DNA neural network computer and application method of DNA neural network

Abstract

The application provides a DNA neural network computer and an application method of the DNA neural network. Each component of the hardware of the DNA neural network computer is dissolved in a liquid to obtain a DNA neural network computer solution, and the hardware of the DNA neural network computer including a non-complete complementary perceptron and a winner-take-all module is formed; and a neural network operation is performed through non-complete complementary hybridization between an input chain and DNA chains in the hardware of the DNA neural network computer after the input chain is added to output a mode category of input information represented by the input chain. The input chain is a lipid-nucleic acid conjugated molecule, the input chain can be removed from the computer solution after the operation is completed, and the processed hardware of the DNA neural network computer can be reused. The DNA neural network computer of the application can complete the calculation without the aid of a fuel chain or an enzyme, the hardware part can be recycled, the calculation cost is lower, and the learning of the DNA neural network is possible.

Description

Technical Field

[0001] This application belongs to the field of biological computing and molecular computing, and in particular relates to a DNA neural network computer and a method for applying DNA neural networks. Background Technology

[0002] DNA neural network computing, proposed by Erik Winfree et al. in 2011, is a novel molecular computing method that aims to simulate biological neural networks for information processing using DNA strand substitution reaction networks. Compared to currently widely used electronic computer neural network computing, DNA neural network computing has significant advantages in parallel operation, energy consumption, data density, and biocompatibility, making it a promising new computing system. Although DNA neural network computing is still in its early stages of development, several studies have successfully utilized DNA neural networks to handle complex problems. For example, in 2018, LuluQian et al. constructed a "Winner Takes All (WTA)" DNA neural network algorithm to achieve handwritten digit recognition; in 2022, Tan Weihong et al. used a DNA neural network algorithm to achieve automated, high-accuracy pathogen diagnosis of acute respiratory infections. These studies demonstrate the powerful information processing capabilities and enormous biomedical application potential of DNA neural network computing.

[0003] Referring to the development history of neural network computing in electronic computers, future DNA neural network computing will evolve into computing systems with machine learning capabilities. However, realizing neural network learning requires inputting large amounts of learning material into the computer in multiple batches, necessitating a reusable computing system. Currently, mainstream DNA neural network computing is primarily based on fully complementary DNA strand substitution reaction networks, requiring the use of fuel chains or enzymes to drive the reaction and complete the computation. This computational process of such DNA neural networks is inherently irreversible, thus preventing the reuse of DNA neural network systems. This not only results in enormous operating costs for DNA computers but also hinders the development of DNA neural network learning.

[0004] Therefore, there is a current need for a DNA neural network that can perform neural network calculations without relying on fuel chains or enzyme-driven reactions, and a DNA neural network computer whose hardware can be recycled at least. Summary of the Invention

[0005] In view of the above problems, this application is proposed to solve the aforementioned problems existing in the prior art.

[0006] The purpose of this application is to provide a DNA neural network computer and a method for applying DNA neural networks, which can complete calculations without the aid of fuel chains or enzymes. Furthermore, the hardware of the DNA neural network computer is recyclable, which greatly reduces the cost of computer use and makes learning of DNA neural networks possible.

[0007] According to a first aspect of this application, a DNA neural network computer is provided. The DNA neural network computer solution is obtained by dissolving the various components of the computer hardware in a pre-modified liquid medium. The solution contains DNA neural network computer hardware including an imperfectly complementary perceptron and a winner-take-all module. After an input chain carrying encoded input information is added to the solution, neural network operations are performed by hybridizing the input chain with the imperfectly complementary components in the hardware to obtain the pattern category of the input information represented by the input chain as the output. Furthermore, the input chain is a lipid-nucleic acid conjugate molecule, wherein the lipid molecule is covalently coupled to the 5' or 3' end of the nucleic acid molecule, allowing the input chain to be removed from the solution after neural network operations to obtain recycled DNA neural network computer hardware for reuse.

[0008] According to a second aspect of this application, a method for applying a DNA neural network is provided. The DNA neural network is used to identify pattern categories of input information. The application method includes: determining the number m of pattern categories of the input information to be identified; constructing a DNA neural network computer as described in various embodiments of this application, such that the constructed DNA neural network contains m incompletely complementary perceptrons; setting the number n of input chain types and the chain structure of each input chain in association with the number m of pattern categories; when it is necessary to identify pattern categories of input information in multiple rounds: encoding the input information in the current round and mapping the encoded input information to a combination of input chains; performing neural network operations based on the combination of input chains by the DNA neural network and outputting the identification result of the pattern category of the input information in the current round; performing a recycling process on the DNA neural network after completing the neural network operation to obtain a DNA neural network after removing the input chains, and using the DNA neural network after removing the input chains for identifying the pattern category of the input information in the next round.

[0009] The DNA neural network computer and its application method according to various embodiments of this application can complete calculations without the aid of fuel chains or enzymes. Instead, it performs neural network operations through incomplete complementary hybridization between DNA chains, which has better reversibility. Furthermore, because the input chain employs a hydrophobic lipid-nucleic acid coupled molecular structure, it can be more easily removed from the DNA neural network computer solution after neural network operations. This allows the recycled DNA neural network computer hardware to be reused for similar neural network operations, thus avoiding the need to discard the costly DNA neural network computer after each operation and significantly reducing its usage cost. In addition, compared to existing technologies, the DNA neural network computer according to this application uses shorter and fewer DNA chains, has a faster computation speed, and is more operable and practical. It provides favorable technical support for the future realization of DNA neural network learning and has broad application prospects in fields such as biosensing, disease diagnosis and treatment, and molecular computing. Attached Figure Description

[0010] In drawings that are not necessarily drawn to scale, the same reference numerals may describe similar parts in different views. The same reference numerals with or without letter suffixes may indicate different instances of similar parts. The drawings illustrate various embodiments generally by way of example rather than limitation, and are used, together with the description and claims, to explain the disclosed embodiments. Where appropriate, the same reference numerals are used in all drawings to refer to the same or similar parts. Such embodiments are illustrative and not intended to be exhaustive or exclusive embodiments of the apparatus or method.

[0011] Figure 1(a) shows the general formula structure of a lipid molecule in a lipid-nucleic acid conjugate molecule of an input chain according to an embodiment of the present application.

[0012] Figure 1(b) shows the specific structure of a lipid molecule in an input chain of a lipid-nucleic acid conjugate according to an embodiment of this application.

[0013] Figure 2 A schematic diagram showing a portion of the structural composition of a DNA neural network computer according to an embodiment of this application is provided.

[0014] Figure 3 A schematic diagram of the DNA neural network computer hardware and input chain according to an embodiment of this application is shown.

[0015] Figure 4(a) illustrates a DNA neural network computer according to an embodiment of this application with an input strand of T. d I1 and T dSchematic diagram of DNA strand displacement reaction at I2.

[0016] Figure 4(b) illustrates a DNA neural network computer according to an embodiment of this application with an input strand of T. d Schematic diagram of DNA strand displacement reaction at I2.

[0017] Figure 5 A schematic diagram illustrating the working principle of the winner-take-all module according to an embodiment of this application is shown.

[0018] Figure 6(a) shows a schematic diagram of the DNA chain reaction during the "You Say, I Guess" game process according to an embodiment of this application.

[0019] Figure 6(b) shows the truth table of the game results under different input conditions of the "You Say, I Guess" game according to an embodiment of this application.

[0020] Figure 6(c) shows the chain reaction results corresponding to the input of the "You Say, I Guess" game according to an embodiment of this application.

[0021] Figure 7 The MALDI-TOF mass spectrum of the product according to an embodiment of this application is shown.

[0022] Figure 8 A schematic diagram illustrating the synthesis process of a lipid-nucleic acid conjugate molecule according to an embodiment of this application is shown.

[0023] Figure 9 This diagram illustrates a DNA neural network performing multiple rounds of computation according to an embodiment of this application. Detailed Implementation

[0024] To enable those skilled in the art to better understand the technical solutions of this application, the application will be described in detail below with reference to the accompanying drawings and specific embodiments. The embodiments of this application will be further described in detail below with reference to the accompanying drawings and specific examples, but these are not intended to limit the scope of this application.

[0025] The terms "first," "second," and similar terms used in this application do not indicate any order, quantity, or importance, but are merely used to distinguish different parts. The use of "first" and "second" is for ease of description and numbering purposes only, and is not intended to imply that "first component" and "second component" must have different physical properties. In fact, "first component" and "second component" can have the same or different structures, without limitation, as long as they are separate components. Furthermore, where the context provides sufficient explanation, "first component" and "second component" may not even be separate components; they can be integrated into the same component, or they can be interchangeable.

[0026] In this application, when a specific device is described as being located between a first device and a second device, an intermediary device may or may not be present between the specific device and the first or second device. When a specific device is described as being connected to other devices, the specific device may be directly connected to the other devices without an intermediary device, or it may not be directly connected to the other devices but may have an intermediary device.

[0027] Words such as "include" or "contain" mean that the element preceding the word covers the element listed after the word, and do not exclude the possibility of covering other elements as well. Words such as "up," "down," "left," and "right" are only used to indicate relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0028] It should also be understood that the term "and / or" in this application is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Furthermore, the character " / " in this application generally indicates that the preceding and following related objects have an "or" relationship.

[0029] In the first embodiment, this application provides a DNA neural network computer that exists in the form of a DNA neural network computer solution and performs related calculations through various reactions occurring in the solution.

[0030] Specifically, the aforementioned DNA neural network computer solution can be obtained by dissolving the various components of the DNA neural network computer hardware in a pre-modified liquid medium. After the multiple components of the DNA neural network computer hardware, i.e., various DNA molecules, are mixed in the liquid, a series of DNA reactions will occur, and ultimately, two functional modules will be formed in the DNA neural network computer solution: an imperfectly complementary perceptron and a winner-take-all module. The imperfectly complementary perceptron and the winner-take-all module formed can be collectively referred to as DNA neural network computer hardware.

[0031] Another essential part that enables a DNA neural network computer to perform calculations is the input chain. For a DNA neural network, the input chain represents the input information intended for computation / recognition using the DNA neural network; in other words, the input chain carries the encoded input information.

[0032] Unlike existing technologies that commonly use fuel chains or enzymes to drive DNA reactions to enable DNA computers to perform calculations, the DNA reaction in this application does not require additional fuel chains or enzymes. Instead, after the input chain carrying encoded input information is added to the DNA neural network computer solution, neural network operations are performed through incomplete complementary hybridization between the input chain and the DNA chains in the DNA neural network computer hardware to obtain the pattern category of the input information represented by the input chain as the output result.

[0033] Therefore, the aforementioned DNA neural network computer hardware is essentially the same as a DNA neural network capable of performing calculations and pattern category recognition based on each input chain. In the following text, the two are not specifically distinguished and can be considered the same concept. Each imperfectly complementary perceptron is equivalent to a neuron in the DNA neural network, and the number of imperfectly complementary perceptrons corresponds to the number of neurons. Multiple parallel imperfectly complementary perceptrons are equivalent to the intermediate layers of the neural network, used to perceive the input layer information represented by the input DNA chains and provide intermediate layer output signals. The Winner-Take-All (WTA) module makes decisions based on the output signals of each neuron and ultimately provides the output signal of the entire DNA neural network, equivalent to the output layer of the neural network. Figure 3 Figure 6(a) and other figures also show schematic diagrams of DNA neural networks that are equivalent to DNA neural network computer hardware.

[0034] Furthermore, in the embodiments according to this application, the input chain is a lipid-nucleic acid coupled molecule, wherein the lipid molecule is covalently coupled to the 5' end of the nucleic acid molecule, and since the reaction between DNA chains does not involve other fuel chains or enzymes, and not all of them are perfectly complementary hybridizations, the process of the incompletely complementary hybridization DNA reaction is reversible. Therefore, the input chain in the DNA neural network computer solution after completing the neural network operation can be removed by techniques such as reversed-phase column chromatography, so that the recovered DNA neural network computer hardware can be reused.

[0035] In other embodiments, lipid molecules of the input chain may also be covalently coupled to the 3' end of the nucleic acid molecule, which is not limited in this application.

[0036] According to embodiments of this application, the input chain employs a lipid-nucleic acid coupled molecular structure, wherein the lipid molecule may be a lipid molecule with a hydrophobic structure. The hydrophobic structure is not particularly limited, as long as it can impart hydrophobicity to the coupled molecule, enabling it to separate from the uncoupled nucleic acid. In some embodiments, the hydrophobic structure may be derived from long-chain alkanes, C6-C50 alkenes, or C6-C50 cycloalkanes, etc. Long-chain alkanes may refer to hydrocarbon chains with more than 6 carbon atoms. In some embodiments, long-chain alkanes may refer to hydrocarbon chains with 6-40, 8-30, or 10-20 carbon atoms.

[0037] Figure 1(a) shows the general formula structure of a lipid molecule in the input chain of a lipid-nucleic acid conjugate according to an embodiment of this application. In some embodiments, the lipid molecule may have the structure shown in Figure 1(a), wherein R1 and R2 may be the same or different and are independently selected from C6-C40 straight-chain or branched alkyl groups, preferably from C8-C30 straight-chain or branched alkyl groups, and even more preferably from C10-C20 straight-chain or branched alkyl groups.

[0038] Figure 1(b) shows the specific structure of a lipid molecule in the lipid-nucleic acid conjugate of the input chain according to an embodiment of this application.

[0039] The following describes the synthesis process of the input strand of the lipid-nucleic acid coupled molecular structure used in the embodiments of this application. This synthesis method provides a sufficiently removable input strand for the reusability of DNA neural network computers. The synthesis process is as follows: (1) Prepare 0.1 M rac-Ψ The activator was prepared in acetonitrile solution, 0.1 M DBU in acetonitrile solution, 0.65 M DBU in acetonitrile solution, and 0.1 M lipid in dichloromethane solution. The structure of the lipid molecule can be shown in Figure 1(b).

[0040] (2) The nucleic acid was synthesized using an automated nucleic acid synthesizer (Qingke Biotechnology Single-Strand Nucleic Acid Synthesizer-192P). The 200 nmol 1000 Å universal CPG (controlled-pore glass) synthesis column was used as the solid-phase support, and the oligonucleotide part was synthesized using traditional methods.

[0041] (3) Rinse the synthesis column once with 0.1 M DBU acetonitrile solution.

[0042] (4) Add 160 µL of acetonitrile solution containing 0.1 M activator and 25 µL of acetonitrile solution containing 0.65 M DBU to each synthesis column and react for 5 minutes. Repeat this operation once. Rinse the synthesis column three times with acetonitrile.

[0043] (5) Add 160 µL of 0.1 M lipid in dichloromethane and 75 µL of 0.65 M DBU in acetonitrile to each synthesis column and react for 20 minutes. Rinse the synthesis column three times with acetonitrile.

[0044] (6) The product was separated from the carrier in an ammonolysis apparatus and dissolved in a solution using tert-butanol:water = 1:4 (0.2 mL) to obtain the target product. The product was further purified using HPLC and ultrafiltration, and identified by MALDI-TOF-MS. The MALDI-TOF mass spectrum of the product is shown below. Figure 7 As shown, the test results only show the peak corresponding to the molecular weight of the target product, indicating that the lipid-nucleic acid conjugate molecule with the target structure and sequence was successfully prepared.

[0045] A schematic diagram of the synthesis process of lipid-nucleic acid coupled molecules is shown below. Figure 8 As shown. First, a nucleic acid moiety with a protecting group was synthesized on CPG via the phosphorimide method; then, via... rac-Ψ An activator activates the hydroxyl groups at the ends of the nucleic acid chain; then a lipid molecule with a hydroxyl group at one end is added to couple with the activated nucleic acid chain hydroxyl group to obtain a lipid-nucleic acid coupled molecule with a protecting group; finally, the protecting group is removed by ammonolysis and the lipid-nucleic acid coupled molecule is cleaved off the CPG.

[0046] The reagents used in the above examples were used directly without special treatment. Solvents were purchased from Sinopharm Group (analytical grade) or Innovent Biologics Inc. Nucleotide monomers and reagents required for the synthesis of lipid-nucleic acid conjugates were purchased from Dinachem Biotechnology Co., Ltd. rac-Ψ The activator was synthesized according to the reference (Science 2021, 373, 1265), and the lipid-free DNA strand was purchased from Hongxun Biotechnology Co., Ltd. Other reagent information is shown in Table 1 below.

[0047] Table 1 Reagent Information

[0048] The DNA neural network computer according to the embodiments of this application employs a computational principle completely different from that of existing DNA neural network computers. The DNA reaction therein does not rely on the driving force of fuel chains or enzymes. Instead, after the input chain is added to the computer solution, neural network operations can be performed through incomplete complementary hybridization between DNA chains, which has better reversibility. Furthermore, because the input chain uses a hydrophobic lipid-nucleic acid coupled molecule structure, it can be more easily removed from the DNA neural network computer solution after completing the neural network operation. This allows the recycled DNA neural network computer hardware to be reused for similar neural network operations, thus avoiding the need to discard the costly DNA neural network computer after each operation, significantly reducing the cost of using the DNA neural network computer. In addition, compared to the prior art, the DNA neural network computer according to this application uses shorter and fewer DNA chains, has a faster computation speed, and is more operable and practical. It also provides favorable technical support for the future realization of DNA neural network learning and has broad application prospects in fields such as biosensing, disease diagnosis and treatment, and molecular computing.

[0049] Figure 2 This diagram illustrates the DNA sequences of various components of a DNA neural network computer hardware according to an embodiment of this application. In the DNA neural network computer according to an embodiment of this application, the number of input strand types n is set in association with the number of pattern categories of the input information and the encoding method of the input information, and the number of incompletely complementary perceptrons m is set correspondingly to the number of pattern categories of the input information, and m is greater than or equal to 2. The following example, with n=4 and m=2, illustrates the various components of the DNA neural network computer hardware.

[0050] With n=4 and m=2, the components of the DNA neural network computer hardware include: There are m weight chains and m output chains corresponding to each non-perfectly complementary perceptron. Annihilation Chain A y A j There are m types of A chains modified with quenching groups and m types of R chains modified with fluorescent groups; where y and j correspond to the numbers of the incompletely complementary perceptrons, and the values ​​of y and j range from 1 to m and are different from each other. Since there are only 2 incompletely complementary perceptrons in this example, we can set y=1 and j=2.

[0051] like Figure 2 As shown, the eight weight chains are as follows: W 11 :TGTCGTAGAGGTTGGTGTTCT (SEQ ID NO:1) W21 :TGTCGTTCTGCATCGTGTTCT (SEQ ID NO:2) W 31 :TGTCGGACTGGTTAGCGTTCT (SEQ ID NO:3) W 41 :TGTCGTACTCTTGCGTGTTCT (SEQ ID NO:4) W 12 :TGTCGTAGAGGTTGGTGTGGA (SEQ ID NO:5) W 22 :TGTCGTTCTGCATCGTGTGGA (SEQ ID NO:6) W 32 :TGTCCGACTGGTTAGCGTGGA (SEQ ID NO:7) W 42 :TGTCGTACTCTTGCGTGTGGA (SEQ ID NO:8) Two output chains corresponding to each of the non-perfectly complementary perceptrons (hereinafter referred to as NCP1 and NCP2), wherein, The output chain O1 corresponding to NCP1 is: AGAACACGAACCAGTACGAAC (SEQ ID NO:9) The output chain O2 corresponding to NCP2 is: TCCACACGAACCAGTACGGGT (SEQ ID NO:10) =1, therefore, there is a total of 1 annihilation chain A1A2: GTTCGTCCTGGTCGGTGCACCCGTCCTGGTCGGTGC (SEQ ID NO: 11) Two types of A chains modified with quenching groups, among which, A1: [BHQ2]GTTCGTCCTGGTCGGTGCCCT (SEQ ID NO:12) A2: [BHQ2]ACCCGTCCTGGTCGGTGCGGT (SEQ ID NO:13) BHQ2 is modified with a quenching group.

[0052] Two R chains with different fluorescent group modifications, among which, R1: AGGGCATCGCTCAGGAGG[Cy3] (SEQ ID NO:14) R2: ACCGCATCGCTCAGGAGG[Cy5] (SEQ ID NO:15) Cy3 and Cy5 are modified with different fluorescent groups.

[0053] Figure 3 A schematic diagram of DNA neural network computer hardware and input chain according to an embodiment of this application is shown. After the various components of the DNA neural network computer described above are dissolved in a pre-modified liquid medium, a DNA neural network computer solution is obtained through sufficient reaction, forming... Figure 3 The shown is a DNA neural network computer hardware 300 containing NCP1, NCP2, and a winner-take-all module.

[0054] In the DNA neural network computer hardware 300, NCP1 contains four structures with the structure W. x1 The incompletely complementary bichain of O1, i.e.: W 11 :O1、W 21 :O1、W 31 :O1 and W 41 :O1, where W 11 :O1 is composed of weighted chain W 11 It is formed by hybridization with the output chain O1, and similarly, each W... x1 The O1 doubly chain is correspondingly composed of the weight chain W. x1 It is formed by hybridization with the output chain O1. Similarly, NCP2 also contains four structures with the W structure. x2 The incompletely complementary bilayer of O2, i.e.: W 12 :O2、W 22 :O2、W 32 O2 and W 42 :O2, respectively composed of weight chain W x2 It is formed by hybridization with the output chain O2, which will not be elaborated here.

[0055] In an embodiment according to this application, the input chain T d I x The nucleic acid sequence contains the downstream tag domain T d Incompletely complementary binding domain I x Where x represents the input chain number, and the value of x ranges from 1 to n. Therefore, in this embodiment, there are 4 input chains: T d I1: ACCCCAACATCTACGACA (SEQ ID NO:16) T d I2:ACAAGATGCCGAACGACA (SEQ ID NO:17) T dI3: ACGGTAACCAGACGGACA (SEQ ID NO:18) T d I4:ACAGGCAAGTGTACGACA (SEQ ID NO:19) Among them, the downstream label field T of each input chain d for Figure 2 The underlined portion in each input chain sequence is labeled "ACA," while the remaining portions of each input chain, excluding the underlined portions, represent the incompletely complementary binding domain I. x .

[0056] In the embodiments according to this application, the corresponding incompletely complementary double strand W xy :O y and W xj :O j Able to connect with input chain T d I x A non-completely complementary chain substitution reaction occurs between them, thereby releasing the corresponding output chain O. y Or O j Therefore, in a specific NCP, the relative concentration of each incompletely complementary double strand can reflect the NCP's sensitivity to different input strands. In other words, the relative concentration of the incompletely complementary double strand W can be adjusted. xy :O y In the incompletely complementary sensor NCP y Adjusting the proportion of each incompletely complementary double chain in the NCP of the incompletely complementary sensor. y For input chain T d I x The perceived weights, and, by adjusting the incompletely complementary bichain W xj :O j In the incompletely complementary sensor NCP j Adjusting the proportion of each incompletely complementary double chain in the NCP of the incompletely complementary sensor. j For input chain T d I x The perceived weights. Furthermore, NCP can be implemented, for example, by adjusting the concentrations of various components of a DNA neural network computer hardware in a liquid. y China-Africa fully complementary double-chain W xy :O y and NCP j China-Africa fully complementary double-chain W xj :O j Adjustment of the proportion.

[0057] Table 2 below shows a concentration ratio of each component of the DNA neural network computer hardware in this embodiment.

[0058] Table 2. Concentration ratios of various components of the DNA neural network computer hardware and the meaning of each DNA strand.

[0059] From Table 2, the weight chain W 11 Weighted chain W 21 Weighted chain W 31 Weighted chain W 41 The concentration of the output chain O1 indicates that the various incompletely complementary double strands W ultimately generated in NCP1... 11 :O1、W 21 :O1、W 31 :O1 and W 41 The ratio of O1 will be 1:1:0:0, meaning that NCP1 accounts for 1% of the input chain T. d I1, T d I2, T d I3 and T d The ratio of the perceived weights for I4 will also be 1:1:0:0. Similarly, from the weight chain W in Table 2... 12 W 22 W 32 W 12 W 42 From the concentration of O2 in the output chain, we can deduce the effect of NCP2 on the input chain T. d I1, T d I2, T d I3 and T d The ratio of the perceptual weights in I4 is 1:0:0:1. Therefore, it can be seen that by changing the relative proportions of different weight chains in a non-perfectly complementary perceptron, the perceptron can memorize and recognize the preset pattern categories of the input information represented by the input chains.

[0060] The above-mentioned incompletely complementary bichain W xy :O y or W xj :O j The percentage in the corresponding NCP is only an example. The actual situation should be set according to the encoding method of each input chain for the input information and the specific input information pattern category that each NCP needs to identify.

[0061] Still back Figure 3 Incompletely complementary bichain W xy :O y and W xj :O j Able to connect with input chain T d I x A non-completely complementary chain substitution reaction occurs between them, thereby releasing the corresponding output chain O. yOr output chain O j The DNA neural network computer hardware according to the embodiments of this application implements the output chain O through a winner-take-all module. y and output chain O j More accurate detection, and finally output fluorescent signals that are easy to identify and thus determine the pattern category of the input information represented by the input chain.

[0062] Specifically, the winner-takes-all module could, for example, include... The structure is A y A j The annihilation chain and m types of report chains corresponding to the output chains of each incompletely complementary perceptron, where the annihilation chain A y A j Used for output chain O y and output chain O j The difference in quantity is amplified, and the output chain O after the quantity difference is amplified is... y and / or output chain O j It undergoes incomplete complementary hybridization with the corresponding reporter double strand and releases the corresponding amount of R. y Chain or R j The input chain generates a corresponding fluorescence signal so that, when the input chain is added to the DNA neural network computer, the pattern category of the input information represented by the input chain is obtained as the output result based on the processing of the fluorescence signal.

[0063] The following is combined with Figure 2 and Figure 3 This study further elucidates the mechanisms of various DNA molecule reactions and neural network operations performed in the DNA neural network computing solution before and after the input strand is added.

[0064] Before the input strands are added to the computing solution, the various components of the DNA neural network computer undergo a series of DNA strand reactions in the solution. This is understandable, similar to NCP... y NCP j The DNA chain reactions associated with various NCPs are basically similar; therefore, the following discussion will only focus on those related to NCPs. y The following example illustrates the associated DNA chain reaction. Weighted strand W xy The DNA sequence contains the upstream tag domain T Wyu Incompletely complementary binding domain W x and the perceptron label field T Wy The output chain O y The DNA sequence contains the sensor tag domain T Oy Incompletely complementary binding domain O and downstream tag domain T Oyd ; through weighted chain Wxy Perceptor label domain T Wy With output chain O y Perceptor label domain T Oy Performing perfect complementary hybridization to make the weight chain W xy Specifically with output chain O y Hybridization to generate incompletely complementary double-stranded W xy :O y .

[0065] Combination Figure 2 As can be seen, in places such as W 11 In the DNA sequences of each weighted strand, the blue underlined portion represents the upstream tag domain, and the brown underlined portion represents the sensor tag domain. Therefore, W 11 W 21 W 31 and W 41 The corresponding upstream tag field T W1u For “TGT”, the corresponding perceptron label field T W1 For "TCT"; W 12 W 22 W 32 and W 42 The corresponding upstream tag field T W2u For “TGT”, the corresponding perceptron label field T W2 The designation is "GGA"; the ununderlined portions in the middle of each weight chain represent incompletely complementary binding domains W1, W2, W3, and W4. In each output chain, the brown underlined portion represents the perceptron label domain, the blue underlined portion represents the downstream label domain, and the ununderlined portion in the middle represents the incompletely complementary binding domain O. Therefore, for example, the perceptron label domain T of output chain O1... O1 For "AGA", the downstream tag field T O1d The perceptron label field T of the output chain O2 is "AAC". O2 For "TCC", the downstream tag field T O1d The identifier is "GGT". Furthermore, in the DNA sequences of each input strand, the underlined portion in blue represents the downstream tag domain T. d Therefore, it is "ACA". The other parts without underscores are incompletely complementary binding domains I. x .from Figure 2 As can be seen from the various incompletely complementary binding domains, the colored parts are mismatched bases. The incompletely complementary hybridization refers to the process by which two DNA strands form a DNA double helix through incompletely complementary base sequences. The number of mismatched bases in the formed incompletely complementary DNA double helix is ​​greater than 0. The specific number of mismatched bases can be optimized according to experimental results, and this application does not make a specific limitation on this.

[0066] With incompletely complementary sensor NCPy For example, NCP y Based on incompletely complementary double-stranded W xy :O y It operates through chain permutation reactions between each input chain, specifically including: via the weight chain W xy upstream tag domain T Wyu Downstream label field T of each input chain d Performing perfect complementary hybridization to promote the input chain T d I x With incompletely complementary bichain W xy :O y Chain substitution reactions occur between them. For example, W in NCP1 11 :O1, its weight chain W 11 upstream tag domain T W1u (“TGT”) can be associated with the downstream tag field T of each input chain. d ("ACA") performs perfect complementary hybridization, thereby promoting the T of each input strand. d I x With incompletely complementary bichain W 11 A chain substitution reaction occurs between O1 and O2.

[0067] In addition, it can also be achieved through the weighted chain W. xy Incompletely complementary binding domain W x Perform incomplete complementary hybridization with the incomplete complementary binding domain O in each output strand and endow with incomplete complementary double strand W. xy :O y With specificity, thus enabling the input chain T d I x After the DNA neural network computer solution is added, the incompletely complementary double-stranded W xy :O y Able to selectively connect to input chain T d I x Incompletely complementary binding domain I x Incomplete complementary hybridization is performed, and the output chain O is replaced and released through a chain substitution reaction. y .

[0068] Figure 4(a) illustrates a DNA neural network computer according to an embodiment of this application with an input strand of T. d I1 and T d A schematic diagram of the DNA strand displacement reaction at I2. Figure 4(b) shows a DNA neural network computer according to an embodiment of this application with an input strand of T. d A schematic diagram of the DNA strand displacement reaction at I2. It is worth noting that the DNA neural network computer in Figures 4(a) and 4(b) was still prepared according to the concentration ratio of each DNA strand as shown in Table 2.

[0069] In Figure 4(a), there is an incompletely complementary double strand W in NCP1. 11 :O1 and W 21 :O1, therefore, the weight chain W 11 The incompletely complementary binding domain W1 undergoes incomplete complementary hybridization with the incompletely complementary binding domain O in each output strand, thus endowing it with incompletely complementary double strands W. 11 :O1 is specific when the input chain T d After I1 enters the computer solution, W 11 O1 can selectively connect to the input chain T d The incompletely complementary binding domain I1 of I1 undergoes incomplete complementary hybridization, displacing and releasing the output chain O1 through a chain substitution reaction; similarly, when the input chain T... d After I2 enters the computer solution, W 21 O1 can selectively connect to the input chain T d The incompletely complementary binding domain I2 undergoes incomplete hybridization, displacing and releasing the output chain O1 via a chain substitution reaction. Similarly, since only incompletely complementary double-stranded W exists in NCP2... 12 O2 and W 42 :O2, therefore, when the input chain is T d I1 and T d After I2 enters the computer solution, only W 12 O2 can selectively interact with the input chain T d The incompletely complementary binding domain of I1 undergoes incomplete hybridization, displacing and releasing the output chain O2 via a chain substitution reaction, while W... 42 O2, on the other hand, does not have a connection with the input chain T. d I1 or T d Due to the specificity of incomplete complementary hybridization of I2, it cannot replace the output chain O2. Therefore, as can be seen from Figure 4(a), the input chain is T. d I1 and T d In the case of I2, the amount of output chain O1 that is replaced and output by NCP1 is twice that of output chain O2 that is replaced and output by NCP2.

[0070] Similar to Figure 4(a), in Figure 4(b), since there is only the input chain T... d I2 enters the computer solution, thus allowing NCP1 to replace and output W with a non-completely complementary double strand. 21 The output chain O1 is equal to O1, while NCP2 cannot replace the output chain O2 at all.

[0071] The differences in the amount / concentration of the output chain corresponding to different NCPs will be further used to determine the pattern category of the input information represented by the input chain. The specific method will be described in detail below.

[0072] Figure 5 This diagram illustrates the working principle of a winner-take-all module according to an embodiment of this application. The annihilation chain A in the winner-take-all module is shown. y A j With output chain O y and output chain O j Related, corresponding to the incompletely complementary perceptron NCP y The report's double-stranded structure is A y :R y This corresponds to the incompletely complementary perceptron NCP. j The report's double-stranded structure is A j :R j . Figure 5 Still using a four-input chain and two NCPs as an example, in this case, the winner-take-all module 500 only needs one annihilation chain A1A2, which is associated with output chains O1 and O2. The report double chain structure corresponding to the incompletely complementary perceptron NCP1 in the winner-take-all module 500 is A1:R1, and the report double chain structure corresponding to the incompletely complementary perceptron NCP2 is A2:R2.

[0073] Furthermore, annihilation chain A j :R j A y Partially corresponds to the incompletely complementary perceptron NCP y The annihilation chain A y A j A j Partially corresponds to the incompletely complementary perceptron NCP j The annihilation chain A y A j It has a single-chain hairpin-like structure that can simultaneously bind non-perfectly complementary sensor NCPs in a 1:1 molar ratio. y Released output chain O y Incompletely complementary sensor NCP j Released output chain O j In other words, in Figure 5 In the annihilation chain A1A2, the A1 part corresponds to the incompletely complementary perceptron NCP1, and the A2 part corresponds to the incompletely complementary perceptron NCP2, and can simultaneously bind O1 and O2 in a 1:1 molar ratio.

[0074] In the incompletely complementary sensor NCP y Release the corresponding output chain O yFurthermore, the incompletely complementary sensor NCP j Release the corresponding output chain O j In the case of the annihilation chain A y A j The structure can be opened and combined with the output chain O. y and output chain O j Both, thus only the remaining uncombined output chain O y and / or output chain O j Continue with the corresponding reporter bistrand undergoing an incomplete complementary chain substitution reaction, thereby releasing the corresponding amounts of R. y and / or R j The chain generates a corresponding fluorescence signal; and in the incompletely complementary sensor NCP y Incompletely complementary sensor NCP j When one of them releases the corresponding output chain, the annihilation chain A y A j The structure can be closed without being linked to the output chain O. y Or output chain O j In this process, the corresponding output strand undergoes incomplete complementary hybridization with the corresponding reporter double strand, thereby releasing the corresponding amount of R. y Chain or R j The chain generates a corresponding fluorescence signal.

[0075] like Figure 5 As shown, when the amount of output chain O1 is greater than the amount of output chain O2, as indicated in box 501, annihilation chain A1A2 can open the structure and combine both output chains O1 and O2, annihilating all output chains O2, leaving only output chain O1. This O1 then undergoes an incomplete complementary chain substitution reaction with the corresponding reporter double strand, releasing a corresponding amount of R1 and generating a fluorescence signal corresponding to R1. When the amount of output chain O1 is less than the amount of output chain O2, the process in box 502 is executed, ultimately releasing a corresponding amount of R2 and generating a fluorescence signal corresponding to R2.

[0076] Combination Figure 5 It can be seen that the annihilation chain A y A j It contains four fields: corresponding to output chain O y downstream tag field T Oyd upstream tag domain T Ayu Where A represents annihilation, y represents the number of the corresponding upstream output chain, and u represents upstream; corresponding to output chain O y The incompletely complementary binding domain O and the incompletely complementary binding domain A correspond to the output chain O. j downstream tag field T Ojd upstream tag domain T AjuWhere A represents annihilation and u represents upstream; corresponding to the output chain O j The incompletely complementary binding domain O and the incompletely complementary binding domain A.

[0077] The report doubly linked list corresponding to each output chain consists of one A y Chain or A j Chain, and R y Chain or R j Chains are formed through incomplete complementary hybridization, such as Figure 5 A1R1 and A2R2 in the output chain O, and can be connected to the output chain O. y Or output chain O j A chain substitution reaction is performed to report the amount of the corresponding upstream output chain; wherein, A y Chain or A j The chain is modified with a quenching group; the R y Chain or R j The chain has fluorescent group modification, from Figure 2 As shown in the DNA sequence, the fluorescent group of R1 is modified to Cy3 (corresponding to Cy3 fluorescence), and the fluorescent group of R2 is modified to Cy5 (corresponding to Cy5 fluorescence).

[0078] The A y Chain or A j The chain contains three fields: the upstream tag field T Ayu or upstream tag domain T Aju A represents annihilation, u represents upstream; the incompletely complementary binding domain A; the report tag domain T. Ay Or report tag field T Aj The chain R y or R j The chain contains two fields: a non-perfectly complementary binding field R and a report label field T. Ry Or report tag field T Rj R represents a report. The upstream label field T... Ayu or upstream tag domain T Aju Able to correspond with output chain O y downstream tag field T Oyd Or output chain O j downstream tag field T Ojd Binding occurs through perfectly complementary hybridization; the incompletely complementary binding domain A can correspondingly bind to R. y The incompletely complementary domain of the chain or R j The incompletely complementary domain R of the chain binds through incompletely complementary hybridization; when O y Or O j When the chain exists, the incompletely complementary binding domain A is associated with the output chain O. y O domain or output chain O j The O domain is combined; the report tag domain TAy Or report tag field T Aj Able to work with R y Chain or R j chain T Ry or T Rj The domains combine through perfectly complementary hybridization.

[0079] Box 503 illustrates the principle by which the annihilation process amplifies the difference in quantity between different output chains when the quantity of output chain O1 is greater than that of output chain O2. Assuming that before entering the winner-take-all module, the quantity of O1 is 15 and the quantity of O2 is 12, the quantity ratio is 1.25, resulting in a small contrast between the pink fluorescent signal representing O1 and the blue fluorescent signal representing O2. After annihilation by, for example, 10 annihilation chains A1A2, the remaining quantity of O1 is 5, and the quantity of O2 is 2. The quantity ratio is now amplified to 2.5, significantly increasing the contrast between the pink fluorescent signal representing O1 and the blue fluorescent signal representing O2.

[0080] Understandably, the number of annihilation chains in the winner-take-all module is not limited to one. Their number and structure are related to the number of NCPs. As an example, suppose there are three NCPs, NCP1, NCP2, and NCP3, with corresponding output chains O1, O2, and O3, respectively. Then, three annihilation chains with different structures, A1A2, A1A3, and A2A3, need to be configured in the winner-take-all module so as to accurately identify the output chain with the largest quantity in the computer solution through pairwise annihilation.

[0081] In some implementations, the concentration adjustment difference amplification effect of the corresponding annihilation chain can also be adjusted. For example, annihilation chain A can be adjusted. y A j Adjust the concentration to match the output chain O y and O j The concentrations of those with higher concentrations are equal (the same applies to O). y and O j (assuming equal concentrations), thereby outputting chain O y and O j The smaller amount is almost completely annihilated, thereby enhancing the amplification effect of the difference in the number of output chains.

[0082] In some implementations, based on the amplified difference in the number of output strands, the R value can be further compared to the value before and after adding the DNA neural network computing solution to the input strand. y Chain or R jThe change in the fluorescence signal corresponding to the chain is used to obtain the pattern category of the input information represented by the input chain and serve as the output result. More specifically, taking Cy3 and Cy5 fluorescence as examples, the intensity values ​​of the fluorescence signals corresponding to chains R1 and R2 before and after completing the neural network calculation can be normalized first using the following formula:

[0083] in This represents the normalized fluorescence intensity value. To report fluorescence intensity values, This represents the initial fluorescence intensity value. The relative changes in Cy3 and Cy5 fluorescence can be determined by the corresponding... The ratio is determined as follows: when This indicates that the Cy3 variation is more significant than the Cy5 variation, and the input information is more similar to the pattern type corresponding to NCP1 (e.g., "Apple") in memory. The threshold value... It is a value slightly greater than 1, for example, it can be set to 1.1 or other applicable values ​​determined experimentally, and this application does not specifically limit it.

[0084] when This indicates that the Cy5 variation is more significant than the Cy3 variation, and the input information is more similar to the memory of the pattern type corresponding to NCP2 (e.g., "Lemon"). The threshold value... The value is slightly less than 1, for example, it can be set to 0.95 or other suitable values ​​determined experimentally, and this application does not specifically limit it.

[0085] when This indicates that the Cy3 variation is similar to the Cy5 variation, making it impossible to distinguish the mode type corresponding to the input information.

[0086] TH1 and TH2 The specific value can be determined based on the experimental results of testing the "winner-take-all" module on a specific sequence. It is related to factors such as the sequence and concentration of each DNA molecule in the DNA neural network computer hardware, which will not be described in detail here.

[0087] The types of input chains, the number of incompletely complementary perceptrons (NCPs), the types of weight chains in the NCPs, the number of annihilation chains, and the number of bits in the input information and / or the encoding method of the input information are set in association with these settings. Furthermore, by increasing the types of input chains, the corresponding types of weight chains, the number of NCPs, and the number of annihilation chains in a matching manner, the complexity of the input information that the DNA neural network computer can process is increased. This complexity of the input information includes at least the number of pattern categories represented by the input information. In other embodiments, for example, by adjusting the encoding method of the input information and / or the mapping method between the encoded input information and the input chains, combined with factors such as the type of input chains, the DNA neural network can achieve different input information recognition accuracies, etc., which are not detailed here.

[0088] After performing neural network calculations based on the input strand using a DNA neural network computer, the input strand can be removed from the DNA neural network computer solution after the calculations are completed using reversed-phase column chromatography to recover the DNA neural network computer hardware. Specifically, for example, the DNA neural network computer solution after the calculations are completed can be transferred to a Sep-Pak solution pre-equilibrated with deionized water. ® The DNA neural network computer hardware is processed on a C18 column, then eluted with deionized water to remove salts from the solution. The components are then eluted from the column with a solution containing eluent. The eluent solution is then removed by nitrogen blowing, rotary evaporation, or lyophilization, yielding the eluted DNA neural network computer hardware with the input strand removed. This technique for recovering DNA neural network computer hardware is unattainable by existing technologies that rely on fuel chains or enzymes to drive fully complementary DNA hybridization reactions.

[0089] In other embodiments, based on the above-described recycling process, the input strand can also be recycled and reused by utilizing different eluents tailored to the characteristics of the input strand, thereby further reducing the operating cost of the DNA neural network computer. The specific process of input strand recycling is not described in detail here.

[0090] Further, the recovered DNA neural network computer hardware can be re-dissolved in a pre-prepared liquid medium, thereby obtaining a DNA neural network computer solution capable of performing similar neural network operations on input strands of the next round (the configurations of the input strands can be the same or different). That is, the high-fidelity incomplete complementary DNA neural network computer can be used in a highly repeatable manner, making the sample-based training of the DNA neural network possible. In other cases, the recovered DNA neural network computer hardware can also be used to prepare DNA neural network computers with other functions, which greatly saves the preparation cost of DNA neural network computers.

[0091] Hereinafter, with reference to FIG. 6(a), FIG. 6(b) and FIG. 6(c), the complete preparation and operation flow of the DNA neural network computer will be described by taking the implementation of the "you describe I guess" game as an example. FIG. 6(a) shows a schematic diagram of DNA strand reactions in the "you describe I guess" game according to an embodiment of the present application; FIG. 6(b) shows a truth table of game results under different input conditions of the "you describe I guess" game according to an embodiment of the present application; FIG. 6(c) shows strand reaction results corresponding to partial input conditions of the "you describe I guess" game according to an embodiment of the present application. Wherein, the DNA neural network computer in FIG. 6(a), FIG. 6(b) and FIG. 6(c) is also obtained by concentration proportioning according to each component of the DNA neural network computer hardware given in Table 2. In addition, in FIG. 6(a), FIG. 6(b) and FIG. 6(c), for convenience of identification, English is adopted in some parts, wherein the equivalence between Chinese and English includes: "Apple" = "苹果", "Lemon" = "柠檬", "Spherical" = "球形的", "Red" = "红色的", "Corticate" = "有皮的", "Yellow" = "黄色的".

[0092] The preparation and use method of the DNA neural network computer according to the embodiment of the present application comprises the following steps: Step 1, newly preparing or preparing a DNA neural network computer solution by using recovered DNA neural network computer hardware.

[0093] For new preparation, stock solutions respectively dissolving each component of the DNA neural network computer hardware are added into a salt solution containing Tween 20 according to a pre-calculated concentration, wherein the salt can be sodium chloride, potassium chloride, magnesium chloride or a mixture thereof. For example, in a 1 M sodium chloride solution containing 0.01% Tween 20, specifically, a certain volume of DNA hardware stock solution can be added into 100 µL of 1 M sodium chloride solution containing 0.01% Tween 20 in a 96-well plate, after mixing, the mixture is allowed to stand at room temperature and wait for the reaction to complete. For example, after 0.5 hours to 5 hours, the DNA neural network computer solution can be obtained.

[0094] Given n types of input chains and m number of incompletely complementary perceptrons, the DNA neural network computer hardware comprises various components including a weight chain W. xy Weighted chain W xj Output chain O y Output chain O j Annihilation Chain A y A j A y Chain, R y Chain, A j Chain, R j The chain, x, ranges from 1 to n, and y and j correspond to the numbers of the incompletely complementary perceptrons, with values ​​ranging from 1 to m and being distinct from each other; in the examples of Figures 6(a)-6(c), the various components of the DNA neural network computer hardware include the weight chain W. 11 -W 41 W 12 -W 42 Output chains O1 and O2, annihilation chains A1 and A2, chains A1 and A2, and chains R1 and R2.

[0095] For example only, the concentration setting requirement is W. xy Total chain concentration greater than or equal to O y W xj Total chain concentration greater than or equal to O j W xy Chain and W xj Chain concentration and input chain TdI x The weights are positively correlated, A y A j Chain concentration greater than or equal to 0 y Chain concentration (O) y Concentration greater than or equal to O j A y Concentration equals R y Equal to O y A j Concentration equals R j Equal to O j .

[0096] The stock solutions for each component of the DNA neural network computer hardware are obtained by dissolving the corresponding DNA strand powder in deionized water. Specifically, during the preparation of the DNA hardware stock solution, each DNA strand powder is dissolved in a certain volume of deionized water, with a final DNA strand concentration of approximately 100 µM, and the exact concentration of each DNA strand is measured.

[0097] When using recycled DNA neural network computer hardware for preparation, the recycled DNA neural network computer hardware can be directly redissolved in a salt solution containing Tween 20. In particular, when preparing DNA neural network computer hardware either brand new or using recycled DNA neural network computer hardware, it is not necessary to pre-anneal the double-stranded structures in each component of the DNA neural network computer hardware as in the prior art, which can greatly simplify the preparation process and shorten the preparation time.

[0098] Step 2, determine the initial fluorescence value of the DNA neural network computer: take out a predetermined volume v of the DNA neural network computer solution, and measure the initial fluorescence intensity value of different fluorescent labels in the DNA neural network computer; after recording the initial fluorescence intensity value of different fluorescent labels, return the taken-out DNA neural network computer solution to the original DNA neural network computer solution, for example, by using a pipette to aspirate and transfer it back to the original solution.

[0099] Step 3, Perform DNA neural network computation: Add an input strand stock solution, greater than or equal to the total concentration of the output strands of all incompletely complementary sensor (NCP) cells in the DNA neural network computer, to the DNA neural network computer solution completed in Step 2. When preparing the input strand stock solution, add each input strand T... d I x DNA powder was dissolved in a specific volume of deionized water to achieve a final DNA strand concentration of approximately 100 µM. The exact concentration of each input DNA strand was then determined. Input strand stock solution was added, mixed thoroughly, and allowed to stand at room temperature for 2 hours until the reaction was complete. A volume (v) of the reacted DNA neural network computer solution was then taken, and the terminal fluorescence intensity of different fluorescent labels in the DNA neural network system was measured.

[0100] As shown in Figure 6(a), each input chain is equivalent to a clue provided in the "You Say, I Guess" game, and the DNA neural network computer can reason based on the clues and ultimately give its guessed answer, determining which of the multiple target objects (corresponding to multiple pattern categories of the input information) it is. Since W 31 :O1 and W 41 The concentration of O1 is 0, therefore in NCP1, only incompletely complementary double-stranded W... 11 :O1 and W 21 O1 can respond to the input chain T respectively. d I1 and T d The input of I2, that is, NCP1 can be used to identify the input chain T which means "Spherical". d I1 and the input chain T, which means "Red". dI2 is used to identify the corresponding output chain O1; similarly, NCP2 can be used to identify the input chain T that means "Spherical". d I1 and the input chain T, which means "Yellow". d I4, and output the corresponding output chain O2.

[0101] Next, the winner-take-all module can output an R1 chain with Cy3 fluorescent label or an R2 chain with Cy5 fluorescent label based on the output chains from NCP1 and NCP2. Then, for example, the fluorescence intensity of Cy3 and Cy5 can be measured using a microplate reader.

[0102] Step 4, Data Processing and Result Output: Based on the terminal fluorescence intensity value and initial fluorescence intensity value of different fluorescent labels, calculate the fluorescence intensity change of different fluorescent labels, and calculate the pattern category of the input information represented by the input chain as the output result based on the ratio between the fluorescence intensity changes of different fluorescent labels.

[0103] For example, when the corresponding input chain is added to the DNA neural network computer solution according to the input chain configuration corresponding to the 4-bit input information encoding in Figure 6(b), the output result of the "You Say, I Guess" game can be output by calculating NS(Cy3) / NS(Cy5) and comparing it with the corresponding threshold range.

[0104] It is worth noting that the DNA neural network computer according to the embodiments of this application can not only identify completely corresponding "clues", but also has a certain fault tolerance capability. As shown in Figure 6(c), when the input information is encoded as (0, 1, 0, 0), there is only one input chain T. d Even when I2 enters the computer solution, it can still give the closest answer, namely, apple; or, if the input information is encoded as (1, 0, 1, 1), with three input chains T d I1, T d I3, T d Even when I4 enters the computer solution, it can still give the closest answer, namely, lemon. As shown in Figures 6(b) and 6(c), for the following four input information codes: (1, 1, 0, 0), (0, 1, 0, 0), (1, 1, 1, 0), and (0, 1, 1, 0), the closest output result after calculation is "apple". For the following four input information codes: (0, 1, 0, 1), (0, 0, 0, 1), (1,0, 1, 1), and (0, 0, 1, 1), the closest output result after calculation is "lemon". The corresponding fault tolerance cases are not listed here.

[0105] After the calculations are completed in step 4, step 5 can be further executed to recycle the DNA neural network computer hardware: the DNA neural network computer hardware is recycled by recovering the DNA neural network computer solution after the neural network operations were completed in step 3, so that the recycled DNA neural network computer hardware can be reused. The specific recycling operation has been described in detail above and will not be repeated here.

[0106] According to embodiments of this application, a method for applying a DNA neural network is also provided. The DNA neural network is used to identify pattern categories of input information. The application method includes: determining the number m of pattern categories of the input information to be identified; constructing the DNA neural network computer described in various embodiments of this application, such that the constructed DNA neural network contains m incompletely complementary perceptrons; setting the number n of input chain types and the chain structure of each input chain in association with the number m of pattern categories; when pattern category identification of multiple rounds of input information is required: encoding the input information of the current round and mapping the encoded input information to a combination of input chains; performing neural network operations based on the combination of input chains using the DNA neural network, and outputting the pattern category identification result of the current round of input information; performing a recycling process on the DNA neural network after completing the neural network operation to obtain a DNA neural network with input chains removed, and using the DNA neural network with input chains removed for pattern category identification of the next round of input information.

[0107] Figure 9 This diagram illustrates a DNA neural network performing multiple rounds of computation according to an embodiment of this application.

[0108] Figure 9 Taking the "I Say, You Guess" game mentioned earlier as an example, a total of five rounds of the game were played. After each round, the DNA neural network was processed and its data was reused in the next round of neural network operations after the pattern category of the input information was identified through neural network operations. Figure 9 The correspondence between the input chain and the recognition result shows that the DNA neural network that has been recycled and reused has the same accuracy as the newly prepared DNA neural network. This proves that the DNA neural network according to the embodiments of this application can be recycled and reused multiple times. Thus, without reducing the computational accuracy, multiple rounds of neural network operations can be completed at a lower time and economic cost, which has great potential in broadening the application scope of DNA neural networks.

[0109] Furthermore, although exemplary embodiments have been described herein, their scope includes any and all embodiments based on this disclosure that have equivalent elements, modifications, omissions, combinations (e.g., schemes involving intersections of various embodiments), adaptations, or alterations. Elements in the claims will be interpreted broadly based on the language used in the claims and are not limited to the examples described in this specification or during the implementation of this application, and such examples will be interpreted as non-exclusive. Therefore, this specification and examples are intended to be considered illustrative only, and the true scope and spirit are indicated by the full scope of the claims and their equivalents.

[0110] The above description is intended to be illustrative and not restrictive. For example, the above examples (or one or more of them) can be used in combination with each other. Other embodiments may be used by those skilled in the art upon reading the above description. Furthermore, in the above detailed description, various features may be grouped together to simplify this disclosure. This should not be construed as an intention that a disclosed feature that is not claimed is necessary for any claim. Rather, the subject matter of the invention may be less than all the features of a particular disclosed embodiment. Thus, the claims are incorporated herein by way of example or embodiment, wherein each claim is an independent, separate embodiment, and these embodiments are contemplated to be combined with each other in various combinations or arrangements. The scope of the invention should be determined by reference to the claims and the full scope of their equivalents.

Claims

1. A DNA neural network computer, characterized in that, A DNA neural network computer solution is obtained by dissolving the various components of the DNA neural network computer hardware in a pre-modified liquid medium, wherein the DNA neural network computer hardware, including an imperfectly complementary sensor and a winner-take-all module, is formed in the DNA neural network computer solution. After the input chain carrying encoded input information is added to the DNA neural network computer solution, neural network operations are performed through incomplete complementary hybridization between the input chain and the DNA chains in the DNA neural network computer hardware to obtain the pattern category of the input information represented by the input chain as the output result; and... The input chain is a lipid-nucleic acid coupled molecule, wherein the lipid molecule is covalently coupled to the 5' or 3' end of the nucleic acid molecule, which enables the input chain in the DNA neural network computer solution after the neural network operation to be removed to obtain the recycled DNA neural network computer hardware for reuse. The number of input chain types, n, is set in relation to the number of pattern categories of the input information and the encoding method of the input information. The number of incompletely complementary perceptrons, m, is set correspondingly to the number of pattern categories of the input information, and m is greater than or equal to 2; input chain T d I x The nucleic acid sequence contains the downstream tag domain T d Incompletely complementary binding domain I x Where x represents the input chain number, and the value of x ranges from 1 to n. The components of the DNA neural network computer hardware include: There are m weight chains and m output chains corresponding to each non-perfectly complementary perceptron. Annihilation Chain A y A j There are m types of A chains modified with quenching groups and m types of R chains modified with different fluorescent groups; where y and j correspond to the numbers of the incompletely complementary sensor, and the values ​​of y and j range from 1 to m and are different from each other; In the aforementioned DNA neural network computer hardware, the incompletely complementary perceptron (NCP) numbered y is... y W contains n structures xy :O y Incompletely complementary bichain, incompletely complementary bichain W xy :O y By weight chain W xy and output chain O y Formed through hybridization and capable of interacting with the input chain T d I x An incompletely complementary hybridization chain substitution reaction occurs, releasing the corresponding output chain O. y By adjusting the incompletely complementary double-stranded W xy :O y In the incompletely complementary sensor NCP y Adjusting the proportion of each incompletely complementary double chain in the NCP of the incompletely complementary sensor. y For input chain T d I x Perceived weights; In the aforementioned DNA neural network computer hardware, the incompletely complementary perceptron (NCP) numbered j is... j W contains n structures xj :O j Incompletely complementary bichain, incompletely complementary bichain W xj :O j By weight chain W xj and output chain O j Formed through hybridization and capable of interacting with the input chain T d I x An incompletely complementary hybridization chain substitution reaction occurs, releasing the corresponding output chain O. j And by adjusting the incompletely complementary double-stranded W xj :O j In the incompletely complementary sensor NCP j Adjusting the proportion of each incompletely complementary double chain in the NCP of the incompletely complementary sensor. j For input chain T d I x Perceived weights; The winner-takes-all module in the DNA neural network computer hardware includes The structure is A y A j The annihilation chain and m types of report chains corresponding to the output chains of each incompletely complementary perceptron, where the annihilation chain A y A j Used for output chain O y and output chain O j The difference in quantity is amplified, and the output chain O after the quantity difference is amplified is... y and / or output chain O j It undergoes incomplete complementary hybridization with the corresponding reporter double strand and releases the corresponding amount of R. y Chain or R j The input chain generates a corresponding fluorescence signal so that, when the input chain is added to the DNA neural network computer, the pattern category of the input information represented by the input chain is obtained as the output result based on the processing of the fluorescence signal.

2. The DNA neural network computer according to claim 1, characterized in that, The weight chain W xy The DNA sequence contains the upstream tag domain T Wyu Incompletely complementary binding domain W x and the perceptron label field T Wy The output chain O y The DNA sequence contains the sensor tag domain T Oy Incompletely complementary binding domain O and downstream tag domain T Oyd ; through weighted chain W xy Perceptor label domain T Wy With output chain O y Perceptor label domain T Oy Performing perfect complementary hybridization to make the weight chain W xy Specifically with output chain O y Hybridization to generate incompletely complementary double-stranded W xy :O y ; The incompletely complementary sensor NCP y Based on incompletely complementary double-stranded W xy :O y It operates through chain substitution reactions between each input chain, specifically Includes: via weight chain W xy upstream tag domain T Wyu Downstream label field T of each input chain d Performing perfect complementary hybridization to promote the input chain T d I x With incompletely complementary bichain W xy :O y Chain substitution reactions occur between them; through the weighted chain W xy Incompletely complementary binding domain W x Perform incomplete complementary hybridization with the incomplete complementary binding domain O in each output strand and endow with incomplete complementary double strand W. xy :O y With specificity, thus enabling the input chain T d I x After the DNA neural network computer solution is added, the incompletely complementary double-stranded W xy :O y Able to selectively connect to input chain T d I x Incompletely complementary binding domain I x Incomplete complementary hybridization is performed, and the output chain O is replaced and released through a chain substitution reaction. y .

3. The DNA neural network computer according to claim 2, characterized in that, The annihilation chain A y A j Used for output chain O y and output chain O j The difference in quantity is amplified, and the output chain O after the quantity difference is amplified is... y and / or output chain O j It undergoes incomplete complementary hybridization with the corresponding reporter double strand and releases the corresponding amount of R. y Chain or R j The chain generates corresponding fluorescence signals, specifically including: Annihilation Chain A y A j With output chain O y and output chain O j Related, corresponding to the incompletely complementary perceptron NCP y The report's double-stranded structure is A y :R y This corresponds to the incompletely complementary perceptron NCP. j The report's double-stranded structure is A j :R j ; The annihilation chain A in the winner-takes-all module y A j A y Partially corresponds to the incompletely complementary perceptron NCP y The annihilation chain A y A j A j Partially corresponds to the incompletely complementary perceptron NCP j The annihilation chain A y A j It has a single-chain hairpin-like structure that can simultaneously bind non-perfectly complementary sensor NCPs in a 1:1 molar ratio. y Released output chain O y Incompletely complementary sensor NCP j Released output chain O j ,in, In the incompletely complementary sensor NCP y Release the corresponding output chain O y Furthermore, the incompletely complementary sensor NCP j Release the corresponding output chain O j In the case of the annihilation chain A y A j The structure can be opened and combined with the output chain O. y and output chain O j Both, thus only the remaining uncombined output chain O y and / or output chain O j Continue with the corresponding reporter bistrand undergoing an incomplete complementary chain substitution reaction, thereby releasing the corresponding amounts of R. y and / or R j The chain generates a corresponding fluorescence signal; and In the incompletely complementary sensor NCP y Incompletely complementary sensor NCP j When one of them releases the corresponding output chain, the annihilation chain A y A j The structure can be closed without being linked to the output chain O. y Or output chain O j In this process, the corresponding output strand undergoes incomplete complementary hybridization with the corresponding reporter double strand, thereby releasing the corresponding amount of R. y Chain or R j The chain generates a corresponding fluorescence signal.

4. The DNA neural network computer according to claim 3, characterized in that, The output chain O is annihilated. y and output chain O j Further amplification of the differences in quantity includes: The effect of amplifying the difference is achieved by adjusting the concentration of the corresponding annihilation chain, including: adjusting annihilation chain A. y A j Adjust the concentration to match the output chain O y and O j The concentrations of those with higher concentrations are equal.

5. The DNA neural network computer according to claim 3 or 4, characterized in that, Performing neural network operations through incomplete complementary hybridization between the input chain and the DNA chain in the DNA neural network computer hardware to obtain the pattern category of the input information represented by the input chain as the output result further includes: Based on the input strand, the R before and after adding the DNA neural network computer solution... y Chain or R j The change in the fluorescence signal corresponding to the chain is used to obtain the pattern category of the input information represented by the input chain and serve as the output result.

6. The DNA neural network computer according to claim 3 or 4, characterized in that, The annihilation chain A y A j It contains four fields: corresponding to output chain O y downstream tag field T Oyd upstream tag domain T Ayu Where A represents annihilation, y represents the number of the corresponding upstream output chain, and u represents upstream; corresponding to output chain O y The incompletely complementary binding domain O and the incompletely complementary binding domain A correspond to the output chain O. j downstream tag field T Ojd upstream tag domain T Aju Where A represents annihilation and u represents upstream; corresponding to the output chain O j The incompletely complementary binding domain O and the incompletely complementary binding domain A; The report doubly linked list corresponding to each output chain consists of one A y Chain or A j Chain, and R y Chain or R j The chain is formed through incomplete complementary hybridization and can interact with the output chain O. y Or output chain O j A chain substitution reaction is performed to report the amount of the corresponding upstream output chain; wherein, A y Chain or A j The chain is modified with a quenching group; the R y Chain or R j The chain is modified with fluorescent groups; and, The A y Chain or A j The chain contains three fields: the upstream tag field T Ayu or upstream tag domain T Aju A represents annihilation, u represents upstream; the incompletely complementary binding domain A; the report tag domain T. Ay Or report tag field T Aj ;and, The chain R y Or R j The chain contains two fields: the incompletely complementary binding field R and the report label field T. Ry Or report tag field T Rj R stands for report; and Upstream tag field T Ayu or upstream tag domain T Aju Able to correspond with output chain O y downstream tag field T Oyd Or output chain O j downstream tag field T Ojd Binding occurs through perfectly complementary hybridization; the incompletely complementary binding domain A can correspondingly bind to R. y The incompletely complementary domain of the chain or R j The incompletely complementary domain R of the chain binds through incompletely complementary hybridization; when O y Or O j When the chain exists, the incompletely complementary binding domain A is associated with the output chain O. y O domain or output chain O j The O domain is combined; the report tag domain T Ay Or report tag field T Aj Able to work with R y Chain or R j chain T Ry or T Rj The domains combine through perfectly complementary hybridization.

7. The DNA neural network computer according to any one of claims 1-4, characterized in that, The types of input chains, the number of incompletely complementary perceptrons (NCPs), the types of weight chains in the NCPs, the number of annihilation chains, and the number of bits of the input information and / or the encoding method of the input information are set in association with each other; and the complexity of the input information that the DNA neural network computer can process is increased by increasing the types of input chains, the corresponding types of weight chains, the number of incompletely complementary perceptrons (NCPs), and the number of annihilation chains in a matching manner, wherein the complexity of the input information includes at least the number of pattern categories represented by the input information.

8. The DNA neural network computer according to any one of claims 1-4, characterized in that, The incomplete complementary hybridization refers to the process by which two DNA strands form a DNA double helix through incomplete complementary base sequences, and the number of mismatched bases in the formed incomplete complementary DNA double helix is ​​greater than 0.

9. The DNA neural network computer according to any one of claims 1-4, characterized in that, By changing the relative proportions of different weight chains in a non-perfectly complementary perceptron, the non-perfectly complementary perceptron can form a memory of the preset pattern category of the input information represented by the input chain and have the ability to recognize it.

10. The DNA neural network computer according to any one of claims 1-4, characterized in that, The process of removing the input strand from the DNA neural network computing solution after neural network computation to obtain recycled DNA neural network computing hardware for reuse includes: The input strands in the DNA neural network computer solution after the calculations are completed are removed using reversed-phase column chromatography to recover the DNA neural network computer hardware. The recovered DNA neural network computer hardware is redissolved in a pre-modified liquid medium to obtain a DNA neural network computer solution capable of performing neural network operations on the next round of input strands; or The recycled DNA neural network computer hardware was used to create DNA neural network computers with other functions.

11. The DNA neural network computer according to claim 10, characterized in that, The recovery of the DNA neural network computer hardware involves using reversed-phase column chromatography to remove the input strands from the completed DNA neural network computer solution. After completing the calculations, the DNA neural network computed solution was transferred to Sep-Pak pre-equilibrated with deionized water. ® The DNA neural network computer hardware is then eluted from the column using a C18 column and then washed with deionized water to remove salts from the solution. The components of the DNA neural network computer hardware are then eluted from the column using a solution containing eluent. The solution containing eluent is then removed by nitrogen blowing, rotary evaporation, or lyophilization to obtain the eluted DNA neural network computer hardware.

12. The DNA neural network computer according to any one of claims 1-4, characterized in that, The lipid molecule is a lipid molecule with a hydrophobic structure, which includes at least a hydrophobic structure derived from long-chain alkanes, C6-C50 alkenes or C6-C50 cycloalkanes, wherein the long-chain alkanes are hydrocarbon chains with more than 6 carbon atoms, or hydrocarbon chains with 6-40, 8-30 or 10-20 carbon atoms.

13. The DNA neural network computer according to any one of claims 1-4, characterized in that, The pre-modified liquid medium is a salt solution containing Tween 20. When the number of input strand types is n and the number of incompletely complementary perceptrons is m, the preparation and use method of the DNA neural network computer includes the following steps: Step 1: Prepare a DNA neural network computer solution using either novel or recycled DNA neural network computer hardware, wherein... In the novel preparation process, stock solutions containing each component of the DNA neural network computer hardware are added to a salt solution containing Tween 20 at pre-calculated concentrations. After mixing, the solution is allowed to stand at room temperature until the reaction is complete to obtain the DNA neural network computer solution. The components of the DNA neural network computer hardware include a weight chain W. xy Weighted chain W xj Output chain O y Output chain O j Annihilation Chain A y A j A y Chain, R y Chain, A j Chain, R j The chain, x, ranges from 1 to n, y and j correspond to the numbers of the incompletely complementary perceptrons, and y and j range from 1 to m and are different from each other; wherein, the stock solution of each component of the DNA neural network computer hardware is obtained by dissolving the corresponding component's DNA strand dry powder in deionized water. When using recycled DNA neural network computer hardware for preparation, the recycled DNA neural network computer hardware can be directly redissolved in a salt solution containing Tween 20. Step 2, determine the initial fluorescence value of the DNA neural network computer: take out a predetermined volume v of the DNA neural network computer solution, and measure the initial fluorescence intensity value of different fluorescent labels in the DNA neural network computer; after recording the initial fluorescence intensity value of different fluorescent labels, return the taken-out DNA neural network computer solution to the DNA neural network computer solution. Step 3, Perform DNA neural network computation: Add an input strand stock solution, greater than or equal to the total concentration of the output strands of all incompletely complementary sensor (NCP) cells in the DNA neural network computer, to the DNA neural network computer solution completed in Step 2. Mix well, allow to stand at room temperature and wait for the reaction to complete. Then, take a volume v of the DNA neural network computer solution after the reaction and measure the terminal fluorescence intensity values ​​of different fluorescent labels in the DNA neural network system. The input strand stock solution is obtained by dissolving the dry powder of the input DNA strand in deionized water. Step 4, Data Processing and Result Output: Based on the terminal fluorescence intensity value and initial fluorescence intensity value of different fluorescent labels, calculate the fluorescence intensity change of different fluorescent labels, and calculate the pattern category of the input information represented by the input chain as the output result based on the ratio between the fluorescence intensity changes of different fluorescent labels.

14. The DNA neural network computer according to claim 13, characterized in that, In step 1, when preparing the DNA neural network computer hardware either novel or using recycled DNA neural network computer hardware, it is not necessary to pre-anneal the double-stranded structures in the various components of the DNA neural network computer hardware.

15. The DNA neural network computer according to claim 13, characterized in that, The method of using the DNA neural network computer further includes: Step 5, Recycling DNA Neural Network Computer Hardware: The DNA neural network computer hardware is recycled by recycling the DNA neural network computer solution after completing the neural network operation in Step 3, so that the recycled DNA neural network computer hardware can be reused.

16. The DNA neural network computer according to claim 14, characterized in that, The method of using the DNA neural network computer further includes: Step 5, Recycling DNA Neural Network Computer Hardware: The DNA neural network computer hardware is recycled by recycling the DNA neural network computer solution after completing the neural network operation in Step 3, so that the recycled DNA neural network computer hardware can be reused.

17. A method for applying a DNA neural network, characterized in that, The DNA neural network is used to identify pattern categories of input information, and the application method includes: Determine the number m of pattern categories of the input information to be identified; Construct a DNA neural network computer according to any one of claims 1-16, such that the constructed DNA neural network comprises m incompletely complementary perceptrons. The number of input chain types n and the chain structure of each input chain are set in relation to the number of pattern categories m; In situations where pattern categories need to be identified from multiple rounds of input information: Encode the input information for the current round and map the encoded input information into a combination of input chains; The DNA neural network performs neural network operations based on the combination of the input strands and outputs the pattern category recognition result of the input information in this round. The DNA neural network after completing the neural network operation is recycled to obtain the DNA neural network after removing the input strands, and the DNA neural network after removing the input strands is used for pattern category recognition of the input information in the next round.

Images