DNA molecular diagnosis system based on RNA expression level, system construction method and application

Abstract

The invention discloses a DNA molecular diagnosis system based on RNA expression level. The diagnosis system comprises an input module, a classification module and an output module, the input module is used for converting collected RNA in serum into DNA molecules and inputting the DNA molecules into the diagnosis system; the classification module comprises a DNA molecular calculation network, and the DNA molecular calculation network implements a silicon-based diagnosis model on a molecular level; and the output module is used for outputting the classification result calculated by the classification module. The invention further discloses a construction method of the DNA molecular diagnosis system and application of the DNA molecular diagnosis system or the construction method, and the DNA molecular diagnosis system or the construction method has wide application value.

Description

Technical Field

[0001] This invention belongs to the fields of nucleic acid molecular nanotechnology, computer science, and cancer diagnostic technology, and relates to a DNA molecular diagnostic system based on RNA expression level, system construction method, and application. Background Technology

[0002] RNA plays a crucial role in the transmission of genetic information and gene expression within organisms. Numerous studies have reported a close correlation between RNA expression levels and the development of diseases, particularly cancer. While RNA expression levels in cancer patients show some similarities, they also differ from those in healthy individuals. Therefore, detecting RNA expression levels in blood (serum) can be used for cancer diagnosis. Many studies have proposed RNA as a tumor biomarker and developed cancer diagnostic models based on differences in RNA expression levels. Existing methods for detecting RNA mainly include reverse transcription quantitative PCR, microarrays, and RNA sequencing, but these still face challenges such as high cost, poor selectivity, and low sensitivity. Furthermore, the analysis of RNA expression levels requires specialized knowledge, limiting the application of this strategy in clinical or home settings.

[0003] As the primary carrier of biological genetic material, DNA can interact with a variety of biomolecules, including DNA, RNA, proteins, and small molecules, enabling the sensing of these molecules. Furthermore, through precise design, DNA computing networks can perform complex computational tasks and analyze the information from detected biomolecules. Therefore, DNA molecular computing can serve as an interface between molecular recognition and information processing, integrating biomarker sensing and analysis. Although researchers have proposed DNA computing platforms for diagnostic purposes, these are limited to single cancers, resulting in targeted diagnosis that is difficult to apply to broad cancer screening. Therefore, it is necessary to provide a DNA molecular diagnostic platform with a wider diagnostic scope, targeting multiple cancers.

[0004] If multiple cancers need to be detected, a more complex diagnostic system needs to be established, requiring a silicon-based diagnostic model with higher computing power and a more complex DNA molecular computing network to be designed. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a DNA molecular diagnostic system based on RNA expression levels. This system detects the expression levels of RNA in blood (serum) and analyzes these levels using a DNA molecular computing network, thereby achieving automated and accurate screening and diagnosis of various cancers.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] This invention provides a DNA molecular diagnostic system based on RNA expression levels, the diagnostic system comprising: an input module, a classification module, and an output module;

[0008] The input module is used to convert RNA in the collected serum into DNA molecules and input them into the diagnostic system.

[0009] The classification module includes a DNA molecule computing network, in which a silicon-based diagnostic model is executed at the molecular level. The DNA molecules input into the classification module represent the RNA expression level data relationship in the silicon-based diagnostic model at the molecular level through their concentration ratios. The RNA expression level data reflects the proportional relationship between RNA concentrations in the serum sample. The RNA collected from the sample is converted into DNA molecules that maintain their concentration ratios after passing through the input module.

[0010] The output module is used to output the classification result calculated by the classification module.

[0011] The silicon-based diagnostic model includes a binary classification model for determining whether a disease is present and / or a multi-classification model for determining what specific disease is present;

[0012] The binary classification model includes support vector machines, but other models capable of binary classification can also be selected in the specific implementation process.

[0013] The multi-classification model includes a fully connected neural network model. In specific implementation, other models capable of multi-classification can also be selected.

[0014] The support vector machine model maps the input RNA expression level data to a high-dimensional space using a kernel function, searches for a hyperplane that maximizes the margin between the two classes of data points, defines the hyperplane using support vectors, and determines the class of the newly input RNA expression level data using the sign of the decision function; and / or,

[0015] The fully connected neural network model includes an input layer, one or more hidden layers, and an output layer. The input layer receives RNA expression level data, one or more hidden layers perform feature extraction and nonlinear transformation, and the output layer uses the Softmax activation function to output the probability of each sample category and selects the sample category with the highest probability for output.

[0016] The DNA molecular computing network includes one or more multiplication computing units, one or more addition computing units, and one or more subtraction computing units. Through the modular splicing of multiple different computing units, the silicon-based diagnostic model is executed at the molecular level.

[0017] The multiplication calculation unit generates a DNA weighted input strand through the reaction of the DNA input strand with the weight molecules. The concentration of the DNA weighted input strand is the product of the concentration of the DNA input strand and the number of the corresponding weight molecules.

[0018] The addition calculation unit converts DNA weighted input strands with the same downstream sequence into the same weighted sum strand under the action of the addition gate. The concentration of the weighted sum strand is the sum of the concentrations of all DNA weighted input strands with positive (or negative) weights.

[0019] The subtraction calculation unit generates a downstream output chain concentration by mutually annihilating the positive weighted sum chain and the negative weighted sum chain. The concentration of the output chain is the concentration of the positive weighted sum chain minus the concentration of the negative weighted sum chain. No output is generated when the concentration of the negative weighted sum chain is higher than that of the positive weighted sum chain. The ReLU activation function is embedded into the subtraction calculation.

[0020] This invention also provides a method for constructing the above-mentioned DNA molecular diagnostic system based on RNA expression levels, the method comprising the following steps:

[0021] Step 1: Construct a silicon-based diagnostic model that includes binary classification and / or multi-class classification models;

[0022] Step 2: Construct a DNA molecular computing network to perform the calculations in the silicon-based diagnostic model described in Step 1 at the molecular level;

[0023] Step 3: Obtain the classification module for disease classification, and then construct the DNA molecular diagnostic system.

[0024] The front and back ends of the classification module are connected to the input module and the output module, respectively, together forming the DNA molecular diagnostic system based on RNA expression level.

[0025] Step one, the construction of the silicon-based diagnostic model further includes the following steps:

[0026] Step 1.1. Obtain microarray data on serum RNA expression levels in patients with different disease types and healthy individuals;

[0027] Step 1.2. Filter the microarray data to obtain the characteristic RNA group;

[0028] Step 1.3. Train the binary classification model and / or multi-class classification model using the characteristic RNA group data to obtain the silicon-based diagnostic model.

[0029] In step 1.2, RNAs with expression levels below 2 in more than half of the samples are screened out; and / or, RNAs with a significance level p greater than 0.01 are screened out; and / or, PCA is used to assess the importance of each RNA for classification; and / or,

[0030] In step 1.3, during the training of the binary classification model, the intercept is set to 0, and the maximum number of iterations is set to 1000–10000; during the training of the multi-class classification model, the learning rate, batch size, and number of epochs are set to 0.001–0.003, 36–64, and 5000–10000, respectively; preferably, the maximum number of iterations is set to 10000; during the training of the multi-class classification model, the learning rate, batch size, and number of epochs are set to 0.003, 36, and 5000, respectively, and the Adam optimizer is used for parameter optimization; during the training of both the binary classification model and / or the multi-class classification model, stratified sampling is performed using a preset data ratio suitable for the actual situation to select the training set and test set.

[0031] In step two, the silicon-based diagnostic model is transformed into a concatenation of three types of computational functions: one or more multiplications, one or more additions, and one or more subtractions, and these computational functions are implemented at the DNA molecule level. Specifically, the computational function executed by the binary support vector machine in the silicon-based diagnostic model can be decomposed into one or more multiplication functions, one or more addition functions, and one subtraction function. Each neuron in the multi-class silicon-based fully connected neural network can be decomposed into one or more multiplication functions, one or more addition functions, and one subtraction function. Since there is usually more than one neuron in the neural network during use, there may also be one or more subtraction functions in the combined state of multiple neurons. The parameters, input, or output values ​​of the silicon-based diagnostic model are encoded using the number or concentration of DNA molecules participating in the reaction; and / or,

[0032] The weight parameters of the silicon-based diagnostic model are encoded by the number of weighted molecules that react with the input DNA strand in the DNA molecule computing network; the input or output values ​​of the silicon-based diagnostic model are encoded by the concentration of the input or output strand used in the DNA molecule computing network; and / or,

[0033] The silicon-based diagnostic model is computed by splicing together the same and / or different computational units in the DNA molecular computing network.

[0034] This invention also provides the application of the above-mentioned DNA molecular diagnostic system based on RNA expression level, or the method for constructing the above-mentioned DNA molecular diagnostic system, in biomarker identification, biological gene expression analysis, etc.

[0035] The beneficial effects of this invention include: This invention proposes a DNA molecular diagnostic system based on RNA expression levels and its construction method. This invention integrates the sensing and analysis of cancer biomarkers, eliminating the need to acquire RNA expression data from samples. Instead, RNA is directly input into the system for analysis, and diagnosis is completed by detecting the fluorescence intensity of the system, without requiring specialized analysis. This DNA molecular diagnostic system, combined with a digital microfluidic device, promises to achieve automated output from blood samples to diagnostic results. It is simple to operate and more suitable for environments such as nursing homes or homes. This DNA molecular diagnostic system can simultaneously diagnose at least three types of cancer, providing a broad cancer screening method compared to existing one-to-one cancer detection methods. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0037] Figure 1 This is a schematic diagram of the overall architecture of the DNA molecular diagnostic system of the present invention.

[0038] Figure 2 This is a schematic diagram of the input module of the present invention.

[0039] Figure 3 This is a schematic diagram of the silicon-based diagnostic model in the classification module of this invention.

[0040] Figure 4 This is a schematic diagram of the confusion matrix of the silicon-based diagnostic model of the classification module in Embodiment 1 of the present invention.

[0041] Figure 5 This is a schematic diagram of the confusion matrix of the diagnostic results of the DNA molecular diagnostic system of the present invention on 20 clinical samples.

[0042] Figure 6 This is a flowchart illustrating the construction of the DNA molecular diagnostic system of the present invention and its application in the diagnosis of multiple cancers.

[0043] Figure 7 This is a flowchart illustrating a method for constructing a silicon-based diagnostic model provided by an embodiment of the present invention.

[0044] Figure 8 This is a schematic flowchart of a method for constructing a DNA computing network provided by an embodiment of the present invention.

[0045] Figure 9This is a schematic diagram of the confusion matrix of the silicon-based diagnostic model of the classification module in Embodiment 4 of the present invention.

[0046] Figure 10 This is a schematic diagram of the confusion matrix of the silicon-based diagnostic model of the classification module in Embodiment 5 of the present invention. Detailed Implementation

[0047] The present invention will be further described in detail below with reference to the specific embodiments and accompanying drawings. Except for the contents specifically mentioned below, the processes, conditions, and experimental methods for implementing the present invention are all common knowledge and general knowledge in the art, and the present invention does not have any particular limitations.

[0048] This invention provides a DNA molecular diagnostic system based on RNA expression levels and its construction method, which can be used for the diagnosis of various cancers. The diagnostic system includes a DNA molecular computational network implementing a silicon-based diagnostic model based on RNA expression information at the molecular level. In practical use, the diagnostic system first acquires microarray data of serum RNA expression levels from healthy individuals and cancer patients; establishes a silicon-based multi-cancer diagnostic model based on the microarray data; designs DNA molecular multiplication, addition, and subtraction computational functions; integrates multiple computational functions according to the computational parameters of the silicon-based diagnostic model to establish a DNA molecular computational network; extracts RNA from serum samples, and obtains ssDNA through reverse transcription, amplification, and transformation; inputs the ssDNA into the DNA molecular computational network to perform diagnosis. This invention obtains a silicon-based diagnostic model based on RNA expression levels and further establishes a DNA molecular diagnostic system, which can directly detect and analyze the RNA expression levels of samples and provide diagnostic results. This method can differentiate between patients with various cancers and healthy individuals, exhibiting good sensitivity and specificity.

[0049] The core innovation of this invention lies in the establishment of an integrated DNA molecular diagnostic system for sensing and analyzing cancer biomarkers, and its construction method. This system can detect and analyze RNA expression levels in serum, directly providing diagnostic results without the need for specialized analysis, making it more suitable for environments such as nursing homes or homes. Furthermore, the system can perform multi-cancer analysis, providing a broad method for cancer screening.

[0050] This invention provides a DNA molecular diagnostic system based on RNA expression levels. The diagnostic system mainly includes an input module, a classification module, and an output module. The classification module includes a DNA molecular computational network that executes a silicon-based diagnostic model at the molecular level.

[0051] Figure 1 This demonstrates the complete architecture of a DNA molecular diagnostic system based on RNA expression levels:

[0052] In the input module, RNA is extracted from serum and, through reverse transcription, amplification, and transformation steps, a set of ssDNA inputs that retain the original RNA expression level information is obtained;

[0053] In the classification module, a silicon-based diagnostic model is first constructed based on RNA expression level microarray data. Then, a DNA molecular computing network is used to realize the silicon-based diagnostic model at the molecular level, and the ssDNA input is analyzed.

[0054] In the output module, the output results of the DNA molecular computing network are obtained through fluorescence signal readout and processing, and then the output of the diagnostic system is obtained.

[0055] In one specific implementation, the silicon-based diagnostic model includes a support vector machine (SVM) algorithm for diagnosing whether a patient has cancer and / or a fully connected neural network (FCNN) model for diagnosing specific cancer types. In practical use, in one implementation, RNA expression level data in a sample is first input into the SVM algorithm for binary classification to determine whether the sample has cancer. If the sample is determined to have cancer, the expression level data is then input into the FCNN model for multi-class classification to determine the final cancer type. The DNA molecular computing network implements the silicon-based diagnostic model at the molecular level, enabling direct detection and analysis at the molecular level, and obtaining the final diagnostic result based on RNA expression levels.

[0056] In another specific implementation, multiple cancer types can also be diagnosed and detected using only a trained multi-class support vector machine, a fully connected neural network model, a convolutional neural network, or other multi-class models.

[0057] Taking the construction method of a multi-cancer diagnostic system as an example, the construction method of the DNA molecular diagnostic system based on RNA expression level of the present invention will be specifically described. The construction method includes the following steps:

[0058] Step 1: Establish a silicon-based diagnostic model based on RNA expression level information, including binary and / or multi-classification models;

[0059] Step 2: The parameters (including mathematical operations and weights) of the silicon-based diagnostic model obtained in Step 1 are encoded in a DNA computing network to realize the operation of the silicon-based diagnostic model at the molecular level.

[0060] Step 3: Obtain the classification module for disease classification, and then construct the DNA molecular diagnostic system.

[0061] Step one, the establishment of the silicon-based diagnostic model includes the following steps:

[0062] Step 1.1: Obtain microarray data on serum RNA expression levels in patients with different cancer types and healthy individuals;

[0063] Step 1.2: Based on the microarray data of serum RNA expression levels obtained in Step 1.1, the characteristic RNA groups of the multi-cancer model are selected using variance screening and principal component analysis (PCA).

[0064] Step 1.3: Train a classifier (including binary classification model and / or multi-class classification model) based on the data of the feature RNA groups selected in Step 1.2 to construct a silicon-based diagnostic model.

[0065] In step 1.1, microarray data of serum RNA expression levels in patients with different cancer types and healthy individuals were obtained from NCBI.

[0066] In step 1.1, to ensure the correctness of the data and the accuracy of the model, the acquisition and processing methods of microarray data in the NCBI database should be consistent.

[0067] Step 1.2, the steps for screening the characteristic RNA set of silicon-based diagnostic models, include:

[0068] 1.2.1. Exclude low-expression miRNAs; RNAs with expression levels below 2 in more than half of the samples were removed from the database; and / or,

[0069] 1.2.2. Use variance screening to delete RNAs with a significance level p greater than 0.01; and / or,

[0070] 1.2.3. PCA was used to assess the importance of each RNA for classification, and the 36 RNAs with the largest weight coefficients in PCA were selected as the characteristic RNA group for multi-cancer diagnosis.

[0071] In step 1.3, the silicon-based diagnostic model includes one or both of two machine learning algorithms: a binary classification SVM algorithm to distinguish between healthy and cancer samples; and a multi-class classification FCNN algorithm to determine the specific type of cancer. The results of both are combined to give the final diagnostic result. In some specific implementations, the multi-class model alone can be used to directly diagnose whether a sample is healthy or has a specific type of cancer.

[0072] In step 1.3, the binary classification SVM algorithm is trained using the svm.LinearSVC module in the Python library sklearn. The multi-class FCNN model is trained on the Keras platform. The model includes an input layer (36 neurons) and a Dense layer, and the output layer uses the Softmax function as its activation function.

[0073] In step 1.3, the intercept is set to 0 and the maximum number of iterations is set to 10000 during the SVM algorithm training process. During the FCNN model training process, the learning rate, batch size, and number of epochs are set to 0.003, 36, and 5000, respectively.

[0074] In step 1.3, during the training of the SVM algorithm and FCNN model, the data is stratified and sampled in an 8:2 ratio to obtain the training set and the validation set.

[0075] In step 1.3, FCNN uses the Adam optimizer to optimize parameters.

[0076] In step 1.3, the metrics for evaluating the performance of the silicon-based diagnostic model include accuracy, sensitivity, specificity, and precision.

[0077] In step 1.3, in order to ensure the performance of the silicon-based diagnostic model, the difference in the number of samples in different categories should not be too large.

[0078] In step two, the silicon-based diagnostic model is implemented at the DNA molecular level, establishing a DNA computational network for multi-cancer diagnosis, including the following steps:

[0079] Step 2.1: Convert the silicon-based diagnostic model into a combination of three types of calculation functions: one or more multiplications, one or more additions, and one or more subtractions.

[0080] Step 2.2: Implement various computational functions at the DNA molecular level;

[0081] Step 2.3: Integrate computational functions and construct a DNA molecule computational network. The weights and other parameters of the silicon-based diagnostic model, as well as the input and output values, are encoded using the number of DNA molecules participating in the reaction or the concentration of DNA molecules.

[0082] In step 2.1, the SVM algorithm executes a linear kernel function, which, according to its calculation formula, can be broken down into one or more multiplication functions, one or more addition functions, and one subtraction function. In the FCNN model, each neuron in the Dense layer performs a multiplication-accumulation operation, which, according to its calculation formula, can be broken down into one or more multiplication functions, one or more addition functions, and one subtraction function. One or more parallel multiplication-accumulation operations are executed in the FCNN model.

[0083] In step 2.2, the DNA-based multiplication function can perform the calculation of multiplying the decimal input by the integer weight. Through the reaction between the DNA input strand and the weight molecules, the concentration of the weighted input strand is generated as a corresponding multiple of the input strand's concentration. The sign of the weight is encoded by the downstream sequence of the output strand.

[0084] In step 2.2, the DNA-based addition function can perform the summation of positive (negative) weighted inputs. DNA weighted input strands with the same downstream sequence are converted into the same weighted sum strand by the addition gate, and the concentration of the weighted sum strand is the same as the sum of the concentrations of the DNA weighted input strands.

[0085] In step 2.2, the DNA-based subtraction function performs subtraction between positive and negative weighted sums. Positive and negative weighted sum chains annihilate each other, and downstream output is only generated when the concentration of the positive weighted sum chain is higher. The DNA-based subtraction function also performs the ReLU activation function operation simultaneously with the subtraction operation.

[0086] In step 2.3, the design of DNA-based computing functions at the molecular level is modular. By adjusting the upstream and downstream sequences, the computing functions can be easily cascaded to construct a DNA computing network and perform molecular diagnostics.

[0087] In step 2.3, the weight parameters of the silicon-based diagnostic model assign different levels of importance to each input, and are encoded in the DNA computational network as the number of weighted molecules that react with the input DNA strand. The bias parameters are encoded using the concentration of the DNA strand.

[0088] In step 2.3, the input and output values ​​in the silicon-based diagnostic model are encoded in the DNA computing network using the concentrations of the input and output strands.

[0089] In practical use, the DNA molecular diagnostic system constructed above can be used for the detection and diagnosis of cancer types. According to the instructions, RNA extraction from serum samples is completed using a kit. The extraction steps are performed according to the instructions of the kit. The obtained RNA is stored at -80℃ for later use.

[0090] Executing the input module involves the following steps:

[0091] Step i: Use a reverse transcription kit to reverse transcribe the RNA extracted from the serum;

[0092] Step ii: Amplify the cDNA strand obtained by reverse transcription using asymmetric polymerase chain reaction (Linear-After-The-Exponential-PCR, LATE-PCR);

[0093] Step iii: Transform the amplified cDNA strand to obtain the ssDNA strand.

[0094] The classification module analyzes ssDNA using the constructed DNA molecule computational network, including the following steps:

[0095] Step a: Input the transformed ssDNA into the DNA computing network;

[0096] Step b: Detect the generated signal.

[0097] The execution output module outputs the diagnostic results from the DNA molecular computing network, including the following methods:

[0098] When the computational model is binary classification, the output signal of the model is compared with the decision value. When the signal is lower than the decision value, the sample belongs to one category; when the signal is higher than the decision value, the sample belongs to the other category.

[0099] When the computational model is multi-class, each category corresponds to an output signal. When a signal is at least twice as high as the other signals, the sample is judged as the category corresponding to the signal.

[0100] In some embodiments of the present invention, the cancer is selected from at least three of the following: brain cancer, lung cancer, hepatocellular carcinoma, breast cancer, stomach cancer, colon cancer, rectal cancer, cervical cancer, ovarian cancer, pancreatic cancer, prostate cancer, thyroid cancer, lymphoma, and skin cancer. Example 1: A DNA molecular multi-cancer diagnostic system based on RNA expression levels.

[0101] This invention provides a DNA molecular multi-cancer diagnostic system based on RNA expression level, the system comprising: an input module, a classification module, and an output module.

[0102] The input module extracts RNA from serum samples and converts it into ssDNA. Figure 2 The molecular reactions during the transformation process are shown, and the specific steps are as follows:

[0103] Step 1: RNA extraction from serum samples was performed using a kit, with all procedures strictly following the kit instructions. The obtained RNA was stored at -80°C for later use.

[0104] Step 2: Add 10 μL of reverse transcription buffer, 2 μL of reverse transcriptase mixture, and 8 μL of extracted whole miRNA to a 200 μL centrifuge tube. Anneal the sample in a PCR instrument using a programmed 37°C for 60 minutes followed by an 85°C 5-minute annealing cycle. (See attached image.) Figure 2 In the above reaction, miRNA binds to the reverse transcription primer, and the primer extends to obtain cDNA under the action of enzymes.

[0105] Step 3: Amplify the 36 cDNAs in 36 centrifuge tubes. For each cDNA, add 1× amplification buffer, 1 μM excess primer, 50 nM restriction primer, and 80 U / mL L Vent (exo-) DNA polymerase to a 200 μL centrifuge tube, resulting in an amplification volume of 20 μL. The amplification program in the PCR instrument is as follows: (1) 95℃ for 5 minutes, (2) 95℃ for 10 seconds, 58℃ for 10 seconds, and 72℃ for 20 seconds, for 40 cycles, (3) 72℃ for 10 minutes, see [link to PCR program]. Figure 2 In the intermediate reaction, the cDNA strand first binds to the restriction primer, the restriction primer extends to form a double strand, and in the subsequent PCR process, the restriction primer is quickly consumed completely, and finally the concentration of the cDNA single strand increases linearly.

[0106] Step 4: Transform the 36 cDNAs into 36 centrifuge tubes. For each cDNA, add 1× amplification buffer (20 mM Tris-HCl, 10 mM (NH4)2SO4, 40 mM potassium chloride, 10 mM MgSO4, 2 μM fusiformin, 200 μg / mL BSA, 0.1% Triton X-100, and 1 mM dNTP) to a 200 μL centrifuge tube, 16 nM cT template strand, 20 U / mL Nt.BstNBI nickase, 180 U / mL Vent(exo-) DNA polymerase, and 2 μL of the amplified cDNA, resulting in an amplification volume of 20 μL. Anneal the cDNA in a PCR instrument at 50°C for 120 minutes using a programmed annealing method. See [link to PCR instructions]. Figure 2 In the reaction below, cDNA binds to the cT strand and is extended and endonucleated under the action of enzymes to obtain ssDNA, which can be used as the input strand for DNA molecular computing networks.

[0107] Step 5: Purify ssDNA using 8% polyacrylamide gel electrophoresis.

[0108] The classification module includes a DNA molecular computational network that implements silicon-based diagnostic models at the molecular level.

[0109] The silicon-based diagnostic model was trained based on an existing RNA expression level database. Figure 3 The study showcases silicon-based diagnostic models, including an SVM algorithm for diagnosing cancer and an FCNN model for diagnosing specific cancer types. The expression level data of a sample is first fed into the SVM algorithm; when the output is greater than 0, the sample is deemed diseased. The expression level data is then fed into the FCNN model to determine the final cancer type.

[0110] Figure 4The test set confusion matrix of the silicon-based diagnostic model is shown. For the SVM algorithm, out of 202 healthy samples, 198 were diagnosed as healthy and 4 as cancer; out of 265 cancer samples, 261 were diagnosed as cancer and 4 as healthy. The model's accuracy is 98.3%, specificity is 98.0%, sensitivity is 99.6%, and precision is 98.5%. For the FCNN model, out of 162 lung cancer samples, 143 were diagnosed as lung cancer, 7 as hepatocellular carcinoma, and 12 as ovarian cancer; out of 32 hepatocellular carcinoma samples, 4 were diagnosed as lung cancer, 27 as hepatocellular carcinoma, and 1 as ovarian cancer; out of 72 lung cancer samples, 7 were diagnosed as lung cancer, 7 as hepatocellular carcinoma, and 58 as ovarian cancer. The model achieved an accuracy of 85.7%, a specificity of 89.4%, a sensitivity of 88.3%, and a precision of 92.9% for lung cancer, a specificity of 94.0%, a sensitivity of 84.4%, and a precision of 65.9% for hepatocellular carcinoma, and a specificity of 93.3%, a sensitivity of 80.6%, and a precision of 81.7% for ovarian cancer.

[0111] The DNA molecular computational network implements a silicon-based diagnostic model at the DNA molecular level, enabling direct detection and analysis to obtain the final diagnostic result based on RNA expression levels. The ssDNA strands converted from RNA in the sample are first input into a DNA-based SVM computational network. When the normalized output signal is greater than 0.5, a high output signal is considered to exist, indicating that the sample is diseased. The ssDNA strands converted from RNA in the sample are then input into a DNA-based FCNN computational network to determine the final cancer type.

[0112] In the classification module, the signals generated by the DNA computing network are detected using analytical instruments such as fluorescence microscopes or ELISA readers.

[0113] The output module outputs the diagnostic results.

[0114] Figure 5 The confusion matrix demonstrates the performance of the DNA molecular diagnostic system. All five healthy samples and five lung cancer samples were correctly diagnosed. One of the five hepatocellular carcinoma samples and one of the five ovarian cancer samples was diagnosed as lung cancer. The overall accuracy of the system is 90%, with all indicators for healthy samples at 100.0%. Specifically, for lung cancer, the specificity is 86.7%, sensitivity is 100%, and precision is 71.4%; for hepatocellular carcinoma, the specificity is 100.0%, sensitivity is 80.0%, and precision is 100.0%; and for ovarian cancer, the specificity is 100.0%, sensitivity is 80.0%, and precision is 100.0%.

[0115] Example 2: A method for constructing a DNA molecular diagnostic system based on RNA expression levels

[0116] This invention also proposes a method for constructing a multi-cancer diagnostic system. Figure 6 The construction method is illustrated, including the following steps:

[0117] Step 1: Establish a silicon-based diagnostic model based on RNA expression information;

[0118] Step 2: The parameters (including mathematical operations and weights) of the model obtained in Step 1 are encoded in a DNA computing network to realize a silicon-based diagnostic model at the molecular level, thereby obtaining the multi-cancer diagnostic system.

[0119] Step 3: Obtain the classification module for disease diagnosis, and then construct the DNA molecular diagnostic system.

[0120] In step one, Figure 7 The method for establishing the silicon-based diagnostic model is demonstrated, including the following steps:

[0121] Step 1.1: Obtain microarray data on serum RNA expression levels of patients with different cancer types and healthy individuals from NCBI;

[0122] Step 1.2: Based on the microarray data of serum RNA expression levels obtained in Step 1.1, the characteristic RNA groups of the multi-cancer model are selected using variance screening and principal component analysis.

[0123] Step 1.3: Train a classifier based on the data of the feature RNA groups selected in Step 1.2 to construct a silicon-based diagnostic model.

[0124] In step 1.3, in one specific implementation, the silicon-based diagnostic model includes two machine learning algorithms: a binary classification SVM algorithm to distinguish between healthy and cancer samples; and a multi-class classification FCNN algorithm to determine the specific type of cancer. The results of both algorithms are combined to provide the final diagnostic result.

[0125] In step 1.3, the metrics for evaluating the performance of the silicon-based diagnostic model include accuracy, sensitivity, specificity, and precision.

[0126] In step two, the silicon-based diagnostic model is implemented at the DNA molecular level, establishing a DNA computational network. Figure 8 The steps for building a DNA molecular computational network for multi-cancer diagnosis are demonstrated, including:

[0127] Step 2.1: Convert the silicon-based diagnostic model into a combination of three types of calculation functions: one or more multiplications, one or more additions, and one or more subtractions.

[0128] Step 2.2: Implement various computational functions at the DNA molecular level;

[0129] Step 2.3: Integrate computational functions and construct a DNA molecule computational network. The weights and other parameters of the silicon-based diagnostic model, as well as the input and output values, are encoded using the number of DNA molecules participating in the reaction or the concentration of DNA molecules.

[0130] The weighting parameter is encoded by the number of weighted molecules that react with the input ssDNA strand. For example, assigning a weight of +3 to the input means that in the DNA molecule computation network, one input ssDNA molecule can react with three weighted molecules to produce three weighted input strands with the same downstream sequence. The sum of their concentrations is three times the concentration of the input ssDNA strand, thus achieving a weighted multiplication of +3.

[0131] Input or output values ​​are encoded using the concentration of the input or output strand in the DNA molecular computing network. For example, in a DNA-based FCNN computing network, the concentration of different categories of output strands corresponds to the output value; the higher the output value, the higher the concentration of the corresponding output strand, and the stronger the detected fluorescence signal.

[0132] In step 2.1, the SVM algorithm executes a linear kernel function, which, according to its calculation formula, can be broken down into one or more multiplication functions, one or more addition functions, and one subtraction function. In the FCNN model, each neuron in the Dense layer performs a multiplication-accumulation operation, which, according to its calculation formula, can be broken down into one or more multiplication functions, one or more addition functions, and one subtraction function. One or more parallel multiplication-accumulation operations are executed in the FCNN model.

[0133] In step 2.2, the DNA-based multiplication function can perform the calculation of multiplying the decimal input by the integer weight. Through the reaction between the DNA input strand and the weight molecules, the concentration of the weighted input strand is generated as a corresponding multiple of the input strand's concentration. The sign of the weight is encoded by the downstream sequence of the output strand.

[0134] In step 2.2, the DNA-based addition function can sum the positive (negative) weighted inputs. DNA strands with the same downstream sequence are converted into the same weighted sum strand by the addition gate, and its concentration is the same as the sum of the input strand concentrations.

[0135] In step 2.2, the DNA-based subtraction function can perform the subtraction of positive weighted sums and negative weighted sums. Positive weighted sums and negative weighted sums annihilate each other, and downstream outputs are only generated when the concentration of positive weighted sums is higher.

[0136] In step 2.3, the molecular design of DNA-based computational functions is modular. By adjusting the upstream and downstream sequences, computational functions can be easily tandemly to construct a DNA computational network and perform molecular diagnostics.

[0137] Example 3

[0138] This embodiment is based on the above embodiment 2, and the similarities with the above embodiment 1 will not be repeated.

[0139] This embodiment introduces a method for establishing a silicon-based diagnostic model, including the following steps:

[0140] This study collected miRNA expression data from serum samples of 1326 women with lung cancer, hepatocellular carcinoma, and ovarian cancer, and 1009 healthy women, and established a model.

[0141] Step 1: Obtain microarray data on serum RNA expression levels in patients with different cancer types and healthy individuals.

[0142] 1.1 Microarray data on serum miRNA expression levels in patients with lung cancer (GSE137140), hepatocellular carcinoma (GSE113740), ovarian cancer (GSE106817), and healthy individuals were obtained from NCBI;

[0143] 1.2 Select female samples from the database;

[0144] 1.3 Randomly select 25% of the data from the healthy sample to maintain a balance in the amount of data across different categories;

[0145] 1.4 The data were logarithmic (base 2) to assess miRNA expression;

[0146] Step 2: Select the characteristic RNA set of multiple cancer models.

[0147] 2.1 Low-expression miRNAs were excluded; RNAs with expression levels below 2 in more than half of the samples were removed from the database.

[0148] 2.2 Using variance screening, 1181 RNAs with a significance level p < 0.01 were selected;

[0149] 2.3 The importance of each RNA for classification was assessed using PCA, and the 36 RNAs with the largest weight coefficients in PCA were selected as the characteristic RNA group for multi-cancer diagnosis.

[0150] Step 3: Train the classifier and build a cancer diagnosis model.

[0151] 3.1 Using expression level data of 36 miRNAs from 2335 samples as the database for the silicon-based computational model, stratified sampling was performed at an 8:2 ratio to obtain the training and validation sets. The binary classification SVM algorithm was trained using the svm.LinearSVC module in the Python library sklearn. During the SVM algorithm training, the intercept was set to 0, and the maximum number of iterations was set to 10000.

[0152] 3.2 Using miRNA expression data from 1326 diseased samples as the database for the silicon-based computational model, stratified sampling was performed at an 8:2 ratio to obtain the training and validation sets. The three-class FCNN model was trained on the Keras platform. The model includes an input layer and a Dense layer, with the Softmax function used as the activation function for the output layer. During FCNN model training, the learning rate, batch size, and epochs were set to 0.003, 36, and 5000, respectively. The Adam optimizer was used for parameter optimization of FCNN.

[0153] Example 4

[0154] This embodiment is based on the above embodiment 3, and the similarities with the above embodiment 1 will not be repeated.

[0155] This embodiment describes a method for building a cancer diagnostic model by training a classifier using other parameters, including the following steps:

[0156] 4.1 Using expression level data of 36 miRNAs from 2335 samples as the database for the silicon-based computational model, stratified sampling was performed at a 7:3 ratio to obtain the training and validation sets. The binary classification SVM algorithm was trained using the svm.LinearSVC module in the Python library sklearn. During the SVM algorithm training, the intercept was set to 0, and the maximum number of iterations was set to 1000.

[0157] 4.2 Using miRNA expression data from 1326 diseased samples as the database for the silicon-based computational model, stratified sampling was performed at an 8:2 ratio to obtain the training and validation sets. The three-class FCNN model was trained on the Keras platform. The model includes an input layer and a Dense layer, with the Softmax function used as the activation function for the output layer. During FCNN model training, the learning rate, batch size, and epochs were set to 0.001, 64, and 10000, respectively. The Adam optimizer was used for parameter optimization of FCNN.

[0158] Figure 9The test set confusion matrix of the silicon-based diagnostic model is shown. For the SVM algorithm, out of 303 healthy samples, 293 were diagnosed as healthy and 10 as cancer; out of 398 cancer samples, 392 were diagnosed as cancer and 6 as healthy. The model's accuracy is 97.7%, specificity is 96.7%, sensitivity is 98.5%, and precision is 97.5%. For the FCNN model, out of 162 lung cancer samples, 146 were diagnosed as lung cancer, 5 as hepatocellular carcinoma, and 13 as ovarian cancer; out of 32 hepatocellular carcinoma samples, 7 were diagnosed as lung cancer and 25 as hepatocellular carcinoma; out of 72 lung cancer samples, 9 were diagnosed as lung cancer, 2 as hepatocellular carcinoma, and 59 as ovarian cancer. The model achieved an accuracy of 86.5%, a specificity of 84.3%, a sensitivity of 89.0%, and a precision of 90.1% for lung cancer, a specificity of 97.0%, a sensitivity of 78.1%, and a precision of 78.1% for hepatocellular carcinoma, and a specificity of 93.4%, a sensitivity of 84.3%, and a precision of 81.9% for ovarian cancer.

[0159] Example 5

[0160] This embodiment is based on the above embodiment 3, and the similarities with the above embodiment 1 will not be repeated.

[0161] This embodiment introduces a method for constructing a cancer diagnostic model using only a fully connected neural network model. The silicon-based diagnostic model directly distinguishes between healthy samples and three types of cancer samples, including the following steps:

[0162] Using expression level data of 36 miRNAs from 2335 samples as the database for the silicon-based computational model, stratified sampling was performed at an 8:2 ratio to obtain the training and validation sets. The four-class classification FCNN model was trained on the Keras platform. The model includes an input layer and a dense layer, with the softmax function used as the activation function for the output layer. During FCNN model training, the learning rate, batch size, and epochs were set to 0.003, 36, and 10000, respectively. The Adam optimizer was used for parameter optimization of FCNN.

[0163] Figure 10The test set confusion matrix of the silicon-based diagnostic model is shown. Among the 162 lung cancer samples in the model, 152 were diagnosed as lung cancer, 2 as hepatocellular carcinoma, 6 as ovarian cancer, and 2 as healthy. Among the 31 hepatocellular carcinoma samples, 6 were diagnosed as lung cancer, 22 as hepatocellular carcinoma, 1 as ovarian cancer, and 2 as healthy. Among the 72 lung cancer samples, 11 were diagnosed as lung cancer, 2 as hepatocellular carcinoma, 56 as ovarian cancer, and 3 as healthy. All 202 healthy samples were diagnosed as healthy. The model achieved an accuracy of 92.9%, a specificity of 94.4%, a sensitivity of 95.0%, and a precision of 89.9% for lung cancer, a specificity of 99.1%, a sensitivity of 71.0%, and a precision of 84.6% for hepatocellular carcinoma, a specificity of 98.2%, a sensitivity of 77.8%, and a precision of 88.9% for ovarian cancer, and a specificity of 98.1%, a sensitivity of 100.0%, and a precision of 97.6% for healthy individuals.

[0164] Those skilled in the art will understand that the embodiments of this application can be provided as methods, systems, computer program products, or cancer diagnosis-related products. Therefore, the silicon-based diagnostic model in the software embodiments of this application can be implemented independently as a hardware embodiment. This application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0165] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), computer program products, and cancer diagnosis-related products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions and / or basic experimental operations. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing device, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that specifies the functions outlined in one or more boxes. It can be designed and implemented in the workflow using different DNA molecular computing network architectures, different DNA molecule sequences, and different computing functions, based on fundamental experimental operations. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0166] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0167] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

[0168] The scope of protection of this invention is not limited to the above embodiments. Any variations and advantages that can be conceived by those skilled in the art without departing from the spirit and scope of this invention are included in this invention and are protected by the appended claims.

Claims

1. A DNA molecular diagnostic system based on RNA expression levels, characterized in that, The diagnostic system includes: an input module, a classification module, and an output module; The input module is used to convert RNA in the collected serum into DNA molecules and input them into the diagnostic system. The classification module includes a DNA molecular computing network, which implements a silicon-based diagnostic model at the molecular level. The output module is used to output the classification result calculated by the classification module.

2. The diagnostic system as described in claim 1, characterized in that, The DNA molecules input into the classification module represent the RNA expression level in the silicon-based diagnostic model at the molecular level through their concentration; the silicon-based diagnostic model includes a binary classification model for determining whether a disease is present and / or a multi-classification model for determining what kind of disease a patient has; The binary classification model includes a support vector machine; the multi-class classification model includes a fully connected neural network model.

3. The diagnostic system as described in claim 2, characterized in that, The support vector machine model maps the input RNA expression level data to a high-dimensional space using a kernel function, searches for a hyperplane that maximizes the margin between the two classes of data points, defines the hyperplane using support vectors, and determines the class of the newly input RNA expression level data using the sign of the decision function; and / or, The fully connected neural network model includes an input layer, one or more hidden layers, and an output layer. The input layer receives RNA expression level data, one or more hidden layers perform feature extraction and nonlinear transformation, and the output layer uses the Softmax activation function to output the probability of each sample category and selects the sample category with the highest probability for output.

4. The diagnostic system as described in claim 1, characterized in that, The DNA molecular computing network includes one or more multiplication computing units, one or more addition computing units, and one or more subtraction computing units. Through the modular splicing of multiple different computing units, the silicon-based diagnostic model is executed at the molecular level. The multiplication calculation unit generates a DNA weighted input chain through the reaction of the DNA input chain with the weight molecules. The concentration of the DNA weighted input chain is the product of the concentration of the DNA input chain and the number of weight molecules corresponding to the input in the network. The addition calculation unit converts DNA weighted input strands with the same downstream sequence into the same weighted sum strand under the action of the addition gate. The concentration of the weighted sum strand is the sum of the concentrations of all DNA weighted input strands with positive or negative weights. The subtraction calculation unit generates a downstream output chain concentration by mutually annihilating the positive weighted sum chain and the negative weighted sum chain. The concentration of the output chain is the concentration of the positive weighted sum chain minus the concentration of the negative weighted sum chain. No output is generated when the concentration of the negative weighted sum chain is higher than that of the positive weighted sum chain. The ReLU activation function is embedded into the subtraction calculation.

5. A method for constructing a DNA molecular diagnostic system based on RNA expression levels, characterized in that, The construction method includes the following steps: Step 1: Construct a silicon-based diagnostic model that includes binary classification and / or multi-class classification models; Step 2: Construct a DNA molecular computing network to perform the computations in the silicon-based diagnostic model described in Step 1 at the molecular level; Step 3: Obtain the classification module for disease classification, and then construct the DNA molecular diagnostic system.

6. The construction method as described in claim 5, characterized in that, The front and back ends of the classification module are connected to the input module and the output module, respectively, together forming the DNA molecular diagnostic system based on RNA expression level.

7. The construction method as described in claim 5, characterized in that, Step one, the construction of the silicon-based diagnostic model further includes the following steps: Step 1.

1. Obtain microarray data on serum RNA expression levels in patients with different disease types and healthy individuals; Step 1.

2. Filter the microarray data to obtain the characteristic RNA group; Step 1.

3. Train the binary classification model and / or multi-class classification model using the characteristic RNA group data to obtain the silicon-based diagnostic model.

8. The construction method as described in claim 7, characterized in that, In step 1.2, RNAs with expression levels below 2 in more than half of the samples are screened out; and / or, RNAs with a significance level p greater than 0.01 are screened out; and / or, PCA is used to assess the importance of each RNA for classification; and / or, In step 1.3, during the training of the binary classification model, the intercept is set to 0, and the maximum number of iterations is set to 1000-10000; during the training of the multi-class classification model, the learning rate, batch size, and number of epochs are set to 0.001-0.003, 36-64, and 5000-10000, respectively, and the Adam optimizer is used for parameter optimization; during the training of both the binary classification model and the multi-class classification model, stratified sampling is performed using a preset data ratio to generate training and test sets.

9. The construction method as described in claim 5, characterized in that, In step two, the silicon-based diagnostic model is transformed into a concatenation of three types of computational functions: one or more multiplications, one or more additions, and one or more subtractions, and these computational functions are implemented at the DNA molecular level; the parameters, input, or output values ​​of the silicon-based diagnostic model are encoded using the number or concentration of DNA molecules participating in the reaction; and / or, The weight parameters of the silicon-based diagnostic model are encoded by the number of weighted molecules that react with the input DNA strand in the DNA molecule computing network; the input or output values ​​of the silicon-based diagnostic model are encoded by the concentration of the input or output strand used in the DNA molecule computing network; and / or, The silicon-based diagnostic model is computed by splicing together the same and / or different computational units in the DNA molecular computing network.

10. The application of the DNA molecular diagnostic system based on RNA expression level as described in any one of claims 1-4, or the method for constructing the diagnostic system as described in any one of claims 5-9, in biomarker identification and biological gene expression analysis.

Images