal amplification technique, RPA recombinase polymerase amplification technique, RCA rolling circle amplification technique, CPA cross-priming amplification technique, SDA strand displacement amplification technique, and HDA helicase-dependent amplification technique. In an embodiment of the present invention, taking the design of LAMP primers as an example, outer primers, inner primers, and loop primers are respectively set. In an embodiment of the present invention, the disease preferably includes liver cancer and / or colorectal cancer. When the disease is liver cancer, the microbial markers for early-stage liver cancer preferably include Prevotella 9, Bacteroides, lachnospira, Blautia, Escherichia-Shigella, Ruminococcus, and Streptococcus. The amplification primers for early-stage liver cancer-related microbial markers have nucleotide sequences as shown in SEQ ID NO:1 to SEQ ID NO:36.When the disease is colorectal cancer, the colorectal cancer-related microbial markers preferably include Blautia, Prevotella 9, Faecalibacterium, Veillonella, Escherichia-Shigella, Alistipes, Clostridium, Porphyromonas, Peptostreptococcus, and Bacteroides. The amplification primers for the colorectal cancer-related microbial markers preferably have nucleotide sequences as shown in SEQ ID NO: 37 to SEQ ID NO: 90.

[0053] The universal disease prediction device provided by the present invention includes an isothermal amplification module for performing an isothermal amplification reaction on different disease-related microbial markers using the amplification primers.

[0054] In the present invention, the isothermal amplification module is carried out using an isothermal amplification device. In the isothermal amplification module, a healthy group and a disease group are included. Each group sets up N reaction systems according to the number N of disease-related microbial markers, and preferably also includes a negative control reaction well and a positive control reaction well. The reaction system dissolution preferably contains components such as lyophilized primers, dNTPs, chromogenic agents, etc. The isothermal amplification reaction system can be referred to the patent record of application number 202011617344. For example, 5% hyaluronic acid is added to each component of the reaction system and lyophilized and stored on the wall of the PCR tube or carriers such as sterile filter paper and sterile non-woven fabric, and can be stored at room temperature. The negative control group and the positive control group are not only the quality control basis for whether the reaction system is contaminated and whether the reagent is ineffective, but also the standard values for the color after the reaction of the isothermal amplification system. The color development principle of the positive control reaction well is that adding a fluorescent dye, a pH-sensitive chromogenic agent, and a metal ion indicator to the reaction system will make the reaction system show different colors at the end point of the reaction; if no chromogenic substance is added, the reaction result can also be judged by the change in turbidity. In either case, signals can be collected by a color acquisition device. In the embodiment of the present invention, a metal ion indicator (hydroxynaphthol blue) is used as the chromogenic agent. The combination of HNB and magnesium ions makes the initial color of the reaction system violet. As the reaction proceeds, magnesium ions react with the precipitated pyrophosphate ions to form magnesium pyrophosphate precipitate, and hydroxynaphthol blue loses magnesium ions, making the system color change to sky blue, while the unreacted system still remains violet. The reaction system of the positive control reaction well is 25 μL, specifically including 12.5 μL of WarmStart LAMP 2× premix, 1 μL of 40 pmol / μL FIPc, 1 μL of 40 pmol / μL BIPc, 0.5 μL of 10 pmol / μL F3c, 0.5 μL of 10 pmol / μL B3c, 2 μL of 20 pmol / μL LoopB, 1 μL of EH template, and 6.5 μL of ddH2O. The nucleotide sequence of FIPc is as shown in SEQ ID NO:90; the nucleotide sequence of BIPc is as shown in SEQ ID NO:91; the nucleotide sequence of F3c is as shown in SEQ ID NO:92; the nucleotide sequence of B3c is as shown in SEQ ID NO:93; the nucleotide sequence of LoopB is as shown in SEQ ID NO:94; the nucleotide sequence of the EH template is as shown in SEQ ID NO:95. The reaction system of the negative control reaction well is to replace the EH template with ddH2O on the basis of the reaction system of the positive control reaction well.

[0055] The general disease prediction device provided by the present invention includes a color development acquisition module for acquiring the color development information of the isothermal amplification reaction of different microbial markers.

[0056] In the present invention, in the color development and acquisition module, it is preferably to use a color acquisition device to photograph and acquire the color of the reaction product in the reaction well. The color acquisition device includes a mobile phone, a camera or a camera head. After being photographed by the color acquisition device, the color development and acquisition module is subjected to denoising, ROI cropping, B-channel extraction, calibration, and normalization processing, and is converted into a grayscale data matrix representing the sample to be tested.

[0057] The general disease prediction device provided by the present invention includes a disease-color development model module for training a disease prediction model based on the color development information.

[0058] In the present invention, the disease prediction model is a disease prediction model trained based on a random forest model. For the construction method of the disease prediction model, it is preferably to divide the digitized color development information matrix obtained by the above LAMP detection scheme for the sample sets including the disease group and the healthy group into a training set and a validation set according to a ratio of 4:1 respectively. The training set is used for training the random forest model. The training of the random forest model preferably uses the sklearn.ensemble.RandomForestClassifier module in python software. The training of the random forest model preferably uses the grid search method to adjust the training parameters of the random forest model, and the random forest model is trained with ten-fold cross-validation. The validation set is used to evaluate the accuracy of the trained disease prediction model. The random forest model includes a set of decision trees for prediction and a matrix of the importance of each feature. The fitting situation of the disease prediction model is evaluated by drawing a learning curve, and the effect of the disease prediction model is evaluated by using the receiver operating characteristic curve. The AUC of the model should be higher than 0.8, and the accuracy should be higher than 0.75. Based on the trained disease prediction model, the risk of disease is judged. The method is to input the detected digitized color development matrix into the set of decision trees in the random forest model, start voting from the root node to the branches, and finally count the voting results of multiple trees ("healthy" or "disease"), and estimate the risk level with the probability value of "disease" in the prediction result. When the probability value of "disease" in the prediction result is above 0.8, it is judged that the disease risk of the sample is extremely high; when the probability value of "disease" in the prediction result is 0.51-0.79, it is judged that the disease risk of the sample is high; when the probability value of "disease" in the prediction result is below 0.5, it is judged that the disease risk of the sample is low.

[0059] The general disease prediction device provided by the present invention includes a calculation and reporting module for feeding back the disease risk result to the operator.

[0060] In the present invention, the calculation and reporting module preferably includes color information processing, uses a disease prediction model trained by the disease-color display module for prediction calculation, and converts the disease risk predicted by the disease prediction model into a report for presentation. The method of color information processing preferably includes the steps of photographing, denoising, cropping the ROI, extracting the B channel, correcting, and normalizing the color of the collected reaction product, and converting it into a grayscale data matrix representing the sample to be tested, thus completing the digital processing. Input the obtained color matrix into the decision tree set of the random forest model, start voting from the root node to the branches, finally count the voting results of multiple trees ["healthy" or "disease"], and estimate the risk level with the probability value of "disease" in the prediction result. The report presentation includes outputting the report result to a mobile phone or other interactive screens. The report includes information such as the collected image, prediction method, model parameters, and prediction results.

[0061] In the present invention, the general disease prediction device further includes a sample processing module for collecting, nucleic acid separating, and loading the sample to be tested. The method of nucleic acid separation is as follows: Add the test sample into a preservation solution tube containing glass beads and shake well, filter with a filter membrane, and mix the filtrate with a lysis solution containing magnetic beads; after standing for 2 - 5 minutes, adsorb with magnetic force for 2 - 5 minutes; discard the solution, wash the magnetic beads with sterile water, and collect the liquid phase to obtain DNA. The diameter of the glass beads is preferably 0.5 mm. The pore size of the filter membrane for filter membrane filtration is preferably 0.45 μm. The test sample preferably includes at least one of the following: sputum, saliva, urine, feces, skin swab, vaginal swab, oral swab, nasal swab, rinse solution, and lavage solution.

[0062] The general disease prediction device provided by the present invention is a device that can be carried out in a non-laboratory environment, meets the requirements of on-site diagnosis, performs LAMP amplification of multiple microbial markers, and establishes a disease model based on the color information of the reaction system for rapid assessment of disease risk.

[0063] The following describes in detail the general disease prediction device based on microbial markers and isothermal amplification provided by the present invention with reference to embodiments, but they should not be construed as limiting the protection scope of the present invention.

[0064] Example 1

[0065] 1. Screening of microbial markers

[0066] According to the patent (202110144203) and the literature DOI: 10.1136 / gutjnl-2017-315084, the key fecal microbial markers for early-stage liver cancer include Prevotella 9, Bacteroides, lachnospira, Blautia, Escherichia-Shigella, Ruminococcus, and Streptococcus.

[0067] 2. Design of Isothermal Amplification Primers for Microbial Markers

[0068] Taking the fecal microbial marker Ruminococcus of early-stage liver cancer as an example, the specific method for obtaining LAMP primers is described;

[0069] Obtain the genomic sequences of this microorganism and the microorganisms to be excluded: Download all complete genomic sequences under the bacterial domain from the FTP of NCBI Genome, extract all complete genomic sequences of the Ruminococcus genus as the target set, and the remaining sequences are all complete genomic sequences of the microorganisms to be excluded, serving as the exclusion set.

[0070] Kmer operation: Use the Kmer statistical tool kmc to obtain the Kmer information of these genomes (Kmer = 99). Select the Kmer information that exists only in Ruminococcus in the target set and does not exist in the exclusion set. Use the short sequence assembly tool skeasa to assemble these Kmers into Contigs, and use the sequence tiling tool Minimap2 to tile them onto the genomes of the taxa to be excluded in turn. Retain the Contigs with more than 2 different sites per 15 bp, and use the blast tool to compare with the genomes of the taxa to be excluded. The ones that can be matched will be removed to obtain Contigs with good specificity.

[0071] Design LAMP primers: Extract the Contigs with good specificity as templates, and use LAVA to design the forward primer F3, reverse primer R3, forward internal primer FIP, reverse internal primer RIP, forward loop primer loopF, and reverse loop primer loopR for the Ruminococcus genus. It should be noted that loopF and loopR cannot be generated during the primer design process for individual biomarkers, but it does not affect the LAMP reaction effect.

[0072] The LAMP primer set for liver cancer microbial markers obtained according to this process is shown in Table 1.

[0073] Table 1 LMAP Primer Set for Liver Cancer Microbial Markers

[0074]

[0075] 3. LAMP system of biomarkers

[0076] Biological samples from healthy subjects and patients were collected, specifically, stool samples from healthy subjects and patients with liver cancer.

[0077] In a laboratory environment, use a fecal genomic extraction kit (or other nucleic acid extraction kits with similar functions) to extract DNA from biological samples, and use Nanodrop for nucleic acid quantification.

[0078] In the field diagnosis environment, use the sample simple processing kit. For fecal samples, the field processing process is as follows: add 0.5g of feces to the preservation solution tube containing 0.5mm glass beads and shake to mix, take 1ml through a 0.45μm syringe-type inorganic filter head, and collect it into the lysis solution. The lysis solution contains DNA enrichment magnetic beads. After standing for 2-5 minutes, place it on the magnetic base that comes with the kit and let it stand for 2-5 minutes. Use a pipette to suck away the liquid, add sterile water eluent, mix it with the magnetic beads, and then suck the liquid to obtain DNA.

[0079] Set up an N+2LMAP reaction system. N is the type of microbial markers corresponding to the disease, and there are 7 for liver cancer. Set up a negative control and a positive control for each disease to standardize the color of the reaction endpoint. The primers and synthetic DNA templates for the Entamoeba Histolytica LAMP reaction in the literature DOI:10.1186 / s12896-020-00629-8 are referenced here. The primer sequences and DNA template sequences are shown in Table 2.

[0080] Table 2 Positive control LAMP system primers and synthetic DNA template sequences

[0081]

[0082] In a laboratory environment, this example uses the "WarmStart LAMP 2× premix" LMAP kit (or other LAMP kits with equivalent functions) of New ENGLAND BioLabs. The reaction systems corresponding to different biomarkers are shown in Table 3:

[0083] Table 3 Reaction system

[0084] LAMP reaction reagent components Volume (μL) WarmStartLAMP 2× Premix 12.5 FIP (40 pmol / μL) 1 BIP (40 pmol / μL) 1 F3 (10 pmol / μL) 0.5 B3 (10 pmol / μL) 0.5 LoopF (20 pmol / μL) 1 LoopB (20 pmol / μL) 1 DNA template 1 <![CDATA[H2O]]> 6.5 Total system 25

[0085] The reaction system of negative control is shown in Table 4.

[0086] Table 4 Reaction system of negative control

[0087] LAMP reaction reagent components Volume (μL) WarmStartLAMP 2× Premix 12.5 FIPc (40 pmol / μL) 1 BIPc (40 pmol / μL) 1 F3c (10 pmol / μL) 0.5 B3c (10 pmol / μL) 0.5 LoopB (20 pmol / μL) 2 <![CDATA[H2O]]> 7.5 Total system 25

[0088] The reaction system of the positive control is shown in Table 5.

[0089] Table 5 Reaction system of the positive control

[0090] LAMP reaction reagent components Volume (μL) WarmStartLAMP 2× Premix 12.5 FIPc (40 pmol / μL) 1 BIPc (40 pmol / μL) 1 F3c (10 pmol / μL) 0.5 B3c (10 pmol / μL) 0.5 LoopB (20 pmol / μL) 2 Synthetic DNA EH 1 <![CDATA[H2O]]> 6.5 Total system 25

[0091] In the on-site diagnosis environment, the LAMP lyophilized reagent described in the patent with the application number 202011617344 was used. Among them, the mixed component of the LAMP reaction reagent and trehalose was solidified onto hairless paper or non-woven fabric and placed at the bottom of the reaction well and vacuum-sealed. Only when it is opened and used, the solvent will be added to restore the activity of the enzyme system.

[0092] Under laboratory conditions, LAMP was performed using a PCR instrument. Using the above amplification system, the amplification conditions were 65 °C for 30 min;

[0093] Under on-site diagnosis conditions, LAMP amplification was performed using a portable and simple temperature control device. Using the above amplification system, the amplification conditions were 65 °C for 30 min.

[0094] 4. Construct a vision-based disease prediction model

[0095] 4.1 Relationship between color and DNA concentration

[0096] To quantitatively judge the concentration (copy number) of the target molecular fragment (template) in the reaction system after reaching the reaction end point, we first investigated the relationship between different template septicemia and color.

[0097] Set up the LAMP reaction, using positive template concentrations of 10 1 -10 6 copy / ul. After the reaction was completed, the target area was obtained by using steps 1-3 of the color information processing. The blue channel intensity (B), red channel intensity (R), B / R, Hue, and Saturation values were obtained and linearly (or non-linearly) fitted with the template concentration respectively. It was found that only the blue channel intensity (B) could achieve a high fit with the template concentration. Therefore, the blue channel intensity (B) was finally selected as the operating parameter. The normalized blue channel intensity was obtained according to the above steps. Using log(normalized intensity) = a·lg(C DNA ) + b, the fitting slope a and intercept b were determined in the range of 10 2 -10 6 copy / ul to determine R 2 . When R 2 ≥0.97, it was determined that this color processing scheme was predictable.

[0098] The optimized color information processing process is as follows:

[0099] 1. Standardized shooting conditions

[0100] Reference object calibration: White (RGB: 0, 0, 0) and blue (RGB: 0, 0, 255) standard color cards are pasted on the reactor for ambient light correction. For the scenario of the CCD camera: Use a fixed light source (such as an LED white light) and a fixed shooting distance (such as 10 cm); When using a mobile phone camera for shooting: Use a dedicated mobile phone software, and guide the user to capture the reactor into a preset contour in the software to ensure that the camera parameters are basically the same.

[0101] 2. Denoising: Use a median filter (3X3 window) to eliminate salt noise;

[0102] 3. Crop ROI: Locate the reaction tube area through color threshold segmentation (H ∈ [190°, 230°] in the HSV space), and crop the ROI to only retain the central 50% area;

[0103] 4. Extract the B channel: Split into three RGB channels, and calculate the median of all pixels in the B channel as the B channel intensity;

[0104] 5. Calibration: Use a white standard color card for calibration, calibration intensity = sample intensity × white standard intensity / white theoretical intensity;

[0105] 6. Normalization: Normalized intensity = (calibration intensity - negative intensity) / (positive intensity - negative intensity);

[0106] 4.2 Training and prediction

[0107] After the LAMP result of each sample is captured by a color acquisition device, it is converted into a column of gray data matrix representing the sample through steps of grayscale conversion, grayscale statistics, and normalization.

[0108] For the liver cancer model, all early liver cancer samples and healthy population samples are divided into a training set and a validation set according to a quantity ratio of 4:1 respectively. Use the sklearn.ensemble.RandomForestClassifier module of the python software to train the random forest model for the training set. Evaluate the fitting situation of the model by plotting a learning curve, and at the same time evaluate the model effect through a receiver operating characteristic curve, and use the validation set to do the final model accuracy evaluation.

[0109] The training of the random forest model includes adjusting the model training parameters using the grid search method and training the random forest model with ten-fold cross-validation. The model training parameters are set as follows: gradient n_estimators = [100, 500, 1000]; max_depth = [1, 2, 3, 7, 9]; max_features = ["log2", "sqrt"]. The method of training the random forest model with ten-fold cross-validation is to use the sklearn.model_selection.GridSearchCV module in Python software to search for the best parameters. Under each parameter combination, the training set is split into 5 sub-samples. One separate sub-sample is reserved as the data for validating the model, and the other 4 sub-samples are used for training. Repeat 10 times, each sub-sample is validated once, and the average of the 10 results is used as the final model result under this parameter combination. Draw a learning curve to evaluate the fitting of the model, and evaluate the model effect through the Receiver Operating Characteristic (ROC) curve.

[0110] The ROC curve is the Receiver Operating Characteristic curve, which is a coordinate graph with the false positive probability (1 - specificity) as the horizontal axis and the true positive probability (sensitivity) as the vertical axis, and the curve drawn from the different results obtained by the subject under specific stimulation conditions using different judgment criteria. Select the best diagnostic threshold value. The closer the ROC curve is to the upper left corner, the higher the accuracy of the test. The point on the ROC curve closest to the upper left corner is the best threshold with the fewest errors, and the total number of false positives and false negatives is the least. Comparison of the disease recognition abilities of two or more different diagnostic tests. When comparing two or more diagnostic methods for the same disease, the ROC curves of each test can be drawn in the same coordinate to visually identify the advantages and disadvantages. The ROC curve closest to the upper left corner represents the most accurate subject work. It can also be compared by calculating the area under the ROC curve (AUC) of each test respectively. The test with the largest AUC has the best diagnostic value.

[0111] In the liver cancer test, a total of 89 fecal samples were collected, including 26 from liver cancer patients and 63 from healthy people. The LAMP reaction was carried out by the above method and after color processing, the random forest model was trained and evaluated using ROC. The area under the ROC curve AUC was 0.85 ( Figure 3 ). From the confusion matrix, the recall was 0.71, the precision was 0.58, and the accuracy was 0.81 ( Figure 4 ).

[0112] Example 2

[0113] 1. Selection of microbial biomarkers

[0114] According to the report of Patent (202111043179.5), the key fecal microbial biomarkers for colorectal cancer include Blautia, Prevotella 9, Faecalibacterium, Veillonella, Escherichia-Shigella, Alistipes, Clostridium, Porphyromonas, Peptostreptococcus, and Bacteroides.

[0115] 2. Design of isothermal amplification primers for microbial biomarkers

[0116] Taking the fecal microbial biomarker Bacteroides of colorectal cancer as an example, the specific method for obtaining LAMP primers is described;

[0117] Obtain the genomic sequences of this microorganism and the microorganisms to be excluded: Download all complete genomic sequences under the bacterial domain from the FTP of NCBI Genome. Extract all complete genomic sequences of the Bacteroides genus as the target set, and the remaining sequences are all complete genomic sequences of the microorganisms to be excluded, serving as the exclusion set.

[0118] Kmer operation: Use the Kmer statistical tool kmc to obtain the Kmer information of these genomes (Kmer = 99). Select the Kmer information that only exists in Bacteroides in the target set and does not exist in the exclusion set. Use the short sequence assembly tool skeasa to assemble these Kmers into Contigs, and use the sequence tiling tool Minimap2 to tile them onto the genomes of the taxa to be excluded in turn. Retain the Contigs with more than 2 different sites every 15 bp. Use the blast tool to compare with the genomes of the taxa to be excluded, and the matching ones will be cleared to obtain Contigs with good specificity.

[0119] Design LAMP primers: Extract the Contig with good specificity as the template, and use LAVA to design the forward primer F3, reverse primer R3, forward internal primer FIP, reverse internal primer RIP, forward loop primer loopF, and reverse loop primer loopR for the Bacteroides genus. Through this method, the LMAP primers for the two disease-related microbial biomarkers in Example 1 can be obtained.

[0120] It should be noted that loopF and loopR cannot be generated during the primer design process of individual biomarkers, but it does not affect the LAMP reaction effect.

[0121] The LMAP primer set for colorectal cancer biomarkers is shown in Table 6.

[0122] Table 6 LMAP primer set for colorectal cancer microbial markers

[0123]

[0124]

[0125]

[0126] 3. LAMP system of biomarkers

[0127] Biological samples were collected from healthy subjects and patients, specifically, stool samples from healthy subjects and patients with colorectal cancer.

[0128] In a laboratory environment, use a fecal genomic extraction kit (or other nucleic acid extraction kits with similar functions) to extract DNA from biological samples, and use Nanodrop for nucleic acid quantification.

[0129] In the field diagnosis environment, use the sample simple processing kit. For fecal samples, the field processing process is as follows: add 0.5g of feces to the preservation solution tube containing 0.5mm glass beads and shake to mix, take 1ml through a 0.45μm syringe-type inorganic filter head, and collect it into the lysis solution. The lysis solution contains DNA enrichment magnetic beads. After standing for 2-5 minutes, place it on the magnetic base that comes with the kit and let it stand for 2-5 minutes. Use a pipette to suck away the liquid, add sterile water eluent, mix it with the magnetic beads, and then suck the liquid to obtain DNA.

[0130] Set up an N+2LMAP reaction system. N is the type of microbial markers corresponding to the disease, and colorectal cancer has 10. Set up a negative control and a positive control for each disease to standardize the color of the reaction endpoint. The primers and synthetic DNA templates for the Entamoeba Histolytica LAMP reaction in the literature DOI:10.1186 / s12896-020-00629-8 are referenced here. The primer sequences and DNA template sequences are the same as Table 2.

[0131] In a laboratory environment, this example uses the "WarmStart LAMP 2× premix" LMAP kit (or other LAMP kits with equivalent functions) of New ENGLAND BioLabs, and the reaction systems corresponding to different biomarkers are the same as in Table 3. The reaction system of the negative control is the same as in Table 4. The reaction system of the positive control is the same as in Table 5.

[0132] In the on-site diagnosis environment, the LAMP freeze-dried reagent described in the patent with the application number 202011617344 is adopted. Among them, the mixed component of the LAMP reaction reagent and trehalose is solidified onto hairless paper or non-woven fabric and placed at the bottom of the reaction well and vacuum-sealed. Only when it is opened for use, the solvent is added to restore the activity of the enzyme system.

[0133] Under laboratory conditions, LAMP is carried out using a PCR instrument, and the amplification system in Tables 3 to 5 is used. The amplification conditions are 65 °C for 30 min.

[0134] Under on-site diagnosis conditions, LAMP amplification is carried out using a portable and simple temperature control device, and the amplification system in Tables 3 to 5 is used. The amplification conditions are 65 °C for 30 min.

[0135] 4. Construct a vision-based disease prediction model

[0136] 4.1 Relationship between color and DNA concentration

[0137] In order to quantitatively judge the concentration (copy number) of the target molecular fragment (template) in the reaction system after reaching the reaction end point, we first investigated the relationship between different template sepsis and color.

[0138] Set up the LAMP reaction, and use positive template concentrations of 10 1 -10 6 copy / ul. After the reaction is completed, the target area is obtained by adopting steps 1-3 of the color information processing. The blue channel intensity (B), red channel intensity (R), B / R, Hue, and Saturation values are obtained, and linear (or non-linear) fitting is carried out with the template concentration respectively. It is found that only the blue channel intensity (B) can achieve a high fitting with the template concentration. Therefore, finally, the blue channel intensity (B) is selected as the operation parameter. Obtain the normalized blue channel intensity according to the above steps, and use log(normalized intensity) = a·lg(C DNA ) + b. Fit the slope a and intercept b in the range of 10 2 -10 6 copy / ul to determine R 2 . When R 2 ≥0.97, it is determined that this color processing scheme has predictability.

[0139] The optimized color information processing process is as follows:

[0140] 1. Standardize the shooting conditions

[0141] Reference calibration: White (RGB: 0, 0, 0) and blue (RGB: 0, 0, 255) standard color cards are pasted on the reactor for ambient light correction. For the scenario of CCD camera: Use a fixed light source (such as an LED white light) and a fixed shooting distance (such as 10 cm); When using a mobile phone camera for shooting: Use a dedicated mobile phone software, and guide the user to capture the reactor into a preset contour in the software to ensure that the camera parameters are basically the same.

[0142] 2. Denoising: Use a median filter (3X3 window) to eliminate salt noise;

[0143] 3. Crop ROI: Locate the reaction tube area through color threshold segmentation (H ∈ [190°, 230°] in the HSV space), and crop the ROI to retain only the central 50% area;

[0144] 4. Extract B channel: Split into three RGB channels, and calculate the median of all pixels in the B channel as the B channel intensity;

[0145] 5. Calibration: Use a white standard color card for calibration, calibration intensity = sample intensity × white standard intensity / white theoretical intensity;

[0146] 6. Normalization: Normalized intensity = (calibration intensity - negative intensity) / (positive intensity - negative intensity);

[0147] 4.2 Training and prediction

[0148] For the colorectal cancer model, the color information matrices containing colorectal cancer samples and healthy population samples and their corresponding disease states ("healthy", "disease") are divided into a training set and a validation set according to a quantity ratio of 4:1 respectively, and the sklearn.ensemble.RandomForestClassifier module of python software is used to train a random forest model for the training set. Evaluate the fitting situation of the model by drawing a learning curve, and evaluate the model effect through the receiver operating characteristic curve, and use the validation set to make a final evaluation of the model accuracy.

[0149] A total of 133 fecal samples were collected, including 41 from colorectal cancer patients and 92 from healthy people. The LAMP reaction was carried out by the method described in this patent and after color processing, a random forest model was trained, and ROC was used for evaluation, and the area under the ROC curve AUC was 0.80 ( Figure 5 ). From the confusion matrix, the recall rate is 0.62, the precision rate is 0.62, and the accuracy rate is 0.76 ( Figure 6 ).

[0150] As can be seen from the results of the above embodiments, the prediction ability of the liver cancer model (0.85) is stronger than that of the colorectal cancer model (0.80). The prediction abilities of both are lower compared to the prediction results in the laboratory environment (usually higher than 0.9). However, considering the prediction ability achieved in the non-laboratory environment and the fact that the solution can be quickly transplanted to the diagnosis of any disease, such results are acceptable, but it is only necessary to accumulate a larger cohort for development and improvement in the future.

[0151] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A universal disease prediction device based on microbial markers and isothermal amplification, characterized in that: Includes the following modules: Microbial marker amplification primer design module, used to design amplification primers for disease-related microbial markers; An isothermal amplification module, used to perform isothermal amplification reaction on different disease-related microbial markers using the amplification primers; Color acquisition module, used to train disease prediction models based on color information; The disease-color model module is used to input the digital color information into the disease prediction model to determine the disease risk results; and a calculation and reporting module, for digitizing the color development results and feeding back the disease risk results to the operator.

2. The universal disease prediction device based on microbial markers and isothermal amplification according to claim 1, characterized in that: The disease-associated microbial markers include microorganisms at at least one level of the following: phylum, class, order, family, genus, species, and strain.

3. The universal disease prediction device based on microbial markers and isothermal amplification according to claim 1, characterized in that: The method for designing amplification primers for disease-related microbial markers comprises the following steps: The genome data of model microorganisms with all complete sequence lengths under the bacterial domain were downloaded from the public database, and all complete genome data under the target microbial marker category were screened out to form a target set, and the remaining genome data formed an exclusion set. The Kmer statistical tool was used to obtain the Kmer information of all genomes in the target set and the exclusion set, and the Kmers that only existed in the target set and not in the exclusion set were selected to assemble into Contigs. The Contigs were sequentially pasted onto the genome of the exclusion set, and primers for isothermal amplification were designed using Contigs with more than two differential sites every 15 bp as templates.

4. The universal disease prediction device based on microbial markers and isothermal amplification according to claim 1, characterized in that: The isothermal amplification includes at least one of the following techniques: LAMP loop-mediated isothermal amplification technique, RPA recombinase polymerase amplification technique, RCA rolling circle amplification technique, CPA cross primer amplification technique, SDA strand displacement amplification technique and HDA helicase-dependent amplification technique.

5. The universal disease prediction device based on microbial markers and isothermal amplification according to claim 1, characterized in that: The isothermal amplification module also includes a negative control reaction well and a positive control reaction well; The reaction system of the positive control reaction well is 25 μL, specifically including 12.5 μL of WarmStart LAMP 2× premix, 1 μL of 40 pmol / μL FIPc, 1 μL of 40 pmol / μL BIPc, 0.5 μL of 10 pmol / μL F3c, 0.5 μL of 10 pmol / μL B3c, 2 μL of 20 pmol / μL LoopB, 1 μL of EH template and 6.5 μL of ddH2O; The nucleotide sequence of the FIPc is as shown in SEQ ID NO:91; The nucleotide sequence of BIPc is as shown in SEQ ID NO:92; The nucleotide sequence of F3c is as shown in SEQ ID NO:93; The nucleotide sequence of B3c is as shown in SEQ ID NO:94; The nucleotide sequence of LoopB is as shown in SEQ ID NO:95; The nucleotide sequence of the EH template is as shown in SEQ ID NO:96; The reaction system of the negative control reaction well is based on the reaction system of the positive control reaction well, with ddH2O replacing the EH template.

6. The universal disease prediction device based on microbial markers and isothermal amplification according to claim 1, characterized in that: In the color acquisition module, after being photographed by a color acquisition device, it is converted into a grayscale data matrix representing the sample to be tested through the steps of denoising, cropping ROI, extracting B channel, correction, and normalization.

7. The universal disease prediction device based on microbial markers and isothermal amplification according to claim 1, characterized in that: The disease prediction model module is based on the random forest model training to obtain the disease prediction model, uses the drawing of learning curves to evaluate the fitting of the disease prediction model, uses the receiver operating characteristic curve to evaluate the effect of the disease prediction model, and uses the isolation set to evaluate the accuracy of the disease prediction model.

8. The universal disease prediction device based on microbial markers and isothermal amplification according to claim 7, characterized in that: The random forest model training is to use a grid search method to adjust the random forest model training parameters, and to train the random forest model using ten cross validations.

9. The universal disease prediction device based on microbial markers and isothermal amplification according to claim 1, characterized in that: The disease includes liver cancer and / or colorectal cancer; When the disease is liver cancer, the amplification primers for disease-related microbial markers are such as nucleotide sequences such as SEQ ID NO: 1 to SEQ ID NO: 36; When the disease is colorectal cancer, the amplification primers for disease-related microbial markers, such as nucleotide sequences, are such as SEQ ID NO: 37 to SEQ ID NO:

90.

10. The universal disease prediction device based on microbial markers and isothermal amplification according to any one of claims 1 to 9, characterized in that: It also includes a sample processing module for collecting samples to be tested, separating nucleic acids and loading samples.

Images