CN115345797A - Chromosome conformation contact map enhancement method and device based on generation of confrontation network - Google Patents
Chromosome conformation contact map enhancement method and device based on generation of confrontation network Download PDFInfo
- Publication number
- CN115345797A CN115345797A CN202211007292.2A CN202211007292A CN115345797A CN 115345797 A CN115345797 A CN 115345797A CN 202211007292 A CN202211007292 A CN 202211007292A CN 115345797 A CN115345797 A CN 115345797A
- Authority
- CN
- China
- Prior art keywords
- chromosome conformation
- sample
- enhanced
- chromosome
- conformation contact
- 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.)
- Pending
Links
- 210000000349 chromosome Anatomy 0.000 title claims abstract description 282
- 238000000034 method Methods 0.000 title claims abstract description 52
- 238000012549 training Methods 0.000 claims abstract description 78
- 230000006870 function Effects 0.000 claims description 32
- 230000002708 enhancing effect Effects 0.000 claims description 23
- 238000012545 processing Methods 0.000 claims description 11
- 210000002230 centromere Anatomy 0.000 claims description 8
- 238000000638 solvent extraction Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 4
- 238000012216 screening Methods 0.000 claims description 4
- 230000003042 antagnostic effect Effects 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 230000017105 transposition Effects 0.000 claims description 3
- 238000012163 sequencing technique Methods 0.000 abstract description 16
- 238000013527 convolutional neural network Methods 0.000 description 11
- 230000004913 activation Effects 0.000 description 10
- 238000010586 diagram Methods 0.000 description 10
- 230000000694 effects Effects 0.000 description 10
- 230000008569 process Effects 0.000 description 9
- 210000004027 cell Anatomy 0.000 description 6
- 238000002474 experimental method Methods 0.000 description 6
- 108010077544 Chromatin Proteins 0.000 description 5
- 210000003483 chromatin Anatomy 0.000 description 5
- 238000010606 normalization Methods 0.000 description 5
- 238000004590 computer program Methods 0.000 description 4
- 230000003993 interaction Effects 0.000 description 4
- 108090000623 proteins and genes Proteins 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000003766 bioinformatics method Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 238000013256 Gubra-Amylin NASH model Methods 0.000 description 2
- 208000026350 Inborn Genetic disease Diseases 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000007429 general method Methods 0.000 description 2
- 208000016361 genetic disease Diseases 0.000 description 2
- 210000003917 human chromosome Anatomy 0.000 description 2
- 230000003902 lesion Effects 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000002265 prevention Effects 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 230000009466 transformation Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 238000009795 derivation Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
- G06T3/4046—Scaling the whole image or part thereof using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformation in the plane of the image
- G06T3/40—Scaling the whole image or part thereof
- G06T3/4053—Super resolution, i.e. output image resolution higher than sensor resolution
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10056—Microscopic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
Abstract
The invention relates to the technical field of chromosome sequencing enhancement, in particular to a chromosome conformation contact map enhancement method and a chromosome conformation contact map enhancement device based on a generated countermeasure network. The invention fully considers the relevance between the base distance and the chromosome conformation contact image, and inputs the base distance as condition information into the generation countermeasure network, so that the generation countermeasure network can treat the chromosome conformation contact images with different base distances in a distinguishing way during training, and further the trained model can carry out targeted resolution enhancement according to the base distance of each chromosome conformation contact image, thereby improving the resolution of the whole chromosome conformation contact image.
Description
Technical Field
The invention relates to the technical field of chromosome sequencing enhancement, in particular to a chromosome conformation contact map enhancement method and a chromosome conformation contact map enhancement device based on a generated countermeasure network.
Background
The human chromosome has a complex multi-level three-dimensional space structure, and the three-dimensional structure of the chromosome is accurately analyzed, so that the mechanism of gene regulation and control is disclosed, and the prevention and treatment of genetic diseases are realized. The High-throughput Chromosome transformation Capture (High-C) technology is a main means for obtaining the three-dimensional structure of chromosomes at present. The general method is to construct a chromosome conformation contact map according to Hi-C sequencing data, and further analyze and count the chromosome conformation contact map to draw a conclusion about whether the chromosome has a lesion. The high-resolution Hi-C sequencing data needs to consume huge labor, material and time costs, but the chromosome conformation contact map constructed according to the high-resolution Hi-C sequencing data can accurately reveal the three-dimensional structure of the chromosome (namely the high-resolution chromosome conformation contact map can accurately reveal the three-dimensional structure of the chromosome); the low-resolution Hi-C sequencing data has the advantages of high speed and low cost, but the chromosome conformation contact map constructed according to the low-resolution Hi-C sequencing data cannot accurately reveal the three-dimensional structure of the chromosome, and sometimes even can lead to wrong conclusion. However, the resolution of the most currently available Hi-C data is 25 to 100Mb, and therefore, it is necessary to develop some methods to improve the resolution of the Hi-C contact map (chromosome conformation contact map). The existing Hi-C contact map enhancement method is a Convolutional Neural Network (CNN) based Hi-C contact map enhancement method. Contact map enhancement methods based on convolutional neural networks typically use mean square error as a loss function, and the enhanced contact maps have the problems of over-smoothing, edge blurring, and exhibit strong distance dependence. For example, in 2018, a HiC contact map enhancement method-HiCPlus, which is implemented based on a convolutional neural network, the edges of the structures such as the topological domains, the chromatin loops and the like in the enhanced contact map are quite blurred, which is quite disadvantageous for the downstream bioinformatics analysis. Because the base distance is not considered in the chromosome conformation contact map enhancement method based on the convolutional neural network, the quality of the enhanced chromosome conformation contact map is poor.
In summary, the prior art enhancement methods result in poor quality of the chromosome conformation contact maps.
Thus, there is a need for improvements and enhancements in the art.
Disclosure of Invention
In order to solve the technical problems, the invention provides a chromosome conformation contact map enhancing method and a device based on a generation countermeasure network, and solves the problem that the enhancing method in the prior art causes poor quality of the chromosome conformation contact map.
In order to achieve the purpose, the invention adopts the following technical scheme:
in a first aspect, the present invention provides a chromosome conformation contact map enhancement method based on generation of a confrontation network, wherein the method comprises:
processing the original chromosome conformation contact map according to the position of each region of the original chromosome conformation contact map in the chromosome to obtain a sample chromosome conformation contact map;
training a generated confrontation network according to the sample chromosome conformation contact image and the base distance of the sample chromosome conformation contact image to obtain the trained generated confrontation network;
inputting the chromosome conformation contact map to be enhanced into the trained generation countermeasure network to obtain the enhanced chromosome conformation contact map output by the trained generation countermeasure network.
In one implementation, the processing the original chromosome conformation contact map according to the positions of the regions of the original chromosome conformation contact map in the chromosome to obtain the sample chromosome conformation contact map comprises:
normalizing the pixel values of the original chromosome conformation contact image to obtain the normalized original chromosome conformation contact image;
partitioning the normalized original chromosome conformation contact image according to a set size to obtain each sample subgraph for representing each region of the original chromosome conformation contact image;
removing sample subgraphs below the diagonal in each sample subgraph and the sample subgraphs at the positions of the chromosome centromere to obtain preselected sample subgraphs;
and screening the sample subgraphs after preselection according to the base distance of the sample subgraphs after preselection to obtain a sample chromosome conformation contact graph.
In one implementation, the screening each of the sample subgraphs after preselection according to the base distance of each of the sample subgraphs after preselection to obtain a sample chromosome conformation contact map comprises:
and removing sample subgraphs with the base distance being more than or equal to a set distance in each sample subgraph after preselection to obtain a sample chromosome conformation contact graph.
In one implementation, the training a generative confrontation network according to the sample chromosome conformation contact map and the base distance of the sample chromosome conformation contact map, and obtaining the trained generative confrontation network, includes:
alternately training a generator to be trained and a discriminator according to the sample chromosome conformation contact pattern and the base distance of the sample chromosome conformation contact pattern to obtain the trained generation confrontation network, wherein the generator and the discriminator form the generation confrontation network.
In one implementation, the alternately training a generator and a discriminator to be trained according to the base distance of the sample chromosome conformation contact map and the sample chromosome conformation contact map to obtain the trained generation countermeasure network, wherein the generator and the discriminator form the generation countermeasure network, and the method includes:
inputting the sample chromosome conformation contact image and the base distance of the sample chromosome conformation contact image into the generator to be trained to obtain an enhanced resolution sample image output by the generator to be trained;
inputting an enhanced resolution sample graph and a real set resolution graph output by the generator to be trained into the discriminator to be trained to obtain a generation loss function output by the discriminator to be trained, wherein the generation loss function is used for representing the difference degree between the enhanced resolution sample graph and the real set resolution graph;
adjusting parameters of the generator to be trained according to the generation loss function to complete a round of training of the generator;
inputting the sample chromosome conformation contact image and the base distance of the sample chromosome conformation contact image into the generator after one round of training to obtain an enhanced resolution sample image output by the generator after one round of training;
inputting the enhanced resolution sample graph and the real set resolution graph output by the generator after one round of training into the discriminator to be trained to obtain an output value of the discriminator to be trained;
calculating a real loss value of the enhanced resolution sample graph output by the generator after one round of training relative to the real set resolution graph;
calculating a loss value of the output value of the discriminator to be trained relative to the output value of the expected discriminator, and recording the loss value as a discrimination loss function;
adjusting parameters of the discriminator to be trained according to the discrimination loss function to obtain a round of training of the discriminator;
and continuing to alternately train the generator and the discriminator according to the generator after one round of training and the discriminator after one round of training until the iteration times reach the set times, and finishing the training of the generated countermeasure network.
In one implementation, the inputting the chromosome conformation contact map to be enhanced to the trained generating confrontation network to obtain the enhanced chromosome conformation contact map output by the trained generating confrontation network includes:
partitioning the chromosome conformation contact map to be enhanced according to a set size to obtain each subgraph to be enhanced;
selecting the subgraph to be enhanced at the diagonal position of the chromosome from each subgraph to be enhanced, and marking as a first subgraph;
selecting the subgraph to be enhanced positioned below the diagonal line of the chromosome from each subgraph to be enhanced and marking as a second subgraph;
selecting the subgraph to be enhanced above the diagonal line of the chromosome from each subgraph to be enhanced, and marking as a third subgraph;
inputting the first subgraph and the third subgraph into a generator in the trained generation countermeasure network to obtain the enhanced first subgraph and the enhanced third subgraph in an enhanced chromosome conformation contact graph output by the generator;
transposing a matrix formed by each pixel value in the second sub-image to obtain the transposed second sub-image;
and inputting the transposed second sub-image into the trained generator to obtain the enhanced second sub-image output by the generator.
In one implementation, the inputting the chromosome conformation contact map to be enhanced to the trained generative confrontation network to obtain the enhanced chromosome conformation contact map output by the trained generative confrontation network, and then further includes:
transposing the enhanced second sub-graph to obtain the transposed enhanced second sub-graph;
and splicing the enhanced first sub-image, the enhanced second sub-image after transposition and the enhanced third sub-image according to the positions of the first sub-image, the second sub-image and the third sub-image in the chromosome conformation contact image to be enhanced to obtain the complete enhanced chromosome conformation contact image.
In a second aspect, an embodiment of the present invention further provides a chromosome conformation contact map enhancing apparatus based on generation of an antagonistic network, where the apparatus includes the following components:
the preprocessing module is used for processing the original chromosome conformation contact map according to the positions of all the areas of the original chromosome conformation contact map in the chromosome to obtain a sample chromosome conformation contact map;
the training module is used for training the generated confrontation network according to the sample chromosome conformation contact image and the base distance of the sample chromosome conformation contact image to obtain the trained generated confrontation network;
and the resolution enhancement module is used for inputting the chromosome conformation contact map to be enhanced into the trained generation countermeasure network to obtain the enhanced chromosome conformation contact map output by the generator in the trained generation countermeasure network.
In a third aspect, an embodiment of the present invention further provides a terminal device, where the terminal device includes a memory, a processor, and a generated countermeasure network-based chromosome conformation contact map enhancement program that is stored in the memory and is executable on the processor, and when the processor executes the generated countermeasure network-based chromosome conformation contact map enhancement program, the processor implements the steps of the generated countermeasure network-based chromosome conformation contact map enhancement method.
In a fourth aspect, the embodiment of the present invention further provides a computer-readable storage medium, where a chromosome conformation contact map enhancement program based on the generation countermeasure network is stored, and when the chromosome conformation contact map enhancement program based on the generation countermeasure network is executed by a processor, the steps of the chromosome conformation contact map enhancement method based on the generation countermeasure network are implemented.
Has the beneficial effects that: the invention fully considers the relevance between the base distance and the chromosome conformation contact image when training to generate the confrontation network, and when training to generate the confrontation network, the base distance is used as condition information to be input into the generated confrontation network, so that the generated confrontation network can treat the chromosome conformation contact images with different base distances in a distinguishing way during training, and further the generated confrontation network after training can be pertinently enhanced according to the base distance of each chromosome conformation contact image when being used for enhancing the resolution of the chromosome conformation contact image, thereby improving the quality of the contact image after the resolution is enhanced by the contact images with various base distances.
Drawings
FIG. 1 is an overall flow chart of the present invention;
FIG. 2 is a contact diagram of chromosome 1 conformation in an example of the present invention;
FIG. 3 is a block diagram of a generator in an embodiment of the invention;
FIG. 4 is a block diagram of an authenticator in an embodiment of the present invention;
FIG. 5 is a flow chart of calculating a discriminator loss function in an embodiment of the invention;
FIG. 6 is a diagram illustrating a mean square error according to an embodiment of the present invention;
FIG. 7 is a diagram illustrating a peak SNR in an embodiment of the present invention;
FIG. 8 is a schematic diagram of structural similarity in an embodiment of the present invention;
FIG. 9 is a low resolution contact map in an embodiment of the present invention;
FIG. 10 is a true high resolution touch map in an embodiment of the invention;
FIG. 11 is a contact graph after generation of enhanced resolution against network output in an embodiment of the present invention;
FIG. 12 is a graph comparing the effect of MSE at different base distances for various models in accordance with embodiments of the present invention;
FIG. 13 is a graph comparing the effect of PSER at different base distances for various models in examples of the present invention;
FIG. 14 is a graph comparing the effect of SSIM at different base distances for various models in accordance with embodiments of the present invention;
FIG. 15 is a graph comparing the effect of MAE at different base distances in cell-type crossing experiments in different models in the examples of the present invention;
FIG. 16 is a graph comparing the effect of PSNR at different base distances in a cell-type crossing experiment for different models in an example of the present invention;
FIG. 17 is a graph comparing the effect of SSIM at different base distances in a transcellular type experiment for different models in an example of the invention;
fig. 18 is a schematic block diagram of an internal structure of a terminal device according to an embodiment of the present invention.
Detailed Description
The technical scheme of the invention is clearly and completely described below by combining the embodiment and the attached drawings of the specification. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Researches show that the human chromosome has a complex multi-stage three-dimensional space structure, the three-dimensional structure of the chromosome is accurately analyzed, the mechanism of gene regulation and control is disclosed, and the prevention and treatment of genetic diseases are further realized. The High-throughput Chromosome transformation Capture (High-C) technology is a main means for obtaining the three-dimensional structure of chromosomes at present. The general method is to construct a chromosome conformation contact map according to Hi-C sequencing data, and further analyze and count the chromosome conformation contact map to draw a conclusion whether the chromosome has a lesion. The high-resolution Hi-C sequencing data needs to consume huge labor, material and time costs, but the chromosome conformation contact map constructed according to the high-resolution Hi-C sequencing data can accurately reveal the three-dimensional structure of the chromosome (namely the high-resolution chromosome conformation contact map can accurately reveal the three-dimensional structure of the chromosome); the low-resolution Hi-C sequencing data has the advantages of high speed and low cost, but the chromosome conformation contact map constructed according to the low-resolution Hi-C sequencing data cannot accurately reveal the three-dimensional structure of the chromosome, and sometimes even can lead to wrong conclusion. However, the resolution of the most currently available Hi-C data is 25 to 100Mb, and therefore, it is necessary to develop some methods to improve the resolution of the Hi-C contact map (chromosome conformation contact map). The existing Hi-C contact map (chromosome conformation contact map) enhancement method is a Hi-C contact map enhancement method based on a Convolutional Neural Network (CNN). The contact map enhancement method based on the convolutional neural network generally uses a mean square error as a loss function, and the enhanced contact map has the problems of excessive smoothness and edge blurring and shows strong distance dependence. For example, in the HiC contact map enhancement method-HiCPlus which is proposed in 2018 and implemented based on a convolutional neural network, the edges of the topological domain, the chromatin loop and other structures in the enhanced contact map are very blurred, which is very disadvantageous for the downstream bioinformatics analysis. Because the base distance is not considered in the chromosome conformation contact map enhancement method based on the convolutional neural network, the quality of the enhanced chromosome conformation contact map is poor.
In order to solve the technical problems, the invention provides a chromosome conformation contact map enhancement method and a device based on a generation countermeasure network, and solves the problem of poor quality of the chromosome conformation contact map caused by the enhancement method in the prior art. In specific implementation, the original chromosome conformation contact map is preprocessed to obtain a sample chromosome conformation contact map, then the sample chromosome conformation contact map and the base distance corresponding to the sample chromosome conformation contact map are input into a generated countermeasure network to be trained to realize the operation of training the generated countermeasure network, after the generated countermeasure network after training is obtained, the chromosome conformation contact map to be enhanced is input into the generated countermeasure network after training, and the generated countermeasure network can enhance the resolution of the chromosome conformation contact map. The invention fully considers the influence of the base distance on the chromosome conformation contact map, so the invention can improve the quality of the chromosome conformation contact map after enhancing the resolution.
For example, the regions where the genes with little effect in the original chromosome conformation contact map are located are removed, the original chromosome conformation contact map can be used as a sample chromosome conformation contact map to train to generate the confrontation network after the regions are removed, and if the sample chromosome conformation contact map a (the base distance a), the sample chromosome conformation contact map B (the base distance B) and the sample chromosome conformation contact map C (the base distance C) are provided, the three maps and the corresponding three base distances are all input into the generated confrontation network, the networks can be trained specifically according to the base distances, so that the confrontation network generated after training can identify the chromosome conformation contact maps with different base distances, the resolution of the chromosome conformation contact maps corresponding to the different base distances can be well enhanced, and finally, the image quality after the resolution is enhanced is improved.
Exemplary method
The chromosome conformation contact map enhancement method based on generation of the countermeasure network in the embodiment can be applied to terminal equipment, and the terminal equipment can be terminal products with computing functions, such as computers and the like. In this embodiment, as shown in fig. 1, the method for enhancing a chromosome conformation contact map based on generation of a confrontation network specifically includes the following steps S100, S200, and S300:
and S100, processing the original chromosome conformation contact map according to the positions of the regions of the original chromosome conformation contact map in the chromosome to obtain a sample chromosome conformation contact map.
In this example, the original chromosome conformation contact map was constructed as follows:
downloading original Hi-C sequencing readings from a GEO database, merging original Hi-C sequencing reading files, performing 1/16 random down-sampling on the merged files to obtain low-resolution sequencing reading files, and constructing a high-resolution Hi-C contact map (a high-resolution map is used for generating a real high-resolution sample for fighting a network as training) and a low-resolution Hi-C contact map (an original chromosome conformation contact map) by using a juicer tool on the low-resolution sequencing reading files in a chromosome unit at a resolution of 10 kb. Because the original chromosome conformation contact map contains unimportant parts (parts which have no significance for researching diseases through chromosomes), the existence of the parts can increase the computing power of the training for generating the countermeasure network, and therefore, the speed of training for generating the countermeasure network can be increased by removing the parts. Step S100 includes steps S101 to S104 as follows:
s101, normalizing the pixel values of all the original chromosome conformation contact images to obtain normalized original chromosome conformation contact image C chrk ：
The reason why all the original chromosome conformation contact maps are subjected to normalization processing is that after normalization processing, the calculation force can be reduced, and the speed of training to generate the confrontation network is improved. In the present embodiment, the high-resolution Hi-C contact map (high-resolution original chromosome conformation contact map) and the low-resolution Hi-C contact map (low-resolution original chromosome conformation contact map) are normalized as follows:
wherein N is chrk Is B chrk Maximum pixel value among all the pixel points in this graph, i.e.
i. j is B chrk The ith row and the jth column in the middle.
M chrk For the original chromosome conformation contact map of the kth chromosome, n chrk Number of contact readings for the kth chromosome, i.e.
N is all N chrk Maximum value of (1), i.e
And S102, partitioning the normalized original chromosome conformation contact image according to a set size to obtain each sample subgraph for representing each region of the original chromosome conformation contact image.
This example is a method of cropping the high-resolution and low-resolution Hi-C contact maps into several 40X 40 (i.e., 400kb X400 kb) size non-overlapping subgraphs (individual sample subgraphs).
And S103, removing sample subgraphs in each sample subgraph except the sample subgraph located above the diagonal line of the original chromosome conformation contact graph and/or removing the sample subgraph of each sample subgraph located at the position of the chromosome centromere to obtain each sample subgraph after preselection.
The diagonal line is the region with the base distance of 0 on the contact map of the original chromosome conformation, and the base distance of the subgraph on the diagonal line is 0, which represents the interaction pair in the corresponding interval of the subgraph.
In one embodiment, all sample subgraphs not including the region above the diagonal of the chromosome and all sample subgraphs including the region of the centromere of the chromosome are removed from all sample subgraphs, leaving the sample subgraphs (each of the sample subgraphs after preselection) suitable for training to generate the countermeasure network. The reason why the sample subgraph is removed is that the region above the diagonal has important research significance and the chromosome centromere region does not have important research significance, and the specific reasons are as follows:
taking the chromosome conformation contact map corresponding to chromosome 1 in fig. 2 as an example, the control aspect of the centromere gene corresponding to the chromosome conformation contact map has little effect and the data thereof is difficult to obtain, so the data set is not used when being constructed; hiC contact map has sparsity, a large amount of dark block-shaped areas are arranged in the areas close to the diagonal lines in the contact map, and correspond to microstructures such as topological domains and chromatin loops, so that the contact map has important significance for further bioinformatics analysis, areas far away from the diagonal lines are usually lighter in color and weak in interaction, and the size of the topological domains is usually within 1 kb; the Hi-C contact map has strict symmetry, and the value at the (x, y) position in the contact map indicates the strength of the contact between the x-th and y-th strands of the base sequence, so that the values at the (x, y) position and the (y, x) position are completely equal.
And S104, removing the sample subgraphs with the base distance being more than or equal to the set distance in each preselected sample subgraph to obtain a sample chromosome conformation contact graph.
Base distance is the number of base pairs that are separated between two points of interaction (i.e., interacting base pairs) on a chromosome. For example, the base distance between two interaction sites is 1Mb, which means that the distance between two sites is 1M (1X 10) 6 ) One base pair.
In this example, a sample subgraph having a base distance of 2Mb or more was removed because a portion having a base distance of more than 2Mb in a contact graph (sample subgraph) is not important to be studied.
In the step S100, the repeated redundant data is removed through the steps S101 to S104, so that the computational power is effectively reduced, and the training speed of training the generation of the countermeasure network is increased. In one embodiment, data on chromosome 23 is discarded to exclude the influence of sex using data on chromosomes 1 to 17 as a training set and data on chromosomes 18 to 22 as a test set.
S200, training the generated confrontation network according to the sample chromosome conformation contact image and the base distance of the sample chromosome conformation contact image, and obtaining the trained generated confrontation network.
In the embodiment, when the generated confrontation network is trained, the contact map and the base distance corresponding to the contact map are input into the generated confrontation network, and the important information of the base distance is utilized, so that the generated confrontation network can treat samples at different base distances differently, and the samples are pertinently enhanced according to local patterns at different base distances. Step S200 includes steps S201 to S209 as follows:
s201, inputting the sample chromosome conformation contact image and the base distance of the sample chromosome conformation contact image into the generator to be trained to obtain an enhanced resolution sample image output by the generator to be trained.
In this example, five sample contact patterns are included, i.e., a sample chromosome conformation contact pattern with a base distance of 0Kb, a sample chromosome conformation contact pattern with a base distance of 40Kb, a sample chromosome conformation contact pattern with a base distance of 80Kb, a sample chromosome conformation contact pattern with a base distance of 120Kb, and a sample chromosome conformation contact pattern with a base distance of 160Kb. Constructing a one-hot code of length 5 to represent the base distance of the sample; copying and splicing the one-hot codes with the length of 5 to form a 5-channel tensor with the size of 5 multiplied by 40; a 5-channel tensor of 5 × 40 × 40 and a sample (sample chromosome conformation contact map) are channel-merged to obtain a 6-channel tensor of 6 × 40 × 40 in size, and this 6-channel tensor is used as an input of the model. For example, a first channel is used for inputting 0Kb, a second channel is used for inputting 40Kb, a third channel is used for inputting 80Kb, a fourth channel is used for inputting 120Kb, a fifth channel is used for inputting 160Kb, a sixth channel is used for inputting a sample chromosome conformation contact map, when the sixth channel has input, the fifth channel also has input, and other channels have no input, the base distance of the input sample chromosome conformation contact map is known to be 160Kb by generating the confrontation network.
In another embodiment, the one-hot coding of the base distance is mapped into a single-channel tensor with the same size as the sample through the full-link layer, and then the single-channel tensor and the sample are subjected to matrix addition or point multiplication, so that the base distance is fused into the sample as the condition information. And the condition information is fused by adopting a full-connection layer method, so that the model convergence process is more stable.
In one embodiment, the generator is as shown in FIG. 3, the generator includes five convolution kernels and five residual modules, except for the last layer, all convolution layers use convolution kernels with a size of 3 × 3 and a step size of 1; after each sample chromosome conformation contact map is input into the generator, the following processes are carried out:
(1) The convolution with the convolution kernel size of 3 multiplied by 3 and the step length of 1 is changed into 64 channels and passes through a ReLU activation function;
(2) 5 residual error modules, wherein each residual error module comprises two convolution and BN normalization and one residual error jump connection, and the number of sample channels is always kept 64;
(3) Convolution and BN normalization with a convolution kernel size of 3 multiplied by 3 and a step size of 1 and skip connection spanning 5 residual modules;
(4) The convolution with the convolution kernel size of 3 multiplied by 3 and the step length of 1 is performed once, and the number of channels is changed into 128;
(5) The convolution with the convolution kernel size of 3 multiplied by 3 and the step length of 1 is performed once, and the number of channels is changed to 256;
(6) And (3) carrying out convolution with a convolution kernel of 1 multiplied by 1 and a step length of 1, wherein the number of channels is 1, and outputting a high-resolution sample through a Tanh activation function.
A high resolution sample (enhanced resolution sample map) is output by (1) to (6) causing the generator to input a low resolution sample and its base distance information.
S202, inputting the enhanced resolution sample image and the real set resolution image output by the generator to be trained into the discriminator to be trained to obtain a generation LOSS function G _ LOSS output by the discriminator to be trained, wherein the generation LOSS function is used for representing the difference degree between the enhanced resolution sample image and the real set resolution image.
The enhanced resolution sample map is obtained by enhancing the resolution of the low-resolution sample chromosome conformation contact map by the generator, while the resolution of the true set resolution map (which is the standard image), i.e., the high-resolution Hi-C contact map in step S101, is the same for both the enhanced resolution sample map and the true set resolution map, but the quality of the two maps (the quality of the maps, such as the pixel variance of the maps) is not the same, so that the discriminator is required to determine the quality difference of the two maps.
In one embodiment of the present invention,
d is a discriminator, c is a base distance, x is an enhanced resolution sample graph output by the generator, D (x | c) is an output result given by the discriminator D to the input sample x under the condition of the input base distance c, MSE _ loss is mean square error loss, VGG _ loss is VGG characteristic loss, TV _ loss is total variation loss, and alpha is a coefficient of the mean square error loss, and the adjustable super-parameter is obtained.
Beta is the coefficient of VGG characteristic loss, and the adjustable hyper-parameter. Gamma is the coefficient of total variation loss and can be regulated to obtain a super-parameter. In one embodiment, ω =0.55, α =0.25, β =0.05, γ =0.15.
Marking the generated samples as 0, the mean square error loss reflecting the pixel-level error between the high resolution samples and the generated samples; the feature layer of the VGG16 network can extract high-level features of the image, such as a topological structure domain, a chromatin loop and other structures, so that VGG feature loss is introduced into a loss function; the total variation loss is a common method for reducing noise in the image field, noise caused by technical problems also exists in the Hi-C contact map, and therefore the total variation loss is introduced to reduce the influence of the noise on the contact map.
S203, adjusting the parameters of the generator to be trained according to the generation loss function, and completing one round of training of the generator.
In one embodiment, the optimizer Adam is used to adjust the parameters of the generator to be trained, with the initial learning rate set to 1e-5 (representing 1 × 10) -5 ) The learning rate in 500 rounds is reduced to 1e-6,1200, and the learning rate in 5e-7 rounds is reduced.
S204, inputting the sample chromosome conformation contact image and the base distance of the sample chromosome conformation contact image into the generator after one round of training to obtain an enhanced resolution sample image output by the generator after one round of training.
S205, inputting the enhanced resolution sample graph and the true set resolution graph output by the generator after one round of training into the discriminator to be trained, and obtaining an output value of the discriminator to be trained.
In one embodiment, the discriminator is a convolutional neural network, as shown in fig. 4, the discriminator includes five convolution kernels, and the loss value is output through the following process:
(1) The convolution with the convolution kernel size of 3 multiplied by 3 and the step length of 1 is performed for one time, the number of channels is changed to 64, and the channels pass through an LeakyReLU activation function;
(2) Convolution with a convolution kernel size of 3 × 3 and a step size of 2 is performed for one time, the number of channels is kept unchanged at 64, and the convolution is normalized by a LeakyReLU activation function and BN;
(3) Convolution with a convolution kernel size of 3 × 3 and a step size of 1 is performed for one time, the number of channels is kept unchanged at 64, and the convolution is normalized by a LeakyReLU activation function and BN;
(4) Convolution with a convolution kernel size of 3 × 3 and a step size of 2 is performed for one time, the number of channels is kept unchanged at 64, and the convolution is normalized by a LeakyReLU activation function and BN;
(5) Convolution with a convolution kernel size of 3 × 3 and a step size of 1 is performed for one time, the number of channels is kept unchanged at 64, and the convolution is normalized by a LeakyReLU activation function and BN;
(6) Convolution with a convolution kernel size of 3 × 3 and a step size of 2 is performed for one time, the number of channels is kept unchanged at 64, and the convolution is normalized by a LeakyReLU activation function and BN;
(7) The convolution with the convolution kernel size of 3 multiplied by 3 and the step length of 1 is performed for one time, the number of channels is changed into 128, and the normalization is performed through a LeakyReLU activation function and BN;
(8) Flattening to a 1-dimensional tensor of length 3200;
(9) The full connection layer is mapped to a 1-dimensional tensor with the length of 512 and passes through a LeakyReLU activation function;
(10) The fully-connected layer is mapped to a 1-dimensional tensor of length 1 (the final output of the discriminator).
The higher the score of the discriminator output to the high resolution (enhanced resolution sample map output by the generator), the more the input samples fit into the true distribution, the higher the score of the discriminator output.
And S206, calculating the real loss value of the enhanced resolution sample graph output by the generator after one round of training relative to the real set resolution graph.
S207, calculating a Loss value of the output value of the discriminator to be trained relative to the output value of the expected discriminator, and recording as a discrimination Loss function D _ Loss.
After a round of training of the generator has been completed through steps S201 to S203, and then the parameters of the generator are fixed, training of the discriminator is started, and after the sample map is input to the generator, the generator outputs an enhanced resolution sample map (the generated high resolution sample in fig. 5), as shown in fig. 5, the enhanced resolution sample map and the real high resolution sample are input to the discriminator, the discriminator has an output value (the output value is used for representing the difference between the enhanced resolution sample map and the real high resolution sample), and a Loss value D _ Loss between the output value of the discriminator and the real Loss value (the real Loss value of the enhanced resolution sample map and the real high resolution sample) is calculated:
d is a discriminator, c is base distance condition information, x is a sample input to the discriminator, which may be a true high resolution sample or a generated sample, P real (x) For distribution of true high resolution samples, P fake (x) To generate the distribution of samples, GP _ loss is the gradient penalty loss, λ is the coefficient of the gradient penalty loss, and the adjustable hyper-parameter (λ is 10). Gradient penalty Loss GP _ Loss is adopted to be matched with Wasserstein Loss so as to meet the Lipschitz limiting condition of the Wasserstein Loss, and the model (generation countermeasure network) can be stably converged.
S208, adjusting the parameters of the discriminator to be trained according to the discrimination loss function to obtain a round of training of the discriminator.
In one embodiment, the parameters of the discriminator are adjusted using the optimizer Adam, the initial learning rate is set to 1e-5, and the learning rate is reduced from 500 rounds to 1e-6,1200 rounds to 5e-7.
S209, according to the generator after one round of training and the discriminator after one round of training, continuing to alternately train the generator and the discriminator until the iteration number reaches the set number, and finishing the training of the generated countermeasure network.
Step S209 is to repeat steps S201 to S208, and train the generator and the discriminator alternately, that is, the generator and the discriminator are trained alternately according to the proportion of 1:1 until the number of iterations reaches 5000 rounds, thereby completing the training of the generator and the discriminator. The purpose of training the discriminator is to enable the discriminator to better supervise the training of the generator.
In another embodiment, the following principle is used to determine whether training for generating the countermeasure network is completed:
in the training process, the parameters of the discriminator are first fixed, and the generator minimizes-D (G (x | c)). Then, the parameters of the generator are fixed and the discriminator maximizes D (x | c). And the maximization and the minimization are alternately iterated, the samples generated by the generator are gradually close to the distribution of the real high-resolution samples, and the capability of the discriminator for distinguishing the authenticity of the picture is gradually improved. By P r (x) Representing the distribution of true high resolution samples, P l (x) Representing the distribution of low resolution samples, P f (x) Representing the distribution of the generated samples. The training process can be expressed as a maximally minimized gaming process for generators and discriminators as follows:
d is a generation loss function of the discriminator output, G is an enhanced resolution sample map of the generator output, c is base distance condition information (base distance),x is a sample, possibly a low resolution sample, a true high resolution sample, a generated sample, P l (x) For distribution of low-resolution samples, P r (x) For the distribution of true high resolution samples (true set resolution map), P f (x) To generate a distribution of samples (enhanced resolution sample map output by the generator), λ is the coefficient of the gradient penalty loss, the adjustable hyper-parameter,
represents the derivation of x;
sample x obeys distribution P representing input r (x) Then, a mathematical expectation is obtained for D (x | c);
when the input sample x obeys the distribution, the mathematical expectation is obtained for-D (G (x | c));
sample x obeys distribution P representing input f (x) When, to
The mathematical expectation is obtained.
In an ideal situation, the game process of the generator and the discriminator reaches a nash equilibrium point (namely, when the generator takes the minimum-D (G (x | c)), and the discriminator takes the maximum D (x | c), the formula reaches a set value, and at this time, the training of generating the countermeasure network is completed), the generator recovers the distribution of the real high-resolution samples, and the discriminator cannot discriminate the true and false of the samples. Finally, a generator network with good sample generating capacity is obtained, the generator network learns the corresponding relation from the distribution of the low-resolution samples to the distribution of the high-resolution samples, the high-resolution samples can be recovered from the low-resolution samples, and the Hi-C contact map is enhanced by using the generator at the moment.
S300, inputting the chromosome conformation contact map to be enhanced into the trained generation countermeasure network to obtain the enhanced chromosome conformation contact map output by the trained generation countermeasure network.
Inputting the chromosome conformation contact map to be enhanced into the generator in the trained generation confrontation network, and the generator can enhance the resolution of the contact map. Step S300 includes steps S301 to S307 as follows:
s301, partitioning the chromosome conformation contact image to be enhanced according to a set size to obtain each subgraph to be enhanced.
In one embodiment, the chromosome conformation contact map to be enhanced is cropped into a plurality of subgraphs to be enhanced with the size of 40 × 40.
S302, selecting the subgraph to be enhanced at the diagonal position of the chromosome from the subgraphs to be enhanced, and marking as a first subgraph; selecting the subgraph to be enhanced positioned below the diagonal line of the chromosome from each subgraph to be enhanced and marking as a second subgraph; and selecting the subgraph to be enhanced above the diagonal line of the chromosome from the subgraphs to be enhanced, and marking as a third subgraph.
S303, inputting the first subgraph and the third subgraph into a generator in the trained generation countermeasure network to obtain the enhanced first subgraph and the enhanced third subgraph in the enhanced chromosome conformation contact graph output by the generator.
And S304, transposing a matrix formed by each pixel value in the second sub-image to obtain the transposed second sub-image.
S305, inputting the transposed second sub-image into the trained generator to obtain the enhanced second sub-image output by the generator.
S306, transposing the enhanced second sub-image to obtain the transposed enhanced second sub-image.
Steps S302 to S306 are that samples on the diagonal line and above are enhanced by directly using a generator; and for the sample below the diagonal line, transposing the sample, enhancing the transposed sample by using a generator, and transposing the enhanced sample to obtain a final result.
And S307, splicing the enhanced first sub-image, the enhanced second sub-image after transposition and the enhanced third sub-image according to the positions of the first sub-image, the second sub-image and the third sub-image in the chromosome conformation contact image to be enhanced to obtain the complete enhanced chromosome conformation contact image.
Step S307 is to splice the samples together according to the original positions to form a complete enhanced Hi-C contact map.
The following experiment verifies that the generation of the countermeasure network (HiC-cGAN) in the present embodiment can effectively improve the quality of the contact map after enhancing the resolution:
training HiC-cGAN on chromosome 1 to 17 of cell type GM12878, testing the model on chromosome 18 to 22 of cell type GM12878, resulted in Mean Square Error (MSE) as shown in fig. 6, peak signal-to-noise ratio (PSNR) as shown in fig. 7, and Structural Similarity (SSIM) as shown in fig. 8. From fig. 6, it can be seen that the chromosome conformation contact maps to be enhanced with different base distances have different mean square errors (mean square error is used to represent the quality of the image) after being enhanced by the generation of the confrontation network, and therefore it is proved that there is a necessary relationship between the base distances and the quality of the contact maps after enhancement, so that the base distances need to be considered when training the generation of the confrontation network. The same is true for peak signal-to-noise ratio and structural similarity.
A subgraph with the base distance of 0kb is selected, and the enhancement effect of HiC-cGAN (the generation countermeasure network in the embodiment) on a Hi-C contact map (a chromosome conformation contact map) is visually displayed. It can be noted that there are some dark block areas (centromere areas) that are only present in the true high resolution contact map as shown in fig. 10, but not in the low resolution contact map as shown in fig. 9 or are lighter in color in the low resolution contact map. These block-shaped dark regions correspond to microscopic chromosome three-dimensional structures such as topological domains, chromatin loops and the like, and have significance for related work of bioinformatics researchers at the downstream of tasks, and the enhanced chromosome conformation contact map generated by the generation countermeasure network of the present embodiment as shown in fig. 11 clearly covers the centromere region, so that the generation countermeasure network of the present embodiment can well improve the quality of the enhanced contact map.
Comparative experiments were conducted on HiC-cGAN (generation countermeasure network in this example) of this example and GAN, cGAN, WGAN-GP, hicGAN of the prior art (training the above model method on chromosomes 1 to 17 of cell type GM12878, and testing model performance on chromosomes 18 to 22 of cell type GM 12878) to obtain Mean Square Error (MSE) shown in fig. 12, peak signal-to-noise ratio (PSNR) shown in fig. 13, and Structural Similarity (SSIM) shown in fig. 14. Since the smaller the mean square error, the closer the generated samples and the high resolution samples representing the model, the better the quality of the generated samples. As can be seen from FIG. 12, the performance of the generated countermeasure network proposed in this example is better than that of all existing models at five base distances, especially at the smaller base distance (0 kb, 40kb).
HiC-cGAN also showed excellent performance in the trans-cell type experiment, firstly training HiC-cGAN and four GAN models on chromosome 1 to 17 of cell type GM12878, respectively, and then testing the performance of different models on chromosome 1 to 17 and chromosome 18 to 22 of GM12878 with the same cell type and two trans-cell types K562 and NHEK, respectively, the experimental results are shown in FIG. 15, FIG. 16 and FIG. 17, and it can be seen from FIG. 15, FIG. 16 and FIG. 17 that HiC-cGAN of this example is due to other four GAN models.
In summary, the invention fully considers the relevance between the base distance and the chromosome conformation contact map when training to generate the countermeasure network, and when training to generate the countermeasure network, the base distance is input into the generated countermeasure network as the condition information, so that the generated countermeasure network can treat the chromosome conformation contact maps with different base distances in a distinguishing way during training, and further the generated countermeasure network after training can be pertinently enhanced according to the base distance of each chromosome conformation contact map when being used for enhancing the resolution of the chromosome conformation contact map, thereby improving the quality of the contact map after the resolution is enhanced by the contact maps with various base distances.
In addition, the invention reduces the data amount adopted by training based on the symmetry and sparsity of the Hi-C contact map, and only adopts the data which is above the diagonal line of the contact map and has the base distance less than 2 kb. The contact pattern has different local patterns under different base distances, the invention fully considers the important information of the base distance and takes the base distance as the condition information to construct the condition generation confrontation network. The loss function is optimized by combining multiple losses such as Wasserstein loss, gradient penalty loss, mean square error loss, VGG characteristic loss, total variation loss and the like, so that the model can be stably converged, and a better enhancement effect is obtained.
Exemplary devices
The embodiment also provides a chromosome conformation contact map enhancing device based on generation of an antagonistic network, which comprises the following components:
the preprocessing module is used for processing the original chromosome conformation contact map according to the positions of all the areas of the original chromosome conformation contact map in the chromosome to obtain a sample chromosome conformation contact map;
the training module is used for training the generated confrontation network according to the sample chromosome conformation contact image and the base distance of the sample chromosome conformation contact image to obtain the trained generated confrontation network;
and the resolution enhancement module is used for inputting the chromosome conformation contact map to be enhanced into the trained generation countermeasure network to obtain the enhanced chromosome conformation contact map output by the generator in the trained generation countermeasure network.
Based on the above embodiments, the present invention further provides a terminal device, and a schematic block diagram thereof may be as shown in fig. 18. The terminal equipment comprises a processor, a memory, a network interface, a display screen and a temperature sensor which are connected through a system bus. Wherein the processor of the terminal device is configured to provide computing and control capabilities. The memory of the terminal equipment comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The network interface of the terminal device is used for connecting and communicating with an external terminal through a network. The computer program is executed by a processor to implement a method of chromosome conformation contact map enhancement based on generation of a countermeasure network. The display screen of the terminal equipment can be a liquid crystal display screen or an electronic ink display screen, and the temperature sensor of the terminal equipment is arranged in the terminal equipment in advance and used for detecting the operating temperature of the internal equipment.
It will be understood by those skilled in the art that the block diagram of fig. 18 is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation to the terminal equipment to which the solution of the present invention is applied, and a specific terminal equipment may include more or less components than those shown in the figure, or may combine some components, or have different arrangements of components.
In one embodiment, a terminal device is provided, the terminal device includes a memory, a processor, and a chromosome conformation contact map enhancing program based on a generation countermeasure network stored in the memory and executable on the processor, and when the processor executes the chromosome conformation contact map enhancing program based on the generation countermeasure network, the following operation instructions are implemented:
processing the original chromosome conformation contact map according to the positions of the areas of the original chromosome conformation contact map in the chromosome to obtain a sample chromosome conformation contact map;
training a generated confrontation network according to the sample chromosome conformation contact image and the base distance of the sample chromosome conformation contact image to obtain the trained generated confrontation network;
inputting the chromosome conformation contact map to be enhanced into the trained generation confrontation network to obtain the enhanced chromosome conformation contact map output by the trained generation confrontation network.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, databases, or other media used in embodiments provided herein may include non-volatile and/or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), rambus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (10)
1. A chromosome conformation contact map enhancement method based on generation of a confrontation network is characterized by comprising the following steps:
processing the original chromosome conformation contact map according to the position of each region of the original chromosome conformation contact map in the chromosome to obtain a sample chromosome conformation contact map;
training a generated confrontation network according to the sample chromosome conformation contact image and the base distance of the sample chromosome conformation contact image to obtain the trained generated confrontation network;
and inputting the chromosome conformation contact map to be enhanced into the trained generator in the generation countermeasure network to obtain an output enhanced chromosome conformation contact map.
2. The method for enhancing the chromosome conformation contact map based on the generation countermeasure network as claimed in claim 1, wherein the step of processing the original chromosome conformation contact map according to the positions of the areas of the original chromosome conformation contact map in the chromosome to obtain the sample chromosome conformation contact map comprises:
normalizing the pixel values of the original chromosome conformation contact image to obtain the normalized original chromosome conformation contact image;
partitioning the normalized original chromosome conformation contact image according to a set size to obtain sample sub-images for representing all regions of the original chromosome conformation contact image;
removing sample subgraphs below the diagonal in each sample subgraph and the sample subgraphs at the positions of the chromosome centromere to obtain preselected sample subgraphs;
and screening the sample subgraph after preselection according to the base distance of the sample subgraph after preselection to obtain a sample chromosome conformation contact map.
3. The method for enhancing chromosome conformation contact map based on generation of confrontation network as claimed in claim 2, wherein said screening the sample subgraphs after preselection according to the base distance of the sample subgraphs after preselection to obtain the sample chromosome conformation contact map comprises:
and removing the sample subgraphs with the base distance being more than or equal to the set distance in the preselected sample subgraph to obtain a sample chromosome conformation contact graph.
4. The method for enhancing the chromosome conformation contact map based on the generation countermeasure network as claimed in claim 1, wherein the training of the generation countermeasure network according to the base distance of the sample chromosome conformation contact map and the sample chromosome conformation contact map to obtain the trained generation countermeasure network comprises:
alternately training a generator to be trained and a discriminator according to the sample chromosome conformation contact pattern and the base distance of the sample chromosome conformation contact pattern to obtain the trained generation confrontation network, wherein the generator and the discriminator form the generation confrontation network.
5. The method for enhancing the chromosome conformation contact map based on the generation countermeasure network as claimed in claim 4, wherein the training of the generator and the discriminator to be trained alternately according to the base distance of the sample chromosome conformation contact map and the sample chromosome conformation contact map to obtain the trained generation countermeasure network, the generator and the discriminator forming the generation countermeasure network comprises:
inputting the sample chromosome conformation contact image and the base distance of the sample chromosome conformation contact image into the generator to be trained to obtain an enhanced resolution sample image output by the generator to be trained;
inputting an enhanced resolution sample graph and a real set resolution graph output by the generator to be trained into the discriminator to be trained to obtain a generation loss function output by the discriminator to be trained, wherein the generation loss function is used for representing the difference degree between the enhanced resolution sample graph and the real set resolution graph;
adjusting parameters of the generator to be trained according to the generation loss function to complete a round of training of the generator;
inputting the sample chromosome conformation contact image and the base distance of the sample chromosome conformation contact image into the generator after one round of training to obtain an enhanced resolution sample image output by the generator after one round of training;
inputting the enhanced resolution sample graph and the real set resolution graph output by the generator after one round of training into the discriminator to be trained to obtain an output value of the discriminator to be trained;
calculating a real loss value of the enhanced resolution sample graph output by the generator after one round of training relative to the real set resolution graph;
calculating the loss value of the output value of the discriminator to be trained relative to the output value of the expected discriminator, and recording as a discrimination loss function;
adjusting parameters of the discriminator to be trained according to the discrimination loss function to obtain a round of training of the discriminator;
and continuing to alternately train the generator and the discriminator according to the generator after one round of training and the discriminator after one round of training until the iteration times reach the set times, and finishing the training of the generated countermeasure network.
6. The method for enhancing chromosome conformation contact map based on generation countermeasure network as claimed in claim 1, wherein the inputting chromosome conformation contact map to be enhanced to the generation countermeasure network trained to obtain the enhanced chromosome conformation contact map output by the generation countermeasure network trained comprises:
partitioning the chromosome conformation contact map to be enhanced according to a set size to obtain each subgraph to be enhanced;
selecting the subgraphs to be enhanced at the diagonal positions of the chromosomes from the subgraphs to be enhanced, and marking the subgraphs as first subgraphs;
selecting the subgraph to be enhanced positioned below the diagonal line of the chromosome from each subgraph to be enhanced and marking as a second subgraph;
selecting the subgraph to be enhanced above the diagonal line of the chromosome from each subgraph to be enhanced, and marking as a third subgraph;
inputting the first subgraph and the third subgraph into a generator in the trained generation countermeasure network to obtain the enhanced first subgraph and the enhanced third subgraph in an enhanced chromosome conformation contact graph output by the generator;
transposing a matrix formed by each pixel value in the second sub-image to obtain the transposed second sub-image;
and inputting the transposed second sub-image into the trained generator to obtain the enhanced second sub-image output by the generator.
7. The method for enhancing chromosome conformation contact map based on generation countermeasure network as claimed in claim 6, wherein the inputting chromosome conformation contact map to be enhanced to the generation countermeasure network trained to obtain the enhanced chromosome conformation contact map output by the generation countermeasure network trained, then further comprising:
transposing the enhanced second sub-graph to obtain the transposed enhanced second sub-graph;
and splicing the enhanced first sub-image, the enhanced second sub-image after transposition and the enhanced third sub-image according to the positions of the first sub-image, the second sub-image and the third sub-image in the chromosome conformation contact image to be enhanced to obtain the complete enhanced chromosome conformation contact image.
8. A chromosome conformation contact map enhancing device based on generation of an antagonistic network is characterized by comprising the following components:
the preprocessing module is used for processing the original chromosome conformation contact map according to the positions of all the areas of the original chromosome conformation contact map in the chromosome to obtain a sample chromosome conformation contact map;
the training module is used for training the generated confrontation network according to the sample chromosome conformation contact image and the base distance of the sample chromosome conformation contact image to obtain the trained generated confrontation network;
and the resolution enhancement module is used for inputting the chromosome conformation contact map to be enhanced into the trained generation countermeasure network to obtain the enhanced chromosome conformation contact map output by the trained generation countermeasure network.
9. A terminal device, characterized in that the terminal device comprises a memory, a processor and a generating countermeasure network-based chromosome conformation contact map enhancement program stored in the memory and operable on the processor, and when the processor executes the generating countermeasure network-based chromosome conformation contact map enhancement program, the steps of the generating countermeasure network-based chromosome conformation contact map enhancement method according to any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium, wherein the computer-readable storage medium has stored thereon a chromosome conformation contact map enhancing program based on a generation countermeasure network, and when the chromosome conformation contact map enhancing program based on the generation countermeasure network is executed by a processor, the computer-readable storage medium implements the steps of the chromosome conformation contact map enhancing method based on the generation countermeasure network according to any one of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211007292.2A CN115345797A (en) | 2022-08-22 | 2022-08-22 | Chromosome conformation contact map enhancement method and device based on generation of confrontation network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211007292.2A CN115345797A (en) | 2022-08-22 | 2022-08-22 | Chromosome conformation contact map enhancement method and device based on generation of confrontation network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115345797A true CN115345797A (en) | 2022-11-15 |
Family
ID=83953203
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211007292.2A Pending CN115345797A (en) | 2022-08-22 | 2022-08-22 | Chromosome conformation contact map enhancement method and device based on generation of confrontation network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115345797A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116777904A (en) * | 2023-08-16 | 2023-09-19 | 笑纳科技(苏州)有限公司 | Different resolution abnormal chromosome generation network and method based on stripe characteristics |
-
2022
- 2022-08-22 CN CN202211007292.2A patent/CN115345797A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116777904A (en) * | 2023-08-16 | 2023-09-19 | 笑纳科技(苏州)有限公司 | Different resolution abnormal chromosome generation network and method based on stripe characteristics |
| CN116777904B (en) * | 2023-08-16 | 2023-10-24 | 笑纳科技(苏州)有限公司 | Different resolution abnormal chromosome generation network and method based on stripe characteristics |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Niu et al. | Single image super-resolution via a holistic attention network | |
| Shen et al. | Networks for joint affine and non-parametric image registration | |
| US7747070B2 (en) | Training convolutional neural networks on graphics processing units | |
| Lu et al. | Geometry constrained sparse coding for single image super-resolution | |
| CN113159143B (en) | Infrared and visible light image fusion method and device based on jump connection convolution layer | |
| Li et al. | Joint spatial–spectral attention network for hyperspectral image classification | |
| CN113421276B (en) | Image processing method, device and storage medium | |
| CN115345797A (en) | Chromosome conformation contact map enhancement method and device based on generation of confrontation network | |
| Gao et al. | Multi-scale patch based representation feature learning for low-resolution face recognition | |
| Sun et al. | FDRN: a fast deformable registration network for medical images | |
| Gendy et al. | Balanced spatial feature distillation and pyramid attention network for lightweight image super-resolution | |
| Yang et al. | Weighted truncated nuclear norm regularization for low-rank quaternion matrix completion | |
| CN111507288A (en) | Image detection method, image detection device, computer equipment and storage medium | |
| Balestriero et al. | A data-augmentation is worth a thousand samples: Exact quantification from analytical augmented sample moments | |
| CN110097503B (en) | Super-resolution method based on neighborhood regression | |
| Wen et al. | Encoder-free multi-axis physics-aware fusion network for remote sensing image dehazing | |
| CN116343052B (en) | Attention and multiscale-based dual-temporal remote sensing image change detection network | |
| Yu et al. | MagConv: Mask-guided convolution for image inpainting | |
| Zeng et al. | Robust unsupervised feature selection via sparse and minimum-redundant subspace learning with dual regularization | |
| CN114359232B (en) | Image change detection method and device based on context covariance matrix | |
| Miao et al. | Quaternion tensor train rank minimization with sparse regularization in a transformed domain for quaternion tensor completion | |
| CN113591528A (en) | Document correction method, device, computer equipment and storage medium | |
| JP7444235B2 (en) | Attention mechanism, image recognition system, feature conversion method and program | |
| CN113298827A (en) | Image segmentation method based on DP-Net network | |
| Zhang et al. | Stochastic modified equations and dynamics of dropout algorithm |
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 |