CN110851954B - Adsorption phase change recognition method of polymer chain on attraction surface based on neural network - Google Patents
Adsorption phase change recognition method of polymer chain on attraction surface based on neural network Download PDFInfo
- Publication number
- CN110851954B CN110851954B CN201910941150.5A CN201910941150A CN110851954B CN 110851954 B CN110851954 B CN 110851954B CN 201910941150 A CN201910941150 A CN 201910941150A CN 110851954 B CN110851954 B CN 110851954B
- Authority
- CN
- China
- Prior art keywords
- polymer chain
- state
- adsorption
- sample
- phase change
- 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.)
- Active
Links
- 229920000642 polymer Polymers 0.000 title claims abstract description 171
- 238000001179 sorption measurement Methods 0.000 title claims abstract description 157
- 230000008859 change Effects 0.000 title claims abstract description 76
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 70
- 238000000034 method Methods 0.000 title claims abstract description 65
- 238000012549 training Methods 0.000 claims abstract description 48
- 238000004088 simulation Methods 0.000 claims abstract description 23
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 17
- 238000012360 testing method Methods 0.000 claims abstract description 13
- 238000002922 simulated annealing Methods 0.000 claims abstract description 10
- 238000003795 desorption Methods 0.000 claims description 35
- 239000000178 monomer Substances 0.000 claims description 21
- 238000013527 convolutional neural network Methods 0.000 claims description 14
- 238000002372 labelling Methods 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 14
- 230000000694 effects Effects 0.000 claims description 8
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 6
- 210000002569 neuron Anatomy 0.000 claims description 4
- 238000011176 pooling Methods 0.000 claims description 4
- 238000000137 annealing Methods 0.000 claims description 3
- 230000002779 inactivation Effects 0.000 claims description 3
- 230000000737 periodic effect Effects 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 abstract description 4
- 238000004364 calculation method Methods 0.000 abstract description 3
- 238000004215 lattice model Methods 0.000 abstract 1
- 230000007704 transition Effects 0.000 description 15
- 238000011160 research Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 238000000342 Monte Carlo simulation Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000010297 mechanical methods and process Methods 0.000 description 2
- 102000004169 proteins and genes Human genes 0.000 description 2
- 108090000623 proteins and genes Proteins 0.000 description 2
- 238000010200 validation analysis Methods 0.000 description 2
- 208000007536 Thrombosis Diseases 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000012620 biological material Substances 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 239000011248 coating agent Substances 0.000 description 1
- 238000000576 coating method Methods 0.000 description 1
- 238000005094 computer simulation Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000005476 size effect Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 208000010110 spontaneous platelet aggregation Diseases 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
- G06N3/08—Learning methods
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16C—COMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
- G16C60/00—Computational materials science, i.e. ICT specially adapted for investigating the physical or chemical properties of materials or phenomena associated with their design, synthesis, processing, characterisation or utilisation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Investigating Strength Of Materials By Application Of Mechanical Stress (AREA)
- Investigating Or Analyzing Materials Using Thermal Means (AREA)
Abstract
The adsorption phase change recognition method based on the neural network for the polymer chain on the attraction surface belongs to the field of theoretical calculation and simulation of the polymer, and solves the problem that the number of analysis samples needed by the existing adsorption phase change recognition method for the polymer chain on the attraction surface is too large. The method comprises the following steps: generating a polymer chain sample set based on a three-dimensional simple square lattice model and a self-avoidance walking algorithm, acquiring conformation information and state information of each polymer chain sample at each temperature based on a simulated annealing algorithm, selecting training samples with a preset proportion in the polymer chain sample set, taking the rest polymer chain samples as test samples, training a neural network by adopting each training sample to obtain a polymer chain adsorption phase change initial identification model, correcting the neural network by adopting each test sample to obtain a polymer chain adsorption phase change final identification model and identifying the polymer chain samples to be identified by adopting the polymer chain adsorption phase change final identification model.
Description
Technical Field
The invention relates to an adsorption phase change recognition method of a high molecular chain on an attraction surface, belonging to the field of theoretical calculation and simulation of high molecules.
Background
The adsorption property of a polymer chain on a surface having adsorption property is closely related to factors such as the shape, size, and solubility of the polymer chain. The adsorption process of a polymer chain on a surface having adsorption properties is an important subject in the physical and biological fields, and the related performance of a part of physical devices and the related characteristics of proteins on the surface can be improved by studying the adsorption effect of the polymer chain on the surface. For example, in biology, by controlling the amount of protein adsorption capacity of the material surface, the related properties of biological materials can be improved, which is beneficial for studying processes such as platelet aggregation adsorption and thrombus formation in medicine. For example, in surface science, the problem of coating different materials on a solid plane surface or generating a surface film can be regarded as an adsorption problem of a high molecular polymer, so that the problem can be solved by studying the adsorption property of a high molecular chain on a surface having adsorption property.
The main content of adsorption property research of the polymer chain on the surface with adsorption property is to identify the adsorption phase change of the polymer chain on the attraction surface. The main content of the existing adsorption phase change recognition method of the polymer chain on the attraction surface is as follows: and simulating and generating a high molecular chain by a computer simulation method, performing thermal motion on the high molecular chain, obtaining relevant simulation parameters of the high molecular chain, and analyzing by a statistical mechanical method to obtain the temperature of the adsorption critical point of the high molecular chain on the surface with adsorption property.
However, the existing adsorption phase change recognition method of the polymer chain on the attraction surface needs a large amount of analysis samples due to the adoption of a statistical mechanical method, so that the early simulation period is overlong, and the requirement on related simulation hardware is high.
Disclosure of Invention
The invention provides an adsorption phase change recognition method of a polymer chain on an attraction surface based on a neural network, which aims to solve the problem that the number of analysis samples required by the existing adsorption phase change recognition method of the polymer chain on the attraction surface is excessive.
The adsorption phase change identification method of the polymer chain on the attraction surface based on the neural network comprises the following steps:
simulating and generating a high molecular chain sample set by adopting a self-avoidance walking algorithm based on a three-dimensional simple square grid model;
acquiring conformation information and state information of each polymer chain sample at each temperature based on a simulated annealing algorithm, wherein the state information comprises an adsorption state and a desorption state;
selecting training samples with a preset proportion in a polymer chain sample set, and taking the rest polymer chain samples as test samples;
training the neural network by adopting the conformational information and the state information of each training sample at each temperature to obtain a polymer chain adsorption phase change initial identification model;
correcting the neural network by adopting the conformational information and the state information of each test sample at each temperature to obtain a final recognition model of the adsorption phase change of the polymer chain;
and respectively inputting the conformational information of the plurality of polymer chain samples to be identified into a polymer chain adsorption phase change final identification model to obtain the state information of each polymer chain sample to be identified, and obtaining adsorption phase change point information when the state information of the plurality of polymer chain samples to be identified is different.
Preferably, the specific process of generating the polymer chain sample set based on the three-dimensional simple square grid model and by adopting the self-avoidance walking algorithm simulation is as follows:
the three-dimensional simple square grid model is a square grid model with a dimension L X ×L Y ×L Z The side length of each simple square lattice is 1, and the monomers of the polymer chains are distributed on lattice points;
setting the chain length of a polymer chain sample as N, namely the polymer chain sample consists of N monomers;
in the simulation generation process of the polymer chain sample, two adjacent monomers in the polymer chain sample are bonded by a bond with a fluctuant bond lengthA link having a bond length of 1,
Or->
An impenetrable surface is provided at z=0 and z=d in the simple square lattice simulation box, respectively, wherein D > N v1 V1 is a three-dimensional Flory index, v1=0.588;
the surface at Z=0 has an attraction effect on all monomers of the polymer chain sample, the surface at Z=D has a volume repulsive effect on all monomers of the polymer chain sample, and the surface at Z=0 is a homogeneous surface or a multi-stripe surface;
the high molecular chain sample meets periodic boundary conditions in the X and Y directions;
when the surface at z=0 is a homogeneous surface, the size of the simple square lattice simulation box in the horizontal direction is set as: l (L) X ＝L Y ＞N v2 V2 is a two-dimensional Flory index, v2=0.75;
when the surface at z=0 is a multi-stripe surface, the stripe width L is set to 4, and the dimension of the simple square lattice simulation box in the horizontal direction is set to: l (L) X ＝L Y ＝144。
Preferably, the specific process of obtaining the conformation information and the state information of each polymer chain sample at each temperature based on the simulated annealing algorithm is as follows:
48 temperatures were set for annealing at each of which the polymer chain sample would experience t=2.5×n 2.13 MCS to reach an equilibrium state, wherein MCS is monte carlo steps, each monte carlo step being expressed as an average of one movement per monomer within the system;
the Metropolis importance sampling method is adopted to judge whether each step of movement of the polymer chain is accepted or not:
assuming that each monomer contacts the adsorption surface an energy e= -1 is obtained, a probability p is used to determine if the motion is accepted:
p＝min{1，exp(-ΔE/K B t) } (1) wherein ΔE represents the energy change before and after each movement, K B Is Boltzmann constant, T is temperature;
in the simulated annealing process, a state marking method or a temperature marking method is adopted to mark the polymer chain sample obtained by simulation:
marking each polymer chain sample by a state marking method, sampling every 1000MCS after the polymer chain sample reaches an equilibrium state at each temperature, marking the polymer chain sample as an adsorption state when a monomer always exists in the polymer chain sample in the 1000MCS and contacts with the surface of a Z=0 position, otherwise marking the polymer chain sample as a desorption state;
the temperature labeling method estimates a temperature range corresponding to the adsorption state and the desorption state of the polymer chain sample according to the adsorption rate distribution, marks the polymer chain sample in the temperature range corresponding to the adsorption state as the adsorption state, and marks the polymer chain sample in the temperature range corresponding to the desorption state as the desorption state.
Preferably, the training of the neural network by adopting the conformation information and the state information of each training sample at each temperature comprises the following specific processes of:
the neural network is a convolutional neural network;
and converting coordinate information of training sample conformations into three-dimensional matrix data, inputting the three-dimensional matrix data into a convolutional neural network, extracting features by a convolutional layer, pooling layer generalized features, full-connection layer combined features and temporarily discarding partial neurons and connections by a discarding layer to prevent overfitting, and finally outputting state information of the training samples.
Preferably, the training of the neural network by adopting the conformation information and the state information of each training sample at each temperature comprises the following specific processes of:
the neural network is a fully connected neural network;
and stretching coordinate information of the training sample conformation into one-dimensional data, inputting the one-dimensional data into a fully connected neural network, extracting features through a plurality of hidden layers, and outputting state information of the training sample, wherein the fully connected neural network adopts a random inactivation and regularization mode to prevent overfitting.
Preferably, the ROC curve is used to assist in determining the accuracy of the neural network recognition result.
Preferably, when the surface at z=0 is a homogeneous surface, the state information of the polymer chain sample identified by the final identification model of the polymer chain adsorption phase change includes a homogeneous adsorption state and a homogeneous desorption state, and the identification information output by the final identification model of the polymer chain adsorption phase change also includes adsorption phase change point information of the homogeneous adsorption state and the homogeneous desorption state;
when the surface at the position z=0 is a multi-stripe surface, the state information of the polymer chain sample identified by the final identification model of the polymer chain adsorption phase change includes a single-stripe adsorption state, a multi-stripe adsorption state and a stripe desorption state, and the identification information output by the final identification model of the polymer chain adsorption phase change also includes adsorption phase change point information of the single-stripe adsorption state and the multi-stripe adsorption state and adsorption phase change point information of the multi-stripe adsorption state and the stripe desorption state.
According to the adsorption phase change identification method of the polymer chain on the attraction surface based on the neural network, a polymer chain sample set is generated by simulation of a self-avoidance walking algorithm based on a three-dimensional simple square grid model, and then conformational information and state information of each polymer chain sample at each temperature are obtained based on a simulated annealing algorithm. And then training a neural network by using a training sample to obtain a polymer chain adsorption phase change initial recognition model, correcting the obtained polymer chain adsorption phase change initial recognition model by using a test sample to obtain a polymer chain adsorption phase change final recognition model, and recognizing the adsorption phase change of the polymer chain on the attraction surface by using the polymer chain adsorption phase change final recognition model. The adsorption phase change recognition method of the polymer chain on the attraction surface adopts the neural network to recognize the adsorption phase change of the polymer chain on the attraction surface, and when the neural network is trained, the number of required training samples is far less than the number of required analysis samples of the existing adsorption phase change recognition method of the polymer chain on the attraction surface due to the self characteristics of the neural network. Therefore, compared with the existing adsorption phase change recognition method of the polymer chain on the attraction surface, the adsorption phase change recognition method of the polymer chain on the attraction surface not only shortens the period of early simulation, but also has lower requirement on hardware required for implementing the simulation and is easier to implement.
Drawings
The adsorption phase change recognition method of the polymer chain based on the neural network on the attraction surface according to the present invention will be described in more detail below based on examples and with reference to the accompanying drawings, wherein:
FIG. 1 is a schematic diagram of a convolutional neural network according to an embodiment;
fig. 2 is a schematic structural diagram of a fully-connected neural network according to an embodiment;
FIG. 3 is a graph showing the change in the adsorption rate of the polymer chain with temperature according to the example;
FIG. 4 is a graph of recognition rate versus Epoch for training samples, number of layers of neural network, and number of training samples employed for each temperature, as referred to in the examples;
FIG. 5 is a graph of recognition results of neural network training according to an embodiment;
FIG. 6 is a schematic diagram showing the adsorption rate of polymer chains on the surface of stripes as a function of temperature and a typical tri-state conformation;
FIG. 7 is a graph of recognition results of neural network training according to an embodiment;
fig. 8 is a flowchart of an implementation of the adsorption phase change recognition method of the polymer chain on the attraction surface based on the neural network according to the embodiment.
Detailed Description
The adsorption phase change identification method of the polymer chain based on the neural network on the attraction surface is further described below with reference to the accompanying drawings.
Examples: the present embodiment is described in detail below with reference to fig. 1 to 8.
The adsorption phase change identification method of the polymer chain based on the neural network on the attraction surface comprises the following steps:
step S1, simulating and generating a polymer chain sample set by adopting a self-avoidance walking algorithm based on a three-dimensional simple square grid model;
s2, obtaining conformation information and state information of each polymer chain sample at each temperature based on a simulated annealing algorithm, wherein the state information comprises an adsorption state and a desorption state;
s3, selecting training samples with a preset proportion from a polymer chain sample set, and taking the rest polymer chain samples as test samples;
s4, training the neural network by adopting the conformation information and the state information of each training sample at each temperature to obtain a polymer chain adsorption phase change initial identification model;
s5, correcting the neural network by adopting the conformational information and the state information of each test sample at each temperature to obtain a final recognition model of the high molecular chain adsorption phase change;
and S6, respectively inputting the conformational information of the plurality of polymer chain samples to be identified into a polymer chain adsorption phase change final identification model to obtain the state information of each polymer chain sample to be identified, and obtaining adsorption phase change point information when the state information of the plurality of polymer chain samples to be identified is different.
The monte carlo simulation method is one of the very classical research methods in the field of polymer simulation, and the sample is generated by adopting the monte carlo simulation method in the embodiment. The polymer chain is generated based on self-avoidance walking algorithm (SAW for short), the chain length is N=160, the bond length takes a value of 1,
The simulation space adopts a three-dimensional simple square sub-space, and an impenetrable plane is respectively arranged at Z=0 and Z=D, > N v1 V1 is a three-dimensional Flory index, v1=0.588.
The surface at z=0 will have an adsorption effect on the monomer, and the surface at z=d will only consider the volume repulsion effect on the monomer, in order to keep the polymer chain away from the surface with adsorption effect.
Setting periodic boundary conditions in two directions of X and Y to simulate the horizontal dimension L of the box X ＝L Y ＞N v2 V2 is a two-dimensional Flory index, v2=0.75.
On the stripe surface with stripe width l=4, we uniformly take a larger simulation box size L for the stripes to meet the periodicity condition X ＝L Y ＝144。
With the simulated annealing algorithm, 48 temperatures were set for annealing, at each of which t=2.5×n would be experienced 2.13 MCS reaches equilibrium.
The Metropolis importance sampling method is used for judging whether each step of movement of the polymer chain is accepted, and the energy epsilon-1 is obtained under the assumption that each monomer contacts the adsorption surface, and then the probability p is used for judging whether the movement is accepted or not, wherein,
p＝min{l，exp(-ΔE/K B T)} (1)
where ΔE represents the energy change before and after each movement, K B Is the boltzmann constant, and T is the temperature.
In the embodiment, the adsorption phase change of the polymer chain is researched by adopting a convolutional neural network and a fully-connected neural network respectively, and 9600 samples are extracted at each temperature for training and testing of the neural network. Gradient updates of the neural network employ a cumulative update algorithm and use a running average and regularization to prevent overfitting.
The convolutional neural network model used in this embodiment is shown in fig. 1. Wherein INPUT represents an INPUT layer, connection represents a Convolution layer, MAXPOOL represents a pooling layer, full Connection represents a Full Connection layer, OUTPUT represents an OUTPUT layer, and the PADDING modes are SAME.
The fully connected neural network model used in this embodiment is shown in fig. 2. Wherein Hidden Layer represents a Hidden Layer, softMax Layer represents a classified Layer, dropout represents a random inactivation operation, regularization represents a Regularization operation, and Dim represents a dimension of an input tensor.
In the field of image recognition, a picture is usually converted into a matrix input neural network. In this embodiment, the coordinate information of the polymer chain conformation is converted into the three-dimensional matrix data of 16×10× 3 as the "RGB image", and is input into the convolutional neural network, the overfitting is prevented by extracting the features of the convolutional layer, the generalizing features of the pooling layer, the combining features of the full-connection layer, and temporarily discarding part of the neurons and connections by the discarding (DropOut) layer, and finally the recognition result of the polymer conformation state is output.
In the embodiment, the coordinate information of the polymer chain conformation is stretched into one-dimensional data with the length of 480, the one-dimensional data is input into a fully-connected neural network, the characteristics are extracted through a plurality of hidden layers, and finally the recognition result of the polymer chain conformation state is output. In fully connected neural networks, dropOut also prevents overfitting by randomly dropping some neurons and connections, and Regularization (Regularization) can prevent overfitting by adding a penalty term to the learned weights, such as L2 Regularization.
In this embodiment, ROC (Receiver Operating Characteristic) curves are adopted to assist in judging the identified performance, and when the area enclosed by the ROC curves is closer to 1, the better the performance of the classifier is explained, that is, the area AUC (Area Under Curve) under the ROC is adopted to judge the performance of the classifier, and the AUC is calculated as follows:
wherein f represents the false positive rate, t represents the true positive rate, the approximate f and t are obtained by sorting the learning possibility of the classifier, then continuously reducing the threshold value from 1, and finally the approximate AUC value is obtained by calculating the approximate scale theorem. The accuracy is the ratio of the learned results consistent with the mark, as shown in formula (3):
wherein N is + The number of samples predicted correctly is represented, and N represents the total number of samples.
The embodiment adopts a state marking method and a state marking method respectivelyThe temperature labeling method is used for labeling the analog obtained sample. The state labeling method labels each sample, and the polymer chain runs at each temperature with t=2.5×n 2.13 After MCS, sampling is performed every 1000MCS, if there is a monomer contact surface all the time in the 1000MCS, we mark as adsorption state, otherwise, it marks as desorption state. A similar method was used for the polymer chain sample state on the surface of the stripes. The temperature marking method is to estimate the general temperature range of the polymer chain state according to the adsorption rate distribution, and mark the samples in the temperature range as the same state. We selected a specific proportion of samples in the dataset for training, the remaining samples for testing and validation.
In the embodiment, the problem of adsorption phase change of the polymer chains on the homogeneous surface is studied by machine learning firstly:
the polymer chain has a Desorption State (DE State) at high temperature, an Adsorption State (AD State) at low temperature and a critical phase transition point between the two states on the homogeneous surface. Fig. 3 is a graph showing the change in adsorption rate of a polymer chain with temperature, wherein the chain length n=160, the polymer chain adsorption state conformation at the temperature t=1.0 is shown in fig. 3 (a), and the polymer chain desorption state conformation at the temperature t=2.0 is shown in fig. 3 (b).
As can be seen from fig. 3, the desorption state is obtained when the adsorption rate is equal to 0 (i.e., there is no monomer contact surface), and the typical conformation thereof is shown in fig. 3 (b). Adsorption occurs when the adsorption rate is non-zero, called the adsorption state, and its typical conformation is shown in fig. 3 (a). The critical phase transition point is about t=1.6, so for the temperature labeling method, the sample of T e [1.1,1.4] is selected as the adsorption state label, and the sample of T e [1.8,3.0] is selected as the desorption state label. The marked samples are then trained and identified using convolutional neural networks and fully-connected neural networks, respectively, where the fully-connected neural networks employ different numbers of hidden layers for the experiment, the results of which are shown in fig. 4.
FIG. 4 illustrates recognition rate of neural network and Epoch of training sample, number of layers of fully connected neural network, anda plot of the number of training samples taken for each temperature. Fig. 4 (a) is a graph showing the relationship between recognition rate and Epoch. SPT (Sample Per Temperature) the number of samples taken at each temperature for training the neural network, the samples being labeled using a state labeling method. Wherein n is h =1 means that the number of hidden layers is 1, and so on. n is n h =1 to n h Each of =3 uses the sample of spt=192 for training and the remaining samples for testing and validation. The inset in fig. 4 (a) depicts the recognition rate and the number of hidden layers n h The recognition rate is the final stable recognition result of each learner. Fig. 4 (b) is a graph showing the relationship between the recognition rate and the number of training samples used for each temperature. The samples are marked by a state marking method, and the hidden layer number is equal to 3. The ordinate is the recognition rate of different training samples when they reach stability under enough epochs, and the test sets are spt=7680 and are not repeated with the training sets.
As can be seen from fig. 4 (a), as the number of hidden layers increases, the recognition rate increases and then tends to stabilize, and when the number of hidden layers is 3 or more, the recognition rate is substantially stabilized at 97.1%. Therefore, the number of hidden layers adopted by default in this embodiment is 3. Training was essentially stable when Epoch > 30.
As can be seen from fig. 4 (b), even though the number of samples (Sample Per Temperature, labeled SPT) extracted for training the neural network at each temperature is small enough, for example, the neural network still has a 91.88% recognition rate when spt=1, it is sufficient to recognize most of the samples. When the SPT is more than or equal to 24, the recognition rate reaches more than 95.5%, which shows that the embodiment can reach higher recognition rate of the polymer conformation by adopting smaller sample number. Then, the present embodiment uses a convolutional neural network for research, and the recognition result thereof is shown in fig. 5.
Fig. 5 is a graph of recognition results of neural network training. The abscissa is temperature, state represents the probability that a sample at each temperature is recognized as a certain State, S represents a State labeling method, T represents a temperature labeling method, AD represents an adsorption State, and DE represents a desorption State. FIG. 5 shows the results of two labeling methods, with a convolutional network identification rate of 98.3%, an AUC value of 0.9989, a fully connected network of 97.6%, and an AUC value of 09982 critical phase transition temperature T of two marking methods C ＝1.5。
According to fig. 5, the convolutional neural network and the fully-connected neural network both obtain a higher recognition rate and a larger AUC value, so that the neural network can better recognize two states of the polymer chains on the homogeneous surface, and the recognition rate of the convolutional neural network is slightly higher. Both methods can be used to determine the adsorption phase transition point, and the obtained critical phase transition temperature is the same and is T C =1.5, slightly less than the critical phase transition temperature T of infinite chain length C This difference is present because there is a finite size effect, the polymer chain length of this example is n=160, and the critical adsorption temperature tends to that of infinite chain length as the chain length increases.
Next, this example investigated the problem of polymer chain adsorption phase change on the surface of stripes using machine learning:
FIG. 6 is a schematic diagram showing the change of the adsorption rate of a polymer chain on the surface of a stripe with temperature and a typical tri-state conformation. Insert 6 (a) is a single stripe adsorption state, temperature t=0.3; insert 6 (b) is a multi-stripe adsorption state, t=0.9; insert 6 (c) is in the desorbed state, t=3.0. Where chain length n=160, stripe width l=4. The stripe direction is perpendicular to the X-axis and extends along the Y-axis, and the selected spatial dimension is 25×120×20. On the surface of the stripe, dark portions are adsorption stripes, and white portions are rejection stripes.
The conformation of the polymer chain on the surface of the stripes relates to three states, namely a Single stripe adsorption State (Single-Stripe Adsorption State), a multi-stripe adsorption State (Muti-Stripe Adsorption State) and a Desorption State (Desorption State), so that the transition between the three states is accompanied by two critical phase transition points.
As can be seen from fig. 6, the adsorption rate of the polymer chain is almost 0 at high temperature, i.e., the desorption state, which is consistent with the desorption state of the homogeneous surface; the adsorption rate of the polymer chain is very high at low temperature, and the polymer chain is adsorbed by the single stripe, and this adsorption state is called a single stripe adsorption state, as shown in fig. 6 (a); the polymer chain is adsorbed on the multi-stripe at the intermediate temperature, the adsorption rate is also between the desorption state and the single-stripe adsorption, and the adsorption state is called as the multi-stripe adsorption state. In the multi-stripe adsorption state, the polymer chains are distributed on different adsorption stripes as shown in fig. 6 (b). For the temperature labeling method, a sample of T epsilon [1.35,1.5] is selected as a desorption state label, a sample of T epsilon [0.75,0.9] is selected as a multi-stripe adsorption state label, and a sample of T epsilon [0.25,0.4] is selected as a single-stripe adsorption state label.
Then, the polymer chain conformation samples on the surface of the stripes were trained and identified in this example, and the results are shown in fig. 7. Fig. 7 is a graph of recognition results of neural network training, in which the abscissa represents temperature, the ordinate represents the probability that a sample at each temperature is recognized as a certain State, S represents a State labeling method, T represents a temperature labeling method, SS represents a single stripe adsorption State, MS represents a multi-stripe adsorption State, and DE represents a desorption State. Wherein the recognition rate of the convolution network is 94.78%, the AUC value is 0.9930, the fully-connected network is 93.85%, the AUC value is 0.9918, and the critical phase transition temperature T of the state marking method is as follows 1 ＝0.55，T 2 Critical phase transition temperature T of =1.1 temperature labeling method 1 ＝0.55，T 2 ＝1.05。
As can be seen from fig. 7, the convolutional neural network and the fully-connected network have higher recognition rates of the polymer conformational states, and the AUC value is very close to 1, which indicates that the neural network can recognize three states of the polymer chains on the surface of the stripes, and the recognition rate of the convolutional neural network is slightly higher than that of the fully-connected network. The critical phase transition temperature obtained by the two sample marking methods is basically the same, wherein the phase transition point from multi-stripe adsorption to single-stripe adsorption is T 1 =0.55, the phase transition point of the desorption state to the polymer multi-stripe adsorption is T 2 ＝1.1。
The adsorption phase change recognition method of the polymer chain on the attraction surface based on the neural network of the embodiment adopts Monte Carlo simulation and the neural network to study the state of the polymer chain on the attraction surface and calculate the adsorption phase change. The research shows that the neural network can identify the desorption and adsorption states of the polymer chains on the homogeneous surface, so that the adsorption phase change of the polymer chains can be determined. A relatively high polymer can be obtained even if a small number of samples are selected at each temperatureChain state recognition rate. The temperature marking method and the state marking method are adopted to mark the polymer chain conformation sample, and the research discovers that the critical phase change points obtained by the two methods are basically the same, and the adsorption phase change point T of the polymer chain on the homogeneous surface C =1.5, the multi-stripe to single stripe phase transition point T at the stripe surface 1 =0.55 and the phase transition point between the desorption state and the adsorption state is T 2 =1.1. In a word, the trained neural network has higher recognition capability on the conformational state of the polymer chain, and the artificial neural network provides a new approach for the simulation calculation research of the polymer physics.
Although the invention herein has been described with reference to particular embodiments, it is to be understood that these embodiments are merely illustrative of the principles and applications of the present invention. It is therefore to be understood that numerous modifications may be made to the illustrative embodiments and that other arrangements may be devised without departing from the spirit and scope of the present invention as defined by the appended claims. It should be understood that the different dependent claims and the features described herein may be combined in ways other than as described in the original claims. It is also to be understood that features described in connection with separate embodiments may be used in other described embodiments.
Claims (5)
1. The adsorption phase change identification method of the polymer chain on the attraction surface based on the neural network is characterized by comprising the following steps of:
simulating and generating a high molecular chain sample set by adopting a self-avoidance walking algorithm based on a three-dimensional simple square grid model;
acquiring conformation information and state information of each polymer chain sample at each temperature based on a simulated annealing algorithm, wherein the state information comprises an adsorption state and a desorption state;
selecting training samples with a preset proportion in a polymer chain sample set, and taking the rest polymer chain samples as test samples;
training the neural network by adopting the conformational information and the state information of each training sample at each temperature to obtain a polymer chain adsorption phase change initial identification model;
correcting the neural network by adopting the conformational information and the state information of each test sample at each temperature to obtain a final recognition model of the adsorption phase change of the polymer chain;
respectively inputting the conformational information of a plurality of polymer chain samples to be identified into a polymer chain adsorption phase change final identification model to obtain the state information of each polymer chain sample to be identified, and obtaining adsorption phase change point information when the state information of the plurality of polymer chain samples to be identified is different;
the specific process for simulating and generating the high polymer chain sample set based on the three-dimensional simple square grid model and by adopting the self-avoidance walking algorithm is as follows:
the three-dimensional simple square grid model is a square grid model with a dimension L X ×L Y ×L Z The side length of each simple square lattice is 1, and the monomers of the polymer chains are distributed on lattice points;
setting the chain length of a polymer chain sample as N, namely the polymer chain sample consists of N monomers;
in the simulation generation process of the polymer chain sample, two adjacent monomers in the polymer chain sample are linked through a bond with a bond length capable of fluctuating, and the bond length value is 1,
Or->
An impenetrable surface is provided at z=0 and z=d in the simple square lattice simulation box, respectively, wherein D is>N v1 V1 is a three-dimensional Flory index, v1=0.588;
the surface at Z=0 has an attraction effect on all monomers of the polymer chain sample, the surface at Z=D has a volume repulsive effect on all monomers of the polymer chain sample, and the surface at Z=0 is a homogeneous surface or a multi-stripe surface;
the high molecular chain sample meets periodic boundary conditions in the X and Y directions;
when the surface at z=0 is a homogeneous surface, the size of the simple square lattice simulation box in the horizontal direction is set as: l (L) X ＝L Y >N v2 V2 is a two-dimensional Flory index, v2=0.75;
when the surface at z=0 is a multi-stripe surface, the stripe width L is set to 4, and the dimension of the simple square lattice simulation box in the horizontal direction is set to: l (L) X ＝L Y ＝144；
The specific process for acquiring the conformational information and the state information of each polymer chain sample at each temperature based on the simulated annealing algorithm comprises the following steps:
48 temperatures were set for annealing at each of which the polymer chain sample would experience t=2.5×n 2.13 MCS to reach an equilibrium state, wherein MCS is monte carlo steps, each monte carlo step being expressed as an average of one movement per monomer within the system;
the Metropolis importance sampling method is adopted to judge whether each step of movement of the polymer chain is accepted or not:
assuming that each monomer contacts the adsorption surface an energy e= -1 is obtained, a probability p is used to determine if the motion is accepted:
p＝min(1，exp(-ΔE/K B T)} (1)
wherein ΔE represents the energy change before and after each movement, K B Is Boltzmann constant, T is temperature;
in the simulated annealing process, a state marking method or a temperature marking method is adopted to mark the polymer chain sample obtained by simulation:
marking each polymer chain sample by a state marking method, sampling every 1000MCS after the polymer chain sample reaches an equilibrium state at each temperature, marking the polymer chain sample as an adsorption state when a monomer always exists in the polymer chain sample in the 1000MCS and contacts with the surface of a Z=0 position, otherwise marking the polymer chain sample as a desorption state;
the temperature labeling method estimates a temperature range corresponding to the adsorption state and the desorption state of the polymer chain sample according to the adsorption rate distribution, marks the polymer chain sample in the temperature range corresponding to the adsorption state as the adsorption state, and marks the polymer chain sample in the temperature range corresponding to the desorption state as the desorption state.
2. The method for identifying the adsorption phase change of the polymer chain on the attraction surface based on the neural network according to claim 1, wherein the specific process of training the neural network by adopting the conformation information and the state information of each training sample at each temperature to obtain the initial identification model of the adsorption phase change of the polymer chain is as follows:
the neural network is a convolutional neural network;
and converting coordinate information of training sample conformations into three-dimensional matrix data, inputting the three-dimensional matrix data into a convolutional neural network, extracting features by a convolutional layer, pooling layer generalized features, full-connection layer combined features and temporarily discarding partial neurons and connections by a discarding layer to prevent overfitting, and finally outputting state information of the training samples.
3. The method for identifying the adsorption phase change of the polymer chain on the attraction surface based on the neural network according to claim 1, wherein the specific process of training the neural network by adopting the conformation information and the state information of each training sample at each temperature to obtain the initial identification model of the adsorption phase change of the polymer chain is as follows:
the neural network is a fully connected neural network;
and stretching coordinate information of the training sample conformation into one-dimensional data, inputting the one-dimensional data into a fully connected neural network, extracting features through a plurality of hidden layers, and outputting state information of the training sample, wherein the fully connected neural network adopts a random inactivation and regularization mode to prevent overfitting.
4. The method for identifying the adsorption phase change of the polymer chain on the attraction surface based on the neural network according to claim 2 or 3, wherein an ROC curve is adopted to assist in judging the accuracy of the identification result of the neural network.
5. The method for identifying the adsorption phase change of the polymer chain on the attraction surface based on the neural network according to claim 4, wherein when the surface at the position z=0 is a homogeneous surface, the state information of the polymer chain sample identified by the final identification model of the polymer chain adsorption phase change comprises a homogeneous adsorption state and a homogeneous desorption state, and the identification information output by the final identification model of the polymer chain adsorption phase change further comprises adsorption phase change point information of the homogeneous adsorption state and the homogeneous desorption state;
when the surface at the position z=0 is a multi-stripe surface, the state information of the polymer chain sample identified by the final identification model of the polymer chain adsorption phase change includes a single-stripe adsorption state, a multi-stripe adsorption state and a stripe desorption state, and the identification information output by the final identification model of the polymer chain adsorption phase change also includes adsorption phase change point information of the single-stripe adsorption state and the multi-stripe adsorption state and adsorption phase change point information of the multi-stripe adsorption state and the stripe desorption state.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910941150.5A CN110851954B (en) | 2019-09-30 | 2019-09-30 | Adsorption phase change recognition method of polymer chain on attraction surface based on neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910941150.5A CN110851954B (en) | 2019-09-30 | 2019-09-30 | Adsorption phase change recognition method of polymer chain on attraction surface based on neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110851954A CN110851954A (en) | 2020-02-28 |
| CN110851954B true CN110851954B (en) | 2023-07-11 |
Family
ID=69597525
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910941150.5A Active CN110851954B (en) | 2019-09-30 | 2019-09-30 | Adsorption phase change recognition method of polymer chain on attraction surface based on neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110851954B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111582436B (en) * | 2020-04-02 | 2024-04-19 | 深圳创维－Rgb电子有限公司 | Adsorption force adjusting method, device, equipment and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104700153A (en) * | 2014-12-05 | 2015-06-10 | 江南大学 | PH (potential of hydrogen) value predicting method of BP (back propagation) neutral network based on simulated annealing optimization |
| CN108804869A (en) * | 2018-05-04 | 2018-11-13 | 深圳晶泰科技有限公司 | Molecular structure based on neural network and chemical reaction energy function construction method |
| CN108984852A (en) * | 2018-06-22 | 2018-12-11 | 温州大学 | Monte-Carlo Simulation Method of the macromolecular chain in adsorption property under external force |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130071837A1 (en) * | 2004-10-06 | 2013-03-21 | Stephen N. Winters-Hilt | Method and System for Characterizing or Identifying Molecules and Molecular Mixtures |
-
2019
- 2019-09-30 CN CN201910941150.5A patent/CN110851954B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104700153A (en) * | 2014-12-05 | 2015-06-10 | 江南大学 | PH (potential of hydrogen) value predicting method of BP (back propagation) neutral network based on simulated annealing optimization |
| CN108804869A (en) * | 2018-05-04 | 2018-11-13 | 深圳晶泰科技有限公司 | Molecular structure based on neural network and chemical reaction energy function construction method |
| CN108984852A (en) * | 2018-06-22 | 2018-12-11 | 温州大学 | Monte-Carlo Simulation Method of the macromolecular chain in adsorption property under external force |
Non-Patent Citations (5)
| Title |
|---|
| hong li ; bin gong ; C. Qian ; He-Bei Gao.Parallel Computing of Polymer Chains Based on Monte Carlo Method.《Applied Mechanics and Materials》.2013,第263-266卷3317-3320. * |
| 张全昌 ; 钱昌吉 ; 孙立望 ; 施双飞 ; 李洪 ; .高分子多链在吸引表面上的吸附特性.温州大学学报(自然科学版).2019,40(02),45-53. * |
| 李洪 ; 艾倩雯 ; 汪鹏君 ; 高和蓓 ; 崔毅 ; 罗孟波 ; .外力驱动作用下高分子链在表面吸附性质的计算机模拟.物理学报.2018,67(16),1-7. * |
| 窦鹏伟 ; 王珍 ; 佘侃侃 ; 樊文玲 ; .基于改进BP神经网络的中药水提液陶瓷膜污染预测研究.中国中医药信息杂志.2017,24(04),92-96. * |
| 谷有众 ; 刘建立 ; 高卫东 ; .基于遗传模拟退火神经网络的纱线强力预测.上海纺织科技.2016,44(01),45-49. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110851954A (en) | 2020-02-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Qian et al. | Introducing self-organized maps (SOM) as a visualization tool for materials research and education | |
| CN105809672B (en) | A kind of image multiple target collaboration dividing method constrained based on super-pixel and structuring | |
| CN107943856A (en) | A kind of file classification method and system based on expansion marker samples | |
| CN110399805A (en) | The Mental imagery Method of EEG signals classification of semi-supervised learning optimization SVM | |
| CN110851954B (en) | Adsorption phase change recognition method of polymer chain on attraction surface based on neural network | |
| CN105354800B (en) | The non-convex compressed sensing image reconstructing method of particle group optimizing based on picture structure | |
| Jaderberg et al. | Quantum self-supervised learning | |
| CN106296734A (en) | Based on extreme learning machine and the target tracking algorism of boosting Multiple Kernel Learning | |
| Easom-McCaldin et al. | Efficient quantum image classification using single qubit encoding | |
| Sturm et al. | Predicting impurity spectral functions using machine learning | |
| Zhang et al. | An efficient modified boosting method for solving classification problems | |
| CN112597921A (en) | Human behavior recognition method based on attention mechanism GRU deep learning | |
| CN109063418A (en) | Determination method, apparatus, equipment and the readable storage medium storing program for executing of disease forecasting classifier | |
| Chen et al. | A weighted bagging LightGBM model for potential lncRNA-disease association identification | |
| CN104573726B (en) | Facial image recognition method based on the quartering and each ingredient reconstructed error optimum combination | |
| CN110110785A (en) | A kind of express mail logistics progress state-detection classification method | |
| CN104376585B (en) | Non-convex compressed sensing image reconstructing method based on picture block structure property policy | |
| Cioroianu et al. | An improved non-affine Arruda-Boyce type constitutive model for collagen networks | |
| Kojima et al. | Search strategy for rare microstructure to optimize material properties of filled rubber using machine learning based simulation | |
| CN111832748A (en) | Electronic nose width learning method for performing regression prediction on concentration of mixed gas | |
| Lion et al. | Life-history evolution of class-structured populations in fluctuating environments | |
| Zhang et al. | Bayesian compositional generalized linear models for analyzing microbiome data | |
| Saha et al. | Novel randomized feature selection algorithms | |
| de Villiers et al. | Autoencoder-based 3D representation learning for industrial seedling abnormality detection | |
| Laosiritaworn et al. | Artificial Neural Network modeling of spin-transition behavior in two-dimensional molecular magnet: The learning by experiences analysis |
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 | ||
| CB03 | Change of inventor or designer information |
Inventor after: Gao Hebei Inventor after: Li Hong Inventor after: Sun Liwang Inventor after: Wang Pengjun Inventor after: Luo Mengbo Inventor before: Li Hong Inventor before: Sun Liwang Inventor before: Gao Hebei Inventor before: Wang Pengjun Inventor before: Luo Mengbo |
|
| CB03 | Change of inventor or designer information |