Amino acid chirality identification method based on artificial intelligence

Abstract

The invention provides an amino acid chiral identification method based on artificial intelligence, which comprises the following steps: mixing amino acid powder to be tested with excipient powder, and pressing into amino acid tablets; a terahertz time-domain spectrometer is adopted to emit terahertz waves to the amino acid tablet and receive feedback signals, and the feedback signals are normalized amplitude signals, transmission signals or absorption signals; inputting the feedback signal into a pre-trained identification model, wherein the identification model is a convolutional neural network model, a radial basis function neural network model or an extreme learning machine model; and determining the amino acid type and chiral result of the to-be-tested amino acid powder based on the output result of the identification model. According to the scheme, on the basis of terahertz signals, tiny differences of amino acid molecular structures can be reflected, the method is high in recognition accuracy, the scheme does not need to combine a chiral reagent with chiral molecules, and the processing efficiency is improved on the premise that the recognition accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of amino acid chirality identification, and in particular to an amino acid chirality identification method based on artificial intelligence. Background Art

[0002] Chirality refers to the asymmetry of an object. If an object cannot be translated, rotated, or aligned with its mirror image, then the object has chirality. Chirality is a common phenomenon in nature. Spiral galaxies in the sky, our human hands, and the double helix structure of DNA molecules all have chirality. Of the 20 amino acids that make up proteins in living organisms, all except glycine (Gly) have chirality. If amino acid molecules with different chirality are regarded as different amino acid molecules, the 20 amino acids that make up proteins in living organisms contain a total of 39 amino acid molecules.

[0003] A pair of chiral molecules that are mirror images of each other are called a pair of enantiomers. They are enantiomers of each other. In spectroscopy, "D" (Dexter) and "L" (Learus) are used to represent the optical activity of chiral media according to their different optical rotation effects. Therefore, enantiomers can also be distinguished by D and L. In a non-chiral environment, enantiomers usually exhibit the same physical and chemical properties, such as melting point, boiling point, density, etc. At this time, enantiomers are considered to be the same substance; but in a chiral environment, enantiomers usually exhibit different or even opposite properties. More importantly, in medical diagnosis and treatment, when an amino acid molecule forms a potent drug, its enantiomers may cause serious side effects.

[0004] Among the existing methods for identifying the chirality of amino acid molecules, chromatography is the most mature and has a high accuracy of identification. However, in chromatography, the identification of any chiral molecule requires the search for a chiral reagent, the reaction of the chiral reagent with the chiral molecule, and finally the completion of the identification of the molecular chirality. The steps are cumbersome, the cycle is long, and the efficiency is low. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides an amino acid chirality identification method based on artificial intelligence to eliminate or improve one or more defects existing in the prior art.

[0006] One aspect of the present invention provides an artificial intelligence-based method for identifying amino acid chirality, the method comprising the steps of:

[0007] The amino acid powder to be tested is mixed with the excipient powder and compressed into amino acid tablets;

[0008] A terahertz time-domain spectrometer is used to emit a terahertz wave to the amino acid pressed tablet, and a feedback signal is received, wherein the feedback signal is a normalized amplitude signal, a transmission signal or an absorption signal;

[0009] Inputting the feedback signal into a pre-trained identification model, wherein the identification model is a convolutional neural network model, a radial basis function neural network model or an extreme learning machine model;

[0010] The amino acid type and chirality results of the amino acid powder to be tested are determined based on the output results of the identification model.

[0011] The above scheme is adopted. In this scheme, the amino acid powder to be tested is prepared into amino acid tablets, and the terahertz signal of the amino acid tablets is obtained using a terahertz time domain test system. The terahertz signal is identified by an identification model to determine the amino acid type and chirality result. This scheme is based on the fact that the terahertz frequency can resonate with the amino acid molecules. The terahertz signal can reflect the subtle differences in the amino acid molecular structure. Compared with the visible spectrum and the infrared spectrum, the method of the present invention has a higher recognition accuracy. Furthermore, this scheme does not require the search for chiral reagents, nor does it require waiting for the chiral reagents to react and combine with the chiral molecules. On the premise of improving the recognition accuracy, the processing efficiency is improved.

[0012] In some embodiments of the present invention, in the step of mixing the amino acid powder to be tested with the excipient powder and compressing them into amino acid tablets:

[0013] Obtain amino acid powder and excipient powder to be tested, and grind the amino acid powder and excipient powder separately using an agate grinding mortar;

[0014] Filter the ground amino acid powder and excipient powder using a cell strainer with a pore size of 100 μm;

[0015] Take the filtered amino acid powder and excipient powder in a mass ratio of 1:5 and place them in a sterile centrifuge tube;

[0016] Use a vortex mixer to mix the amino acid powder and excipient powder until the two are fully mixed;

[0017] The mixed amino acid powder and excipient powder are placed in a mold and pressed at a pressure of 500 kg / cm 2 , press at room temperature for 60 seconds to obtain 1 mm thick amino acid tablets.

[0018] In some embodiments of the present invention, the excipient powder is a powder of polytetrafluoroethylene, polyethylene or polymethylpentene.

[0019] In some embodiments of the present invention, in the step of taking the filtered amino acid powder and the excipient powder in a mass ratio of 1:5 and placing them in a sterile centrifuge tube, 300 mg of the filtered amino acid powder and 1500 mg of the excipient powder are taken.

[0020] In some embodiments of the present invention, in the step of placing the mixed amino acid powder and excipient powder in a mold, 160 mg of the mixture of the mixed amino acid powder and excipient powder is placed in the mold.

[0021] In some embodiments of the present invention, in the step of inputting the feedback signal into the pre-trained identification model, if a normalized amplitude signal is used, it is calculated based on the following formula:

[0022]

[0023]

[0024] Wherein, A(ω) represents the normalized amplitude signal, f(t) represents the terahertz time domain signal obtained by the terahertz time domain spectrometer, and F(ω) represents the terahertz frequency domain signal obtained by Fourier transforming the time domain signal of the amino acid tablet.

[0025] In some embodiments of the present invention, in the step of inputting the feedback signal into the pre-trained identification model, if the transmission signal is used, it is calculated based on the following formula:

[0026]

[0027] Where T(ω) represents the transmission signal, F ref (ω) represents the terahertz frequency domain signal without amino acid pellets, F sam (ω) represents the terahertz frequency domain signal of the amino acid pellet.

[0028] In some embodiments of the present invention, in the step of inputting the feedback signal into the pre-trained identification model, if the absorption signal is used, it is calculated based on the following formula:

[0029]

[0030] Wherein, n(ω) represents the refractive index of the amino acid pressed sheet, d is the thickness of the amino acid pressed sheet, T(ω) represents the transmission signal, and α(ω) represents the absorption signal.

[0031] In some embodiments of the present invention, if the identification model adopts a convolutional neural network model, the convolutional neural network model includes an input layer, two convolutional layers, two ReLU activation layers and a pooling layer, and the convolution kernel sizes of the two convolutional layers are both 2×1, one convolutional layer has 16 convolution kernels, and the other convolutional layer has 32 convolution kernels; the pooling layer size is 2×1.

[0032] In some embodiments of the present invention, if the identification model adopts a radial basis function neural network model, the radial basis width is set to 100, and the hidden layer contains 1950 radial basis neurons.

[0033] In some embodiments of the present invention, if the identification model adopts an extreme learning machine model, the number of hidden layer nodes of the model is 10.

[0034] The second aspect of the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps implemented by the aforementioned artificial intelligence-based amino acid chirality identification method.

[0035] Additional advantages, purposes, and features of the present invention will be described in part in the following description, and will become apparent to those skilled in the art after studying the following, or may be learned from the practice of the present invention. The purposes and other advantages of the present invention can be specifically pointed out and obtained in the specification and the accompanying drawings.

[0036] Those skilled in the art will appreciate that the objectives and advantages that can be achieved with the present invention are not limited to the above specific description, and the above and other objectives that can be achieved by the present invention will be more clearly understood from the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The drawings described herein are used to provide a further understanding of the present invention, constitute a part of this application, and do not constitute a limitation of the present invention.

[0038] Figure 1 A schematic diagram of an embodiment of the method for identifying amino acid chirality based on artificial intelligence of the present invention;

[0039] Figure 2 A schematic diagram of the steps for preparing amino acid tablets according to the present invention;

[0040] Figure 3 It is a schematic diagram of the architecture of the present invention;

[0041] Figure 4 Schematic diagram of terahertz time domain signal;

[0042] Figure 5 is a schematic diagram of a normalized amplitude signal;

[0043] Figure 6 is a schematic diagram of transmission signal;

[0044] Figure 7 is a schematic diagram of the absorption signal;

[0045] Figure 8 is a schematic diagram of 20 normalized amplitude signals of the amino acid Leu;

[0046] Fig. 9 is a schematic diagram of the 20 transmission signals of the amino acid Leu;

[0047] Fig.10 is a schematic diagram of 20 absorption signals of amino acid Leu;

[0048] Fig.11 It is a structural diagram of the convolutional neural network model;

[0049] Fig.12 It is a structural diagram of the extreme learning machine model;

[0050] Fig.13 is a schematic diagram of the accuracy of amino acid classification;

[0051] Fig.14 Schematic diagram of the accuracy of amino acid chirality identification;

[0052] Fig.15 Schematic diagram of the time it takes to classify amino acids;

[0053] Fig.16 Schematic diagram of the time-consuming process of amino acid chirality identification. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments and the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0055] In the prior art, in medical diagnosis and treatment, when an amino acid molecule forms a potent drug, its enantiomers may cause serious side effects; for example, studies have shown that L-serine (L-Ser) plays a protective role in the treatment of nervous system diseases and neurodegenerative damage, and the D-serine (D-Ser) content in human urine and blood reflects the health level of human kidneys. If it is necessary to treat nervous system diseases, L-serine should be used instead of D-serine; L-arginine (L-Arg) can promote the secretion of multiple hormones, and D-arginine (D-Arg) treats physiological disorders caused by excessive release of growth hormone. If it is necessary to promote the secretion of body hormones, L-arginine should be used instead of D-arginine.

[0056] Amino acids are the cornerstones of protein synthesis and are the structural elements and energy sources necessary for living organisms to maintain life activities. Therefore, accurately and efficiently identifying the chirality of the 20 standard amino acid molecules (distinguishing the D and L forms of the same amino acid) has important scientific and practical value in food, materials and chemistry, especially in disease diagnosis and treatment. The Chinese, English, English abbreviations and chirality of the 20 standard amino acids that make up proteins are shown in Table 1.

[0057] Table 1

[0058] Serial number Amino Acids English and abbreviations Chirality 1 Alanine Alanine, Ala D-Ala, L-Ala 2 Arginine Arginine, Arg D-Arg, L-Arg 3 Aspartic acid Aspartic acid, Asp D-Asp, L-Asp 4 Asparagine Asparagine, Asn D-Asn, L-Asn 5 Cysteine Cysteine, Cys D-Cys, L-Cys 6 Glutamine Glutamine, Gln D-Gln, L-Gln 7 Glutamate Glutamic acid, Glu D-Glu, L-Glu 8 Glycine Glycine, Gly Gly, achiral 9 Histidine Histidine, His D-His, L-His 10 Isoleucine Isoleucine, Ile D-Ile, L-Ile 11 Leucine Leucine, Leu D-Leu, L-Leu 12 Lysine Lysine, Lys D-Lys, L-Lys 13 Methionine Methionine, Met D-Met, L-Met 14 Phenylalanine Phenylalanine, Phe D-Phe, L-Phe 15 Proline Proline, Pro D-Pro, L-Pro 16 Serine Serine, Ser D-Ser, L-Ser 17 Threonine Threonine, Thr D-Thr, L-Thr 18 Tryptophan Tryptophan, Trp D-Trp, L-Trp 19 Tyrosine Tyrosine, Tyr D-Tyr, L-Tyr 20 Valine Valine, Val D-Val, L-Val

[0059] Thanks to the explosive growth in computing power brought about by the rapid development of computer technology, artificial intelligence technologies represented by machine learning and deep learning have developed rapidly in recent years and have been widely used in computer vision, natural language processing and other fields, promoting profound changes in basic learning represented by physics, chemistry and biology. Compared with the human brain, artificial intelligence technology can make better use of massive data, learn and summarize data patterns from massive data and analyze subtle differences between data, which brings new opportunities and challenges to the identification of amino acid chirality.

[0060] The research on the chirality identification of amino acid molecules has a long history, and there are many methods for the chirality identification of amino acid molecules. Common methods for identifying the chirality of amino acid molecules can be summarized into four categories: chromatography, spectroscopy, mass spectrometry and sensor method.

[0061] Chromatography combines chiral reagents and enantiomers to form diastereomers with different physical and chemical properties, and uses the differences in the physical and chemical properties of diastereomers to achieve chiral identification and separation. In chromatography, a specific chiral reagent is required to combine with the enantiomer to form diastereomers. Generally speaking, a matching chiral reagent is required to identify an amino acid isomer.

[0062] Spectroscopic methods include circular dichroism (CD), nuclear magnetic resonance spectroscopy (NMRS), ultraviolet-visible-infrared spectroscopy, etc.

[0063] The principle of circular dichroism spectroscopy is that chiral substances absorb circularly polarized light of different rotation directions differently. By irradiating the chiral substance with left-handed circularly polarized light and right-handed circularly polarized light respectively, and obtaining the difference in the absorption of the chiral substance to left-handed circularly polarized light and right-handed circularly polarized light, information such as the chirality and three-dimensional spatial configuration of the substance can be obtained. Nuclear magnetic resonance spectroscopy: chiral reagents and enantiomers combine to form diastereomers, so that enantiomers with the same physicochemical properties become diastereomers with different physicochemical properties. The chemical shift or peak area of ​​the nuclear magnetic resonance spectrum of the diastereomer will change. Different chemical shift or peak area variables can be identified to achieve chirality identification. The spectra of chiral materials in the ultraviolet-visible-infrared band will be different. By identifying different characteristic peaks in the spectrum, the chirality of the substance can be identified.

[0064] Mass spectrometry (MA) uses electrons or other particles to bombard chiral molecules to convert them into charged gas-phase ions, and accelerate the charged gas-phase ions through a specific electric field or magnetic field. Gas-phase ions with the same mass-to-charge ratio experience the same amount of deflection under the action of the electric field or magnetic field. Different chiral molecules will have different mass-to-charge ratios when converted into gas-phase ions, and thus will experience different deflections in the electric field or magnetic field. The chiral molecules can be identified based on the different deflection displacements.

[0065] The sensor method uses chiral materials as sensors for chiral molecules. After the chiral materials and chiral molecules interact, some physical and chemical properties will change, such as the dielectric constant. These changes in physical and chemical properties can be converted into information outputs such as optical signals and electrical signals according to certain rules. By analyzing the changes in physical and chemical properties, chirality identification can be achieved.

[0066] Among the existing methods for identifying the chirality of amino acid molecules, chromatography is the most mature and has a high accuracy of identification. However, chromatography has the following disadvantages: 1. It requires a specific chiral reagent to be combined with the chiral molecule. If multiple chiral molecules need to be identified, multiple types of chiral reagents need to be found. There are mature chiral reagents for some chiral molecules, but for amino acid molecules, there are only a few chiral reagents. It is difficult to find chiral reagents corresponding to 39 kinds of amino acid chiral molecules. This method is not suitable for the identification of a large number of chiral molecules and cannot complete the identification of 39 kinds of amino acid chiral molecules; 2. The combination of chiral reagents and chiral molecules will destroy the original molecular structure of amino acid molecules; 3. The identification efficiency is low. The identification of any chiral molecule requires the search for chiral reagents, the reaction of chiral reagents with chiral molecules to convert into diastereomers, the separation of diastereomers, and the identification of chiral molecules. The steps are cumbersome and costly, especially the search for chiral reagents is very time-consuming.

[0067] Spectroscopy is usually used as a supplement to other chiral identification methods. In chromatography, circular dichroism spectroscopy requires the use of left-handed circularly polarized and right-handed circularly polarized light to illuminate chiral molecules respectively. Compared with linearly polarized light, the generation and detection of circularly polarized light is more complicated, the optical system components are numerous, and only some substances can be detected. There are currently no cases of using circular dichroism spectroscopy for large-scale chiral detection of amino acid molecules. Circular dichroism spectroscopy is usually used in combination with metasurface sensors. Circular dichroism spectroscopy is not suitable for chiral detection of multiple types of amino acid molecules. Nuclear magnetic resonance spectroscopy requires specific chiral reagents to combine with chiral molecules to form diastereomers. To identify 39 types of amino acid chiral molecules, 39 chiral reagents are required. Similar to chromatography, this severely limits the types of amino acid chiral molecules that can be identified by nuclear magnetic resonance spectroscopy. The identification accuracy of ultraviolet spectroscopy, visible spectroscopy, and infrared spectroscopy is low, and is greatly affected by the purity of the sample.

[0068] Mass spectrometry requires the use of high-speed particles such as electrons to bombard sample molecules, converting them into charged gas-phase ions, which will destroy the original molecular structure of chiral molecules. To identify all 39 amino acid isomers, 39 suitable high-speed particles need to be found, and the 39 amino acid gas-phase ions need to be matched with suitable electric or magnetic fields. The experiment has high time cost, high trial and error cost, and the mass spectrometer is expensive. The sensor method requires specific sensors to match chiral molecules, and different chiral molecules usually require different sensors. Taking the metasurface sensor as an example, once the sensor is used once, its performance is affected, and the cost of each metasurface sensor is relatively high.

[0069] In summary, if one wants to identify 39 amino acid enantiomers, the existing methods have the disadvantages of high cost, complicated identification process, and low identification accuracy. Based on this, the present invention proposes an efficient, convenient, fast, and low-cost amino acid chiral identification method: an amino acid chiral identification method based on artificial intelligence. Compared with chromatography, this method does not require 39 chiral reagents, which reduces costs; compared with spectroscopy, the accuracy of chiral identification is significantly improved without increasing costs; compared with mass spectrometry, this method does not require complex trial and error costs and time costs, nor does it require the purchase of expensive experimental equipment. This method applies artificial intelligence technology and terahertz time-domain spectroscopy technology to amino acid chiral identification. Compared with existing methods, this method has lower economic and time costs and higher chiral identification accuracy.

[0070] It should also be noted that, in order to avoid obscuring the present invention due to unnecessary details, only structures and / or processing steps closely related to the scheme of the present invention are shown in the drawings, while other details that are not closely related to the present invention are omitted.

[0071] like Figure 1 and Figure 3As shown, the present invention proposes an amino acid chirality identification method based on artificial intelligence, the steps of the method comprising:

[0072] Step S100, mixing the amino acid powder to be tested with the excipient powder, and pressing them into amino acid tablets;

[0073] In a specific implementation process, the excipient powder is a powder of polytetrafluoroethylene, polyethylene or polymethylpentene, preferably polytetrafluoroethylene.

[0074] In the specific implementation process, the amino acid tablets are round tablets.

[0075] Step S200, using a terahertz time-domain spectrometer to emit terahertz waves to the amino acid pressed tablet, and receiving a feedback signal, wherein the feedback signal is a normalized amplitude signal, a transmission signal or an absorption signal;

[0076] In the specific implementation process, terahertz (THz) waves refer to electromagnetic waves with a frequency between 0.1THz-10THz.

[0077] In the specific implementation process, the feedback signal is a reflection signal or a transmission signal of the amino acid tablet, and in this solution, it is preferably a transmission signal.

[0078] In the specific implementation process, since the water vapor in the air has a strong absorption effect on terahertz waves, before the test, the terahertz test system is first sealed, and then dry air or nitrogen is filled into the system to achieve the purpose of reducing the humidity of the test system. When the humidity of the system drops to less than 5%, the test can be started. During the entire test process, the humidity of the test system needs to be maintained below 5%, and dry air or nitrogen needs to be continuously filled. During the test, the amino acid tablet needs to be kept perpendicular to the terahertz wave vector. In order to obtain a large amount of data, an amino acid terahertz frequency domain feature database is constructed for training artificial intelligence models. On the premise of keeping the tablet plane and the terahertz wave vector perpendicular, the tablet can be slightly displaced in the vertical plane, so that the terahertz wave is irradiated at different positions on the tablet to obtain a large amount of data, reducing the number of tablets that need to be prepared.

[0079] When the humidity of the terahertz test system is lower than 5%, and no amino acid tablet is placed, the terahertz frequency domain signal of the terahertz test system is measured when it is unloaded, and marked as the reference signal, that is, Figure 4 The solid line in the middle indicates the reference; the amino acid tablet is placed, and the terahertz frequency domain signal of the amino acid tablet is measured, which is marked as the tablet signal, that is, Figure 4 The tablet is indicated by the dotted line.

[0080] Step S300, inputting the feedback signal into a pre-trained identification model, wherein the identification model is a convolutional neural network model, a radial basis function neural network model or an extreme learning machine model;

[0081] In some embodiments of the present invention, the signal value of the feedback signal is constructed as an input vector and input into a pre-trained identification model.

[0082] Step S400, determining the amino acid type and chirality of the amino acid powder to be tested based on the output result of the identification model.

[0083] The above scheme is adopted. In this scheme, the amino acid powder to be tested is prepared into amino acid tablets, and the terahertz signal of the amino acid tablets is obtained using a terahertz time domain test system. The terahertz signal is identified by an identification model to determine the amino acid type and chirality result. This scheme is based on the fact that the terahertz frequency can resonate with the amino acid molecules. The terahertz signal can reflect the subtle differences in the amino acid molecular structure. Compared with the visible spectrum and the infrared spectrum, the method of the present invention has a higher recognition accuracy. Furthermore, this scheme does not require the search for chiral reagents, nor does it require waiting for the chiral reagents to react and combine with the chiral molecules. On the premise of improving the recognition accuracy, the processing efficiency is improved.

[0084] like Figure 2 As shown, in some embodiments of the present invention, in the step of mixing the amino acid powder to be tested with the excipient powder and pressing them into amino acid tablets:

[0085] Step S110, obtaining amino acid powder and excipient powder to be tested, and grinding the amino acid powder and excipient powder respectively using an agate grinding mortar;

[0086] In the specific implementation process, when using an agate grinding mortar to grind the amino acid powder and the excipient powder separately, the agate grinding mortar must be strictly controlled to ensure the purity of the ground powder and prevent it from being contaminated by other substances.

[0087] Step S120, filtering the ground amino acid powder and excipient powder using a cell filter with a pore size of 100 μm;

[0088] Step S130, taking the filtered amino acid powder and excipient powder in a mass ratio of 1:5 and placing them in a sterile centrifuge tube;

[0089] In the specific implementation process, an analytical balance was used to weigh 300 mg of the ground and filtered amino acid powder and 1500 mg of polytetrafluoroethylene powder into sterile centrifuge tubes, respectively, and the mass ratio of the amino acid powder to polytetrafluoroethylene was 1:5.

[0090] By adopting the above scheme, the mass ratio of amino acid powder and polytetrafluoroethylene can be adjusted according to the integrity and smoothness of the tablets; the amino acid powder weighed is not fixed to 300 mg, and can be adjusted according to the capacity of the sterile centrifuge tube.

[0091] Step S140, using a vortex mixer to mix the amino acid powder and the excipient powder to make the two fully mixed;

[0092] Step S150, placing the mixed amino acid powder and excipient powder in a mold and pressing them at a pressure of 500 kg / cm 2 , press at room temperature for 60 seconds to obtain 1 mm thick amino acid tablets.

[0093] In some embodiments of the present invention, the excipient powder is a powder of polytetrafluoroethylene, polyethylene or polymethylpentene.

[0094] In some embodiments of the present invention, in the step of obtaining the amino acid powder and excipient powder to be tested, an analytical balance is used to weigh 320 mg of amino acid powder and 1610 mg of polytetrafluoroethylene, respectively. Since polytetrafluoroethylene has low absorption in the terahertz band and is almost transparent, polytetrafluoroethylene is preferably used as an excipient.

[0095] In some embodiments of the present invention, in the step of taking the filtered amino acid powder and the excipient powder in a mass ratio of 1:5 and placing them in a sterile centrifuge tube, 300 mg of the filtered amino acid powder and 1500 mg of the excipient powder are taken.

[0096] In some embodiments of the present invention, in the step of placing the mixed amino acid powder and excipient powder in a mold, 160 mg of the mixture of the amino acid powder and the excipient powder is placed in the mold.

[0097] In the specific implementation process, the unloaded terahertz frequency domain signal (reference signal) and the terahertz frequency domain signal with the pressed sheet (pressed sheet signal) obtained by measuring and processing the terahertz time domain spectrometer are further processed to obtain relevant frequency domain characteristic signals such as Figure 5 , Figure 6 or Figure 7 As shown. The acquired time domain signal is converted into the following formula:

[0098]

[0099] Perform Fourier transform to obtain the terahertz frequency domain signal of the sample. In the formula, f(t) is the original time domain signal of the sample tested by the terahertz time domain spectroscopy test system.

[0100] In some embodiments of the present invention, in the step of inputting the feedback signal into the pre-trained identification model, if a normalized amplitude signal is used, then the following formula is used to calculate:

[0101]

[0102] Where A(ω) represents the normalized amplitude signal. The normalized amplitude signal obtained by processing the D-leucine (D-Leu) measurement data is as follows: Figure 5 shown.

[0103] In some embodiments of the present invention, in the step of inputting the feedback signal into the pre-trained identification model, if the transmission signal is used, it is calculated based on the following formula:

[0104]

[0105] Where T(ω) represents the transmission signal, F ref (ω) represents the frequency domain signal without amino acid tablets, F sam (ω) represents the frequency domain signal of the amino acid tablet.

[0106] The transmission signal obtained by processing the measurement data of D-Leu is as follows Figure 6 shown.

[0107] In some embodiments of the present invention, in the step of inputting the feedback signal into the pre-trained identification model, if the absorption signal is used, it is calculated based on the following formula:

[0108]

[0109] Where n(ω) represents the refractive index of the amino acid tablet, d is the thickness of the amino acid tablet, T(ω) represents the transmission signal, α(ω) represents the absorption signal, and the absorption signal obtained by processing the measurement data of D-Leu is as follows: Figure 7 shown.

[0110] like Fig.11 As shown, in some embodiments of the present invention, if the identification model adopts a convolutional neural network model, the convolutional neural network model includes an input layer, two convolutional layers, two ReLU activation layers and a pooling layer, and the convolution kernel sizes of the two convolutional layers are both 2×1, one convolutional layer has 16 convolution kernels, and the other convolutional layer has 32 convolution kernels; the pooling layer size is 2×1.

[0111] In the specific implementation process, CNN, Convolutional Neural Network, is a type of feedforward neural network that includes convolution calculations and has a deep structure. It is one of the representative algorithms for deep learning. A convolutional neural network mainly includes five layers: data input layer, convolution operation layer, ReLU activation layer, pooling layer and fully connected layer; among them, the convolution operation layer mainly performs convolution operations on the data to extract local features of the data; the ReLU activation function layer mainly implements nonlinear transformations in neural networks and introduces nonlinearity in CNN; the pooling layer mainly implements feature dimensionality reduction, reduces the data scale, and preserves the most important features; the fully connected layer converts the feature matrix into a one-dimensional feature vector to achieve the connection output between data features and prediction results. Convolutional neural networks share convolution kernels, have no pressure on high-dimensional data processing, do not need to manually select features, have good feature classification effects, and are widely used in computer vision, natural language processing and other fields. In the method of the present invention, the constructed CNN model is as follows Fig.11 As shown in the figure, it contains an input layer, two convolutional layers, two ReLU activation layers, and a pooling layer. The convolution kernel size of the two convolutional layers is 2×1, convolutional layer 1 has 16 convolution kernels, and convolutional layer 2 has 32 convolution kernels; the pooling layer size is 2×1, and the pooling step size is [2, 1].

[0112] In some embodiments of the present invention, if the identification model adopts a radial basis function neural network model. RBFNN, Radial Basis Function Neural Network, is a single hidden layer, feedforward neural network based on function approximation. The core of RBFNN is to use the radial basis function as a mapping function to map the input data in a linearly inseparable low-dimensional space to a linearly separable high-dimensional space. The radial basis function expression is:

[0113]

[0114] In the formula, represents sample data, is the center, ||·|| represents a norm operation, generally using the Euclidean norm. When the distances from different sample data to the center are equal, the corresponding function values ​​are also equal. Usually the Gaussian kernel is used as the kernel function of the radial basis function. The mathematical expression of the Gaussian kernel is:

[0115]

[0116] Where σ represents the radial basis width. In the method of the present invention, Gaussian radial basis function is selected as the kernel function of RBFNN, the radial basis width is set to 100, and the hidden layer contains 1950 radial basis neurons. RBFNN has attracted great attention from researchers in various fields for its simple structure, strong nonlinear approximation ability and good generalization ability, and has been widely used in many research fields such as function approximation, speech recognition and automatic control.

[0117] like Fig.12 As shown, in some embodiments of the present invention, if the identification model adopts an extreme learning machine.

[0118] In the specific implementation process, ELM, Extreme Learning Machine, randomly selects the input layer weights and hidden layer biases, and then determines the output layer weights by minimizing the loss function. This loss function is usually composed of the training error term and the regularization term of the output layer weight norm, and is calculated and analyzed based on the Moore-Penrose (MP) generalized inverse matrix theory. Compared with traditional neural networks, ELM has a faster learning speed while ensuring accuracy. ELM has been widely used in pattern recognition, regression analysis, clustering algorithms and other fields. For a Fig.12 As shown in the single hidden layer neural network, assume that there are N arbitrary samples (X i , t i ), where X i =[x i1 ,x i2 ,…,x im ] T ∈R n , t i =[t i1 ,t i2 ,…,t im ] T ∈R n . For a single hidden layer neural network with L hidden layer nodes, it can be expressed as shown in the following formula:

[0119]

[0120] In the formula, g(x) is the activation function, W i =[w i,1 ,w i,2 ,…,w i,n ] T is the input weight, β i is the output weight, b i is the bias of the i-th hidden layer unit, W i ·X j W i and X jThe goal of learning a single hidden layer neural network is to minimize the output error, which can be expressed as:

[0121]

[0122] That is, there is β i , W i and b i Make

[0123]

[0124] It can be expressed as a matrix:

[0125] Hβ=T;

[0126] In the formula, H is the output of the hidden layer node, β is the output weight, T is the expected output, and the matrix form is as follows

[0127]

[0128] To train a single hidden layer neural network, we want to find and Make

[0129]

[0130] Among them, i = 1, ..., L, which is equivalent to minimizing the loss function

[0131]

[0132] In ELM, enter the weight W i and bias b i The output matrix H of the hidden layer is determined randomly, and the training of the single hidden layer neural network is transformed into solving a linear system Hβ = T. The output weight β can be determined:

[0133]

[0134] is the Moore-Penrose generalized inverse of the matrix H, and the solution is The norm of is the smallest and unique. In the method of the present invention, an ELM with 10 hidden layer nodes is constructed.

[0135] In the process of training the identification model, the measurement data of a certain amino acid at a certain time can be obtained through steps S100 and S200, including the normalized amplitude signal (Amp), transmission signal (Tra), and absorption signal (Abs) of the time domain signal and the frequency domain characteristic signal. By continuously repeating steps S100 and S200, multiple frequency domain characteristic signals of 20 amino acids and 39 enantiomers can be obtained. A tablet of leucine Leu (including D-Leu and L-Leu) is made and measured, and a total of 20 time domain signals are obtained, and 20 frequency domain characteristic signals Amp, Tra, and Abs are obtained. Figure 8 , Fig. 9 and Fig.10 Three frequency domain characteristic signals Amp, Tra, and Abs of the amino acid Leu are shown respectively. Among them, there are 20 frequency domain characteristic signals of each type, 10 for D-Leu and 10 for L-Leu, D-Leu is represented by a solid line, and L-leu is represented by a dotted line. By continuously repeating steps S100 and S200, multiple frequency domain characteristic signals of 20 amino acids and 39 enantiomers can be obtained, a training data set can be constructed, and the identification model can be trained.

[0136] In the specific implementation process, terahertz (THz) waves refer to electromagnetic waves with a frequency between 0.1THz and 10THz. Terahertz waves have the characteristics of wide coverage, low single photon energy, non-ionization and biomedical harmlessness. They have great applications in biomedicine, high-resolution imaging, non-destructive testing and other fields.

[0137] Since the vibration and rotation of amino acid molecules and the weak interactions between molecules are within the THz frequency band, terahertz waves can interact with amino acid molecules to produce resonance. The characteristic resonance peaks can reflect the subtle differences in the structure of amino acid molecules. By identifying such subtle differences, the classification and chiral identification of amino acid chiral molecules can be achieved.

[0138] Terahertz time-domain spectroscopy (THz-TDS) uses THz pulses to irradiate samples, detect the change of THz electric field intensity over time after reflection or transmission of the sample (obtain frequency domain information through Fourier transform), and obtain information such as the molecular structure of the sample by analyzing the reflection or transmission signal. Terahertz time-domain spectroscopy has the advantages of high signal-to-noise ratio, wide detection bandwidth and high detection sensitivity. By analyzing the broadband terahertz pulses carrying medium information, it can extract information such as the absorption coefficient and transmission of the medium material.

[0139] With the help of powerful computing and data processing capabilities of computers, artificial intelligence technology can find commonalities and subtle differences between data from massive data; there are subtle differences in the spectral signals of amino acid enantiomers, which are almost indistinguishable by the human eye, and artificial intelligence technology can summarize and generalize massive data and identify subtle differences between data. Therefore, the terahertz spectrum signals of a large number of amino acid molecules can be used to train artificial intelligence models and algorithms, and then the trained artificial intelligence models and algorithms can be used to analyze the terahertz spectrum signals, classify the terahertz spectrum signals of amino acid chiral molecules, and find subtle differences in the terahertz frequency domain signals of amino acid enantiomers, thereby realizing amino acid molecule classification and chirality identification.

[0140] This scheme prepares amino acids into tablets that are easy to test, uses THz-TDS to test amino acid tablets, obtains the time domain signal of the amino acid tablets, and then performs Fourier transform and other operations to obtain the terahertz spectrum information of the amino acid tablets. Finally, the trained artificial intelligence model is used to analyze the terahertz spectrum signal to identify the type of amino acids and the chirality of amino acids. This method does not require the assistance of any other chiral reagents and sensors, and overcomes the shortcomings of chromatography and sensor methods that require matching multiple chiral reagents or sensors; thanks to the unique frequency band of terahertz waves, terahertz waves resonate with amino acid molecules, and the transmitted (or reflected) terahertz signals can reflect the subtle differences in amino acid molecules. The chirality identification accuracy of this method is higher than that of visible spectroscopy and infrared spectroscopy; the THz-TDS used in this method to measure the terahertz signal of amino acid tablets has lower economic and time costs than nuclear magnetic resonance spectroscopy and mass spectrometry.

[0141] Experimental example

[0142] To verify the feasibility and effectiveness of the method of the present invention, we made 10 tablets for each amino acid enantiomer, measured different positions of each tablet 10 times, and obtained 100 time domain signals for each amino acid enantiomer. Through the data processing of step S200, each enantiomer had 100 frequency domain characteristic signals Amp, Tra and Abs. Therefore, 20 amino acids and 39 enantiomers obtained 3900 frequency domain characteristic signals Amp, Tra, and Abs, and a total of 3900×3=11700 data constituted the feature database.

[0143] When using CNN, RBFNN and ELM for amino acid classification and chirality identification, it is necessary to first train the model with the data in the feature database, and then use the trained model to predict the type and chirality of the amino acid. Each frequency domain feature signal of each enantiomer has 100 data, and these 100 numbers are divided into training data and test data in a 1:1 ratio; therefore, the scale of training and test data for each frequency domain feature signal is 1950:1950. During training and testing, each frequency domain feature signal is trained and tested separately, independently of each other and does not interfere with each other. When training and testing the artificial intelligence model, due to the random division of the data, there is a certain fluctuation in each training and test result. For this reason, multiple repeated experiments are required to verify the effectiveness of the method of the present invention.

[0144] Here, this scheme conducted 20 repeated experiments, recorded the test accuracy and total time of each experiment (including the time for training 1950 data and testing 1950 data), obtained the accuracy and time of three models CNN, RBFNN, and ELM using frequency domain feature signals Amp, Tra, and Abs for amino acid classification and chirality identification, and drew a box plot. Among them, Fig.13 The box plot shows the accuracy of amino acid classification (i.e., distinguishing an amino acid as one of the 20 standard amino acids). It can be seen from the figure that the constructed RBFNN model has the highest prediction accuracy using the Tra spectrum feature signal; for the same artificial intelligence model, the accuracy of using Tra to predict the type of amino acid is higher than that of the feature signals Amp and Abs. Fig.14 The box plots showing the accuracy of amino acid chirality identification show that, similar to classification, the constructed RBFNN model also has the highest accuracy in prediction using the Tra spectrum feature signal, and the accuracy of chirality identification using the feature signal Tra is higher than that using the feature signals Amp and Abs.

[0145] Affected by the computer configuration and operating status, the time taken for each training and test will fluctuate slightly. We conducted experiments while keeping the computer in a stable state as much as possible. The absolute time taken for the experiment is affected by the computer performance, but the relative relationship between the time taken for different models remains stable. The CNN model requires a series of operations such as convolution and pooling on the data, so it takes the longest time. RBFNN is a single hidden layer neural network that needs to find the best center point for each Gaussian kernel, so it takes less time than CNN. ELM only requires the Moore-Penrose generalized inverse of the hidden layer transfer matrix H to complete the training, so it takes the shortest time. Fig.15 and Fig.16 From the time statistics, whether it is amino acid classification or amino acid chirality identification, ELM takes less than 1 second, RBFNN takes about 6 seconds, and CNN takes about 11 seconds.

[0146] The beneficial effects of this program include;

[0147] 1. Chromatography and nuclear magnetic resonance spectroscopy require 39 chiral reagents that match amino acid isomers for identification, so it is difficult to achieve chiral identification of 39 amino acid isomers in a short time. The method of the present invention can identify 20 amino acids and 39 amino acid enantiomers at one time without the assistance of other chiral reagents, and is more feasible;

[0148] 2. The sensor method requires a specific sensor to identify the chirality of amino acids. If 39 enantiomers need to be identified, multiple sensors are required. The price of chiral sensors is high. The method of the present invention does not require complex sensors, has lower economic costs and higher efficiency.

[0149] 3. The method of the present invention uses a terahertz time domain test system to obtain the terahertz signal of the pressed tablet. Since the frequency of terahertz can resonate with the amino acid molecule, the terahertz signal can reflect the subtle differences in the structure of the amino acid molecule. Compared with the visible spectrum, infrared spectrum, etc., the recognition accuracy of the method of the present invention is higher;

[0150] 4. The method of the present invention uses three artificial intelligence models, ELM, RBFNN, and CNN, to identify the chirality of amino acids. Compared with the existing technology for identifying the chirality of amino acid molecules, the method of the present invention has a higher recognition accuracy and a shorter time consumption;

[0151] 5. When the method of the present invention trains and tests the artificial intelligence model, the ratio of training data to test data is 1:1. Compared with other artificial intelligence technologies, the demand for training data is lower, which means that the method of the present invention requires less training data to achieve the ideal effect, and less training data means lower economic costs;

[0152] 6. Mass spectrometry uses a mass spectrometer for testing and analysis. In mass spectrometry, the bombarding particles need to have appropriate kinetic energy to collide with the molecules to be tested and convert the molecules to be tested into charged gas-phase ions. It is also necessary to set up a suitable deflection electric field or magnetic field to distinguish the gas-phase ions. This requires a large number of experiments to obtain the strength of the deflection electric field (or magnetic field) and the kinetic energy of the bombarding particles. The field strength of the deflection electric field and the kinetic energy of the bombarding particles required for each amino acid are different, which means that the time cost of mass spectrometry is very high, while each test data of the method of the present invention is valid, and the method of the present invention requires less time for amino acid chirality identification.

[0153] The embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps implemented by the aforementioned amino acid chirality identification method based on artificial intelligence are implemented. The computer-readable storage medium can be a tangible storage medium, such as a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a floppy disk, a hard disk, a removable storage disk, a CD-ROM, or any other form of storage medium known in the technical field.

[0154] It should be understood by those skilled in the art that the exemplary components, systems and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software or a combination of the two. Whether it is performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present invention are programs or code segments used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link via a data signal carried in a carrier.

[0155] It should be clear that the present invention is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present invention.

[0156] In the present invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with features of other embodiments or replace features of other embodiments.

[0157] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the embodiments of the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An artificial intelligence-based amino acid chirality identification method, characterized in that: The steps of the method include: The amino acid powder to be tested is mixed with the excipient powder and compressed into amino acid tablets; A terahertz time-domain spectrometer is used to emit a terahertz wave to the amino acid pressed tablet, and a feedback signal is received, wherein the feedback signal is a normalized amplitude signal, a transmission signal or an absorption signal; Inputting the feedback signal into a pre-trained identification model, wherein the identification model is a convolutional neural network model, a radial basis function neural network model or an extreme learning machine model; The amino acid type and chirality results of the amino acid powder to be tested are determined based on the output results of the identification model.

2. The method for identifying amino acid chirality based on artificial intelligence according to claim 1, characterized in that: In the step of mixing the amino acid powder to be tested with the excipient powder and compressing them into amino acid tablets: Obtain amino acid powder and excipient powder to be tested, and grind the amino acid powder and excipient powder separately using an agate grinding mortar; Filter the ground amino acid powder and excipient powder using a cell strainer with a pore size of 100 μm; Take the filtered amino acid powder and excipient powder in a mass ratio of 1:5 and place them in a sterile centrifuge tube; Use a vortex mixer to mix the amino acid powder and excipient powder until the two are fully mixed; The mixed amino acid powder and excipient powder are placed in a mold and pressed at a pressure of 500 kg / cm 2 , press at room temperature for 60 seconds to obtain 1 mm thick amino acid tablets.

3. The method for identifying amino acid chirality based on artificial intelligence according to claim 2, characterized in that: The excipient powder is powder of polytetrafluoroethylene, polyethylene or polymethylpentene.

4. The method for identifying amino acid chirality based on artificial intelligence according to claim 2, characterized in that: In the step of taking the filtered amino acid powder and the excipient powder in a mass ratio of 1:5 and placing them in a sterile centrifuge tube, 300 mg of the filtered amino acid powder and 1500 mg of the excipient powder are taken.

5. The method for identifying amino acid chirality based on artificial intelligence according to claim 2, characterized in that: In the step of placing the mixed amino acid powder and excipient powder in a mold, 160 mg of the mixture of the mixed amino acid powder and excipient powder is placed in the mold.

6. The method for identifying amino acid chirality based on artificial intelligence according to claim 1, characterized in that: In the step of inputting the feedback signal into the pre-trained identification model, if a normalized amplitude signal is used, it is calculated based on the following formula: Wherein, A(ω) represents the normalized amplitude signal, f(t) represents the terahertz time domain signal obtained by the terahertz time domain spectrometer, and F(ω) represents the terahertz frequency domain signal obtained by Fourier transforming the time domain signal of the amino acid tablet.

7. The method for identifying amino acid chirality based on artificial intelligence according to claim 1, characterized in that: In the step of inputting the feedback signal into the pre-trained identification model, if the transmission signal is used, it is calculated based on the following formula: Where T(ω) represents the transmission signal, F ref (ω) represents the terahertz frequency domain signal without amino acid pellets, F sam (ω) represents the terahertz frequency domain signal of the amino acid pellet.

8. The method for identifying amino acid chirality based on artificial intelligence according to claim 1, characterized in that: In the step of inputting the feedback signal into the pre-trained identification model, if the absorption signal is used, it is calculated based on the following formula: Wherein, n(ω) represents the refractive index of the amino acid pressed sheet, d is the thickness of the amino acid pressed sheet, T(ω) represents the transmission signal, and α(ω) represents the absorption signal.

9. The method for identifying amino acid chirality based on artificial intelligence according to claim 1, characterized in that: If the identification model adopts a convolutional neural network model, the convolutional neural network model includes an input layer, two convolutional layers, two ReLU activation layers and a pooling layer. The convolution kernel sizes of the two convolutional layers are both 2×1, one convolutional layer has 16 convolution kernels, and the other convolutional layer has 32 convolution kernels; the pooling layer size is 2×1.

10. The method for identifying amino acid chirality based on artificial intelligence according to claim 1, characterized in that: If the identification model adopts a radial basis function neural network model, the radial basis width of the radial basis function neural network model is set to 100, and the hidden layer contains 1950 radial basis neurons; if the identification model adopts an extreme learning machine model, the number of hidden layer nodes of the extreme learning machine model is 10.

Images