Breast tumor diagnosis method based on molecular learning vector quantization neural network

Abstract

The present invention proposes a breast tumor diagnosis method based on a molecular learning vector quantization neural network, the steps are: determine the functions of the input layer, competition layer and linear output layer of the learning vector quantization neural network, respectively compared with the DNA chain displacement reaction; design the functional modules corresponding to the input layer, competition layer and linear output layer of the learning vector quantization neural network according to the DNA chain displacement reaction; determine the structure and small fulcrum structure of the DNA chain of the auxiliary and reactant in the reaction process of each functional module, and determine the base sequence of the structural domain in each DNA chain; cascade each functional module into a DNA-based learning vector quantization neural network according to the structure of the learning vector quantization neural network, and set the concentration; obtain female breast tumor cases from the breast cancer database for preprocessing; use the DNA-based learning vector quantization neural network through Visual DSD to diagnose the benign or malignant female breast tumor cases after preprocessing. The molecular learning vector quantization neural network designed by the present invention has great potential in early cancer diagnosis and personalized precision medicine, and successfully diagnoses the benign and malignant nature of female breast tumors.

Description

Technical Field

[0001] The invention relates to the technical field of DNA molecular circuits, and in particular to a breast tumor diagnosis method based on a molecular learning vector quantization neural network. Background Art

[0002] Nowadays, breast cancer has no obvious clinical symptoms in the early stage of onset and is not easy to detect. Breast cancer has obvious heterogeneity, and its pathological characteristics are composed of malignant cells and normal cells. Therefore, the diagnosis and treatment of breast cancer also face huge challenges. The vigorous development of nanomedicine has provided strong support for the diagnosis and treatment of breast cancer. Liposomes, magnetic nanoparticles, nanogels, etc. have been used in clinical treatment. However, these nanomedicine technologies lack the ability to autonomously calculate biological samples and cannot show sufficiently intelligent behavior when conducting diagnosis and treatment.

[0003] DNA computers have attracted great attention from scientists due to their small size, large storage, low energy consumption and high parallel computing capabilities. They are composed of a series of biological molecules in solution form, and mainly perform calculations through biochemical reactions to achieve rapid processing of molecular information. DNA chains have the characteristics of high programmability, huge parallelism and nanoscale. They are an easily accessible and robust nanoscale engineering material and have been widely used in NP-complete computing problems, biosensor devices, nanorobots, drug delivery, disease diagnosis and nanocircuits. DNA logic circuits based on cancer diagnosis and treatment have been developed and have shown fewer side effects and higher therapeutic effects in the face of cancer treatment. However, the lower flexibility and learning ability expose the difficulties it faces in dealing with large-scale and complex tasks.

[0004] In recent years, neural networks have become popular. Artificial neural networks have strong learning capabilities and have demonstrated their superiority in being able to think independently when faced with complex and changing problems. Through learning and training, neural networks can effectively extract key features and patterns from input data. This enables artificial neural networks to demonstrate outstanding capabilities in many fields such as image recognition, speech recognition, natural language processing, and medical diagnosis. DNA-based artificial neural networks have been developed and have demonstrated their powerful computing power and huge application potential. Combined with the biological non-exclusion of DNA molecules, it has promoted its wider application in disease diagnosis and nanomedicine.

[0005] DNA strand displacement technology (DSDT) provides a rich and intelligent tool for DNA computing. Compared with other DNA self-assembly techniques, DSDT has unique advantages. According to the Watson-Crick base complementarity principle, DNA strand displacement (DSD) can react autonomously at room temperature without any other catalyst. In addition, DSD can dynamically demonstrate the interaction process of information and predict the reaction results. Compared with other DNA self-assembly techniques, DSD is easier to be used by programs. In the DSD process, the small branch point domain on the single-stranded DNA species is paired with the small branch point domain exposed on the double-stranded DNA species that is complementary to it, and the branch structure migration domains of the two match, then a new DNA single strand can be displaced and a new DNA double strand can be generated. A free DNA single strand can represent a signal, and through the DNA strand displacement reaction, the boundary between another single strand and the double strand can be replaced by complementary toe connection and branch migration, so that new signals can be easily obtained. Summary of the invention

[0006] In view of the technical problem that the existing DNA logic circuits have low flexibility and learning ability when diagnosing cancer, the present invention proposes a breast tumor diagnosis method based on molecular learning vector quantization neural network. A learning vector quantization neural network is constructed through DNA chain displacement reaction, which is then applied to the diagnosis of benign and malignant female breast tumors. The results are verified by Visual DSD software. The results show that the benign and malignant diagnosis experiment of female breast tumors can be successfully completed with an accuracy rate of 94%.

[0007] In order to achieve the above object, the technical solution of the present invention is implemented as follows: a method for diagnosing breast tumors based on molecular learning vector quantization neural network, the steps of which are as follows:

[0008] Step 1, determine the functions of the input layer, competition layer and linear output layer of the learning vector quantization neural network, and compare them with the DNA strand displacement reaction respectively;

[0009] Step 2, according to the DNA chain displacement reaction, the functional modules corresponding to the input layer, the competition layer and the linear output layer of the learning vector quantization neural network are designed respectively, and the chemical reactions of the functional modules are determined;

[0010] Step 3: Determine the structure of the DNA chain and the small support structure of the auxiliary and reactant in the reaction process of each functional module, and determine the base sequence of the domain in each DNA chain. Use Visual DSD to write the required DNA chain and perform functional verification;

[0011] Step 4: cascade each functional module into a DNA-based learning vector quantization neural network according to the structure of the learning vector quantization neural network, program it into Visual DSD for concentration setting, and perform functional verification;

[0012] Step 5, obtaining female breast tumor cases from the breast cancer database and performing preprocessing;

[0013] Step six, Visual DSD uses a DNA-based learning vector quantization neural network to diagnose benign or malignant female breast tumor cases after preprocessing.

[0014] Preferably, the learning vector quantization neural network includes a competition layer and a linear output layer, a full connection is adopted between the input layer and the competition layer, a partial connection is adopted between the competition layer and the linear output layer, the number of neurons in the competition layer is always greater than the number of neurons in the linear output layer, each competition layer neuron is only connected to one linear output layer neuron, and the connection weight is always 1; each linear output layer neuron can be connected to multiple competition layer neurons, and the values ​​of the competition layer neurons and the linear output layer neurons can only be 1 or 0; when the input value is sent in through the input layer, the competition layer neuron closest to the input vector is activated and the state is "1", the state of other competition layer neurons is "0", the state of the linear output layer neuron connected to the activated competition layer neuron is "1", and the states of other linear output layer neurons are all "0";

[0015] The reaction mechanism of the DNA chain displacement is a chain displacement reaction based on the small pivot domain. The small pivot domain on the single-stranded DNA species is paired and combined with the complementary small pivot domain exposed on the double-stranded DNA species to displace a new DNA single strand and generate a new DNA double strand. If the newly generated DNA single strand and the small pivot domain on the DNA double strand are still complementary, and the branch structure migration domain on the small pivot domain matches, this reaction is a reversible DNA chain displacement reaction. The DNA chain displacement reaction is a spontaneous, dynamic, cascaded reversible reaction process.

[0016] Preferably, the functional modules of the input layer include an input activation module, a subtraction annihilation module, and an absolute value summation module; the functional modules of the competition layer include a signal inversion module, an inversion summation module, and an annihilation module; and the functional modules of the linear output layer include a report summation module and a report reaction module;

[0017] The chemical reaction expression of the input activation module is:

[0018]

[0019] Among them, Xi represents the activation factor, XFi represents the fuel chain, Ci j * represents input substrate, Wij * represents weight substrate, Ci j Represents the activated input, Wi j Represents the activated weight;

[0020] The chemical reaction expression of the subtractive annihilation module is:

[0021]

[0022] K f >>K s

[0023] Among them, K f is the rate constant of the subtractive annihilation reaction, K s is the absolute value summation reaction rate constant, where K f >>K s SGi j represents the subtraction gate, and waste represents the waste chain;

[0024] The chemical reaction expression of the absolute value summation module is:

[0025]

[0026] Among them, D j represents the distance factor;

[0027] The chemical reaction expression of the signal reversal module is:

[0028] D j +SRG jk →SF jk

[0029]

[0030] Among them, SRG jk Represents the signal inversion gate, SF jk represents the inversion factor;

[0031] The chemical reaction expression of the inversion summation module is:

[0032] SF jk +RSG k →S k

[0033] Among them, RSG k represents the inverting sum gate, S k Represents a reversal signal;

[0034] The chemical reaction expression of the annihilation module is:

[0035]

[0036]

[0037] Among them, K f is the rate constant of the annihilation reaction, Anh kj Represents the annihilation chain;

[0038] The chemical reaction expression of the report summation module is:

[0039]

[0040] Among them, K s is the rate constant for the reported summation reaction, and K s < <K f , R.E. kl Representatives request and report gate, RY kl represents the reporting signal chain;

[0041] The chemical reaction expression of the reporter reaction module is:

[0042] RY kl +R l →Y l

[0043] Among them, R l Represents the reporting gate, Y l Represents output.

[0044] Preferably, the input activation module is a catalytic reaction that activates factor X i Before adding to the reaction, input chain C ij With weight chain W ij The concentrations of the input substrate C ij *With weight substrate W ij *Inside; only activated factor X i After addition, factor X is activated i With input substrate C ij * and weight substrate W ij *Perform a reversible strand displacement reaction to generate input strand C ij , weight chain W ij and an identical intermediate, intermediate with fuel chain XF i The reversible DNA strand displacement reaction is repeated to regenerate activated factor X i , so that as long as factor X is activated i Exists and fuel chain XF i is sufficient for the reaction to convert all the input substrate C ij *With weight substrate W ij * All are converted into input chain C ijWith weight chain W ij ;

[0045] The subtractive annihilation module takes the input chain C ij With weight chain W ij Subtract each other through annihilation reaction, where each group of input chain C ij With weight chain W ij They will continue to annihilate each other until only the input chain C remains. ij With weight chain W ij One of the reactions stops; the first step of the subtraction annihilation module is a reversible reaction, input chain C ij Or weight chain W ij A DNA strand in the subtraction gate SG ij The exposed foothold in the middle binds and migrates to the center; the second step is an irreversible reaction, input chain C ij Or weight chain W ij The other DNA chain in the first step combines with another exposed foothold on the intermediate product chain generated by the first step reaction and migrates to the midpoint, finally ij Breaking down into two waste chains;

[0046] The absolute value summation module refers to the input chain C remaining after the annihilation reaction is completed. ij Or weight chain W ij Sum the concentrations and add them to the distance factor D j The absolute value sum module is divided into two reactions. The first is the input chain C remaining in the reaction of the subtraction annihilation module. ij Or weight chain W ij Respectively with the input summing gate SumC ij Or weight sum gate SumW ij Perform a reversible DNA strand displacement reaction to generate the input summation strand IS j and weighted sum chain WS j , then enter the summing chain IS j and weighted sum chain WS j Then respectively with the sum gate Sum j The irreversible DNA strand displacement reaction is carried out to change the distance factor D j The irreversible DNA strand displacement reaction drives the first step of the subtractive annihilation module in the positive direction until all input strands C ij With weight chain W ij All have been reacted;

[0047] The signal reversal module and the reversal summation module are both irreversible DNA strand displacement reactions. The double-stranded products of the DNA strand displacement reactions no longer contain exposed footholds, and the product of the previous reaction is the reactant of the next reaction. The signal reversal module converts the distance factor D j As input, the signal inversion gate SRG jk DNA strand displacement reaction is performed, and the distance factor D j The concentration is evenly distributed to different reversal factors SF jk The inversion summation module is to convert different inversion factors SF jk Perform concentration summation and sum the concentration to the inverted signal chain S with the same subscript 'k' k middle;

[0048] The annihilation module is to link different inversion signals S k With the inverted signal chain S j Perform paired annihilation reactions until the signal chain S is reversed j With the inverted signal chain S k Only one of the reactions will stop; the annihilation module will reverse the signal chain S k Combined to the annihilation chain Anh kj The branch moves to the midpoint of the double-stranded domain if only the inverted signal chain S k exists, the process is completely reversible, and no molecules are consumed; if another reverse signal chain S j Also exists and can be combined into the annihilation chain Anh kj On the other side, the branch moves to the midpoint of the double-stranded domain; when the signal chain S is reversed k and invert the signal chain S j When they reach the midpoint at the same time, the annihilation chain Anh kj Splitting into two waste molecules, neither of which has an exposed foothold and cannot interact with any other molecule;

[0049] The DNA strand displacement reaction of the reporter summation module is a reversible reaction, which consumes the summation reporter gate RE. kl The reaction of the report summation reaction module is irreversible, the product of the previous reaction is the reactant of the next reaction, and the reaction is promoted in the positive direction; the report summation module is to sum the report gate RE kl With the inverted signal chain S k Perform DNA strand displacement reaction to generate reporter signal chain RY kl , where the summation reporting gate RE kl Branch migration domain Y lCorresponding to different linear outputs, it reflects the partial connection mode in the linear output layer of the learning vector quantization neural network; different branch migration domains Y l and the corresponding inverted signal chain S k Combine to realize the function of partial connection;

[0050] The reaction of the reporter reaction module is an irreversible strand displacement reaction, the product of the previous reaction is the reactant of the next reaction, and the reaction is driven in the positive direction; the reporter reaction module realizes the migration of the same branch domain Y l Sum and extract the output of the reaction to report the release of fluorescent output signal; report signal chain RY kl With reporting chain R l Irreversible strand displacement reaction to generate output strand Y l , separating the fluorophore and quencher labeled strands in the reporter gate, resulting in fluorescence enhancement, and the reaction result is obtained by detecting the fluorescence level of the remaining DNA species.

[0051] Preferably, the toehold used in the annihilation reaction of the subtractive annihilation module is G, which has 7 nucleotides; the first part of the absolute value summation module uses the input chain C ij With weight chain W ij The toehold H on the nucleotide sequence has 5 nucleotides;

[0052] The number and structure of each DNA chain are set by the number of input values ​​of the constructed molecular neural network and the corresponding modules, and the base sequence is generated by Visual DSD.

[0053] Preferably, the activated factor X i There are ten lines in total: X1, X2, X3, X4, X5, X6, X7, X8, X9, X 10 , the structures are<X1 G^> ,<X2 G^> ,<X3 G^> ,<X4 G^> ,<X5 G^> ,<X6 G^> ,<X7 G^> ,<X8 G^> ,<X9 G^> ,<X10 G^> ; Wherein, X1-X10 represent the names of 10 input branch migration domains, G represents the name of the small pivot domain, ^ represents the small pivot domain, and <> represents a single chain; Input substrate C ij * There are 30 items, namely C 11 * , C 12 * , C 13 * , C 14 * , C 15 * , C16 * 、C 17 * 、C 18 * 、C 19 * 、C 110 * 、C 21 * 、C 22 * 、C 23 * 、C 24 * 、C 25 * 、C 26 * 、C 27 * 、C 28 * 、C 29 * 、C 210 * 、C 31 * 、C 32 * 、C 33 * 、C 34 * 、C 35 * 、C 36 * 、C 37 * 、C 38 * 、C 39 * 、C 310 *, the structures are<H^C11> [G^X1]{G^*},<H^C12> [G^X2]{G^*},<H^C13> [G^X3]{G^*},<H^C14> [G^X4]{G^*},<H^C15> [G^X5]{G^*},<H^C16> [G^X6]{G^*},<H^C17> [G^X7]{G^*},<H^C18> [G^X8]{G^*},<H^C19> [G^X9]{G^*},<H^C110> [G^X10]{G^*},<H^C2> [G^X1]{G^*},<H^C22> [G^X2]{G^*},<H^C23> [G^X3]{G^*},<H^C24> [G^X4]{G^*},<H^C25> [G^X5]{G^*},<H^C26> [G^X6]{G^*},<H^C27> [G^X7]{G^*},<H^C28> [G^X8]{G^*},<H^C29> [G^X9]{G^*},<H^C210> [G^X10]{G^*},<H^C31> [G^X1]{G^*},<H^C32> [G^X2]{G^*},<H^C33> [G^X3]{G^*},<H^C34> [G^X4]{G^*},<H^C35> [G^X5]{G^*},<H^C36> [G^X6]{G^*},<H^C37> [G^X7]{G^*},<H^C38> [G^X8]{G^*},<H^C39> [G^X9]{G^*},<H^C310> [G^X10]{G^*}; where H represents the small branch domain, C11-C19, C21-C29, C31-C39, C110, C210, and C310 represent the names of the branch migration domains of 30 input substrates, * represents complementarity, {} represents single-stranded, and [] represents double-stranded; weight substrate W ij * There are 30 DNA chains, namely W 11 * , W 12 * , W 13 * , W 14 * , W 15 * , W 16 * , W 17 * , W 18 * , W 19 * , W 110 * , W21 * ,IN 22 * ,IN 23 * ,IN 24 * ,IN 25 * ,IN 26 * ,IN 27 * ,IN 28 * ,IN 29 * ,IN 210 * ,IN 31 * ,IN 32 * ,IN 33 * ,IN 34 * ,IN 35 * ,IN 36 * ,IN 37 * ,IN 38 * ,IN 39 * ,IN 310 *, the structures are<H^W11> [G^X1]{G^*},<H^W12> [G^X2]{G^*},<H^W13> [G^X3]{G^*},<H^W14> [G^X4]{G^*},<H^W15> [G^X5]{G^*},<H^W16> [G^X6]{G^*},<H^W17> [G^X7]{G^*},<H^W18> [G^X8]{G^*},<H^W19> [G^X9]{G^*},<H^W110> [G^X10]{G^*},<H^W2> [G^X1]{G^*},<H^W22> [G^X2]{G^*},<H^W23> [G^X3]{G^*},<H^W24> [G^X4]{G^*},<H^W25> [G^X5]{G^*},<H^W26> [G^X6]{G^*},<H^W27> [G^X7]{G^*},<H^W28> [G^X8]{G^*},<H^W29> [G^X9]{G^*},<H^W210> [G^X10]{G^*},<H^W31> [G^X1]{G^*},<H^W32> [G^X2]{G^*},<H^W33> [G^X3]{G^*},<H^W34> [G^X4]{G^*},<H^W35> [G^X5]{G^*},<H^W36> [G^X6]{G^*},<H^W37> [G^X7]{G^*},<H^W38> [G^X8]{G^*},<H^W39> [G^X9]{G^*},<H^W310> [G^X10]{G^*}; where W11-W19, W21-W29, W31-W39, W110, W210, and W310 represent the names of the branch migration domains of the 30 weight substrates, respectively;

[0054] Fuel Chain XF i There are 10 items, namely XF1, XF2, XF3, XF4, XF5, XF6, XF7, XF8, XFuel9, XF 10 , the structures are<G^X1> ,<G^X2> ,<G^X3> ,<G^X4> ,<G^X5> ,<G^X6> ,<G^X7> ,<G^X8> ,<G^X9> ,<G^X10> ;

[0055] Subtraction gate chain SG ij There are 30 items in total, namely SG 11 SG 12 SG 13 SG 14 SG 15 SG 16 SG 17 SG18 , SG 19 , SG 110 , SG 21 , SG 22 , SG 23 , SG 24 , SG 25 , SG 26 , SG 27 , SG 28 , SG 29 , SG 210 , SG 31 , SG 32 , SG 33 , SG 34 , SG 35 , SG 36 , SG 37 , SG 38 , SG 39 , SG 310 , with structures <G^*>[C11*W11]{G^*}, <G^*>[C12*W12]{G^*}, <G^*>[C13*W13]{G^*}, <G^*>[C14*W14]{G^*}, <G^*>[C15*W15]{G^*}, <G^*>[C16*W16]{G^*}, <G^*>[C17*W17]{G^*}, <G^*>[C18*W18]{G^*}, <G^*>[C19*W19]{G^*}, <G^*>[C110*W110]{G^*}, <G^*>[C21*W21]{G^*}, <G^*>[C22*W22]{G^*}, <G^*>[C23*W23]{G^*}, <G^*>[C24*W24]{G^*}, <G^*>[C25*W25]{G^*}, <G^*>[C26*W26]{G^*}, <G^*>[C27*W27]{G^*}, <G^*>[C28*W28]{G^*}, <G^*>[C29*W29]{G^*}, <G^*>[C210*W210]{G^*}, <G^*>[C31*W31]{G^*}, <G^*>[C32*W32]{G^*}, <G^*>[C33*W33]{G^*}, <G^*>[C34*W34]{G^*}, <G^*>[C35*W35]{G^*}, <G^*>[C36*W36]{G^*}, <G^*>[C37*W37]{G^*}, <G^*>[C38*W38]{G^*}, <G^*>[C39*W39]{G^*}, <G^*>[C310*W310]{G^*};

[0056] Input sum gate SumC ij A total of 30 items, including SumC 11 、SumC 12 、SumC 13 、SumC 14 、SumC 15 、SumC 16 、SumC 17 、SumC 18 、SumC 19 、SumC 110 、SumC 21 、SumC 22 、SumC 23 、SumC 24 、SumC 25 、SumC 26 、SumC 27 、SumC 28 、SumC 29 、SumC 210 、SumC 31 、SumC 32 、SumC 33 、SumC 34 、SumC 35 、SumC 36 、SumC 37 、SumC 38 、SumC 39 、SumC 310 , the structures are {H^*}[C11 T^] <d1>、{H^*}[C12 T^] <d1>、{H^*}[C13 T^] <d1>、{H^*}[C14 T^] <d1>、{H^*}[C15 T^] <d1>、{H^*}[C16 T^] <d1>,{H^*}[C17 T^] <d1>、{H^*}[C18 T^] <d1>、{H^*}[C19 T^] <d1>、{H^*}[C110 T^] <d1>、{H^*}[C21 T^] <d2>、{H^*}[C22 T^] <d2>、{H^*}[C23 T^] <d2>、{H^*}[C24 T^] <d2>、{H^*}[C25 T^] <d2>、{H^*}[C26 T^] <d2>、{H^*}[C27 T^] <d2>、{H^*}[C28 T^] <d2>、{H^*}[C29 T^] <d2>,{H^*}[C210 T^] <d2>、{H^*}[C31T^] <d3>、{H^*}[C32 T^] <d3><h2 style=";text-align:left;direction:ltr">{H^*}[C33 T^]<h2 style=";text-align:left;direction:ltr"> <d3>、{H^*}[C34 T^] <d3>、{H^*}[C35 T^] <d3>、{H^*}[C36 T^] <d3>、{H^*}[C37 T^] <d3>、{H^*}[C38 T^] <d3>、{H^*}[C39 T^] <d3>、{H^*}[C310 T^] <d3>; Where T represents the name of the small pivot domain, D1, D2, and D3 represent the distance factors D j The name of the branch migration domain;

[0057] Weight sum gate SumW ij A total of 30 items, including SumW 11 、SumW 12 、SumW 13 、SumW 14 、SumW 15 、SumW 16 、SumW 17 、SumW 18 、SumW 19 、SumW 110 、SumW 21 、SumW 22 、SumW 23 、SumW 24 、SumW 25 、SumW 26 、SumW 27 、SumW 28 、SumW 29 、SumW 210 、SumW 31 、SumW 32 、SumW 33 、SumW 34 、SumW 35 、SumW 36 、SumW 37 、SumW 38 、SumW 39 、SumW 310 , the structures are {H^*}[W11T^] <d1>、{H^*}[W12 T^] <d1>、{H^*}[W13 T^] <d1>、{H^*}[W14 T^] <d1>、{H^*}[W15T^] <d1>、{H^*}[W16 T^] <d1>、{H^*}[W17 T^] <d1>、{H^*}[W18 T^] <d1>,{H^*}[W19T^] <d1>、{H^*}[W110 T^] <d1>、{H^*}[W21 T^] <d2>、{H^*}[W22 T^] <d2>、{H^*}[W23T^] <d2>、{H^*}[W24 T^] <d2>、{H^*}[W25 T^] <d2>、{H^*}[W26 T^] <d2>、{H^*}[W27T^] <d2>、{H^*}[W28 T^] <d2>、{H^*}[W29 T^] <d2>、{H^*}[W210 T^] <d2>、{H^*}[W31T^] <d3>、{H^*}[W32 T^] <d3>、{H^*}[W33 T^] <d3>、{H^*}[W34 T^] <d3>、{H^*}[W35T^] <d3>、{H^*}[W36 T^] <d3>、{H^*}[W37 T^] <d3>、{H^*}[W38 T^] <d3>、{H^*}[W39T^] <d3>、{H^*}[W310 T^] <d3>;

[0058] Sum Gate j There are 3 items in total, namely Sum1, Sum2, Sum3, and their structures are {T^*}<H^> [D1]<H^> 、{T^*}<H^> [D2]<H^> 、{T^*}<H^> [D3]<H^> ;Signal reversal gate SRG jk There are 6 items in total, namely SRG 12 , SRG 13 , SRG 21 , SRG 23 , SRG 31 , SRG 32 , the structures are<G^L2> [H^D1]{H^*},<G^L3> [H^D1]{H^*},<G^L1> [H^D2]{H^*},<G^L3> [H^D2]{H^*},<G^L1> [H^D3]{H^*},<G^L2> [H^D3]{H^*}; where L1, L2, and L3 represent the inverted product numbers corresponding to different signal inversion gates; the inverted sum gate RSG k There are 3 lines in total, namely RSG1, RSG2, RSG3, and their structures are<H^S1> [G^L1]{H^*},<H^S2> [G^L2]{H^*},<H^S3> [G^L3]{H^*}; where S1, S2, and S3 represent the names of the branch migration domains of the inverted object; annihilation chain Anh kj There are 3 items in total, namely Anh 12 、Anh 23 、Anh 31 , the structures are<G^*> [S1*S2]{G^*},<G^*> [S2*S3]{G^*},<G^*> [S3*S1]{G^*}; Report sum gate RE kl There are 3 items in total, namely RE 11 ,RE 22 ,RE 32 , the structures are {H^*}[S1 H^] <y1>,{H^*}[S2H^] <y2>,{H^*}[S3 H^] <y2>; Report gate R l There are 2 lines in total, namely R1 and R2, and their structures are {H^*}[Y1] and {H^*}[Y2] respectively.

[0059] Preferably, the number of DNA chains and the structure of DNA chains required by the input activation module, the subtraction annihilation module, the absolute value summation module, the signal inversion module, the inversion summation module, the annihilation module, the report summation module and the report reaction module are respectively: activation factor X i With input substrate C ij The concentration of * is set as the average of 10 quantitative features of tumor cell nuclei, with a i nM; weight substrate W ij *The concentration uses the weight values ​​trained on the silicon-based computer, with b ij Expressed in nM; fuel chain XFuel i The concentration is set to [200% × (a i +b ij )]nM; subtraction gate SG ij The concentration is set to [105%×min(a i ,b ij )]nM; input sum gate SumC ij SumW ij The concentrations are all set to [105% × | a i -b ij |]nM; Sum gate Sum j The concentration is set to Use d j nM represents; all signal inversion gates SRG jk The concentration setting is [55%×max(d j )]nM; inverting sum gate RSG k The concentration is set to [100% × max (d j )]nM; Annihilation Gate kj The concentration is set to [55% × (max (d j )-min(d j ))]nM; Report summing gate RE kl The concentration is set to [100% × max (d j )]nM; Report gate R l The concentration is set to [200% × max (d j )]nM.

[0060] Preferably, the benign or malignant nature of the breast tumor is diagnosed by establishing a relationship between 10 quantitative characteristics of the breast tumor cell nucleus obtained by fine needle puncture and the benign or malignant nature of the cell; the 10 quantitative characteristics of the breast tumor cell nucleus are the radius, texture, circumference, area, smoothness, compactness, depression, number of depression points, symmetry, and fracture degree of the cell nucleus.

[0061] Preferably, a set of data for each female breast tumor case includes the average value, standard deviation and worst value of 10 quantitative features of each cell nucleus in the sampled tissue; the average value of the 10 quantitative features of the cell nucleus microscopic image of the breast tumor lesion tissue is used as the activation factor X of the DNA-based learning vector quantization neural network. i , breast tumor diagnosis results as output chain Y l , output chain Y l The value of is 1 or 0. It is defined that when the output chain Y1 is 1, the breast tumor is diagnosed as benign, and the output chain Y2 is 0; when the output chain Y2 is 1, the breast tumor is diagnosed as malignant, and the output chain Y1 is 0.

[0062] Preferably, 519 groups of data are randomly selected as training sets, and the remaining 50 groups of data are test sets. The designed DNA-based learning vector quantization neural network is trained with the data of the training set, and then the data of the test set is tested and the test results are analyzed; K-fold cross-validation is used to determine that the optimal number of neurons is 3. This DNA-based learning vector quantization neural network has the fastest training speed and the highest accuracy in breast tumor diagnosis.

[0063] Compared with the prior art, the beneficial effects of the present invention are as follows: first, the structure and function of the network input layer, hidden layer and output layer of the learning vector quantization neural network are determined; according to the structure and function of each layer of the learning vector quantization neural network, the chemical reaction network (CRN) reaction formula corresponding to the DNA chain displacement reaction is written; according to the CRN reaction formula, the DNA chain structure and small pivot structure of the auxiliary and reactant in the reaction process of each layer of the network are determined, and the base sequence of the domain in each DNA chain is determined; the network function of each layer is verified using Visual DSD software, and each layer of the network is cascaded into a learning vector quantization neural network, and functional verification is performed; breast cancer tumor case samples are obtained, and silicon-based computers are used for training to obtain the optimal weight and number of neurons in the competition layer; Visual DSD software is used to perform molecular learning vector quantization neural networks for benign and malignant diagnosis experiments of breast tumors. The molecular learning vector quantization neural network designed by the present invention has great potential in early cancer diagnosis and personalized precision medicine, and provides new ideas for the application of molecular circuits in the medical field. The designed molecular learning vector quantization neural network successfully diagnoses the benign and malignant nature of female breast tumors, and the accuracy rate reaches 94%. The computing power of DNA-based molecular learning vector quantization neural networks surpasses traditional DNA logic circuits, demonstrating intelligence embedded in molecular circuits. This behavior of embedding cognitive capabilities into molecular circuits has opened up new avenues for the development of molecular medicine. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0065] Figure 1 It is a flow chart of the present invention.

[0066] Figure 2 Schematic diagram of the DNA-based learning vector quantization neural network model and reaction principle of the present invention; wherein, (a) is the DNA-based learning vector quantization neural network model; (b) is the DNA chain displacement reaction principle.

[0067] Figure 3 It is a reaction flow chart of each layer of the DNA-based learning vector quantization neural network of the present invention; wherein, (a) is the reaction flow chart of the input layer of the DNA-based learning vector quantization neural network; (b) is the reaction flow chart of the competition layer of the DNA-based learning vector quantization neural network; (c) is the reaction flow chart of the output layer of the DNA-based learning vector quantization neural network.

[0068] Figure 4 The present invention is a schematic diagram of the chain displacement reaction process of each module of the DNA-based learning vector quantization neural network; wherein, (a) is the DNA chain displacement reaction process of the input activation module; (b) is the DNA chain displacement reaction process of the subtractive annihilation module; (c) is the reaction process of the subtractive annihilation module with negative weights; (d) is the DNA chain displacement reaction process of the absolute value summation module; (f) is the DNA chain displacement realization of the three-input input-takes-all network, and the reaction here includes the reaction processes of the signal inversion module, the inversion summation module, the annihilation module, the report summation module and the report reaction module.

[0069] Figure 5 The diagram shows the diagnosis results of benign and malignant breast tumors based on the DNA-based learning vector quantization neural network of the present invention; wherein (a) is the diagnosis result of a malignant case, and (b) is the diagnosis result of a benign case. DETAILED DESCRIPTION

[0070] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0071] like Figure 1 As shown, a method for diagnosing breast tumors based on molecular learning vector quantization neural network, the steps are as follows:

[0072] Step 1: Determine the functions of the input layer, competition layer, and linear output layer of the learning vector quantization neural network, and compare them with the DNA chain displacement reaction to consider their feasibility.

[0073] The learning vector quantization neural network consists of three layers of neurons: input layer, competition layer and linear output layer. Figure 2 As shown in (a), the input layer and the competitive layer are fully connected, and the competitive layer and the linear output layer are partially connected. The number of neurons in the competitive layer is always greater than the number of neurons in the linear output layer. Each competitive layer neuron is only connected to one linear output layer neuron, and the connection weight is always 1. However, each linear output layer neuron can be connected to multiple competitive layer neurons. The values ​​of competitive layer neurons and linear output layer neurons can only be 1 or 0. When the input value is sent to the network through the input layer, the competitive layer neuron closest to the input vector is activated, and its state is "1", and the states of other competitive layer neurons are "0". Therefore, the state of the linear output layer neuron connected to the activated competitive layer neuron is also "1", while the states of other linear output layer neurons are all "0". Figure 2 (a) X1-X i Represents the input value, D1-D j Represents the distance factor, W 11 -W ij represents the weight value, LTA represents the loser-takes-all network response mechanism, S1-S K Represents the inversion factor produced by the reaction, Y1-Y l Represents the output value.

[0074] The reaction mechanism of DNA strand displacement is based on the strand displacement reaction of the small branch domain. The principle of DNA strand displacement reaction is as follows Figure 2 As shown in (b), the small branch domain on the single-stranded DNA species is paired with the small branch domain exposed on the double-stranded DNA species that is complementary to it, and the branch structure migration domains of the two match, then a new DNA single strand can be displaced and a new DNA double strand can be generated. If the small branch domains on the newly generated DNA single strand and the DNA double strand are still complementary, and the branch structure migration domains on them match, then this reaction is a reversible chain displacement reaction. That is, the newly generated DNA single strand and DNA double strand can continue to undergo chain displacement reactions to regenerate the initial DNA single strand and DNA double strand, otherwise it is irreversible. Here, the digital quantity is converted into the concentration quantity of the DNA chain to construct a model of a learning vector quantization neural network based on DNA. The DNA chain displacement reaction is a spontaneous, dynamic, cascadable, reversible reaction process.

[0075] Step 2: According to the DNA chain displacement reaction, the functional modules corresponding to the input layer, competition layer and linear output layer of the learning vector quantization neural network are designed respectively, and the chemical reactions of the functional modules are determined.

[0076] like Figure 3 The figure shows the reaction flow chart of each layer of the DNA-based learning vector quantization (LVQ) neural network. According to the reaction characteristics of the DNA chain displacement reaction and the corresponding functions of each layer of neurons in the learning vector quantization neural network, the learning vector quantization neural network is divided into eight functional modules, namely, input activation module, subtraction annihilation module, absolute value summation module, signal inversion module, inversion summation module, annihilation module, report summation module and report reaction module. The first three modules are shown in Figure 3 As shown in (a), they correspond to the input layer of the LVQ neural network. The fourth to sixth modules are as follows Figure 3 (b) corresponds to the competitive layer in the LVQ neural network. The seventh and eighth modules are shown in Figure 3 (c) corresponds to the linear output layer of the LVQ neural network.

[0077] The chemical reaction expression input to the activation module is:

[0078]

[0079] Among them, X i XF stands for activating factor i represents the fuel chain, C ij * represents input substrate, W ij * represents weighted substrate, C ij represents the activated input, W ij Represents the activated weight.

[0080] The chemical reaction expression of the subtractive annihilation module is:

[0081]

[0082] No waste molecules were shown in the reaction, K f is the rate constant of the subtractive annihilation reaction, K s is the absolute value summation reaction rate constant, where K f >>K s Among them, SG ij stands for subtraction gate and waste stands for waste chain.

[0083] The chemical reaction expression of the absolute value summation module is:

[0084]

[0085] Where D j Represents the distance factor.

[0086] The chemical reaction expression of the signal reversal module is:

[0087]

[0088] Among them, SRG jk Represents the signal inversion gate, SF jk Represents the inversion factor.

[0089] The chemical reaction expression of the inverse summation module is:

[0090] SF jk +RSG k →S k (5)

[0091] Among them, RSG k represents the inverting sum gate, S k Represents a reversal signal.

[0092] The chemical reaction expression of the annihilation module is:

[0093]

[0094] Among them, K f is the rate constant of the annihilation reaction, which is the same as the subtractive annihilation module, Anh kj Represents the annihilation chain.

[0095] The chemical reaction expression reported by the summation module is:

[0096]

[0097] Among them, K s is the rate constant for the reported summation reaction, which is the same as the absolute value summation module, and K s < <K f , R.E. kl Representatives request and report gate, RY kl Stands for Report Signal Chain.

[0098] The chemical reaction expression of the reporting reaction module is:

[0099] RY kl +R l →Y l (8)

[0100] Among them, R l Represents the reporting gate, Y l Represents output.

[0101] The input activation module refers to the activation factor X i Before adding to the reaction, input chain C ij With weight chain W ij The concentrations of the input substrate C ij *With weight substrate W ij *Inside. Only activated factor X i After adding, the input and weight will be generated, and then the next reaction will be carried out. ij With weight chain W ij Subtract each other through annihilation reaction, where each group of input chain C ij With weight chain W ij They will continue to annihilate each other until only the input chain C remains. ij With weight chain W ij The absolute value sum module is to input the remaining input chain C after the annihilation reaction ends. ij Or weight chain W ij Sum the concentrations and sum them to the distance factor D with the same subscript "j" j In summary, the first three modules correspond to the input layer in the learning vector quantization neural network. Since it is difficult to directly construct the Euclidean distance solution model between the input layer vector and the competition layer neurons through the reaction between DNA chains, we have proposed an equivalent Manhattan distance solution model to solve the distance calculation problem between the input layer vector and the competition layer neurons. The fourth to sixth modules construct a loser-takes-all network to implement the competition layer of the learning vector quantization neural network. The loser-takes-all means that the output signal is "1" if and only if the corresponding input has the smallest analog value among all inputs. The first is the signal inversion module, which refers to the product of the absolute value summation module, that is, the distance factor D j As the input of the loser-takes-all network, it is then connected to the signal reversal gate SRG jk DNA strand displacement reaction is performed, and the distance factor D j The concentration is evenly distributed to different reversal factors SF jk The inversion summation module is to convert different inversion factors SF jk Sum the concentrations and add them to the inverted signal chain S with the same subscript 'k' k In this way, the signal is inverted, and the distance factor D j The species with the lowest concentration inverts the signal chain S k The species with the highest concentration in the . Similarly, all distance factors D j The concentrations of the species are reversed. The next step is the annihilation module, which is similar to the subtraction annihilation module, that is, the different inversion signal chains S k With the inverted signal chain S j Perform paired annihilation reactions until the signal chain S is reversed j With the inverted signal chain S k Only one of the reactions will stop, and the remaining species is the winner. These three modules realize the loser-takes-all function, which is equivalent to solving the competition layer neurons with the smallest distance to the input vector in the learning vector quantization neural network. Finally, there are two modules corresponding to the linear output layer of the learning vector quantization neural network. The first is the sum report module, the sum report gate RE kl With the inverted signal chain S k Perform strand displacement reaction to generate reporter signal chain RY kl , where the summation reporting gate RE kl Branch migration domain Y l Corresponding to different linear outputs, it reflects the partial connection mode in the linear output layer of the learning vector quantization neural network. Different branch migration domains Y l The corresponding inverted signal chain S k Finally, there is the reporting reaction module, which implements two functions. One is to migrate the same branch domain Y l The first step is to sum the results, and the second step is to extract and report the output results of the reaction. Through this step of the reaction, the fluorescent output signal can be released, which is convenient for the use of the detection output results.

[0102] like Figure 4 As shown, the gray circle with arrows indicates the direction of the catalytic cycle, the black arrows and white arrows indicate the forward and backward directions of the reaction steps, respectively, and the line segment without an arrow indicates that the reaction direction is irreversible. In all DNA chains (including the reactions below), the zigzag lines represent short branch point domains, containing 5 or 7 nucleotides. Straight lines represent long (15 or 20 nucleotides) branch migration domains, and arrows mark their 3' ends. Each domain is labeled with a name and assigned a unique DNA sequence, and the asterisk in the name indicates sequence complementarity.

[0103] Input activation module such as Figure 4 As shown in (a), it is a catalytic reaction. The input activation module is a catalytic reaction. The activation factor X i With input substrate C ij * and weight substrate W ij * A reversible strand displacement reaction is performed to generate the input strand C ij , weight chain W ij and an identical intermediate, intermediate with fuel chain XF i The reversible DNA strand displacement reaction is repeated to regenerate activated factor X i , so that as long as factor X is activated i Exists and fuel chain XF i is sufficient for the reaction to convert all the input substrate C ij * With weight substrate W ij * All are converted into input chain C ij With weight chain W ij , where activation factor X i The concentration of is set to be the same as the input substrate concentration. Figure 4 As shown in (b), the first step of the subtractive annihilation module is a reversible reaction, and the input chain C ij Or weight chain W ij With the subtraction gate SG ij The second step is an irreversible reaction. The other DNA strand, the input strand C ij Or weight chain W ij Combine with another foothold exposed on the intermediate product chain generated by the first step reaction and migrate to the midpoint, finally subtracting the gate SG ij Decomposed into two waste chains, the foothold used by the annihilation reaction here is G, which has 7 nucleotides. The absolute value sum module is as follows Figure 4 As shown in (d), the absolute value summation module is divided into two reactions. The first is the input chain C remaining in the reaction of the subtraction annihilation module. ij Or weight chain W ij and the input summing gate SumC ij Or weight sum gate SumW ij Perform a reversible strand displacement reaction to generate the input summation strand IS j and weighted sum chain WS j , then enter the summing chain IS j and weighted sum chain WS j Then respectively with the sum gate Sum j The irreversible DNA strand displacement reaction is carried out to change the distance factor D j The double-stranded product of the reaction here no longer contains exposed footholds, so the reaction is irreversible. The irreversible reaction will drive the first reversible reaction of the subtraction annihilation module in the positive direction until all input chains C ij With weight chain W ij All are reacted. Figure 4 As shown in (c), when the weight is negative, the branch migration domain W ij Set to DNA strand NW ij , which will result in a DNA strand NW representing a negative weight ij No longer participates in the reaction of the subtraction annihilation module. The corresponding input chain C ij When participating in the reaction of the subtractive annihilation module, since there is only one reactant, the reaction only carries out the first part of the reversible reaction, so the input chain C will not be consumed at all. ij This implements the absolute value sum module of the input and negative weights. The first part of the absolute value sum module uses the input chain C ij With weight chain W ij Toehold H on the DNA strand has 5 nucleotides. It is well known that the speed of the DNA chain displacement reaction increases exponentially with the increase of the toehold length. Previous studies have shown that the difference in reaction speed between a toehold of four nucleotides and a toehold of seven nucleotides is astonishingly thousands of times. The speed difference between toehold G and toehold H here is hundreds of times, so the reaction speed of the absolute value summation module will be much smaller than the reaction speed of the subtraction annihilation module. The reactants and products of the signal inversion module, inversion summation module, annihilation module, report summation module and report reaction module are as follows: Figure 4 (e) As shown. The signal reversal module and the inversion summation module are both irreversible DNA strand displacement reactions. The double-stranded products of the DNA strand displacement reaction no longer contain exposed footholds, and the product of the previous reaction is the reactant of the next reaction. All reactions occur naturally in sequence. The annihilation module is similar to the subtraction annihilation module. A reverse signal chain S k Can be combined with Annihilation Chain Anh kj The branch migrates to the midpoint of the double-stranded domain. k exists, then the process is completely reversible and no molecules are consumed. However, if another reverse signal chain S j If S is also present, it can bind to the other branch opposite the annihilation chain and branch to the midpoint of the double-stranded domain. k and invert the signal chain S j When they reach the midpoint at the same time, the annihilation chain Anh kj The DNA strand displacement reaction of the reporter summation module is a reversible reaction, which consumes the summation reporter gate RE. kl The reaction of the reporter reaction module is an irreversible chain displacement reaction. The product of the previous reaction is the reactant of the next reaction, and the reaction is driven in the positive direction.

[0104] The last is the reporting reaction module, which is used to extract the reaction results and generate the reporting signal chain RY kl With reporting chain R l Irreversible strand displacement reaction to generate output strand Y l , separating the fluorophore and quencher labeled strands in the reporter gate, resulting in fluorescence enhancement, and by detecting the fluorescence level of the remaining DNA species, the reaction result can be quickly obtained.

[0105] Step three, determine the structure and small support structure of the DNA chain of the auxiliary and reactant in the reaction process of each functional module, and determine the base sequence of the structural domain in each DNA chain, and then use Visual DSD to write the DNA chain required for each module and perform functional verification. The number and structure of each DNA chain are input through the constructed molecular neural network, and then set correspondingly to each module, and the base sequence is generated by Visual DSD.

[0106] Activator chain X i There are ten lines in total: X1, X2, X3, X4, X5, X6, X7, X8, X9, X 10 , whose structures are<X1G^> ,<X2 G^> ,<X3 G^> ,<X4 G^> ,<X5 G^> ,<X6 G^> ,<X7 G^> ,<X8 G^> ,<X9 G^> ,<X10 G^> , where X1-X10 represent the names of the 10 input branch migration domains, G represents the name of the small pivot domain, ^ represents the small pivot domain, <> represents a single chain, and the chain direction is from 5' to 3'. Input substrate C ij * There are 30 items, namely C 11 * , C 12 * , C 13 * , C 14 * , C 15 * , C 16 * , C 17 * , C 18 * , C 19 * , C 110 * , C 21 * , C 22 * , C 23 * , C 24 * , C 25 * , C 26 * , C 27 * , C 28 * , C 29 * , C 210 * , C 31 * , C 32 * , C 33 * , C 34 * , C 35 * , C 36 * , C 37 * , C 38 * , C 39 * , C 310 * , whose structures are<H^C11> [G^X1]{G^*},<H^C12> [G^X2]{G^*},<H^C13> [G^X3]{G^*},<H^C14> [G^X4]{G^*},<H^C15> [G^X5]{G^*},<H^C16> [G^X6]{G^*},<H^C17> [G^X7]{G^*},<H^C18> [G^X8]{G^*},<H^C19> [G^X9]{G^*},<H^C110> [G^X10]{G^*},<H^C2> [G^X1]{G^*},<H^C22> [G^X2]{G^*},<H^C23> [G^X3]{G^*},<H^C24> [G^X4]{G^*},<H^C25> [G^X5]{G^*},<H^C26> [G^X6]{G^*},<H^C27> [G^X7]{G^*},<H^C28> [G^X8]{G^*},<H^C29> [G^X9]{G^*},<H^C210> [G^X10]{G^*},<H^C31> [G^X1]{G^*},<H^C32> [G^X2]{G^*},<H^C33> [G^X3]{G^*},<H^C34> [G^X4]{G^*},<H^C35> [G^X5]{G^*},<H^C36> [G^X6]{G^*},<H^C37> [G^X7]{G^*},<H^C38> [G^X8]{G^*},<H^C39> [G^X9]{G^*},<H^C310> [G^X10]{G^*}. Among them, H represents the name of the small branch domain, C11-C19, C21-C29, C31-C39, C110, C210, and C310 represent the names of the branch migration domains of 30 input substrates, * represents complementarity, and only chains with and without * can be complementary, {} represents a single chain, and the chain direction is from 3' to 5', and [] represents a double chain.

[0107] Weight substrate W ij * There are 30 DNA chains, namely W 11 * , W 12 * , W 13 * , W 14 * , W 15 * , W 16 * , W 17 * ,IN 18 * ,IN 19 * ,IN 110 * ,IN 21 * ,IN 22 * ,IN 23 * ,IN 24 * ,IN 25 * ,IN 26 * ,IN 27 * ,IN 28 * ,IN 29 * ,IN 210 * ,IN 31 * ,IN 32 * ,IN 33 * ,IN 34 * ,IN 35 * ,IN 36 * ,IN 37 * ,IN 38 * ,IN 39 * ,IN 310 * , whose structures are<H^W11> [G^X1]{G^*},<H^W12> [G^X2]{G^*},<H^W13> [G^X3]{G^*},<H^W14> [G^X4]{G^*},<H^W15> [G^X5]{G^*},<H^W16> [G^X6]{G^*},<H^W17> [G^X7]{G^*},<H^W18> [G^X8]{G^*},<H^W19> [G^X9]{G^*},<H^W110> [G^X10]{G^*},<H^W2> [G^X1]{G^*},<H^W22> [G^X2]{G^*},<H^W23> [G^X3]{G^*},<H^W24> [G^X4]{G^*},<H^W25> [G^X5]{G^*},<H^W26> [G^X6]{G^*},<H^W27> [G^X7]{G^*},<H^W28> [G^X8]{G^*},<H^W29> [G^X9]{G^*},<H^W210> [G^X10]{G^*},<H^W31> [G^X1]{G^*},<H^W32> [G^X2]{G^*},<H^W33> [G^X3]{G^*},<H^W34> [G^X4]{G^*},<H^W35> [G^X5]{G^*},<H^W36> [G^X6]{G^*},<H^W37> [G^X7]{G^*},<H^W38> [G^X8]{G^*},<H^W39> [G^X9]{G^*},<H^W310> [G^X10]{G^*}. Wherein, W11-W19, W21-W29, W31-W39, W110, W210, and W310 represent the names of the branch migration domains of the 30 weight substrates, respectively.

[0108] Fuel Chain XF i There are 10 items, namely XF1, XF2, XF3, XF4, XF5, XF6, XF7, XF8, XFuel9, XF 10 , whose structures are<G^X1> ,<G^X2> ,<G^X3> ,<G^X4> ,<G^X5> ,<G^X6> ,<G^X7> ,<G^X8> ,<G^X9> ,<G^X10> .

[0109] Subtraction gate chain SG ij There are 30 items in total, namely SG 11 SG 12 SG 13 SG 14 SG 15 SG 16 SG 17 SG 18 SG 19 SG 110 SG 21 SG 22 SG 23 SG 24 SG 25 SG 26 SG 27 SG 28 SG 29 SG 210 SG 31 SG 32 SG 33 SG 34 SG 35 SG 36 SG 37 SG 38 SG 39 SG 310 The structures are<G^*> [C11*W11]{G^*},<G^*> [C12*W12]{G^*},<G^*> [C13*W13]{G^*},<G^*> [C14*W14]{G^*},<G^*> [C15*W15]{G^*},<G^*> [C16*W16]{G^*},<G^*> [C17*W17]{G^*},<G^*> [C18*W18]{G^*},<G^*> [C19*W19]{G^*},<G^*> [C110*W110]{G^*},<G^*> [C21*W21]{G^*},<G^*> [C22*W22]{G^*},<G^*> [C23*W23]{G^*},<G^*> [C24*W24]{G^*},<G^*> [C25*W25]{G^*},<G^*> [C26*W26]{G^*},<G^*> [C27*W27]{G^*},<G^*> [C28*W28]{G^*},<G^*> [C29*W29]{G^*},<G^*> [C210*W210]{G^*},<G^*> [C31*W31]{G^*},<G^*> [C32*W32]{G^*},<G^*> [C33*W33]{G^*},<G^*> [C34*W34]{G^*},<G^*> [C35*W35]{G^*},<G^*> [C36*W36]{G^*},<G^*> [C37*W37]{G^*},<G^*> [C38*W38]{G^*},<G^*> [C39*W39]{G^*},<G^*> [C310*W310]{G^*}.

[0110] Input sum gate SumC ij A total of 30 items, including SumC 11 、SumC 12 、SumC 13 、SumC 14 、SumC 15 、SumC 16 、SumC 17 、SumC 18 、SumC 19 、SumC 110 、SumC 21 、SumC 22 、SumC 23 、SumC 24 、SumC 25 、SumC 26 、SumC 27 、SumC 28 、SumC 29 、SumC 210 、SumC 31 、SumC 32 、SumC 33 、SumC 34 、SumC 35 、SumC 36 、SumC 37 、SumC 38 、SumC 39 、SumC 310 . Their structures are {H^*}[C11 T^] <d1>、{H^*}[C12 T^] <d1>、{H^*}[C13 T^] <d1>、{H^*}[C14 T^] <d1>、{H^*}[C15 T^] <d1>、{H^*}[C16 T^] <d1>,{H^*}[C17 T^] <d1>、{H^*}[C18 T^] <d1>、{H^*}[C19 T^] <d1>、{H^*}[C110 T^] <d1>、{H^*}[C21 T^] <d2>、{H^*}[C22 T^] <d2>、{H^*}[C23 T^] <d2>、{H^*}[C24 T^] <d2>、{H^*}[C25 T^] <d2>、{H^*}[C26 T^] <d2>、{H^*}[C27 T^] <d2>、{H^*}[C28 T^] <d2>、{H^*}[C29 T^] <d2>,{H^*}[C210 T^] <d2>、{H^*}[C31T^] <d3>、{H^*}[C32 T^] <d3><h2 style=";text-align:left;direction:ltr">{H^*}[C33 T^]<h2 style=";text-align:left;direction:ltr"> <d3>、{H^*}[C34 T^] <d3>、{H^*}[C35 T^] <d3>、{H^*}[C36 T^] <d3>、{H^*}[C37 T^] <d3>、{H^*}[C38 T^] <d3>、{H^*}[C39 T^] <d3>、{H^*}[C310 T^] <d3>Where T represents the name of the small pivot domain, D1, D2, and D3 represent the distance factors D j The name of the branch migration domain.

[0111] Weight sum gate SumW ij A total of 30 items, including SumW 11 、SumW 12 、SumW 13 、SumW 14 、SumW 15 、SumW 16 、SumW 17 、SumW 18 、SumW 19 、SumW 110 、SumW 21 、SumW 22 、SumW 23 、SumW 24 、SumW 25 、SumW 26 、SumW 27 、SumW 28 、SumW 29 、SumW 210 、SumW 31 、SumW 32 、SumW 33 、SumW 34 、SumW 35 、SumW 36 、SumW 37 、SumW 38 、SumW 39 、SumW 310 , whose structures are {H^*}[W11T^] <d1>、{H^*}[W12 T^] <d1>、{H^*}[W13 T^] <d1>、{H^*}[W14 T^] <d1>、{H^*}[W15T^] <d1>、{H^*}[W16 T^] <d1>、{H^*}[W17 T^] <d1>、{H^*}[W18 T^] <d1>,{H^*}[W19T^] <d1>、{H^*}[W110 T^] <d1>、{H^*}[W21 T^] <d2>、{H^*}[W22 T^] <d2>、{H^*}[W23T^] <d2>、{H^*}[W24 T^] <d2>、{H^*}[W25 T^] <d2>、{H^*}[W26 T^] <d2>、{H^*}[W27T^] <d2>、{H^*}[W28 T^] <d2>、{H^*}[W29 T^] <d2>、{H^*}[W210 T^] <d2>、{H^*}[W31T^] <d3>、{H^*}[W32 T^] <d3>、{H^*}[W33 T^] <d3>、{H^*}[W34 T^] <d3>、{H^*}[W35T^] <d3>、{H^*}[W36 T^] <d3>、{H^*}[W37 T^] <d3>、{H^*}[W38 T^] <d3>、{H^*}[W39T^] <d3>、{H^*}[W310 T^] <d3>.

[0112] Sum Gate j There are 3 items in total, namely Sum1, Sum2, Sum3, and their structures are {T^*}<H^> [D1]<H^> 、{T^*}<H^> [D2]<H^> 、{T^*}<H^> [D3]<H^> . Signal inversion gate SRG jk There are 6 items in total, namely SRG 12 , SRG 13 , SRG 21 , SRG 23 , SRG 31 , SRG 32 The structures are<G^L2> [H^D1]{H^*},<G^L3> [H^D1]{H^*},<G^L1> [H^D2]{H^*},<G^L3> [H^D2]{H^*},<G^L1> [H^D3]{H^*},<G^L2> [H^D3]{H^*}, where L1, L2, and L3 represent the inverted product numbers corresponding to different signal inversion gates. Reverse sum gate RSG k There are 3 lines in total, namely RSG1, RSG2, and RSG3. Their structures are<H^S1> [G^L1]{H^*},<H^S2> [G^L2]{H^*},<H^S3> [G^L3]{H^*}. Where S1, S2, and S3 represent the names of the branch migration domains of the inverted object. kj There are 3 items in total, namely Anh 12 、Anh 23 、Anh 31 The structures are<G^*> [S1*S2]{G^*},<G^*> [S2*S3]{G^*},<G^*> [S3*S1]{G^*}. Report sum gate RE kl There are 3 items in total, namely RE 11 ,RE 22 ,RE 32 . Their structures are {H^*}[S1 H^] <y1>、{H^*}[S2 H^] <y2>、{H^*}[S3H^] <y2>Among them, the reporting gate R l There are 2 chains, R1 and R2. Their structures are {H^*}[Y1] and {H^*}[Y2]. Input chain C ij , weight chain W ij , waste chain waste, distance factor chain D j , reversal factor chain SF jk , invert the signal chain S k and S j , output chain Y l All DNA chains are products of the reaction and are contained in their corresponding reactants.

[0113] The number and structure of DNA chains required for input activation module, subtraction annihilation module, absolute value summation module, signal inversion module, inversion summation module, annihilation module, report summation module and report reaction module are: activation factor X i With input substrate C ij The concentration of * is set to the average of 10 quantitative features of tumor cell nuclei, here a i nM. Weight substrate W ij The concentration of * uses the weight value trained on the silicon-based computer, here b ij Expressed in nM. Fuel chain XFuel i The concentration is set to [200% × (a i +b ij )]nM; subtraction gate SG ij The concentration is set to [105%×min(a i ,b ij )]nM; input sum gate SumC ij SumW ij The concentrations are all set to [105% × | a i -b ij |]nM; Sum gate Sum j The concentration is set to Here we use d j nM represents it; all signal inversion gates SRG jk The concentration setting is [55%×max(d j )]nM; inverting sum gate RSG k The concentration is set to [100% × max (d j )]nM; Annihilation Gate kj The concentration is set to [55% × (max (d j )-min(d j ))]nM; Report summing gate RE kl The concentration is set to [100% × max (d j )]nM; Report gate R l The concentration is set to [200% × max (d j )]nM. By setting the concentration in this way, the reaction can be more standardized, the reaction speed can be accelerated, and the accuracy of the reaction can be improved.

[0114] Step 4: According to the structure of the learning vector quantization neural network, each module is cascaded into a DNA-based learning vector quantization neural network. The modules are cascaded by setting the structure of the input chain of each module to be the same as the output chain structure of the previous module, and programmed into Visual DSD. The concentration is set according to the concentration setting formula in step 3, and then the function is verified.

[0115] Step 5, obtain female breast tumor cases from the Wisconsin Cancer Breast Cancer Database, and perform data processing. Medical research shows that breast tumors can be divided into benign and malignant according to the difference in characteristic data between breast tumor lesion tissue cells and normal tissue cells obtained by fine needle aspiration. Therefore, by establishing the relationship between 10 quantitative features of the breast tumor cell nucleus obtained by fine needle aspiration and the benign and malignant nature of the cell, the benign and malignant nature of the breast tumor can be diagnosed. The 10 quantitative features of the breast tumor cell nucleus are the radius, texture, perimeter, area, smoothness, compactness, concavity, number of concavity points, symmetry, and fracture degree of the cell nucleus. Data of 569 breast tumor cases were obtained from the Wisconsin Breast Cancer Database, of which 357 were benign cases and 212 were malignant cases. A set of data for each case includes a total of 30 data of the mean value, standard deviation, and worst value (the mean value of the 3 maximum data of each feature) of the 10 quantitative features of each cell nucleus in the sampled tissue. Each group of data in the data file is divided into 32 fields. The first field is the case number; the second field is the diagnosis result: B is benign and M is malignant; the third to twelfth fields are the average values ​​of the 10 quantitative features of the microscopic images of the cell nuclei of the tumor lesion tissue in the case; the 13th to 22nd fields are the corresponding standard deviations; the 23rd to 32nd fields are the corresponding worst values. Here, the data from the 1st to 12th fields are selected for the experiment, that is, the case number, the diagnosis result, and the average value of the 10 quantitative features of the cell nuclei of the case. The average value of the 10 quantitative features of the microscopic images of the cell nuclei of the breast tumor lesion tissue is used as the activation factor X of the network. i , breast tumor diagnosis results are used as the output chain Y of the network l , where chain Y l The value of is 1 or 0. Here we define that when Y1 is 1, the breast tumor is diagnosed as benign, and Y2 is 0; when Y2 is 1, the breast tumor is diagnosed as malignant, and Y1 is 0. The present invention randomly selects 519 groups of data as training sets, and the remaining 50 groups of data are test sets. The designed DNA-based learning vector quantization neural network is trained with the data of the training set, and then the data of the test set is tested and the test results are analyzed. Here, the optimal number of neurons is determined to be 3 by using K-fold cross validation. At this time, the training speed of the DNA-based learning vector quantization neural network is the fastest, and the accuracy of breast tumor diagnosis is the highest.

[0116] Step 6: Use Visual DSD to use the DNA-based learning vector quantization neural network to diagnose benign or malignant data of female breast tumor cases.

[0117] like Figure 5 The output diagram of the reaction is shown below. Figure 5 (a) is the diagnosis result of malignant cases, Figure 5 (b) is the diagnosis result of benign cases, where Y1 indicates a benign diagnosis and Y2 indicates a malignant diagnosis. As expected, the DNA-based learning vector quantization neural network constructed by the present invention successfully performed benign and malignant diagnosis on breast tumor case data with an accuracy rate of 94%. And it is completely consistent with the diagnosis result of the computer, which shows that the DNA-based learning vector quantization neural network designed by the present invention is completely feasible and can realize complex prediction functions.

[0118] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for diagnosing breast tumors based on molecular learning vector quantization neural network, characterized in that: The steps are as follows: Step 1, determine the functions of the input layer, competition layer and linear output layer of the learning vector quantization neural network, and compare them with the DNA strand displacement reaction respectively; Step 2, according to the DNA chain displacement reaction, the functional modules corresponding to the input layer, the competition layer and the linear output layer of the learning vector quantization neural network are designed respectively, and the chemical reactions of the functional modules are determined; Step 3: Determine the structure of the DNA chain and the small support structure of the auxiliary and reactant in the reaction process of each functional module, and determine the base sequence of the domain in each DNA chain. Use Visual DSD to write the required DNA chain and perform functional verification; Step 4: cascade each functional module into a DNA-based learning vector quantization neural network according to the structure of the learning vector quantization neural network, program it into Visual DSD for concentration setting, and perform functional verification; Step 5, obtaining female breast tumor cases from the breast cancer database and performing preprocessing; Step 6, using Visual DSD to use DNA-based learning vector quantization neural network to diagnose benign or malignant female breast tumor cases after preprocessing; The functional modules of the input layer include an input activation module, a subtraction annihilation module, and an absolute value summation module; the functional modules of the competition layer include a signal inversion module, an inversion summation module, and an annihilation module; and the functional modules of the linear output layer include a report summation module and a report reaction module; The input activation module is a catalytic reaction that activates factor X i Before adding to the reaction, input chain C ij With weight chain W ij The concentrations of the input substrate C ij *With weight substrate W ij *Inside; only activated factor X i After addition, factor X is activated i With input substrate C ij * and weight substrate W ij * A reversible strand displacement reaction is performed to generate the input strand C ij , weight chain W ij and an identical intermediate, intermediate with fuel chain XF i The reversible DNA strand displacement reaction is repeated to regenerate activated factor X i , so that as long as factor X is activated i Exists and fuel chain XF i is sufficient for the reaction to convert all of the input substrate C ij * With weight substrate W ij * All are converted into input chain C ij With weight chain W ij ; The subtractive annihilation module takes the input chain C ij With weight chain W ij Subtract each other through annihilation reaction, where each group of input chain C ij With weight chain W ij They will continue to annihilate each other until only the input chain C remains. ij With weight chain W ij One of the reactions stops; the first step of the subtraction annihilation module is a reversible reaction, input chain C ij Or weight chain W ij A DNA strand in the subtraction gate SG ij The exposed foothold in the middle binds and migrates to the center; the second step is an irreversible reaction, input chain C ij Or weight chain W ij The other DNA chain in the first step combines with another exposed foothold on the intermediate product chain generated by the first step reaction and migrates to the midpoint, finally ij Breaking down into two waste chains; The absolute value summation module refers to the input chain C remaining after the annihilation reaction is completed. ij Or weight chain W ij Sum the concentrations and add them to the distance factor D j middle; The absolute value summation module is divided into two reactions. The first is the subtraction annihilation module, which annihilates the remaining input chain C. ij Or weight chain W ij Respectively with the input summing gate SumC ij Or weight sum gate SumW ij Perform a reversible DNA strand displacement reaction to generate the input summation strand IS j and weighted sum chain WS j , then enter the summing chain IS j and weighted sum chain WS j Then respectively with the sum gate Sum j The irreversible DNA strand displacement reaction is carried out to change the distance factor D j The irreversible DNA strand displacement reaction drives the first step of the subtractive annihilation module in the positive direction until all input strands C ij With weight chain W ij All have been reacted; The signal reversal module and the reversal summation module are both irreversible DNA strand displacement reactions. The double-stranded products of the DNA strand displacement reactions no longer contain exposed footholds, and the product of the previous reaction is the reactant of the next reaction. The signal reversal module converts the distance factor D j As input, the signal inversion gate SRG jk DNA strand displacement reaction is performed, and the distance factor D j The concentration is evenly distributed to different reversal factors SF jk The inversion summation module is to convert different inversion factors SF jk Perform concentration summation and sum the concentration to the inverted signal chain S with the same subscript 'k' k middle; The annihilation module is to connect different inversion signal chains S k With the inverted signal chain S j Perform paired annihilation reactions until the signal chain S is reversed j With the inverted signal chain S k Only one of the reactions will stop; the annihilation module will reverse the signal chain S k Combined with annihilation chain Anh kj The branch moves to the midpoint of the double-stranded domain if only the inversion signal chain S k exists, the process is completely reversible, and no molecules are consumed; if another reverse signal chain S j Also exists, combined with the annihilation chain Anh kj On the other side, the branch moves to the midpoint of the double-stranded domain; when the signal chain S is reversed k and invert the signal chain S j When they reach the midpoint at the same time, the annihilation chain Anh kj Splitting into two waste molecules, neither of which has an exposed foothold and cannot interact with any other molecule; The DNA strand displacement reaction of the reporter summation module is a reversible reaction, which consumes the summation reporter gate RE. kl The report summing module is to sum the report gate RE kl With the inverted signal chain S k Perform DNA strand displacement reaction to generate reporter signal chain RY kl , where the summation reporting gate RE kl Branch migration domain Y l Corresponding to different linear outputs, it reflects the partial connection mode in the linear output layer of the learning vector quantization neural network; different branch migration domains Y l and the corresponding inverted signal chain S k Combine to realize the function of partial connection; The reaction of the reporter reaction module is an irreversible strand displacement reaction, the product of the previous reaction is the reactant of the next reaction, and the reaction is driven in the positive direction; the reporter reaction module realizes the migration of the same branch domain Y l Sum and extract the output of the reaction to report the release of fluorescent output signal; report signal chain RY kl With reporting chain R l Irreversible strand displacement reaction to generate output strand Y l , separating the fluorophore and quencher labeled strands in the reporter gate, resulting in fluorescence enhancement, and the reaction result is obtained by detecting the fluorescence level of the remaining DNA species.

2. The method for breast tumor diagnosis based on molecular learning vector quantization neural network according to claim 1, characterized in that: The learning vector quantization neural network includes an input layer, a competition layer and a linear output layer. The input layer and the competition layer are fully connected, and the competition layer and the linear output layer are partially connected. The number of neurons in the competition layer is always greater than the number of neurons in the linear output layer. Each competition layer neuron is only connected to one linear output layer neuron, and the connection weight is always 1; each linear output layer neuron is connected to multiple competition layer neurons, and the values ​​of the competition layer neurons and the linear output layer neurons can only be 1 or 0; when an input value is sent in through the input layer, the competition layer neuron closest to the input vector is activated and has a state of "1", and the states of other competition layer neurons are "0", and the linear output layer neurons connected to the activated competition layer neurons have a state of "1", while the states of other linear output layer neurons are all "0"; The reaction mechanism of the DNA chain displacement is a chain displacement reaction based on the small pivot domain. The small pivot domain on the single-stranded DNA species is paired and combined with the complementary small pivot domain exposed on the double-stranded DNA species to displace a new DNA single strand and generate a new DNA double strand. If the newly generated DNA single strand and the small pivot domain on the DNA double strand are still complementary, and the branch structure migration domain on the small pivot domain matches, this reaction is a reversible DNA chain displacement reaction. The DNA chain displacement reaction is a spontaneous, dynamic, cascaded reversible reaction process.

3. The method for diagnosing breast tumors based on molecular learning vector quantization neural network according to claim 1 or 2, characterized in that: The chemical reaction expression of the input activation module is: Among them, X i XF stands for activating factor i represents the fuel chain, C ij * represents input substrate, W ij * represents weighted substrate, C ij represents the activated input chain, W ij Represents the activated weight; The chemical reaction expression of the subtractive annihilation module is: K f >>K s Among them, K f is the rate constant of the subtractive annihilation reaction, K s is the absolute value summation reaction rate constant, where K f >>K s SG ij represents the subtraction gate, and waste represents the waste chain; The chemical reaction expression of the absolute value summation module is: Among them, D j represents the distance factor; The chemical reaction expression of the signal reversal module is: D j +SRG jk →SF jk Among them, SRG jk Represents the signal inversion gate, SF jk represents the inversion factor; The chemical reaction expression of the inversion summation module is: SF jk +RSG k →S k Among them, RSG k represents the inverting sum gate, S k Represents a reversal signal; The chemical reaction expression of the annihilation module is: Among them, K f is the rate constant of the annihilation reaction, Anh kj Represents the annihilation chain; The chemical reaction expression of the report summation module is: Among them, K s is the rate constant for the reported summation reaction, and K s < <K f , R.E. kl Representatives request and report gate, RY kl represents the reporting signal chain; The chemical reaction expression of the reporter reaction module is: RY kl +R l →Y l Among them, R l Represents the reporting gate, Y l Represents output.

4. The method for breast tumor diagnosis based on molecular learning vector quantization neural network according to claim 3, characterized in that: The annihilation reaction of the subtractive annihilation module uses the base G, which has 7 nucleotides; the first part of the absolute value summation module uses the input chain C ij With weight chain W ij The toehold H on the nucleotide sequence has 5 nucleotides; The number and structure of each DNA chain are set by the number of input values ​​of the constructed molecular neural network and the corresponding modules, and the base sequence is generated by Visual DSD.

5. The method for breast tumor diagnosis based on molecular learning vector quantization neural network according to claim 4, characterized in that: The activation factor X i There are ten lines in total: X1, X2, X3, X4, X5, X6, X7, X8, X9, X 10 , the structures are<X1G^> ,<X2G^> ,<X3G^> ,<X4G^> ,<X5G^> ,<X6G^> ,<X7G^> ,<X8G^> ,<X9G^> ,<X10G^> ; Wherein, X1-X10 represent the names of 10 input branch migration domains, G represents the name of the small pivot domain, ^ represents the small pivot domain, and <> represents a single chain; Input substrate C ij * There are 30 items, namely C 11 * , C 12 * , C 13 * , C 14 * , C 15 * , C 16 * , C 17 * , C 18 * , C 19 * , C 110 * , C 21 * , C 22 * , C 23 * , C 24 * , C 25 * , C 26 * , C 27 * , C 28 * , C 29 * , C 210 * , C 31 * , C 32 * , C 33 * , C 34 * , C 35 * , C 36 * , C 37 * , C 38 * , C 39 * , C 310 * , the structures are<H^C11> [G^X1]{G^*},<H^C12> [G^X2]{G^*},<H^C13> [G^X3]{G^*},<H^C14> [G^X4]{G^*},<H^C15> [G^X5]{G^*},<H^C16> [G^X6]{G^*},<H^C17> [G^X7]{G^*},<H^C18> [G^X8]{G^*},<H^C19> [G^X9]{G^*},<H^C110> [G^X10]{G^*},<H^C2> [G^X1]{G^*},<H^C22> [G^X2]{G^*},<H^C23> [G^X3]{G^*},<H^C24> [G^X4]{G^*},<H^C25> [G^X5]{G^*},<H^C26> [G^X6]{G^*},<H^C27> [G^X7]{G^*},<H^C28> [G^X8]{G^*},<H^C29> [G^X9]{G^*},<H^C210> [G^X10]{G^*},<H^C31> [G^X1]{G^*},<H^C32> [G^X2]{G^*},<H^C33> [G^X3]{G^*},<H^C34> [G^X4]{G^*},<H^C35> [G^X5]{G^*},<H^C36> [G^X6]{G^*},<H^C37> [G^X7]{G^*},<H^C38> [G^X8]{G^*},<H^C39> [G^X9]{G^*},<H^C310> [G^X10]{G^*}; where H represents the small branch domain, C11-C19, C21-C29, C31-C39, C110, C210, and C310 represent the names of the branch migration domains of 30 input substrates, * represents complementarity, {} represents single-stranded, and [] represents double-stranded; weight substrate W ij * There are 30 DNA chains, namely W 11 * , W 12 * , W 13 * , W 14 * , W 15 * , W 16 * , W 17 * , W 18 * ,IN 19 * ,IN 110 * ,IN 21 * ,IN 22 * ,IN 23 * ,IN 24 * ,IN 25 * ,IN 26 * ,IN 27 * ,IN 28 * ,IN 29 * ,IN 210 * ,IN 31 * ,IN 32 * ,IN 33 * ,IN 34 * ,IN 35 * ,IN 36 * ,IN 37 * ,IN 38 * ,IN 39 * ,IN 310 * , the structures are<H^W11> [G^X1]{G^*},<H^W12> [G^X2]{G^*},<H^W13> [G^X3]{G^*},<H^W14> [G^X4]{G^*},<H^W15> [G^X5]{G^*},<H^W16> [G^X6]{G^*},<H^W17> [G^X7]{G^*},<H^W18> [G^X8]{G^*},<H^W19> [G^X9]{G^*},<H^W110> [G^X10]{G^*},<H^W2> [G^X1]{G^*},<H^W22> [G^X2]{G^*},<H^W23> [G^X3]{G^*},<H^W24> [G^X4]{G^*},<H^W25> [G^X5]{G^*},<H^W26> [G^X6]{G^*},<H^W27> [G^X7]{G^*},<H^W28> [G^X8]{G^*},<H^W29> [G^X9]{G^*},<H^W210> [G^X10]{G^*},<H^W31> [G^X1]{G^*},<H^W32> [G^X2]{G^*},<H^W33> [G^X3]{G^*},<H^W34> [G^X4]{G^*},<H^W35> [G^X5]{G^*},<H^W36> [G^X6]{G^*},<H^W37> [G^X7]{G^*},<H^W38> [G^X8]{G^*},<H^W39> [G^X9]{G^*},<H^W310> [G^X10]{G^*}; where W11-W19, W21-W29, W31-W39, W110, W210, and W310 represent the names of the branch migration domains of the 30 weight substrates, respectively; Fuel Chain XF i There are 10 items, namely XF1, XF2, XF3, XF4, XF5, XF6, XF7, XF8, XF9, XF 10 , the structures are<G^X1> ,<G^X2> ,<G^X3> ,<G^X4> ,<G^X5> ,<G^X6> ,<G^X7> ,<G^X8> ,<G^X9> ,<G^X10> ; Subtraction gate chain SG ij There are 30 in total, namely SG 11 , SG 12 , SG 13 , SG 14 , SG 15 , SG 16 , SG 17 , SG 18 , SG 19 , SG 110 , SG 21 , SG 22 , SG 23 , SG 24 , SG 25 , SG 26 , SG 27 , SG 28 , SG 29 , SG 210 , SG 31 , SG 32 , SG 33 , SG 34 , SG 35 , SG 36 , SG 37 , SG 38 , SG 39 , SG 310 , the structures are<G^*> [C11*W11]{G^*},<G^*> [C12*W12]{G^*},<G^*> [C13*W13]{G^*},<G^*> [C14*W14]{G^*},<G^*> [C15*W15]{G^*},<G^*> [C16*W16]{G^*},<G^*> [C17*W17]{G^*},<G^*> [C18*W18]{G^*},<G^*> [C19*W19]{G^*},<G^*> [C110*W110]{G^*},<G^*> [C21*W21]{G^*},<G^*> [C22*W22]{G^*},<G^*> [C23*W23]{G^*},<G^*> [C24*W24]{G^*},<G^*> [C25*W25]{G^*},<G^*> [C26*W26]{G^*},<G^*> [C27*W27]{G^*},<G^*> [C28*W28]{G^*},<G^*> [C29*W29]{G^*},<G^*> [C210*W210]{G^*},<G^*> [C31*W31]{G^*},<G^*> [C32*W32]{G^*},<G^*> [C33*W33]{G^*},<G^*> [C34*W34]{G^*},<G^*> [C35*W35]{G^*},<G^*> [C36*W36]{G^*},<G^*> [C37*W37]{G^*},<G^*> [C38*W38]{G^*},<G^*> [C39*W39]{G^*},<G^*> [C310*W310]{G^*}; Input sum gate SumC ij A total of 30 items, including SumC 11 、SumC 12 、SumC 13 、SumC 14 、SumC 15 、SumC 16 、SumC 17 、SumC 18 、SumC 19 、SumC 110 、SumC 21 、SumC 22 、SumC 23 、SumC 24 、SumC 25 、SumC 26 、SumC 27 、SumC 28 、SumC 29 、SumC 210 、SumC 31 、SumC 32 、SumC 33 、SumC 34 、SumC 35 、SumC 36 、SumC 37 、SumC 38 、SumC 39 、SumC 310 , the structures are {H^*}[C11T^] <d1>、{H^*}[C12T^] <d1>、{H^*}[C13T^] <d1>,{H^*}[C14T^] <d1>,{H^*}[C15T^] <d1>,{H^*}[C16T^] <d1>,{H^*}[C17T^] <d1>、{H^*}[C18T^] <d1>、{H^*}[C19T^] <d1>、{H^*}[C110T^] <d1>、{H^*}[C21T^] <d2>、{H^*}[C22T^] <d2>、{H^*}[C23T^] <d2>、{H^*}[C24T^] <d2>、{H^*}[C25T^] <d2>、{H^*}[C26T^] <d2>、{H^*}[C27T^] <d2>、{H^*}[C28T^] <d2>、{H^*}[C29T^] <d2>,{H^*}[C210T^] <d2>、{H^*}[C31T^] <d3>、{H^*}[C32T^] <d3><h2 style=";text-align:left;direction:ltr">{H^*}[C33T^]<h2 style=";text-align:left;direction:ltr"> <d3>、{H^*}[C34T^] <d3>、{H^*}[C35T^] <d3>、{H^*}[C36T^] <d3>、{H^*}[C37T^] <d3>、{H^*}[C38T^] <d3>、{H^*}[C39T^] <d3>、{H^*}[C310T^] <d3>; Where T represents the name of the small pivot domain, D1, D2, and D3 represent the distance factors D j The name of the branch migration domain; Weight sum gate SumW ij A total of 30 items, including SumW 11 、SumW 12 、SumW 13 、SumW 14 、SumW 15 、SumW 16 、SumW 17 、SumW 18 、SumW 19 、SumW 110 、SumW 21 、SumW 22 、SumW 23 、SumW 24 、SumW 25 、SumW 26 、SumW 27 、SumW 28 、SumW 29 、SumW 210 、SumW 31 、SumW 32 、SumW 33 、SumW 34 、SumW 35 、SumW 36 、SumW 37 、SumW 38 、SumW 39 、SumW 310 , the structures are {H^*}[W11T^] <d1>、{H^*}[W12T^] <d1>、{H^*}[W13T^] <d1>、{H^*}[W14T^] <d1>、{H^*}[W15T^] <d1>、{H^*}[W16T^] <d1>、{H^*}[W17T^] <d1>、{H^*}[W18T^] <d1>,{H^*}[W19T^] <d1>、{H^*}[W110T^] <d1>,{H^*}[W21T^] <d2>、{H^*}[W22T^] <d2>、{H^*}[W23T^] <d2>、{H^*}[W24T^] <d2>、{H^*}[W25T^] <d2>、{H^*}[W26T^] <d2>、{H^*}[W27T^] <d2>、{H^*}[W28T^] <d2><h2 style=";text-align:left;direction:ltr">、{H^*}[W29T^]<h2 style=";text-align:left;direction:ltr"> <d2>、{H^*}[W210T^] <d2>、{H^*}[W31T^] <d3><h2 style=";text-align:left;direction:ltr">{H^*}[W32T^]<h2 style=";text-align:left;direction:ltr"> <d3>、{H^*}[W33T^] <d3>、{H^*}[W34T^] <d3>、{H^*}[W35T^] <d3>、{H^*}[W36T^] <d3>、{H^*}[W37T^] <d3>、{H^*}[W38T^] <d3>、{H^*}[W39T^] <d3>、{H^*}[W310T^] <d3> ；< / d3> Sum Gate j There are 3 items in total, namely Sum1, Sum2, Sum3, and their structures are {T^*}<H^> [D1]<H^> 、{T^*}<H^> [D2]<H^> 、{T^*}<H^> [D3]<H^> ;Signal reversal gate SRG jk There are 6 items in total, namely SRG 12 , SRG 13 , SRG 21 , SRG 23 , SRG 31 , SRG 32 , the structures are<G^L2> [H^D1]{H^*},<G^L3> [H^D1]{H^*},<G^L1> [H^D2]{H^*},<G^L3> [H^D2]{H^*},<G^L1> [H^D3]{H^*},<G^L2> [H^D3]{H^*}; where L1, L2, and L3 represent the inverted product numbers corresponding to different signal inversion gates; the inverted sum gate RSG k There are 3 lines in total, namely RSG1, RSG2, RSG3, and their structures are<H^S1> [G^L1]{H^*},<H^S2> [G^L2]{H^*},<H^S3> [G^L3]{H^*}; where S1, S2, and S3 represent the names of the branch migration domains of the inverted object; annihilation chain Anh kj There are 3 items in total, namely Anh 12 、Anh 23 、Anh 31 , the structures are<G^*> [S1*S2]{G^*},<G^*> [S2*S3]{G^*},<G^*> [S3*S1]{G^*}; Report sum gate RE kl There are 3 items in total, namely RE 11 ,RE 22 ,RE 32 , the structures are {H^*}[S1H^] <y1>,{H^*}[S2H^] <y2>、{H^*}[S3H^] <y2>; Report gate R l There are 2 lines in total, namely R1 and R2, and their structures are {H^*}[Y1] and {H^*}[Y2] respectively.

6. The method for breast tumor diagnosis based on molecular learning vector quantization neural network according to claim 5, characterized in that: The number of DNA chains and the structure of DNA chains required for the input activation module, subtraction annihilation module, absolute value summation module, signal inversion module, inversion summation module, annihilation module, report summation module and report reaction module are as follows: activation factor X i With input substrate C ij The concentration of * is set as the average of 10 quantitative features of tumor cell nuclei, with a i nM; weight substrate W ij *The concentration uses the weight values ​​trained on the silicon-based computer, with b ij nM to express; fuel chain XF i The concentration is set to [200% × (a i +b ij )]nM; subtraction gate SG ij The concentration is set to [105%×min(a i ,b ij )]nM; input sum gate SumC ij SumW ij The concentrations are all set to [105% × | a i -b ij |]nM; Sum gate Sum j The concentration is set to Use d j nM represents; all signal inversion gates SRG jk The concentration setting is [55%×max(d j )]nM; inverting sum gate RSG k The concentration is set to [100% × max (d j )]nM; Annihilation Gate kj The concentration is set to [55% × (max (d j )-min(d j ))]nM; Report summing gate RE kl The concentration is set to [100% × max (d j )]nM; Report gate R l The concentration is set to [200% × max (d j )]nM.

7. The method for diagnosing breast tumors based on molecular learning vector quantization neural network according to any one of claims 1, 2, 4-6, characterized in that: By establishing the relationship between the 10 quantitative characteristics of breast tumor cell nuclei obtained by fine needle puncture and the benign and malignant nature of the cells, the benign and malignant nature of the breast tumor can be diagnosed; the 10 quantitative characteristics of breast tumor cell nuclei are the radius, texture, circumference, area, smoothness, compactness, depression, number of depression points, symmetry, and fracture degree of the cell nucleus.

8. The method for breast tumor diagnosis based on molecular learning vector quantization neural network according to claim 7, characterized in that: A set of data for each female breast tumor case includes the average, standard deviation and worst value of 10 quantitative features of each cell nucleus in the sampled tissue; the average value of the 10 quantitative features of the cell nucleus microscopic image of the breast tumor lesion tissue is used as the activation factor X of the DNA-based learning vector quantization neural network. i , breast tumor diagnosis results as output chain Y l , output chain Y l The value of is 1 or 0. It is defined that when the output chain Y1 is 1, the breast tumor is diagnosed as benign, and the output chain Y2 is 0; when the output chain Y2 is 1, the breast tumor is diagnosed as malignant, and the output chain Y1 is 0.

9. The method for breast tumor diagnosis based on molecular learning vector quantization neural network according to claim 8, characterized in that: 519 groups of data were randomly selected as training sets, and the remaining 50 groups of data were used as test sets. The designed DNA-based learning vector quantization neural network was trained with the data of the training set, and then the data of the test set was tested and the test results were analyzed. K-fold cross validation was used to determine that the number of neurons in the competition layer was 3.

Images