CN112348178A - Artificial neural network calculation model construction method based on DNA strand displacement - Google Patents
Artificial neural network calculation model construction method based on DNA strand displacement Download PDFInfo
- Publication number
- CN112348178A CN112348178A CN202011235689.8A CN202011235689A CN112348178A CN 112348178 A CN112348178 A CN 112348178A CN 202011235689 A CN202011235689 A CN 202011235689A CN 112348178 A CN112348178 A CN 112348178A
- Authority
- CN
- China
- Prior art keywords
- gate
- dna
- neuron
- gate module
- strand displacement
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 108020004414 DNA Proteins 0.000 title claims abstract description 65
- 238000006073 displacement reaction Methods 0.000 title claims abstract description 40
- 102000053602 DNA Human genes 0.000 title claims abstract description 39
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 37
- 238000004364 calculation method Methods 0.000 title description 4
- 238000010276 construction Methods 0.000 title description 2
- 210000002569 neuron Anatomy 0.000 claims abstract description 57
- 238000000034 method Methods 0.000 claims abstract description 16
- 238000009825 accumulation Methods 0.000 claims abstract description 5
- 238000005094 computer simulation Methods 0.000 claims abstract description 5
- 238000003062 neural network model Methods 0.000 claims abstract description 3
- 238000006243 chemical reaction Methods 0.000 description 13
- 238000010586 diagram Methods 0.000 description 9
- 238000005516 engineering process Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 5
- 238000004088 simulation Methods 0.000 description 5
- 238000001514 detection method Methods 0.000 description 4
- 238000002474 experimental method Methods 0.000 description 3
- 102000039446 nucleic acids Human genes 0.000 description 2
- 108020004707 nucleic acids Proteins 0.000 description 2
- 150000007523 nucleic acids Chemical class 0.000 description 2
- 238000012408 PCR amplification Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 238000001962 electrophoresis Methods 0.000 description 1
- 239000007850 fluorescent dye Substances 0.000 description 1
- 238000001215 fluorescent labelling Methods 0.000 description 1
- 238000000338 in vitro Methods 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 230000001404 mediated effect Effects 0.000 description 1
- 238000012821 model calculation Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Neurology (AREA)
- Measuring Or Testing Involving Enzymes Or Micro-Organisms (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method for constructing an artificial neural network computational model based on DNA strand displacement, which comprises the following steps: taking the DNA single chain as the input of a multiplication gate, constructing a multiplication gate module by using a DNA chain replacement mode, and carrying out multiplication of neuron input and weight; taking the DNA single strand as the input of an adder gate, constructing an adder gate module by using a DNA strand displacement mode, and accumulating neurons; taking the DNA single strand as the input of a subtraction gate, and constructing a subtraction gate module by using a DNA strand displacement mode so as to carry out an accumulation process when the weight of the neuron is negative; constructing a threshold gate module by using a DNA strand displacement mode and using the DNA single strand as the input of the threshold gate; thereby realizing the neuron threshold function; cascading a multiplication gate module, an addition gate module, a subtraction gate module and a threshold gate module to construct a single artificial neuron model; and (4) cascading the single neuron models to construct a feed-forward neural network model based on DNA strand displacement.
Description
Technical Field
The invention relates to the field of biological computation, in particular to a method for constructing an artificial neural network computation model based on DNA strand displacement.
Background
In vitro DNA molecule programming by utilizing a toehold mediated strand displacement reaction has the characteristics of spontaneity, stability, accuracy and the like. Based on these characteristics, DNA molecular programming has made some progress in the fields of digital logic computing circuits, analog computing circuits, DNA nano-robots, neural networks, and the like. Currently, the development of DNA systems with specific functions to solve specific problems is the main way to program DNA molecules today. DNA strands have a strong specific recognition property because they follow the strict Watson Crick principle when they are subjected to strand displacement. The DNA strand displacement can react under the normal temperature condition, the reaction is driven to be carried out mainly by intermolecular acting force, and any abstract chemical reaction kinetics can be realized through the cascade of the DNA strand displacement, so that the molecular programming by utilizing the DNA strand displacement has more experimental operability and flexibility. The DNA strand displacement can realize mathematical operations such as addition, subtraction, multiplication and division of four fundamental operations in mathematics, and can also construct a DNA molecular reaction network to realize more complex models, such as a neural network, a feedback control network, an oscillator, a timer and the like. The realization of more complex model calculations by DNA molecular reaction networks is the direction of DNA strand displacement development. The nucleic acid molecule has natural biocompatibility, and the reaction on the nanometer scale can be directly carried out in cells, so that the nucleic acid molecule has great application potential for disease detection and treatment.
There have been several efforts to implement simple artificial neural networks using DNA molecular programming. Qian et al realized Hopfield neural networks of four neurons based on a 'seesaw' model, and showed that the neural network realized by DNA molecules can also realize the function of associative memory. In 2016, Lakin et al proposed a DNA circuit which utilizes a buffer DNA chain replacement network to realize a gradient descent algorithm self-adaptive molecular framework and simulates supervised learning of a linear function. Qian et al in 2018 cascade-connected a neural network of 'winning' by using 'seesaws' gate circuits, and realize the recognition of '0-9' handwritten numbers. These results indicate that DNA strand displacement has great application potential in molecular neural networks.
The detection problem in the DNA strand displacement technology is a problem of limiting DNA calculation, and in the current research on DNA strand displacement, the main method is to use the conventional biological detection technology, mainly including fluorescence labeling technology, electrophoresis technology, PCR amplification technology, molecular beacon technology and the like. These detection techniques have a problem in that the concentration of a DNA strand cannot be accurately detected. Another problem is leakage during the experiment, which is an undesirable reaction when performing DNA strand displacement experiments, and leakage restriction is a problem that needs to be considered by the experimenter when designing DNA strand displacement experiments.
Disclosure of Invention
According to the problems in the prior art, the invention discloses a method for constructing an artificial neural network computational model based on DNA strand displacement, which specifically comprises the following steps:
taking the DNA single chain as the input of a multiplication gate, constructing a multiplication gate module by using a DNA chain replacement mode, and carrying out multiplication of neuron input and weight;
taking the DNA single strand as the input of an adder gate, constructing an adder gate module by using a DNA strand displacement mode, and accumulating neurons;
taking the DNA single strand as the input of a subtraction gate, and constructing a subtraction gate module by using a DNA strand displacement mode so as to carry out an accumulation process when the weight of the neuron is negative;
constructing a threshold gate module by using a DNA strand displacement mode and using the DNA single strand as the input of the threshold gate; thereby realizing the neuron threshold function;
cascading a multiplication gate module, an addition gate module, a subtraction gate module and a threshold gate module to construct a single artificial neuron model;
and (4) cascading the single neuron models to construct a feed-forward neural network model based on DNA strand displacement.
Further, the multiplication of the input and the weight is performed based on a multiplication gate module, wherein the weight is set to be a positive number or a negative number.
Further, the adder gate module integrates the output of the multiplier gate with the output of the adder gate, and adds a positive value to a positive value and a negative value to a negative value.
Further, the threshold gate module subtracts the positive value output of the addition gate from the value of the threshold gate.
Further, the subtraction gate module subtracts the positive value output of the addition gate from the negative value output of the addition gate in a working state to obtain the output of a single neuron.
Further, the neuron cascade of the artificial neural network computational model realizes a feedforward neural network. Specifically, a basic operation module is cascaded into a single neuron model, and then the single neuron model is cascaded to form a feedforward neural network.
Due to the adoption of the technical scheme, the method for constructing the artificial neural network calculation model based on the DNA strand displacement realizes the operation module based on the DNA strand displacement, combines the single neuron module based on the operation module, realizes the connection of large-scale neurons through the cascade connection of the single neurons, and realizes the feedforward neural network, so that the method provides a beneficial thought for constructing the complicated DNA strand displacement neural network.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments described in the present application, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a schematic diagram of a single neuron model according to the present invention;
FIG. 2 is a schematic diagram of the reaction process of the summing gate module of the present invention;
FIG. 3 is a schematic diagram of a reaction process of a subtraction gate module according to the present invention;
FIG. 4 is a schematic diagram of a reaction process of a multiplication gate module according to the present invention;
FIG. 5 is a schematic diagram of a reaction process of a threshold gate module according to the present invention;
FIG. 6 is a flow chart of the basic operation module cascaded into a single neuron module according to the present invention;
FIG. 7 is a schematic diagram of an XOR neural network structure according to the present invention;
FIG. 8 is a diagram of the simulation result of the neural network of the XOR logic operation of the present invention;
FIG. 9 is a schematic diagram of a logical operation neural network of the full adder according to the present invention;
FIG. 10 is a diagram of the simulation result of the full adder logic operation neural network of the present invention.
Detailed Description
In order to make the technical solutions and advantages of the present invention clearer, the following describes the technical solutions in the embodiments of the present invention clearly and completely with reference to the drawings in the embodiments of the present invention:
FIG. 6 shows a method for constructing an artificial neural network computational model based on DNA strand displacement, which comprises the following steps:
step 1: a multiplication gate module is constructed based on the DNA strand displacement technique, and as shown in FIG. 4, the input of the neuron is multiplied by the weight to obtain the output Ia of the multiplication gatei. When the weight is positive, the output positive value Ia is obtainedi+When the weight is negative, the output negative value Ia is obtainedi-。
Step 2: based on the DNA strand displacement structure addition gate module, as shown in FIG. 2, outputs Ia to the multiplication gate modulei+And Iai-Performing addition operation to obtain output Sp of addition gate module+And Sp-。
And step 3: implementation of threshold gate modules based on DNA Strand Displacement techniques As shown in FIG. 5, the output Sp of the addition gate module is+Reacting with a threshold gate module and outputting Sp after the reaction is finished+The value is obtained.
And 4, step 4: the subtraction gate module is implemented based on DNA strand displacement, and as shown in FIG. 3, the input of the subtraction gate module is the output Sp of the threshold gate+And the output Sp of the summing gate module-The subtraction gate module makes the input value Sp+Minus the input value Sp-The remaining output Sp+Is the output of a single neuron, and the process completes the operation of the single neuron. The building of a single neuron is completed through the cascade connection of basic operation modules, and an abstract single neuron model is shown in figure 1.
And 5: the hierarchical connection of single neurons to form a feed-forward neural network based on DNA strand displacement is shown in fig. 7 and 8.
Example 1: feed-forward neural network for XOR logical operation
The single neuron modules form a multi-layer feedforward neural network through hierarchical connection as shown in FIG. 7, and the first layer weight is set as followsFirst layer thresholdSecond layerSecond layer bias th3=0。
From the whole reaction process, the input chain x1And the weight value w1Performing multiplication operation, the multiplication gate will output x1w1And one and x1One chain of equal amount x'1Same as x2And w2Performing a multiplication operation outputs x2w2And a and input x2Equal further chain x'2. Output chain x of multiplication gate1w1And x2w2The accumulation function of the neurons is realized through an addition gate. Output S of the adder gate1Obtaining the output chain Y of the first neuron through a threshold value gate1。
Additionally outputs a replicated output chain x 'of the multiplier gate'1And x'2Multiplying the input of the next neuron of the same layer by the weight to obtain x'1w3And x'2w4Similarly to the first neuron, the threshold gate obtains the output Y of the second neuron after passing through the addition gate2。
We set upPut Y1With a negative weight w5Result of multiplication Y1 w5As the decrement of the decrement gate. Output Y of the second neuron of the first layer2And the weight value w6Multiplying to obtain an output chain Y2w6,Y2w6Reacts with the threshold gate and then with Y1 w5Are subtracted. An output Z chain is obtained, and the output is determined according to the quantity of the output Z when the reaction is finished.
The result of the neural network simulation with exclusive-or logic operation is shown in fig. 8.
Example 2 implementation of full-adder logic operation by feedforward neural network
The cascade of multiple neurons can also enable more complex operations, where we use five neurons to form a three-input two-output two-layer neural network as shown in fig. 9. The neural network can realize the logic operation of the full adder. Similar to the implementation of an XOR operation neural network, the network is built by using a single neuron, and after a weight and a threshold are designed, the building of the network can be completed by directly using single neuron cascade.
The full adder operation feedforward neural network has two layers, the first layer has three neurons, the second layer has two neurons, the adjacent layers are connected in a full connection mode, and the network has five neurons. A full adder neural network architecture implementation is shown in fig. 9. First layer weight settingThreshold valueSecond tier weight settingThreshold setting
The way that DNA strand displacement implements a full adder is similar to the way that XOR is implemented, with three inputs x1,x2,x3Respectively at w1,w2,w3Multiplying to obtain an output chain x1w1,x2w2,x3w3And x'1,x'2,x'3。x1w1,x2w2,x3w3The accumulation function of the neuron is realized as the input of the addition gate, and the accumulated sum is subtracted from the sum threshold value to obtain the output, x 'of the first neuron'1,x'2,x'3The weights of the second neuron are multiplied, and the output of the second neuron is obtained through an addition gate and a threshold gate. The third neuron is implemented in the same manner. The output of the neurons of layer 1 serve as the input chain for the neurons of layer 2. And multiplying the input chain by the weight, accumulating, thresholding, adding a gate, and subtracting a gate to obtain an output value. The full-adder neural network simulation results are shown in fig. 10.
The simulation experiment is carried out on the model by means of Visual DSD under the operating environments of Intel (R) CPU3.6GHz, 4.0GB memory and Windows 10.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered to be within the technical scope of the present invention, and the technical solutions and the inventive concepts thereof according to the present invention should be equivalent or changed within the scope of the present invention.
Claims (6)
1. A method for constructing an artificial neural network computational model based on DNA strand displacement is characterized by comprising the following steps:
taking the DNA single chain as the input of a multiplication gate, constructing a multiplication gate module by using a DNA chain replacement mode, and carrying out multiplication of neuron input and weight;
taking the DNA single strand as the input of an adder gate, constructing an adder gate module by using a DNA strand displacement mode, and accumulating neurons;
taking the DNA single strand as the input of a subtraction gate, and constructing a subtraction gate module by using a DNA strand displacement mode so as to carry out an accumulation process when the weight of the neuron is negative;
constructing a threshold gate module by using a DNA strand displacement mode and using the DNA single strand as the input of the threshold gate; thereby realizing the neuron threshold function;
cascading a multiplication gate module, an addition gate module, a subtraction gate module and a threshold gate module to construct a single artificial neuron model;
and (4) cascading the single neuron models to construct a feed-forward neural network model based on DNA strand displacement.
2. The method of claim 1, wherein: and multiplying the input by a weight value based on the multiplication gate module, wherein the weight value is set to be a positive number or a negative number.
3. The method of claim 1, wherein: the adder gate module integrates the output of the multiplier gate from the adder gate, adds a positive value to a positive value, and adds a negative value to a negative value.
4. The method of claim 1, wherein: the threshold gate module performs subtraction operation on the positive value output of the addition gate and the value of the threshold gate.
5. The method of claim 1, wherein: and the subtraction gate module subtracts the positive value output of the addition gate and the negative value output of the addition gate under the working state to obtain the output of a single neuron.
6. The method of claim 1, wherein: the neuron cascade of the artificial neural network computing model realizes a feedforward neural network.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011235689.8A CN112348178B (en) | 2020-11-06 | 2020-11-06 | Artificial neural network calculation model construction method based on DNA strand displacement |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011235689.8A CN112348178B (en) | 2020-11-06 | 2020-11-06 | Artificial neural network calculation model construction method based on DNA strand displacement |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112348178A true CN112348178A (en) | 2021-02-09 |
CN112348178B CN112348178B (en) | 2024-03-29 |
Family
ID=74429674
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011235689.8A Active CN112348178B (en) | 2020-11-06 | 2020-11-06 | Artificial neural network calculation model construction method based on DNA strand displacement |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112348178B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113762513A (en) * | 2021-09-09 | 2021-12-07 | 沈阳航空航天大学 | DNA neuron learning method based on DNA strand displacement |
CN115358389A (en) * | 2022-09-01 | 2022-11-18 | 清华大学 | Neural network training error reduction method, neural network training error reduction device, electronic device, and medium |
WO2024041137A1 (en) * | 2022-08-25 | 2024-02-29 | 大连大学 | Method for implementing ultra-sensitive brink control over delayed enzymatic reaction on basis of dna strand displacement |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110544511A (en) * | 2019-09-02 | 2019-12-06 | 郑州轻工业学院 | four-input factorial addition operation molecular circuit design method based on DNA strand displacement |
CN111275161A (en) * | 2020-02-20 | 2020-06-12 | 大连大学 | Competitive neural network framework based on DNA strand displacement |
WO2020131587A2 (en) * | 2018-12-21 | 2020-06-25 | Microsoft Technology Licensing, Llc | Neural networks implemented with dsd circuits |
-
2020
- 2020-11-06 CN CN202011235689.8A patent/CN112348178B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020131587A2 (en) * | 2018-12-21 | 2020-06-25 | Microsoft Technology Licensing, Llc | Neural networks implemented with dsd circuits |
CN110544511A (en) * | 2019-09-02 | 2019-12-06 | 郑州轻工业学院 | four-input factorial addition operation molecular circuit design method based on DNA strand displacement |
CN111275161A (en) * | 2020-02-20 | 2020-06-12 | 大连大学 | Competitive neural network framework based on DNA strand displacement |
Non-Patent Citations (2)
Title |
---|
YANFENG WANG,ET AL.: "Molecular Logic Gates Based on Localized DNA Strand Displacement", 《JOURNAL OF COMPUTATIONAL AND THEORETICAL NANOSCIENCE》, 31 December 2016 (2016-12-31) * |
卢晓建,等: "基于 Seesaw 门的双向联想记忆神经网络", 《广州大学学报( 自然科学版)》, 29 February 2020 (2020-02-29) * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113762513A (en) * | 2021-09-09 | 2021-12-07 | 沈阳航空航天大学 | DNA neuron learning method based on DNA strand displacement |
CN113762513B (en) * | 2021-09-09 | 2023-09-29 | 沈阳航空航天大学 | DNA neuron learning method based on DNA strand displacement |
WO2024041137A1 (en) * | 2022-08-25 | 2024-02-29 | 大连大学 | Method for implementing ultra-sensitive brink control over delayed enzymatic reaction on basis of dna strand displacement |
CN115358389A (en) * | 2022-09-01 | 2022-11-18 | 清华大学 | Neural network training error reduction method, neural network training error reduction device, electronic device, and medium |
Also Published As
Publication number | Publication date |
---|---|
CN112348178B (en) | 2024-03-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112348178B (en) | Artificial neural network calculation model construction method based on DNA strand displacement | |
Zhou et al. | A survey on evolutionary construction of deep neural networks | |
Canals et al. | A new stochastic computing methodology for efficient neural network implementation | |
Zhang et al. | AS-NAS: Adaptive scalable neural architecture search with reinforced evolutionary algorithm for deep learning | |
De Campos et al. | Optimization of neural networks through grammatical evolution and a genetic algorithm | |
Raza | Fuzzy logic based approaches for gene regulatory network inference | |
CN111275161B (en) | Competing neural network framework based on DNA strand displacement | |
CN111027672A (en) | Time sequence prediction method based on interactive multi-scale recurrent neural network | |
Wang et al. | Traffic prediction with missing data: A multi-task learning approach | |
CN110544511B (en) | Four-input factorial addition operation molecular circuit design method based on DNA strand displacement | |
Alzaeemi et al. | Agent-based Modeling in doing Logic Programming in Fuzzy Hopfield Neural Network. | |
Ji et al. | Competitive decomposition-based multiobjective architecture search for the dendritic neural model | |
CN108470212B (en) | Efficient LSTM design method capable of utilizing event duration | |
Przybył et al. | The method of hardware implementation of fuzzy systems on FPGA | |
KR20200020117A (en) | Deep learning apparatus for ANN with pipeline architecture | |
Zahoor et al. | Evolutionary computation technique for solving Riccati differential equation of arbitrary order | |
George et al. | DNA implementation of fuzzy inference engine: towards DNA decision-making systems | |
Thabit et al. | Implementation three-step algorithm based on signed digit number system by using neural network | |
Abdullah et al. | Dual kidney-inspired algorithm for water quality prediction and cancer detection | |
Zhou et al. | Pattern classification and prediction of water quality by neural network with particle swarm optimization | |
Xing Li et al. | Representing diverse mathematical problems using neural networks in hybrid intelligent systems | |
Oh et al. | A new approach to self-organizing multi-layer fuzzy polynomial neural networks based on genetic optimization | |
Li et al. | Broad-Deep Network-Based Fuzzy Emotional Inference Model with Personal Information for Intention Understanding | |
Kaul et al. | Deep Learning-based Advancement in Fuzzy Logic Controller | |
Asgari et al. | Digital multiplier‐less implementation of high‐precision SDSP and synaptic strength‐based STDP |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |