Correction model transfer methods, electronic devices and storage media

Abstract

This invention discloses a calibration model transfer method, electronic device, and storage medium. The method includes acquiring three-dimensional fluorescence data of each sample in a calibration set and a spiked prediction set under L instruments; selecting three-dimensional fluorescence data of K samples in the calibration set under L instruments, combining the three-dimensional fluorescence data along the instrument dimensions to form four-dimensional data, and constructing a four-linear component model; based on the four-linear component model, transferring the three-dimensional fluorescence data of each sample in the calibration set under a certain instrument to a target instrument using a transfer formula; constructing a calibration model based on the transferred three-dimensional fluorescence data, and using the calibration model to predict the quantitative results of each sample in the spiked prediction set under the target instrument. This invention can achieve calibration model transfer of three-dimensional fluorescence data from multiple instruments with a single modeling step, enabling rapid and accurate transfer; when unknown interference is introduced into the samples of the calibration set, satisfactory quantitative results can be obtained by selecting appropriate components to eliminate the interference.

Description

Technical Field

[0001] This invention belongs to the field of model transfer technology, and particularly relates to a method for transferring calibration models of three-dimensional fluorescence data between multiple instruments, an electronic device, and a computer-readable storage medium. Background Technology

[0002] Excitation-emission matrix fluorescence (EX-EFL) is widely used in the food, environmental, and pharmaceutical fields due to its speed, sensitivity, and low cost. When target analytes and matrix interferences coexist in an analytical system, multidimensional calibration methods are required to obtain satisfactory quantitative results. Using mathematical separation instead of physical or chemical separation during the separation of target analytes and matrix interferences can save significant manpower and resources. For convenience, multiple instruments are often used in practice to acquire large amounts of three-dimensional fluorescence data. Due to differences between instruments, calibration models developed for one instrument are often not applicable to sample predictions for another instrument. However, developing a calibration model for each instrument is very time-consuming. Therefore, there is an urgent need to propose a method for transferring calibration models within three-dimensional fluorescence data.

[0003] Instrument standardization methods involve standardizing model coefficients, predicted values, and spectral responses using mathematical formulas. Model coefficient standardization can be applied to situations where there are significant spectral differences between samples; it involves recalculating the model coefficients using spectra measured on the secondary instrument, but no real information is transferred from the primary instrument to the secondary instrument.

[0004] One of the most widely used methods for correcting predicted values ​​is the simple univariate slope and bias correction (SBC), which works well when data from different instruments are small and systematic. Spectral response normalization is a widely used strategy, employing methods including univariate approaches such as single-wavelength normalization (SWS) and multivariate approaches such as direct normalization (DS) or piecewise direct normalization (PDS). DS or PDS finds a transition matrix that transforms the spectrum of a sample measured on a secondary instrument into the corresponding spectrum on the primary instrument, effectively making the spectrum equivalent to being measured on the primary instrument.

[0005] Recent studies have shown that correction transfer has been successfully applied to many instruments, including near-infrared spectroscopy (NIR), excitation-emission matrix fluorescence (EEM), liquid chromatography-mass spectrometry (LC-MS), high-performance liquid chromatography-diode array detector (HPLC-DAD), and nuclear magnetic resonance spectroscopy (NMR). For example, the parametric-free framework for near-infrared correction enhancement based on the correlation constraint method (J. Zhang, et al. Analytica Chimica Acta 1142 (2021) 169-178.) solves the problem of spectral inconsistency in near-infrared spectra under different environments and improves the predictive ability of the correction model; Chen et al. proposed Spectral Spatial Transformation (SST) (LMLZPChen, et al. The Analyst 136(1)(2011) p. 98-106.), which constructs a data matrix by aligning spectra measured on different instruments; Liu et al. proposed the idea of ​​standardizing near-infrared spectra measured by multiple instruments based on the Alternating Trilinear Decomposition (ATLD) algorithm (Y. Liu, et al. Analytica Chimica Acta 836 (2014) 18-23.); Malli et al. proposed several techniques for correction transfer in settings where transfer standards cannot be measured and only a few reference measurement data are available in changed measurement / environment / sample settings (B. Malli, et al. Chemometrics and Intelligent Laboratory Systems). 161(2017)49-60.); Thygesen et al. compared two-dimensional and three-dimensional correction transfer methods for three-dimensional fluorescence spectroscopy (J.Thygesen, et al. Analytica Chimica Acta 705(1-2)(2011)81-87.), finding that the PARAFAC model with only four transfer samples could achieve good results, but it was highly dependent on the selection of the transfer set, and pointed out that the performance of the three-dimensional correction transfer method was slightly better than that of the two-dimensional method; Sun et al. proposed a chemometrics-assisted correction transfer strategy (XDSun, et al. Chemometrics and Intelligent Laboratory Systems 194(2021.)) to quantify three agrochemicals in environmental samples using three-dimensional fluorescence technology; Valverde et al. proposed using multivariate correction transfer combined with partial least squares method to determine tetracycline in surface water (RSValverde, et al. Analytica Chimica Acta 705(1-2)(2011)81-87.), finding that the PARAFAC model with only four transfer samples could achieve good results, but it was highly dependent on the selection of the transfer set, and pointed out that the performance of the three-dimensional correction transfer method was slightly better than that of the two-dimensional method; 562(1)(2006)85-93.), to correct the effect of solid-phase preconcentration on photochemically induced fluorescence signal; Gu et al. modeled LC-MS data using the PDS-ATLD method (HWGu, et al.).(al. Journal of Chromatography A 1407(2015)157-168.), which solved the signal instability to maintain the second-order advantage of resolution and detected multiple analytes in complex systems; Chen et al. proposed the idea of ​​PDS-assisted second-order correction method to solve the signal instability in HPLC-DAD system (Y. Chen, et.al. Journal of Chromatography A 1667(2022)462851.); Vaughan et al. suggested the development of correction transfer model (AAVaughan, et.al. Analytical Chemistry 84(22)(2012)9848-9857.), and successfully mapped the response of one LC-MS instrument to another instrument, enabling them to merge data from different samples analyzed by different instruments; García et al. proposed PDS-assisted multivariate curve resolution-alternating least squares method and expanded partial least squares method (G. García, et.al. Journal of Chromatography A 1407(2015)157-168.). 1179(2)(2008)106-114.), then the residual bilinear decomposition algorithm was used to solve the matrix effect, and eight tetracycline antibiotics in the effluent were identified by solid-phase extraction; Lindner et al. first studied the partial least squares regression correction transfer between high-field (600MHz) NMR and benchtop NMR equipment (43 and 60MHz) (S.Lindner, et al. Analytical Chemistry 94(9)(2022)3997-4004.).

[0006] Most of the methods described above can only transfer calibration models from one instrument to another or from different situations. However, in practical applications, it may be necessary to transfer calibration models between more than two instruments. Methods for transferring calibration models directly for high-dimensional data are rare; these methods usually require converting the data into vectors. Furthermore, existing model transfer methods only include the target analyte in their calibration sets, without considering interference that may be introduced during the measurement process. Summary of the Invention

[0007] The purpose of this invention is to provide a calibration model transfer method, electronic device, and storage medium to solve the problem of calibration model transfer between two or more instruments, and to take into account the problem of interference that may be introduced during the measurement process.

[0008] This invention solves the above-mentioned technical problems through the following technical solution: a correction model transfer method, the method comprising the following steps:

[0009] Obtain the three-dimensional fluorescence data of each sample in the calibration set and the spiked prediction set using L instruments; where L≥3.

[0010] Three-dimensional fluorescence data of K samples from the calibration set were selected from L instruments, and a four-dimensional linear component model was constructed based on the three-dimensional fluorescence data along the four-dimensional data composed of the instrument dimensions.

[0011] The quadlinear component model is analyzed using a quadlinear decomposition algorithm. The three-dimensional fluorescence data of each sample in the calibration set, obtained from a certain instrument, is then transferred to the target instrument using a transfer formula. The specific expression of the transfer formula is as follows:

[0012] X pq =Adiag(d (q) )((Adiag(d (p) )) + X p (B T ) + B T

[0013] Among them, X pq d represents the three-dimensional fluorescence data obtained after transferring the three-dimensional fluorescence data of each sample in the sample set from the p-th instrument to the q-th target instrument; A is the normalized excitation spectral array; diag() is a function that constructs a diagonal matrix from the vectors, with all off-diagonal elements being 0; d (q) d (p) These are the vectors in the qth and pth rows of the relative instrument response intensity matrix D; X p To calibrate the three-dimensional fluorescence data of each sample under the p-th instrument; B is the normalized emission spectrum array;

[0014] A calibration model is constructed based on the three-dimensional fluorescence data obtained from the transfer, and the calibration model is used to predict the quantitative results of each sample in the spiked prediction set under the target instrument.

[0015] Furthermore, the specific implementation method for obtaining the three-dimensional fluorescence data of each sample in the calibration set and the spiked prediction set under L instruments is as follows:

[0016] Prepare samples for the calibration set and the spiked prediction set;

[0017] The three-dimensional fluorescence data of each sample in the calibration set and the spiked prediction set are measured using each instrument, thus obtaining the three-dimensional fluorescence data of each sample in the calibration set and the spiked prediction set under L instruments.

[0018] Furthermore, the specific implementation process for configuring each sample in the calibration set and the spiked prediction set is as follows:

[0019] Donepezil hydrochloride and trazodone hydrochloride were dissolved in methanol to obtain a stock solution; a certain amount of the stock solution was taken and diluted with methanol to 100 ng / mL. -1 The working solution is obtained;

[0020] Take 5 mL of plasma into a 50 mL centrifuge tube, add 10 mL of acetonitrile to remove protein; then sonicate the plasma for 30 min at 4000 rpm. -1 Centrifuge at a certain speed for 15 min; filter the centrifuged plasma sample through a 0.45 μm organic nylon membrane, and store the supernatant at 4 °C.

[0021] The sample concentrations are designed according to the uniform design table, and the samples in the calibration set are prepared using the working solution, and the samples in the spiking prediction set are prepared using the working solution and the supernatant.

[0022] Preferably, the concentration range of donepezil hydrochloride is 150-990 ng / mL. -1 The concentration range of the trazodone hydrochloride is 100-700 ng / mL. -1 .

[0023] Furthermore, the instrument is a fluorescence spectrometer, and the parameters of the fluorescence spectrometer are set as follows during the measurement:

[0024] The excitation wavelength is 250-350 nm, the emission wavelength is 300-550 nm, and the scan speed is 30000 nm min. -1 The detector voltage is 700V, and the excitation / emission slit width is 5nm / 5nm.

[0025] Furthermore, the specific implementation method for obtaining the three-dimensional fluorescence data of each sample in the calibration set and the spiked prediction set under L instruments is as follows:

[0026] The three-dimensional fluorescence data of each sample in the calibration set and the spiked prediction set were simulated using MATLAB under L instruments.

[0027] Furthermore, the specific implementation process of constructing the four-linear-component model is as follows:

[0028] The blank background and scattering of the three-dimensional fluorescence data are removed, and the processed three-dimensional fluorescence data are used to form a four-dimensional data array along the instrument dimensions;

[0029] The number of components in the four-linear component model is determined using the kernel consistency diagnostic method, and the number of components in the four-linear component model is increased.

[0030] The four-dimensional data array is decomposed using the AQLD algorithm to obtain the normalized excitation spectrum array A, the normalized emission spectrum array B, the normalized relative concentration array C, and the relative instrument response intensity array D.

[0031] Furthermore, the update formulas for the normalized excitation spectral array A, the normalized emission spectral array B, the normalized relative concentration array C, and the relative instrument response intensity array D are as follows:

[0032]

[0033]

[0034]

[0035]

[0036] Where I, J, K and L represent excitation wavelength, emission wavelength, number of samples and number of instruments, respectively; and Both are incompletely expanded matrices; diagm() extracts the diagonal elements of the matrix within the parentheses and generates a column vector; a (i) b (j) c (k) and d (l) These are the i, j, k, and l-th row vectors of matrices A, B, C, and D, respectively.

[0037] Furthermore, when interfering substances are added to the sample set measured by the r-th instrument, the specific expression of the transfer formula is:

[0038] X * r =X′ r -A′diag(d′)diag(x′)B′ T

[0039] X * rs =A * diag(d * (s) (A) * diag(d * (r) )) + X * r (B *T ) + B *T

[0040] Where, X′ rThe sample data consists of three-dimensional fluorescence data of each sample in the sample set after the addition of interfering substances, measured on the r-th instrument; A′ and B′ are the interfering substance signal matrices in the normalized excitation and emission spectral arrays, respectively; c′ is the interfering substance row vector in the normalized relative concentration array; d′ is the interfering substance row vector in the relative instrument response intensity array; X * r For X′ r Three-dimensional fluorescence data after removing interference response; A * B * These are the signal matrices corresponding to the target analyte after removing the column vectors containing interfering substances from the normalized excitation and emission spectral arrays, respectively; d * (s) The row vector corresponding to the s-th instrument after removing the row vector containing the interfering object from the relative instrument response intensity matrix; d * (r) The row vector corresponding to the r-th instrument after removing the row vector containing the interfering object from the relative instrument response intensity matrix; X * rs For X′ r The interference-free three-dimensional fluorescence data is transferred to the s-th target instrument to obtain the three-dimensional fluorescence data; the normalized excitation spectrum array, normalized emission spectrum array, normalized relative concentration array and relative instrument response intensity array are all obtained by decomposing the four-dimensional data array using the AQLD algorithm.

[0041] The present invention also provides an electronic device, which includes a processor and a memory, wherein the memory stores a computer program, and when the processor calls the computer program in the memory, it executes the steps in any of the correction model transfer methods provided by the present invention.

[0042] The present invention also provides a computer-readable storage medium having a computer program stored thereon, the computer program being loaded by a processor to perform the steps in the correction model transfer method.

[0043] Beneficial effects

[0044] Compared with the prior art, the advantages of the present invention are as follows:

[0045] The present invention provides a calibration model transfer method, electronic device, and storage medium. The method transfers three-dimensional fluorescence data measured by other instruments to the target instrument through a transfer formula. Then, the calibration model constructed using the transferred three-dimensional fluorescence data predicts the quantitative results of each sample in the spiked prediction set of the target instrument. The calibration model transfer of three-dimensional fluorescence data from multiple instruments can be achieved in one modeling process, and the transfer can be carried out quickly and accurately. When unknown interference is introduced into the samples of the calibration set, satisfactory quantitative results can be obtained by selecting appropriate components. Attached Figure Description

[0046] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only one embodiment of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a schematic diagram of the correction model transfer method in an embodiment of the present invention;

[0048] Figure 2 The diagram shows the decomposition of the simulation data analyzed by the ATLD algorithm in this embodiment of the invention, where (A) is the analytical normalized excitation spectrum profile, (B) is the analytical normalized emission spectrum profile, and (C) is the relative concentration curve.

[0049] Figure 3 The AQLD analysis results of the simulated data in the embodiments of the present invention are shown, where (A) is the analytical normalized excitation spectrum profile, (B) is the analytical normalized emission spectrum profile, (C) is the normalized relative concentration curve and (D) is the relative instrument response intensity curve.

[0050] Figure 4 The diagram shows the decomposition of real data analyzed by the ATLD algorithm in this embodiment of the invention, where (A) is the analytical normalized excitation spectrum profile, (B) is the analytical normalized emission spectrum profile, and (C) is the relative concentration curve.

[0051] Figure 5 The AQLD analysis results of real data in the embodiments of the present invention are shown, where (A) is the analytical normalized excitation spectrum profile, (B) is the analytical normalized emission spectrum profile, (C) is the normalized relative concentration curve and (D) is the relative instrument response intensity curve.

[0052] Figure 6 The above are contour plots of three-dimensional fluorescence data of sample C07 in this embodiment of the invention. (A), (B), and (C) are contour plots of three-dimensional fluorescence data obtained by the first, second, and third instruments, respectively. (D) is the contour plot of three-dimensional fluorescence data after the data is transferred from the second instrument to the first instrument. (E) is the contour plot of three-dimensional fluorescence data after the data is transferred from the third instrument to the first instrument.

[0053] Figure 7 Models 11 and 21 in the embodiments of the present invention CMT And Model 31 CMT Elliptical joint confidence interval plot of quantitative results, where (A) is donepezil hydrochloride and (B) is trazodone hydrochloride. Detailed Implementation

[0054] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] The technical solutions of this application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0056] Three-dimensional fluorescence spectroscopy, due to its advantages of high sensitivity, simplicity, speed, and online monitoring capabilities, has been widely applied in environmental, food, and other fields. Recent studies have shown that correction transfer has been successfully applied to many instruments, including near-infrared spectroscopy (NIR), excitation-emission matrix fluorescence (EEM), liquid chromatography-mass spectrometry (LC-MS), high-performance liquid chromatography-diode array detector (HPLC-DAD), and nuclear magnetic resonance spectroscopy (NMR). Currently, there is a lack of research on correction model transfer methods for three-dimensional fluorescence data across multiple instruments. Most methods only transfer the model from one instrument to another or from different situations, but in practical applications, correction model transfer between two or more instruments is usually required. Correction model transfer methods directly targeting high-dimensional data are rare; most of the data in these methods needs to be converted to vectors. Furthermore, existing model transfer methods only include the target analyte in their correction sets, without considering special cases that may be introduced by unknown interferences. Therefore, this invention proposes a correction model transfer method for three-dimensional fluorescence data across multiple instruments, and its performance is demonstrated through simulation of three-dimensional fluorescence data and plasma three-dimensional fluorescence data. Meanwhile, in order to address the issue of unknown interference that may be introduced into the calibration set during the measurement process, interference was artificially introduced into the calibration set that originally contained only the target analyte, in order to demonstrate the method's ability to resist the introduction of unknown interference.

[0057] Example 1

[0058] This embodiment provides a correction model transfer method, such as Figure 1 As shown, the method includes the following steps:

[0059] 1. Obtain the three-dimensional fluorescence data of each sample in the calibration set and the spiked prediction set under L instruments.

[0060] There are two specific ways to obtain the three-dimensional fluorescence data of each sample in the calibration set and the spiked prediction set under L instruments: one is to use instruments to measure each sample in the prepared calibration set and the spiked prediction set; the other is to use MATLAB to simulate the three-dimensional fluorescence data of each sample in the calibration set and the spiked prediction set under L instruments.

[0061] Specifically, the process of obtaining three-dimensional fluorescence data by measuring each sample in the prepared calibration set and the spiked prediction set using instruments is as follows:

[0062] 1.1 The specific process for configuring the calibration set and the spiked prediction set is as follows:

[0063] 1.11 Dissolve donepezil hydrochloride and trazodone hydrochloride in methanol to obtain a stock solution; take a certain amount of the stock solution and dilute it to 100 ng / mL with methanol. -1 The working solution is obtained;

[0064] 1.12 Take 5 mL of plasma into a 50 mL centrifuge tube, add 10 mL of acetonitrile to remove protein; then sonicate the plasma for 30 min at 4000 rpm. -1 Centrifuge at a certain speed for 15 min; filter the centrifuged plasma sample through a 0.45 μm organic nylon membrane, and store the supernatant at 4 °C.

[0065] 1.13 According to the uniform design table U7 * (7 4 The sample concentrations were designed, and samples in the calibration set were prepared using the working solution, as well as samples in the spiking prediction set using the working solution and the supernatant.

[0066] The stock solution is a high-concentration standard solution. Diluting the stock solution by a certain factor yields the working solution. Using the diluted working solution to prepare samples in the calibration set and the spiked prediction set can reduce experimental errors. The purpose of using a uniform design is to reduce signal collinearity.

[0067] 1.2 The three-dimensional fluorescence data of each sample in the calibration set and the spiked prediction set were measured using each instrument, thus obtaining the three-dimensional fluorescence data of each sample in the calibration set and the spiked prediction set under L instruments. The blank solution was ultrapure water, and measurements were taken three times before, during, and after the experiment to avoid the influence of solvent response and Raman scattering.

[0068] In this embodiment, the prepared sample sets include C01-C07, each containing donepezil hydrochloride and trazodone hydrochloride; the prepared spiking prediction sets include P01-P05, each containing 150 μL of processed plasma, donepezil hydrochloride, and trazodone hydrochloride. The concentrations of donepezil hydrochloride and trazodone hydrochloride in each sample of the sample sets and spiking prediction sets are set as shown in Table 1.

[0069] Table 1. Concentration Design of Calibration Set and Spiked Prediction Set in Plasma

[0070]

[0071] Samples marked with an asterisk (*) are selected as samples for model correction transfer.

[0072] In this embodiment, when L=3, the three-dimensional fluorescence data of each sample in the calibration set and the spiked prediction set are measured using the first instrument, the three-dimensional fluorescence data of each sample in the calibration set and the spiked prediction set are measured using the second instrument, and the three-dimensional fluorescence data of each sample in the calibration set and the spiked prediction set are measured using the third instrument.

[0073] In another specific embodiment of the present invention, the three-dimensional fluorescence data of each sample in the calibration set and the scald prediction set are simulated using MATLAB under three instruments, that is, the data is simulated using the peak function in MATLAB software. In this embodiment, the simulated data is shown in Table 2.

[0074] Table 2 MATLAB code for simulation data

[0075]

[0076] Notes on Table 2: Function y=peak(a,s,x,I); y=zeros(I,1); for i=1:I; y(i)=a*exp(-(ix).^2 / s.^2); end.

[0077] The peak function generates a Gaussian curve, represented by a column vector y, where a represents the height of the Gaussian curve, s represents the standard deviation of the Gaussian curve, x represents the mean of the Gaussian curve, and I represents the length of the Gaussian curve; A is the normalized excitation spectrum array, B is the normalized emission spectrum array, C is the normalized relative concentration array, and D is the relative instrument response intensity array.

[0078] In this embodiment, an F-7000 fluorescence spectrometer was used, equipped with a 150W xenon lamp and connected to a computer for data reading. During the measurement, the fluorescence spectrometer parameters were set as follows: excitation wavelength 250-350 nm (2 nm intervals), emission wavelength 300-550 nm (2 nm intervals), and scan speed 30000 nm / min. -1The detector voltage is 700V, and the excitation / emission slit width is 5nm / 5nm.

[0079] 2. Select three-dimensional fluorescence data of K samples from the calibration set under L instruments, and form a four-dimensional data matrix along the instrument dimension based on the three-dimensional fluorescence data. Then, construct a four-linear component model based on the four-dimensional data matrix.

[0080] In this embodiment, K=3. As shown in Table 1, the samples marked with * are selected as the samples for the calibration model transfer, that is, the three-dimensional fluorescence data of samples C01, C03, and C07 in the calibration set are selected under L instruments.

[0081] Based on the three-dimensional fluorescence data of samples C01, C03, and C07 in the calibration set under L instruments, a quadrature component model was constructed using the AQLD algorithm (alternating quadrature decomposition algorithm) which has a "second-order advantage" (see XDQing, et al. Chemometrics and Intelligent Laboratory Systems 132 (2014) 8-17.). That is, the AQLD algorithm was used to perform quadrature decomposition on the four-dimensional data matrix. The specific implementation process is as follows:

[0082] 2.1 Remove blank background and scattering from the three-dimensional fluorescence data, and form a four-dimensional data array from the processed three-dimensional fluorescence data along the instrument dimensions;

[0083] 2.2 The number of components in the four-linear component model is determined using the kernel consistency diagnostic method, and the number of components in the four-linear component model is increased to eliminate the introduced interference;

[0084] 2.3 The four-dimensional data array is decomposed using the AQLD algorithm to obtain the normalized excitation spectrum array A, the normalized emission spectrum array B, the normalized relative concentration array C, and the relative instrument response intensity array D.

[0085] 3. Based on the calibration model in step 2, the three-dimensional fluorescence data of each sample in the calibration set under a certain instrument are transferred to the target instrument using the transfer formula.

[0086] In this embodiment, the specific expression of the transfer formula is:

[0087] X pq =Adiag(d (q) )((Adiag(d (p) )) + X p (B T ) + B T (1)

[0088] Among them, X pqd represents the three-dimensional fluorescence data obtained after transferring the three-dimensional fluorescence data of each sample in the sample set from the p-th instrument to the q-th target instrument; A is the normalized excitation spectral array; diag() is a function that constructs a diagonal matrix from the vectors, with all off-diagonal elements being 0; d (q) d (p) These are the vectors in the qth and pth rows of the relative instrument response intensity matrix D; X p To calibrate the three-dimensional fluorescence data of each sample under the p-th instrument; B is the normalized emission spectrum array.

[0089] For example, when L=3, the transfer formula for transferring the three-dimensional fluorescence data of each sample in the calibration set from the second and third instruments to the first instrument (i.e., the target instrument) is as follows:

[0090] X 21 =Adiag(d (1) )((Adiag(d (2) )) + X2(B T ) + B T (2)

[0091] X 31 =Adiag(d (1) )((Adiag(d (3) )) + X3(B T ) + B T (3)

[0092] Where X2 and X3 are the three-dimensional fluorescence data of each sample in the sample set under the second and third instruments, respectively; d (1) d (2) and d (3) These are the vectors in the 1st, 2nd, and 3rd rows of the relative instrument response intensity matrix D, i.e., the relative instrument response intensity vectors of the 1st, 2nd, and 3rd instruments; X 21 X represents the three-dimensional fluorescence data obtained after transferring the three-dimensional fluorescence data of each sample in the sample set from the second instrument to the first instrument. 31 The three-dimensional fluorescence data of each sample in the sample set is obtained after the three-dimensional fluorescence data under the third instrument is transferred to the first instrument.

[0093] Before using the r-th instrument for measurement, interference is added to each sample in the calibration set to demonstrate that the method of this invention has the ability to resist unknown interference, that is, the introduction of unknown interference into the samples during subsequent instrument measurements will not affect the performance of the calibration model transfer method. When interference is added to the sample set measured by the r-th instrument, the specific expression of the transfer formula is:

[0094] X * r =X′ r -A′diag(d′)diag(c′)B′ T (4)

[0095] X * rs =A * diag(d * (s) (A) * diag(d * (r) )) + X * r (B *T ) + B *T (5)

[0096] Where, X′ r The sample data consists of three-dimensional fluorescence data of each sample in the sample set after the addition of interfering substances, measured on the r-th instrument; A′ and B′ are the interfering substance signal matrices in the normalized excitation and emission spectral arrays, respectively; c′ is the interfering substance row vector in the normalized relative concentration array; d′ is the interfering substance row vector in the relative instrument response intensity array; X * r For X′ r Three-dimensional fluorescence data after removing interference response; A * B * These are the signal matrices corresponding to the target analyte after removing the column vectors containing interfering substances from the normalized excitation spectral array and the normalized emission spectral array, respectively; d * (s) The row vector corresponding to the s-th instrument after removing the row vector containing the interfering object from the relative instrument response intensity matrix; d * (r) The row vector corresponding to the r-th instrument after removing the row vector containing the interfering object from the relative instrument response intensity matrix; X * rs For X′ r The three-dimensional fluorescence data obtained after transferring the interference-free three-dimensional fluorescence data to the target instrument is as follows: the normalized excitation spectrum array, normalized emission spectrum array, normalized relative concentration array, and relative instrument response intensity array are all obtained by decomposing the four-dimensional data according to the AQLD algorithm.

[0097] For example, before using a third instrument to measure each sample in the calibration set, the interfering agent arbutin is artificially added to each sample, and the transfer formula is:

[0098] X * 3=X′3-A′diag(d′)diag(c′)B′T (6)

[0099] X * 31 =A * diag(d * (1) (A) * diag(d * (3) )) + X * 3(B *T ) + B *T (7)

[0100] After data transfer, the ATLD algorithm (Alternating Trilinear Decomposition Algorithm) is used to decompose the transferred data to obtain A. CMT B CMT and C CMT The relative concentration (C) of the target analyte was obtained. CMT The concentrations and actual concentrations were used to construct regression curves to predict the concentrations of the target analyte in the spiked prediction set samples under the first instrument. Among them, A... CMT B CMT and C CMT These represent the normalized excitation spectrum matrix, normalized emission spectrum matrix, and relative concentration matrix obtained after the data decomposition following the transfer, respectively.

[0101] 4. Construct a calibration model based on the three-dimensional fluorescence data obtained from the transfer, and use the calibration model to predict the quantitative results of each sample in the spiked prediction set under the target instrument.

[0102] A calibration model was constructed from the converted three-dimensional fluorescence data. The kernel consistency diagnostic method was used to determine the number of calibration model components. Simultaneously, the number of calibration model components was increased to eliminate introduced interference. The ATLD algorithm was used to decompose the data, obtaining the normalized excitation spectrum, normalized emission spectrum, and relative concentration spectrum. Regression curves were constructed based on the obtained relative concentration information of the target analyte and the true concentration information to predict the quantitative results of spiked samples under the target instrument. For example, this can be used to obtain the quantitative results of donepezil hydrochloride and trazodone hydrochloride in plasma.

[0103] Example 2

[0104] The effectiveness and reliability of the correction model transfer method of the present invention will be illustrated by taking three instruments as an example.

[0105] 1. Experimental Instruments and Materials

[0106] Instruments: F-7000 fluorescence spectrometer, ultrasonic instrument, centrifuge, computer.

[0107] Materials: donepezil hydrochloride and trazodone hydrochloride (>98%), methanol (HPLC), plasma.

[0108] 2. Experimental Methods

[0109] 2.1 Parameter Settings

[0110] Three Hitachi F-7000 fluorescence spectrometers were used to collect three-dimensional fluorescence data. Each instrument was equipped with a 150W xenon lamp and connected to a computer for data reading. The instrument parameters were set as follows: excitation wavelength 250-350 nm (2 nm intervals), emission wavelength 300-550 nm (2 nm intervals), and scan speed 30,000 nm / min. -1 The detector voltage is 700V, and the excitation / emission slit width is 5nm / 5nm.

[0111] All data processing was performed on a computer running Windows 10. The AQLD algorithm, ATLD algorithm, and data transfer were implemented in MATLAB.

[0112] 2.2 Sample solution preparation method

[0113] The sample solution was prepared as shown in step 1 of Example 1.

[0114] 2.3 Model Classification

[0115] In this embodiment, three F-7000 fluorescence spectrometers (the first, second, and third) are used. A set of calibration samples and a set of spiked prediction samples obtained from the same fluorescence spectrometer can constitute one dataset. The three-dimensional fluorescence data obtained from the three instruments can respectively form three calibration models (models 11, 22, and 33). Using the calibration model constructed based on the three-dimensional fluorescence data obtained from the second or third instrument for prediction of the spiked prediction sample data obtained from the first instrument can be regarded as two calibration models (model 21 or model 31). At the same time, two new models are constructed according to the method proposed in this invention, which respectively include model 21 constructed from the three-dimensional fluorescence data after calibration and transfer in the second instrument and the spiked prediction sample data obtained in the first instrument. CMT Model 31 was constructed using the three-dimensional fluorescence data after correction and transfer from the third instrument and the spiked prediction set sample data obtained from the first instrument. CMT .

[0116] 2.4 Theory

[0117] 2.4.1 Quadlinear Component Model

[0118] Measuring K samples with an F-7000 fluorescence spectrometer yields three-dimensional fluorescence data of size I×J×K. I, J, and K represent the excitation wavelength, emission wavelength, and number of samples, respectively. Arranging the obtained three-dimensional fluorescence data along the instrument's dimension L forms a four-dimensional data array of size I×J×K×L. X . X Each element x in ijkl It can be represented as:

[0119]

[0120] Where i = 1, 2, ..., I; j = 1, 2, ..., J; k = 1, 2, ..., K; l = 1, 2, ..., L; N is the number of components, including target analytes, unknown interferences, and even noise. I It is a four-dimensional superdiagonal kernel data array of size N×N×N×N, with the superdiagonal elements being 1 and the other elements being 0. A I×N B J×N C K×N and D L×N They are respectively X The normalized excitation spectrum array, normalized emission spectrum array, normalized relative concentration array, and instrument relative response intensity array. E It is a four-dimensional residual array of size I×J×K×L.

[0121] 2.4.2 Alternating Quadlinear Decomposition Algorithm

[0122] The Alternating Quadlinear Decomposition (AQLD) algorithm was proposed by Qing et al. (XDQing, et al. Chemometrics and Intelligent Laboratory Systems 132 (2014) 8-17.). It has the fastest convergence speed among similar iterative quadlinear decomposition algorithms. Its objective function is established based on the incomplete extended matrix form of the quadlinear model. The update formulas for A, B, C, and D are as follows:

[0123]

[0124]

[0125]

[0126]

[0127] Where I, J, K, and L represent excitation wavelength, emission wavelength, number of samples, and number of instruments, respectively; A, B, C, and D represent normalized excitation spectral array, normalized emission spectral array, normalized relative concentration array, and instrument relative response intensity array, respectively. and It is obtained through the following two steps: First, X I×J×K×L Expanding along different dimensions yields four different three-dimensional data matrices. X J×KL×I , X K×LI×J , X L×IJ×K and X I×JK×L Then extract them separately. X J×KL×I The i-th front slice matrix, X K×LI×J The j-th front slice matrix, X L×IJ×K The k-th front slice matrix and X I×JK×L The l-th front slice matrix yields the incompletely expanded matrix described above; diagm() extracts the diagonal elements of the matrix within the parentheses and generates a column vector; a (i) b (j) c (k) and d (l) These are the i, j, k, and l-th row vectors of matrices A, B, C, and D, respectively. is the Khatri-Rao product; + denotes the Moore-Penrose inverse of the matrix; T is the transpose operation. The AQLD algorithm can be seen as an extension of the ATLD algorithm in four-dimensional form. Its computation is essentially based on slice matrices, and therefore it also has the advantages of the ATLD algorithm, such as "second-order advantage" and fast convergence speed. It is not sensitive to too many components and initial values ​​(HLWu, et al. Journal of the Chemometrics 12(1)(1998)1–26.). However, some studies have also shown that data with high noise levels are more likely to affect the accuracy of the algorithm.

[0128] 2.4.3 Correction Model

[0129] Three-dimensional fluorescence data (X-ray fluorescence) of three calibration sets (transfer subsets) were obtained using three instruments. L1 X L2 and X L3The three-dimensional fluorescence data arrays obtained from the three instruments are formed into a four-dimensional data array along the instrument dimensions. Then, the four-dimensional data array is decomposed using the AQLD algorithm with "second-order advantage". The interference introduced is eliminated by increasing the number of components, resulting in a normalized excitation spectrum array A, a normalized emission spectrum array B, a normalized relative concentration array C, and a relative instrument response intensity array D. Combining the four matrices obtained from the above decomposition, all three-dimensional fluorescence data of the calibration set under the second and third instruments are transferred to the first instrument using formulas (2) and (3) in Example 1. Then, the transferred data is decomposed using the ATLD algorithm to obtain A. CMT B CMT and C CMT The relative concentration (C) of the target analyte was obtained. CMT The actual concentrations were used to construct regression curves to predict the concentrations of the target analytes in the spiked prediction set of the first instrument.

[0130] The method of this invention can standardize the three-dimensional fluorescence data of multiple instruments through a single modeling process. In addition, when unknown interference is introduced during the measurement of the calibration sample by the third instrument, the method can eliminate the interference by selecting more components during the modeling process, thereby not affecting the effect of the calibration model transfer.

[0131] 2.5 Results

[0132] 2.5.1 Simulation Data

[0133] 2.5.1.1 Three-dimensional fluorescence spectroscopy analysis

[0134] In previous model transfer studies, the calibration set contained only the target analyte and no interference. However, when unknown interferences are introduced into the calibration set during model transfer, conventional calibration methods may fail to yield satisfactory results. To address this issue, this invention designs a calibration set for simulated data (i.e., data generated by MATLAB simulations) that includes two target analytes and one interference, while the spiked prediction set includes the target analyte and the actual matrix. In simulated fluorescence data, the spectra of the target analyte overlap with those of the unknown interference and the actual matrix, potentially leading to unsatisfactory results from conventional calibration methods.

[0135] 2.5.1.2 Quantitative Analysis of Simulation Data

[0136] In cases where complex matrices and target analytes coexist and their spectra overlap, the ATLD algorithm is used to decompose three-dimensional fluorescence data and extract normalized excitation, emission, and relative concentration spectra. Figure 2It can be seen that the spectra of the target analyte overlap with those of the interfering substances, and the actual spectrum of the target analyte is consistent with the spectrum obtained from decomposition. In Tables 3 and 4, the average recoveries of the two target analytes in models 11, 22, and 33 are close to 100%, and the quantitative results are satisfactory, resolving the problem of spectral overlap between the target analyte and the matrix. However, the models established in the second and third instruments are no longer applicable to the prediction of the spiked prediction set samples under the first instrument. The average recoveries and root mean square errors of prediction in models 21 and 31 both deviate from the normal range, and cannot accurately predict the quantitative results of the spiked prediction set samples under the first instrument.

[0137] Table 3. Prediction results of simulation data under different models

[0138]

[0139] Notes to Table 3:

[0140] Where RMSEP is the root mean square error of prediction, N p It is the number of predicted samples. It is the predicted concentration of the analyte, c n This is the actual concentration of the analyte.

[0141] Table 4. Prediction results of two components in the simulation data analyzed by the ATLD algorithm.

[0142]

[0143] 2.5.1.3 Spectral transfer measured on different instruments

[0144] The AQLD algorithm is used to process a four-dimensional data array of size I×J×K×L to obtain the normalized excitation spectrum array A, the normalized emission spectrum array B, the normalized relative concentration array C, and the relative instrument response intensity array D. The spectral intensity differences between the 1st, 2nd, and 3rd instruments are shown below. Figure 3 As shown in (D), there are obvious differences among the three instruments. The spectral intensity of the second instrument is lower than that of the first and third instruments, while the spectral intensities of the first and third instruments are similar. To further demonstrate the performance of the method of this invention in accurately quantifying interference introduced during measurements by other instruments during model transfer, interference was mixed into the calibration set samples when the third instrument measured the calibration set samples in the simulation experiment. Using the AQLD algorithm with "second-order dominance," the interference was eliminated by selecting more components, thus not affecting the performance of the calibration model transfer.

[0145] Figure 3 (A) Figure 3 (B) and Figure 3(C) Shows the normalized excitation spectrum profile, normalized emission spectrum profile, and relative concentration plot after corrected transfer using the ATLD algorithm. Quantitative results, shown in Table 3, obtained through corrected model transfer of three transfer subsets, demonstrate satisfactory average recoveries, indicating good transfer from instruments 2 and 3 to instrument 1. However, comparisons with models 11 and 21 show... CMT and 31 CMT The quantitative results obtained show that all models yield satisfactory results, and compared to models 21 and 31, the calibration model transfer method improves the predictive power of the calibration models. The results clearly demonstrate that spectra measured on any one of the three instruments can handle the model transfer problem well (from 7 calibration samples to 3 calibration samples), and acceptable quantitative results can still be obtained while reducing experimental workload. Furthermore, by fully utilizing the "second-order advantage" of the AQLD algorithm and selecting appropriate component numbers, satisfactory quantitative results can also be obtained when unknown disturbances are introduced during the calibration sample measurement process.

[0146] 2.5.2 Real Data

[0147] 2.5.2.1 Three-dimensional fluorescence spectroscopy analysis

[0148] Rayleigh scattering in fluorescence spectra is typically removed using interpolation. The fluorescence spectra of donepezil hydrochloride and trazodone hydrochloride overlap with those of the plasma matrix. If conventional analytical methods such as fluorescence spectroscopy, chromatography, spectrophotometry, and mass spectrometry are used, the experiments may require significant time, manpower, and resources for plasma sample processing. This processing may also result in the loss of donepezil hydrochloride and trazodone hydrochloride, ultimately affecting the measurement results. Considering these factors, in cases where the target analyte and the actual sample matrix significantly overlap, the ATLD algorithm, with its "second-order advantage," can be used to accurately quantify the content of donepezil hydrochloride and trazodone hydrochloride in plasma samples.

[0149] 2.5.2.2 Grouping fractions

[0150] Nuclear concordance diagnostics were used to estimate the group number. When the group number N is greater than 4, the nuclear concordance diagnostic curve drops sharply. Therefore, the group number was chosen to be 4 in this experiment, including two target analytes, matrix interference, and artificial interference; the artificial interference was arbutin, which was added during the determination of the calibration set samples on the third instrument.

[0151] 2.5.2.3 Quantitative Analysis of Real Data

[0152] Although the spectra of donepezil hydrochloride and trazodone hydrochloride overlap with those of human plasma matrix, the ATLD algorithm can be used to accurately quantify donepezil hydrochloride and trazodone hydrochloride in plasma samples. Normalized excitation spectral profiles, normalized emission spectral profiles, and relative concentration plots are shown below. Figure 4 As shown, the relative concentrations of donepezil hydrochloride and trazodone hydrochloride obtained were regressed against their corresponding actual concentrations. Then, the relative concentrations of donepezil hydrochloride and trazodone hydrochloride in the spiked prediction set samples were substituted into the regression formula. Quantitative results of donepezil hydrochloride and trazodone hydrochloride in the spiked prediction set samples were obtained in models 11, 22, and 33, as shown in Tables 5 and 6. The mean recovery ± standard deviation in model 22 was 95.7 ± 2.0% and 101.3 ± 10.7%, respectively. The models established by the second and third instruments were no longer applicable to the prediction of the spiked prediction samples measured by the first instrument. This can be seen from the results of models 21 and 31, which deviated significantly from the 100% mean recovery rate, and their root mean square error of prediction was also much greater than that of model 11.

[0153] Table 5. Prediction results of plasma data under different models

[0154]

[0155] Table 6. Prediction results of plasma analysis using the ATLD algorithm

[0156]

[0157] Notes to Table 6:

[0158] SEN n =m n {[(A exp T (IA unx A unx T A exp )*(B exp T (IB unx B unx T B exp )] -1} nn -1 / 2 (14)

[0159] Among them, SEN n For sensitivity, A and B represent the qualitative contour matrices obtained from the decomposition, * denotes the Hadamard product, and m nThis represents the total analyte signal at a unit concentration of the target analyte; the subscripts nn, exp, and unx represent the (n, n) diagonal element, predicted component, and uncorrected component of the qualitative profile matrix, respectively, and I represents an identity matrix.

[0160] 2.5.2.4 Spectral transfer measured on different instruments

[0161] To address the issue of transferring three-dimensional fluorescence spectral data from different instruments, this invention proposes a correction model transfer method, reducing the heavy experimental workload and cost. The spectral intensity differences between the first, second, and third instruments are as follows: Figure 5 As shown in (D). To further demonstrate the performance of this method in accurately quantifying interference introduced during other instrument measurements during model transfer, arbutin, an interfering substance, was mixed into each sample of the calibration set when the calibration set samples were measured by the third instrument. By using the AQLD algorithm, which has a "second-order advantage", more components were selected to eliminate the interference, thus not affecting the performance of the calibration model transfer.

[0162] Figure 5 (A) Figure 5 (B) and Figure 5 (C) Displays the normalized excitation spectrum profile, normalized emission spectrum profile, and relative concentration plot. Transfer is performed using a calibration model across three transfer subsets. Figure 6 The three-dimensional fluorescence spectra of sample C07 in the calibration set were obtained at the first instrument, after being transferred from the second instrument to the first instrument, and after being transferred from the third instrument to the first instrument. The correlation coefficients of the fluorescence spectra of sample C07 before and after the transfer were calculated, and all were greater than 0.99, indicating that the spectral data transfer results were satisfactory. Model 21 CMT and 31 CMT The quantitative results are shown in Table 5. The average recovery rate and root mean square error of prediction are better than those of models 21 and 31, indicating a good transfer effect from the second or third instrument to the first instrument. (Models 11 and 21) CMT and 31 CMT The elliptic joint confidence interval plot of the quantitative results is shown below. Figure 7 As shown, their center points are all within the elliptical joint confidence interval, indicating that models 11 and 21... CMT and 31 CMTThe quantitative results showed no significant difference, demonstrating that the proposed method can solve the calibration model transfer problem in multi-instrument 3D fluorescence data, enabling accurate quantification of target analytes in complex systems. The results clearly show that the calibration model transfer method in multi-instrument systems can significantly improve the predictive ability of the calibration model, achieving acceptable quantitative results even with a reduced number of calibration set samples (from 7 to 3). Furthermore, when unknown interferences are introduced during the measurement of the calibration set samples, satisfactory quantitative results are obtained by selecting a strategy with more components, thanks to the algorithm's "second-order advantage."

[0163] In summary, this invention proposes a calibration model transfer method based on the Alternating Quadrlinear Algorithm (AQLD) for quantitative analysis of target analytes in complex systems. This method can be applied to multiple 3D fluorescence data in a single analysis, and both simulated and real data demonstrate its advantages. In these data sets where the 3D fluorescence spectral shapes of the target analytes do not vary significantly, the proposed method can achieve calibration model transfer for multiple instruments' 3D fluorescence data with only one modeling step. Furthermore, when unknown interferences are introduced during the measurement of the calibration set samples, satisfactory quantitative results can be obtained by selecting appropriate component numbers and then reconstructing the data to eliminate interference. This is due to the "second-order advantage" provided by the proposed method. This invention can be extended to the calibration model transfer of high-order data; moreover, the method is fast, flexible, and meets the requirements of green analytical chemistry.

[0164] The above description only discloses specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or modifications that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method of correcting model transfer, characterized by, The method comprises the following steps: Obtaining three-dimensional fluorescence data of each sample in the calibration set and the spiked prediction set under L instruments; wherein, L≥3; Selecting three-dimensional fluorescence data of K samples in the calibration set under L instruments, and constructing a four-linear component model according to four-dimensional data formed by the three-dimensional fluorescence data along the instrument dimension; Using a four-linear decomposition algorithm to analyze the four-linear component model, and using a transfer formula to transfer three-dimensional fluorescence data of each sample in the calibration set under a certain instrument to a target instrument; the specific expression of the transfer formula is: wherein, is the three-dimensional fluorescence data of each sample in the calibration set under the pth instrument after being transferred to the qth target instrument; is the normalized excitation spectrum matrix; is the function of constructing a vector into a diagonal matrix, and the elements on the off-diagonal line are all 0; , are the qth and pth row vectors of the relative instrument response intensity matrix D respectively; is the three-dimensional fluorescence data of each sample in the calibration set under the pth instrument; is the normalized emission spectrum matrix; Constructing a calibration model according to the three-dimensional fluorescence data obtained by transfer, and predicting quantitative results of each sample in the spiked prediction set under the target instrument by using the calibration model.

2. The calibration model transfer method of claim 1, wherein, The specific implementation mode of obtaining three-dimensional fluorescence data of each sample in the calibration set and the spiked prediction set under L instruments is: Preparing each sample in the calibration set and the spiked prediction set; Using each instrument to measure three-dimensional fluorescence data of each sample in the calibration set and the spiked prediction set, so as to obtain three-dimensional fluorescence data of each sample in the calibration set and the spiked prediction set under L instruments.

3. The method of claim 2, wherein, The specific implementation process of preparing each sample in the calibration set and the spiked prediction set is: Dissolve donepezil hydrochloride and trazodone hydrochloride in methanol to obtain a stock solution; take a certain amount of the stock solution and dilute with methanol to 100 ng mL -1 to obtain a working solution; Take 5 mL plasma into 50 mL centrifuge tube, add 10 mL acetonitrile to remove protein; then ultrasonically treat the plasma for 30 min and centrifuge at 4000 r min -1 speed for 15 min; filter the centrifuged plasma sample with 0.45 μm organic nylon membrane, and store the obtained supernatant at 4 ℃; Designing sample concentrations according to a uniform design table, and preparing samples in the calibration set by using the working solution, and preparing samples in the spiked prediction set by using the working solution and the supernatant.

4. The correction model transfer method according to claim 3, wherein The concentration of the donepezil hydrochloride ranges from 150 to 990 ng mL -1 The concentration of the trazodone hydrochloride ranges from 100 to 700 ng mL -1 .

5. The correction model transfer method according to any one of claims 1 to 4, characterized by, The instrument is a fluorescence spectrometer, and the parameter settings of the fluorescence spectrometer during measurement are: Excitation wavelength 250-350 nm, emission wavelength 300-550 nm, scan speed 30000 nm min -1 , detector voltage 700 V, excitation / emission slit width 5 nm / 5 nm.

6. The calibration model transfer method according to claim 1, characterized in that, The specific implementation mode of obtaining three-dimensional fluorescence data of each sample in the calibration set and the spiked prediction set under L instruments is: using MATLAB to simulate three-dimensional fluorescence data of each sample in the calibration set and the spiked prediction set under L instruments.

7. The method of claim 1, wherein, The specific implementation process of constructing the four-linear component model is: Removing blank background and scattering of the three-dimensional fluorescence data, and forming a four-dimensional data array from the processed three-dimensional fluorescence data along the instrument dimension; Determining the component number of the four-linear component model by using a kernel consistency diagnosis method, and increasing the component number of the four-linear component model; Using an AQLD algorithm to decompose the four-dimensional data array, so as to obtain a normalized excitation spectrum array A, a normalized emission spectrum array B, a normalized relative concentration array C, and a relative instrument response intensity array D.

8. The correction model transfer method according to claim 7, wherein, The update formulae of the normalized excitation spectrum array A, the normalized emission spectrum array B, the normalized relative concentration array C, and the relative instrument response intensity array D are: where I, J, K and L represent excitation wavelength, emission wavelength, number of samples and number of instruments, respectively; , , and are incomplete expanded matrices; diagm() indicates extracting diagonal elements of the matrix in the parentheses and generating a column vector; a （i） , b （j） , c （k） and d （l） are the i, j, k and l row vectors of A, B, C and D matrices, respectively.

9. The correction model transfer method of claim 1, wherein, When adding an interferent to the sample set measured on the rth instrument, the specific expression of the transfer formula is: in, The three-dimensional fluorescence data of each sample in the sample set after adding interfering substances, under the r-th instrument; , These are the interference signal matrices in the normalized excitation spectral array and the normalized emission spectral array, respectively. The row vector of interfering substances in the normalized relative concentration matrix; The vector representing the row of interfering objects in the relative instrument response intensity array; To Three-dimensional fluorescence data after removing interference response; , These are the signal matrices corresponding to the target analyte after removing the column vectors containing interfering substances from the normalized excitation spectral array and the normalized emission spectral array, respectively. The row vector corresponding to the s-th instrument after removing the row vector containing the interfering object from the relative instrument response intensity matrix; The row vector corresponding to the r-th instrument after removing the row vector containing the interfering object from the relative instrument response intensity matrix; To The three-dimensional fluorescence data obtained after removing interference response is transferred to the s-th target instrument; the normalized excitation spectrum array, normalized emission spectrum array, normalized relative concentration array and relative instrument response intensity array are all obtained by decomposing the four-dimensional data array using the AQLD algorithm.

10. An electronic device, comprising: The electronic device comprises a processor and a memory, the memory stores a computer program, and the processor invokes the computer program in the memory to execute the steps in any one of the calibration model transfer methods provided in claims 1-9.

11. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is loaded by the processor to execute the steps in the calibration model transfer method in any one of claims 1-9.

Images