CN115586152A - Intelligent spectrum quantitative analysis method based on convolutional neural network visualization - Google Patents
Intelligent spectrum quantitative analysis method based on convolutional neural network visualization Download PDFInfo
- Publication number
- CN115586152A CN115586152A CN202211282175.7A CN202211282175A CN115586152A CN 115586152 A CN115586152 A CN 115586152A CN 202211282175 A CN202211282175 A CN 202211282175A CN 115586152 A CN115586152 A CN 115586152A
- Authority
- CN
- China
- Prior art keywords
- contribution
- spectrum
- neural network
- analysis method
- convolutional neural
- 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
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
- G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
- G01N21/31—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/62—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light
- G01N21/63—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light optically excited
- G01N21/65—Raman scattering
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Biochemistry (AREA)
- Analytical Chemistry (AREA)
- Chemical & Material Sciences (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Theoretical Computer Science (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Investigating, Analyzing Materials By Fluorescence Or Luminescence (AREA)
Abstract
The invention provides an intelligent spectrum quantitative analysis method based on convolutional neural network visualization, which solves the problems that the existing spectrum intelligent analysis method is lack of objective quantitative evaluation and cannot objectively quantify the contribution degree of each characteristic peak in spectrum data to different types of sample differentiation; s1: obtaining a contribution heat map; s2: calculating contribution weight, obtaining the full width at half maximum of a target characteristic peak in the spectrum, and determining a wave number range corresponding to the full width at half maximum; finding the contribution value corresponding to the wave number in the heat map, and taking the contribution value as a molecule; taking the sum of the contribution values in the wave number range corresponding to the full width at half maximum of all the characteristic peaks as a denominator; and dividing the numerator by the denominator to obtain the contribution weight of the target characteristic peak to the final classification result of the target object.
Description
Technical Field
The invention relates to the technical field of spectrum quantitative analysis, in particular to an intelligent spectrum quantitative analysis method based on convolutional neural network visualization.
Background
The method is widely applied to chemical engineering, medicine, biochemistry, environmental protection and other aspects, and has the characteristics of simple and rapid operation, high sensitivity, good precision and accuracy, wide linear effective range and low detection limit.
The traditional spectrum intelligent analysis method based on deep learning lacks quantitative evaluation, can only give out whether spectrum peaks have statistical significance in different types of samples, and cannot give out quantitative contribution degree of each Raman peak to different types of sample discrimination.
Disclosure of Invention
The invention provides an intelligent spectrum quantitative analysis method based on convolutional neural network visualization, which solves the problems that the existing spectrum intelligent analysis method is lack of objective quantitative evaluation and cannot objectively quantify the contribution degree of each characteristic peak in spectrum data to different types of sample differentiation.
An intelligent spectrum quantitative analysis method based on convolutional neural network visualization comprises the following steps:
s1: obtaining a contribution heat map, obtaining feature vectors of classification results by utilizing a one-dimensional convolutional neural network, taking a gradient global average value of each feature vector to the classification results as a weight value of the feature vector based on visualization of a gradient weighting class mapping method, and obtaining activation intensity distribution of an input spectrum to the classification results through a back propagation method to obtain the contribution heat map;
s2: calculating contribution weight, obtaining the full width at half maximum of a target characteristic peak in the spectrum, and determining a wave number range corresponding to the full width at half maximum; finding the contribution value corresponding to the wave number in the heat map, and taking the contribution value as a molecule; taking the sum of the contribution values in the wave number range corresponding to the full width at half maximum of all the characteristic peaks as a denominator; and dividing the numerator by the denominator to obtain the contribution weight of the target characteristic peak to the final classification result of the target object.
Further, the process of obtaining the feature vector of the classification result by using the one-dimensional convolutional neural network includes:
the method comprises the following steps: processing the spectral data, namely averagely dividing the spectral data of all the target objects into a plurality of parts, wherein one part is used as a test sample and is classified into a test set; the rest parts are as follows 3:1, dividing the training sample and the verification sample into a training sample and a verification sample, performing data expansion on the training sample and the verification sample, and respectively classifying the training sample and the verification sample into a training set and a verification set; and preprocessing the spectra of the training set, the verification set and the test set in the following modes: denoising and normalizing;
step two: the establishment of the one-dimensional convolution neural network structure is sequentially set as follows: the first convolution layer is used for extracting input spectral features, the size of a convolution kernel is 3, and the step length is 2; each residual error layer comprises a convolution module and an identity module, the convolution module comprises two convolution layers and a shortcut connection, the identity module comprises two convolution layers, and the convolution kernel size of each convolution layer in each residual error layer is 3; global average pooling; a full connectivity layer and a Softmax activation function;
step three: training and performance evaluation of the one-dimensional convolutional neural network, setting the learning rate to be 0.0001 during model training, setting the training times to be 10000, and automatically storing the current optimal model by comparing the classification accuracy of the model on a verification set in the training process.
Further, the spectrum comprises: raman spectrum, absorption spectrum.
Further, the data expansion method in the first step comprises: and removing original noise from the original spectrum by a Savitzky-Golay smoothing algorithm, and then adding Poisson random noise to obtain a forged spectrum required by expansion.
Further, the activation function of the convolution layer in the first convolution layer and the residual is a non-linear ReLU activation function.
Further, the input channel dimension, the kernel size, and the output channel dimension of the convolution layer in the first convolution layer, the convolution layer in the convolution module, and the convolution layer in the identity module are respectively: 1,3, 64;64,3, 10;10,3, 10;10,3, 10;10,3, 10;10,3, 30;30,3, 30;30,3, 30;30,3, 30.
Further, the quantitative analysis method analyzes and searches potential markers for classification and identification by quantitatively evaluating the contribution degree of each characteristic peak.
Further, the quantitative analysis method comprises the following steps: pathogen genotype detection, cancer genotyping detection.
The technical effects are as follows:
1) Visualizing the classification working mechanism of the 1D-CNN based on the difference of biochemical components of the target and a gradient weighting activation mapping method to obtain a contribution heat map of Raman spectrum information of different target objects to final classification decision, quantifying the contribution weight of a main Raman characteristic peak to a final classification result, and finding that knocking out different histone deacetylase genes can cause different characteristic markers which can represent the most similar target objects, wherein the difference of the biochemical components can be related to aspects such as virulence, drug resistance and metabolic function;
2) The visualization result based on Grad-CAM shows the characteristic regions searched by the 1D-CNN model for judging different classes, for example, in the corresponding contribution heat maps, the characteristic regions present obvious 'strip-shaped' distribution, which indicates the most important characteristic markers of the class of target objects, and the substances are basically not overlapped with the most important characteristic markers of other strains, so that the different classes of target objects can be correctly identified.
Drawings
FIG. 1 is a flow chart of the classification and identification of strains of different genotypes based on the 1D-CNN method;
FIG. 2 is a graph showing the results of preprocessing of spectral data;
FIG. 3 is a 1D-CNN network structure;
FIG. 4 is a parameter set for a 1D-CNN network;
FIG. 5 is a method of visualization of a one-dimensional convolutional neural network based on a gradient-weighted class activation mapping method;
FIG. 6 is a heat map of the mean Raman spectra after normalization of different genotypic strains versus the contribution of non-genotypic strains to the final classification decision for 1D-CNN.
Detailed Description
The present invention will be described in detail below by way of examples and test examples. The examples are given to better illustrate the content and advantages of the invention, but it should not be understood that the content of the invention is limited to the examples. Those skilled in the art who have the benefit of this disclosure will realize additional modifications and adaptations to the embodiments described herein without departing from the scope of the invention.
Example 1: an intelligent spectrum quantitative analysis method based on convolutional neural network visualization.
The spectral data processing and analyzing method based on the one-dimensional convolutional neural network (1D-CNN) is summarized as follows:
the embodiment aims at cryptococcus neoformans standard strain H99 and six different histone deacetylase gene knockout strains, and the specific method is as follows: based on a gradient weighting activation mapping method, a classification mechanism of a one-dimensional convolutional neural network is visualized, a contribution heat map of Raman spectrum information of different genotype strains corresponding to a final classification decision is obtained, and a pathogen genotype classification detection mechanism based on a Raman spectrum technology is obtained.
The flow chart of the classification and identification of different genetic strains based on the 1D-CNN method is shown in the attached figure 1.
The method specifically comprises the following steps.
The method comprises the following steps: the pre-processing of the spectral data is carried out,
averagely dividing the spectral data of the target into 5 parts, selecting one part as a test specimen, and dividing the rest four parts into 3 parts: 1, dividing the test sample into a training sample and a verification sample, then adding noise to the training sample and the verification sample to perform data expansion, and not performing data expansion on the test sample.
Wherein, the process of data expansion is as follows: and removing original noise from the original spectrum to be expanded by utilizing a Savitzky-Golay smoothing algorithm, and then adding Poisson noise to the smoothed spectrum to obtain the expanded spectrum.
Specifically, each target in this embodiment includes 150 raman spectra, and the total of seven targets is 1020 raman spectra, and 150 spectral data of each target are divided into 5 parts on average; the original spectrum to be extended is shown in fig. 2 (a), the smoothed spectrum is shown in fig. 2 (b), the poisson noise is shown in fig. 2 (c), the extended spectrum is shown in fig. 2 (d), the extended spectrum-smoothed spectrum is shown in fig. 2 (e), and the smoothed spectrum and the extended spectrum-smoothed spectrum are shown in fig. 2 (f).
As can be seen from fig. 2 (a) -2 (d), the spectrum after noise addition has a certain difference from the original spectrum; as can be seen from fig. 2 (b), 2 (e), and 2 (f), the difference still exists after the spectral noise reduction processing.
Finally, the training set containing the raman spectra of the seven targets was expanded to 6300 spectra and the validation set to 2100 spectra.
And (3) carrying out training or testing after noise reduction, background fluorescence removal and normalization on the spectrums of the training set, the verification set and the test set.
Step two: the establishment of a 1D-CNN network structure,
the 1D-CNN network structure used in this embodiment is as shown in fig. 3, where one-dimensional spectral data is input, 2318 × 1 one-dimensional spectral data is input in this embodiment, and enters a convolutional layer 1, where a convolutional kernel is 3 and a step length is 2, and it is used to perform feature extraction on an input spectrum, an activation function of the convolutional layer is a nonlinear ReLU (Rectified Linear Unit, reLU) activation function, and a Max pooling layer (Max pooling) with a size of 3 is used; next, two residual error layers are provided, each residual error layer comprises a convolution module and an identity module, the convolution module comprises two convolution layers and a shortcut connection, the identity module comprises two convolution layers, the sizes of the convolution cores are all 3, and the activation function of the convolution layers in the residual error layers is a nonlinear ReLU activation function; followed by global average pooling; and finally, using a full connection layer and a Softmax activation function to finish the classification of the spectra of different targets.
The parameters of the specific network structure are shown in fig. 4, where the parameters of each convolutional layer represent the input channel dimension, the kernel size, and the output channel dimension, respectively; the solid line connections between convolutional layers indicate the same number of channels, and the dashed line connections indicate different numbers of channels, which is the shortcut connection part in ResNet.
Step three: 1D-CNN training and performance evaluation,
the method is carried out under a framework built by Tensorflow, the used language is Python 3.7, the learning rate is set to be 0.0001 during model training, the training times are 10000, and the current optimal model is automatically stored by comparing the classification accuracy of the model on a verification set in the training process.
As mentioned in the first step, the training model adopts a five-fold poor verification method to evaluate the performance of the classifier, and after five-fold cross verification, the accuracy is obtained according to the confusion matrix as the final result of the discrimination and classification.
Step four: visualization based on a gradient-weighted class activation mapping method (Grad-CAM),
a gradient weighted class activation mapping method (Grad-CAM) utilizes the feature vectors of the classification results obtained by the one-dimensional convolution neural network, and takes the gradient global average value of the classification results on each feature vector as the weight value of the feature vector; subsequently, based on the weight value of each feature vector, obtaining the activation intensity distribution of the input spectrum for the classification result by a back propagation method, i.e. a heat map of the contribution of each raman peak in the raman spectrum to the final classification decision, as shown in fig. 5; and finally, combining Raman spectrum peak attribution analysis to obtain the contribution of biochemical components in different target objects to the final classification decision.
Wherein, the contribution degree heat map has a bright color area (red-yellow area) to represent that the raman spectrum information therein has a larger contribution degree to the final classification decision; the dark areas (blue areas) represent where the raman spectral information contributes less to the final classification decision.
From the results, the color distributions of the contribution heat maps corresponding to the raman spectra of the target objects are not the same, but the color distributions of the contribution heat maps have certain regularity in the same target object, as shown in fig. 6.
Specifically, the method comprises the following steps: in the contribution heat maps corresponding to the two mutant strains Clr61 Δ and Clr62 Δ, the bright color regions appeared in a distinct "striped" distribution. Wherein, in a contribution heat map corresponding to the Clr61 delta mutant strain, the region is mainly distributed in 750-950 cm -1 Within the range; clThe regions of the r62 delta mutant strain which have the greatest contribution to the final classification decision are mainly distributed in the range of 1200-1350 cm -1 Within the range.
The color distribution of the contribution degree heat maps of the Hda1 delta and the Hos1 delta mutant strains is relatively concentrated, and in the corresponding contribution degree heat map of the Hda1 delta mutant strain, the region with larger contribution is mainly distributed in 400-650 cm -1 And 1400-1500 cm -1 In the range, the region of the mutant strain of Hos1 delta which greatly contributes to the final classification decision is concentrated in 200-300 cm -1 And 500 to 650cm -1 Within the range.
In the heat map of contribution degree of the Hos2 delta and Hos3 delta mutant strains, the distribution of bright color areas is dispersed a little, wherein the heat map of the Hos2 delta mutant strain shows that the Raman information with larger contribution is mainly distributed between 700 cm and 800cm -1 And 1000 to 1130cm -1 Within the range; while the area with larger contribution is mainly distributed in 1340-1370 cm in the heat map of the Hos3 delta mutant strain -1 And 200 to 300cm -1 Within the range.
In the heat map of contribution degree of H99 wild strain, the region with larger contribution is concentrated in the low wavenumber range of the spectrum, mainly 930-950 cm -1 、540～650cm -1 And 200 to 330cm -1 。
By combining the analysis result of the Raman spectrum peak attribution with the contribution heat map of the target, the contribution weight of the main characteristic peak of the Raman spectrum of each type of target to the final classification decision can be quantified, as shown in the following table:
TABLE 1 contribution weight of Raman spectrum characteristic peaks of strains with different genotypes to final classification result of 1D-CNN model
And combining the contribution heat maps of each genotype sample to form the total distribution of the contribution heat maps corresponding to the genotypes, and further calculating the contribution weight of each characteristic peak in each genotype to the final classification result.
For example, 541cm in H99 genotype is calculated -1 And (3) processing the contribution weight of the characteristic peak to the final classification result, specifically comprising the following steps: firstly, 541cm is obtained in an average Raman spectrogram of H99 -1 Determining the half-height width of the characteristic peak so as to determine the wave number range corresponding to the half-height width; then obtaining the contribution value corresponding to the wave number range in the heat map, and taking the contribution value as a molecule; using the sum of the contribution values in the wave number range corresponding to the full width at half maximum of all the characteristic peaks as a denominator to obtain 541cm in the H99 genotype -1 The contribution weight of the characteristic peak to such final classification result.
Of course, the method of this embodiment can be applied not only to raman spectroscopy, but also to absorption spectroscopy including infrared spectroscopy having similar properties.
In addition, for scenarios where marker identification is required, such as cancer detection spectroscopy, it is also applicable, which facilitates the discovery of potential diagnostic markers.
Claims (8)
1. An intelligent spectrum quantitative analysis method based on convolutional neural network visualization is characterized by comprising the following steps: s1: obtaining a contribution heat map, obtaining feature vectors of classification results by utilizing a one-dimensional convolutional neural network, taking a gradient global average value of each feature vector to the classification results as a weight value of the feature vector based on visualization of a gradient weighting class mapping method, and obtaining activation intensity distribution of an input spectrum to the classification results through a back propagation method to obtain the contribution heat map; s2: calculating contribution weight, obtaining the full width at half maximum of a target characteristic peak in the spectrum, and determining a wave number range corresponding to the full width at half maximum; finding the contribution value corresponding to the wave number in the heat map, and taking the contribution value as a molecule; taking the sum of the contribution values in the wave number range corresponding to the full width at half maximum of all the characteristic peaks as a denominator; and dividing the numerator by the denominator to obtain the contribution weight of the target characteristic peak to the final classification result of the target object.
2. The intelligent quantitative spectrum analysis method based on convolutional neural network visualization as claimed in claim 1, wherein the process of obtaining feature vectors of classification results by using one-dimensional convolutional neural network comprises:
the method comprises the following steps: processing the spectral data, namely averagely dividing the spectral data of all the target objects into a plurality of parts, wherein one part is used as a test sample and is classified into a test set; the rest parts are as follows 3:1, dividing the training sample and the verification sample into a training sample and a verification sample, performing data expansion on the training sample and the verification sample, and respectively classifying the training sample and the verification sample into a training set and a verification set; and preprocessing the spectra of the training set, the verification set and the test set in the following modes: denoising and normalizing;
step two: the establishment of the one-dimensional convolution neural network structure is sequentially set as follows: the first convolution layer is used for extracting input spectral features, the size of a convolution kernel is 3, and the step length is 2; each residual error layer comprises a convolution module and an identity module, the convolution module comprises two convolution layers and a shortcut connection, the identity module comprises two convolution layers, and the convolution kernel size of each convolution layer in each residual error layer is 3; global average pooling; a full connectivity layer and a Softmax activation function;
step three: training and performance evaluation of the one-dimensional convolutional neural network, setting the learning rate to be 0.0001 during model training, setting the training times to be 10000, and automatically storing the current optimal model by comparing the classification accuracy of the model on a verification set in the training process.
3. An intelligent spectral quantitative analysis method based on convolutional neural network visualization as claimed in claim 1 or 2, wherein the spectrum comprises: raman spectrum, absorption spectrum.
4. The intelligent spectrum quantitative analysis method based on convolutional neural network visualization as claimed in claim 2, wherein the data expansion method in step one is: and removing original noise from the original spectrum by a Savitzky-Golay smoothing algorithm, and then adding Poisson random noise to obtain a forged spectrum required by expansion.
5. The intelligent spectral quantitative analysis method based on convolutional neural network visualization of claim 2, wherein the activation function of the convolutional layer in the first convolutional layer and the residual is a nonlinear ReLU activation function.
6. The intelligent spectral quantitative analysis method based on convolutional neural network visualization as claimed in claim 2, wherein the input channel dimension, kernel size and output channel dimension of the convolutional layer in the first convolutional layer, convolutional layer in the convolutional module and convolutional layer in the identity module are respectively: 1,3, 64;64,3, 10;10,3, 10;10,3, 10;10,3, 10;10,3, 30;30,3, 30;30,3, 30;30,3, 30.
7. The intelligent spectral quantitative analysis method based on convolutional neural network visualization as claimed in claim 1, wherein the quantitative analysis method is used for analyzing and searching potential markers for classification and identification by quantitatively evaluating the contribution degree of each characteristic peak.
8. The intelligent spectral quantitative analysis method based on the convolutional neural network visualization as claimed in claim 1, wherein the quantitative analysis method comprises: pathogen genotype detection, cancer genotyping detection.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211282175.7A CN115586152A (en) | 2022-10-19 | 2022-10-19 | Intelligent spectrum quantitative analysis method based on convolutional neural network visualization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211282175.7A CN115586152A (en) | 2022-10-19 | 2022-10-19 | Intelligent spectrum quantitative analysis method based on convolutional neural network visualization |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115586152A true CN115586152A (en) | 2023-01-10 |
Family
ID=84779749
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211282175.7A Pending CN115586152A (en) | 2022-10-19 | 2022-10-19 | Intelligent spectrum quantitative analysis method based on convolutional neural network visualization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115586152A (en) |
-
2022
- 2022-10-19 CN CN202211282175.7A patent/CN115586152A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11144576B2 (en) | Target class feature model | |
| CN108629365B (en) | Analysis data analysis device and analysis data analysis method | |
| JP7057913B2 (en) | Big data analysis method and mass spectrometry system using the analysis method | |
| JP6715451B2 (en) | Mass spectrum analysis system, method and program | |
| CN108398416A (en) | A kind of mix ingredients assay method based on laser Raman spectroscopy | |
| CN111833310B (en) | Surface defect classification method based on neural network architecture search | |
| CN112466399A (en) | Method for predicting mutagenicity of chemicals through machine learning algorithm | |
| CN107025387B (en) | Method for cancer biomarker identification | |
| CN108827909B (en) | Rapid soil classification method based on visible near infrared spectrum and multi-target fusion | |
| CN114611582A (en) | Method and system for analyzing substance concentration based on near infrared spectrum technology | |
| US7991223B2 (en) | Method for training of supervised prototype neural gas networks and their use in mass spectrometry | |
| CA2817220C (en) | Automatic chemical assay classification | |
| CN115586152A (en) | Intelligent spectrum quantitative analysis method based on convolutional neural network visualization | |
| CN117078960A (en) | Near infrared spectrum analysis method and system based on image feature extraction | |
| CN116864011A (en) | Colorectal cancer molecular marker identification method and system based on multiple sets of chemical data | |
| Khalilabad et al. | Fully automatic classification of breast cancer microarray images | |
| CN112330622B (en) | Hyperspectral image band selection method based on ground feature maximum discrimination | |
| CN114141316A (en) | Method and system for predicting biological toxicity of organic matters based on spectrogram analysis | |
| CN113469084B (en) | Hyperspectral image classification method based on contrast generation countermeasure network | |
| CN113138181B (en) | Method for grading quality of fen-flavor wine base | |
| CN115620818A (en) | Protein mass spectrum peptide fragment verification method based on natural language processing | |
| KR101708715B1 (en) | Device and method for analyzing gene expressing response data of model experiment for actual human response | |
| CN113791062A (en) | Method for judging fixed substance type based on Raman spectrum | |
| CN112735532A (en) | Metabolite identification system based on molecular fingerprint prediction and application method thereof | |
| Bidaut et al. | WaveRead: automatic measurement of relative gene expression levels from microarrays using wavelet 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 |