WO2020258515A1 - 多柱连续流层析设计及分析的方法 - Google Patents
多柱连续流层析设计及分析的方法 Download PDFInfo
- Publication number
- WO2020258515A1 WO2020258515A1 PCT/CN2019/104488 CN2019104488W WO2020258515A1 WO 2020258515 A1 WO2020258515 A1 WO 2020258515A1 CN 2019104488 W CN2019104488 W CN 2019104488W WO 2020258515 A1 WO2020258515 A1 WO 2020258515A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- continuous flow
- chromatography
- flow chromatography
- parameters
- model
- Prior art date
Links
- 238000004587 chromatography analysis Methods 0.000 title claims abstract description 285
- 238000000034 method Methods 0.000 title claims abstract description 238
- 238000013461 design Methods 0.000 title claims abstract description 121
- 238000004458 analytical method Methods 0.000 title claims abstract description 79
- 230000035515 penetration Effects 0.000 claims abstract description 219
- 230000008569 process Effects 0.000 claims abstract description 185
- 238000013528 artificial neural network Methods 0.000 claims abstract description 105
- 238000005457 optimization Methods 0.000 claims abstract description 38
- 230000007246 mechanism Effects 0.000 claims description 119
- 239000011159 matrix material Substances 0.000 claims description 75
- 102000004169 proteins and genes Human genes 0.000 claims description 63
- 108090000623 proteins and genes Proteins 0.000 claims description 63
- 238000012549 training Methods 0.000 claims description 47
- 238000004364 calculation method Methods 0.000 claims description 41
- 238000009826 distribution Methods 0.000 claims description 38
- 238000002474 experimental method Methods 0.000 claims description 38
- 238000000926 separation method Methods 0.000 claims description 32
- 238000010586 diagram Methods 0.000 claims description 25
- 238000009792 diffusion process Methods 0.000 claims description 20
- 238000011156 evaluation Methods 0.000 claims description 19
- 230000014759 maintenance of location Effects 0.000 claims description 12
- 238000003325 tomography Methods 0.000 claims description 9
- 238000013210 evaluation model Methods 0.000 claims description 8
- 210000002569 neuron Anatomy 0.000 claims description 5
- 230000008859 change Effects 0.000 claims description 2
- 238000012546 transfer Methods 0.000 description 31
- 239000002245 particle Substances 0.000 description 27
- 238000001179 sorption measurement Methods 0.000 description 26
- 238000004140 cleaning Methods 0.000 description 16
- 238000010494 dissociation reaction Methods 0.000 description 14
- 230000005593 dissociations Effects 0.000 description 14
- 239000007788 liquid Substances 0.000 description 13
- 229920006395 saturated elastomer Polymers 0.000 description 13
- 239000007790 solid phase Substances 0.000 description 12
- 238000010828 elution Methods 0.000 description 6
- 239000007791 liquid phase Substances 0.000 description 6
- 230000008929 regeneration Effects 0.000 description 6
- 238000011069 regeneration method Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 4
- 239000012515 MabSelect SuRe Substances 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 210000005036 nerve Anatomy 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000011112 process operation Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 108091003079 Bovine Serum Albumin Proteins 0.000 description 1
- 229940125644 antibody drug Drugs 0.000 description 1
- 238000010364 biochemical engineering Methods 0.000 description 1
- 229940098773 bovine serum albumin Drugs 0.000 description 1
- 238000004113 cell culture Methods 0.000 description 1
- 238000013375 chromatographic separation Methods 0.000 description 1
- 238000010924 continuous production Methods 0.000 description 1
- 238000004185 countercurrent chromatography Methods 0.000 description 1
- 238000012938 design process Methods 0.000 description 1
- 238000004255 ion exchange chromatography Methods 0.000 description 1
- 238000012067 mathematical method Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000013064 process characterization Methods 0.000 description 1
- 238000011165 process development Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000011144 upstream manufacturing Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N30/00—Investigating or analysing materials by separation into components using adsorption, absorption or similar phenomena or using ion-exchange, e.g. chromatography or field flow fractionation
- G01N30/02—Column chromatography
- G01N30/86—Signal analysis
- G01N30/8658—Optimising operation parameters
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N30/00—Investigating or analysing materials by separation into components using adsorption, absorption or similar phenomena or using ion-exchange, e.g. chromatography or field flow fractionation
- G01N30/02—Column chromatography
- G01N30/86—Signal analysis
- G01N30/8693—Models, e.g. prediction of retention times, method development and validation
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B01—PHYSICAL OR CHEMICAL PROCESSES OR APPARATUS IN GENERAL
- B01D—SEPARATION
- B01D15/00—Separating processes involving the treatment of liquids with solid sorbents; Apparatus therefor
- B01D15/08—Selective adsorption, e.g. chromatography
- B01D15/10—Selective adsorption, e.g. chromatography characterised by constructional or operational features
- B01D15/18—Selective adsorption, e.g. chromatography characterised by constructional or operational features relating to flow patterns
- B01D15/1864—Selective adsorption, e.g. chromatography characterised by constructional or operational features relating to flow patterns using two or more columns
- B01D15/1871—Selective adsorption, e.g. chromatography characterised by constructional or operational features relating to flow patterns using two or more columns placed in series
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Definitions
- the present invention relates to protein chromatography separation technology in the fields of biochemical engineering and bioengineering, in particular to a method for multi-column continuous flow chromatography design and analysis, including a method for realizing multi-column continuous flow chromatography design and analysis based on a chromatography model and A method based on artificial neural network to realize multi-column continuous flow chromatography design and analysis.
- the continuous manufacturing process has been widely used in the petrochemical, food and chemical industries, but in the field of biotechnology, continuous production technology started late and the technology is not mature enough.
- a new type of continuous chromatography separation technology-multi-column periodic counter-current chromatography also known as continuous flow chromatography, has been successfully applied to protein separation, especially antibody drug production
- the protein A affinity capture process The traditional antibody downstream separation process is a three-step batch chromatography process based on protein A affinity capture. The process efficiency is limited and it is difficult to match the rapid growth of the upstream cell culture process yield.
- protein A affinity media is expensive, and the media availability of traditional batch chromatography is only about 60%. It is necessary to increase the availability of protein A affinity media and reduce media costs.
- the basic principle of continuous flow chromatography is to load the sample through double columns in series, use the second column to accept the protein that the first column penetrates, and the first column to terminate the sample loading at the appropriate penetration point and switch
- the second chromatographic column is loaded, the first chromatographic column is eluted and regenerated, and the multiple columns alternately realize continuous operation, thereby improving the process yield and medium utilization, reducing buffer consumption and equipment scale.
- the patent (US Patent 10099156 B2) describes a protein capture method with two columns in series.
- the patent (US Patent 2012/0091063 A1) proposes a three-column continuous flow device and applies it to the separation of a mixture containing monoclonal antibodies and bovine serum albumin.
- the patent (US Patent 2017/0016864 A1) proposes a multi-column continuous flow protein capture method, which includes two-column and three-column serial loading, and the process is optimized.
- the artificial neural network system appeared in the 1940s. It is formed by connecting many neurons with adjustable connection weights. It has the capabilities of large-scale parallel processing, distributed information storage, and self-organization and self-learning.
- artificial neural networks are used for tomographic separation process analysis, and have evolved from simple pattern recognition to network prediction based on mechanism models.
- patents US Patent 5,121,443; EP0395481A2
- Du et al. J. Chromatogr. A, 2007, 1145: 165-174 proposed that artificial neural networks can be used to predict the penetration curve of protein adsorption.
- one aspect of the present invention is to provide a method for multi-column continuous flow chromatography design and analysis based on a chromatography model, aiming to fit and accurately predict the penetration curve generated by the experiment, and Based on the analysis of multiple control parameters in continuous flow chromatography, its influence on the process yield and medium utilization, combined with the chromatography mechanism model, the continuous flow chromatography model in multiple operation modes forms an efficient and comprehensive system model, which assists in multiple operations.
- Process analysis and optimization design of column continuous flow chromatography is to provide a method for multi-column continuous flow chromatography design and analysis based on a chromatography model, aiming to fit and accurately predict the penetration curve generated by the experiment, and Based on the analysis of multiple control parameters in continuous flow chromatography, its influence on the process yield and medium utilization, combined with the chromatography mechanism model, the continuous flow chromatography model in multiple operation modes forms an efficient and comprehensive system model, which assists in multiple operations. Process analysis and optimization design of column continuous flow chromatography.
- Another aspect of the present invention is to provide a method for realizing the design and analysis of multi-column continuous flow chromatography based on artificial neural network, which aims to quickly fit and predict the penetration curve generated by the experiment, and is based on continuous flow chromatography.
- Multiple control parameters analyze their influence on process yield and media utilization, combine the advantages of mechanism model and artificial neural network to form an efficient and comprehensive system model to assist process analysis and optimization design of multi-column continuous flow chromatography .
- the present invention adopts the following technical solutions on the one hand:
- a method for multi-column continuous flow chromatography design and analysis based on a chromatography mechanism model including the following steps:
- Step 101 Fitting the penetration curve, substituting the experimental operation parameters into the tomographic mechanism model, and fitting the penetration curve obtained from the experiment to obtain the mechanism model parameters;
- Step 102 Prediction of penetration curve, limit the scope of chromatographic operation, and substitute the mechanism model parameters and chromatographic operation parameters obtained in step 101 into the chromatographic mechanism model to obtain one-column and two-column series penetration under different flow rates and different protein concentrations. Through the curve.
- Step 103 process analysis of continuous flow chromatography, substitute the penetration curve and continuous flow basic operating parameters predicted in step 102 into the continuous flow chromatography model to obtain the design parameters and evaluation parameters of the continuous flow chromatography process, and analyze the continuous flow chromatography
- the influence of changes in operating parameters on performance indicators such as process yield and medium utilization of multi-column continuous flow chromatography;
- step 104 the operation space of continuous flow chromatography is optimized. Based on specific separation goals and requirements, an appropriate process yield and medium utilization are determined, and the operation space of optimized continuous flow chromatography design parameters is obtained through the analysis in step 103.
- the penetration curve fitting in step 101 further includes the following steps:
- the penetration curve prediction in step 102 further includes the following steps:
- the process analysis of continuous flow chromatography includes the following steps:
- the optimization of the operating space of continuous flow chromatography includes the following steps:
- a parameter matrix is generated, the process yield is calculated for all parameter points in the matrix, the process yield matrix is obtained, the matrix is linearly interpolated, and the process yield distribution diagram under different operating conditions is drawn, Used for continuous flow chromatography process analysis and optimization;
- a parameter matrix is generated, the media utilization rate is calculated for all parameter points in the matrix, and the media utilization rate matrix is obtained. After linear interpolation of the matrix, the distribution map of the media utilization rate under different operating conditions is drawn , Used for continuous flow chromatography process analysis and optimization;
- the tomographic mechanism model is a general rate model considering parallel diffusion.
- the method for implementing multi-column continuous flow chromatography design and analysis based on a chromatography mechanism model is characterized in that the continuous flow chromatography model is a continuous flow design model established according to different operation modes, and the continuous flow
- the evaluation parameters of chromatography mainly include process yield and media utilization. Different operation modes include two columns, three columns, four columns, and N columns, with N>4.
- a method for realizing multi-column continuous flow chromatography design and analysis based on artificial neural network including the following steps:
- Step 201 the first artificial neural network training, the mechanism model and experimental method are used to establish the tomographic penetration curve data set and the mechanism model parameter set, the penetration curve data set is used as input, the mechanism model parameter set is output, and the first training is obtained.
- Artificial neural networks
- step 202 the second artificial neural network is trained.
- the mechanism model and experimental methods are used to establish a tomographic penetration curve data set and a mechanism model parameter set, taking the mechanism model parameter set as input and the penetration curve data set as output, and the second training is obtained.
- Artificial neural networks
- Step 203 penetration curve fitting, linear interpolation is performed on the penetration curve obtained in the experiment to obtain the characteristic points of the penetration curve, the characteristic points and experimental operating parameters are substituted into the first artificial neural network as input, and the mechanism model parameters are obtained by fitting calculation ;
- Step 204 penetration curve prediction, substituting the mechanism model parameters obtained in step 203 into the second artificial neural network, and obtaining penetration curves of different flow rates and different protein concentrations according to the prediction range of the tomographic parameters, and performing the calculation with the penetration curve experimental data
- the error is greater than 5%, re-train the first artificial neural network and the second artificial neural network, and re-run steps 203 and 204;
- Step 205 process analysis of continuous flow chromatography, substituting the penetration curve and continuous flow basic operating parameters predicted in step 204 into the continuous flow chromatography model to obtain the design parameters and evaluation parameters of the continuous flow chromatography process, and analyze the continuous flow chromatography
- the influence of changes in operating parameters on performance indicators such as process yield and medium utilization of multi-column continuous flow chromatography;
- step 206 the operation space of continuous flow chromatography is optimized. Based on specific separation goals and requirements, an appropriate process yield and medium utilization are determined, and an optimized continuous flow chromatography operation space is obtained through analysis in step 205.
- the first artificial neural network training in step 201 and the penetration curve fitting in step 203 further include the following steps:
- Extract and normalize the characteristic points on the penetration curve determine the number of neuron nodes and the number of network layers, use the characteristic points of the penetration curve and the mechanism model parameters as the input set and output set, and train the artificial neural network.
- the penetration curve data obtained in the experiment is linearly interpolated to obtain the characteristic points of the penetration curve, which are substituted into the trained first artificial neural network for calculation, and the characteristic model parameters in the mechanism model are obtained.
- the second artificial neural network training in step 202 and the penetration curve prediction in step 204 further include the following steps:
- Extract and normalize the feature points on the penetration curve determine the number of neuron nodes and the number of network layers, use the mechanism model parameters and the penetration curve feature points as the input set and output set, respectively, train the neural network, denoted as The second artificial neural network;
- the chromatography operation parameter matrix is generated within a certain retention time and protein concentration range, combined with the mechanism model parameters, and substituted into the trained second artificial neural network for calculation, and the penetration curve under different chromatography operation conditions is predicted.
- the process of re-training the first artificial neural network and the second artificial neural network is: using the mechanism model to fit the experimental penetration curve to obtain the mechanism model parameters under the experimental conditions, and in the mechanism model A new mechanism model parameter set is randomly generated within the parameter ⁇ 30% interval, and a new penetration curve set is obtained by substituting it into the mechanism model.
- the new penetration curve set is merged into the original penetration curve database, and the first is performed separately.
- the process analysis of continuous flow chromatography includes the following steps:
- the obtained continuous flow chromatography process design parameters and flow arrangement scheme are substituted into the continuous flow chromatography evaluation model, and the process yield and medium utilization of the multi-column continuous flow chromatography are calculated.
- the operation space optimization of continuous flow chromatography includes the following steps:
- a parameter matrix is generated, the process yield is calculated for all parameter points in the matrix, the process yield matrix is obtained, the matrix is linearly interpolated, and the process yield distribution diagram under different operating conditions is drawn, Used for continuous flow chromatography process analysis and optimization;
- a parameter matrix is generated, the media utilization rate is calculated for all parameter points in the matrix, and the media utilization rate matrix is obtained. After linear interpolation of the matrix, the distribution map of the media utilization rate under different operating conditions is drawn , Used for continuous flow chromatography process analysis and optimization;
- the tomographic mechanism model is a general rate model considering parallel diffusion.
- the continuous flow chromatography model is a continuous flow design model established according to different operation modes.
- the evaluation parameters of the continuous flow chromatography mainly include process productivity and medium utilization.
- the different operation modes include two-column and three-column , Four pillars, N pillars, N>4.
- the general tomographic mechanism model is used to generate the training set and test set of the artificial neural network, which greatly reduces the experimental workload and increases the reliability of the artificial neural network;
- the artificial neural network is used to replace the mechanism model to fit and predict the penetration curve, which speeds up the calculation while maintaining the calculation accuracy, which is conducive to the analysis and optimization of a wide range of parameters;
- the fitting effect can be checked during the use process. If the error is found to be large, the mechanism model is used to generate a new training set, and the neural network is retrained to form an intelligent self-learning system;
- FIG. 1 is a schematic diagram of the steps of a method for implementing multi-column continuous flow chromatography design and analysis based on a chromatography model according to an embodiment of the present invention
- Figure 2 is a comparison of the experimental penetration curve and the model fitting penetration curve in Example 1 of the present invention
- Example 3 is a schematic diagram of two penetration curves in Example 2 of the present invention.
- Example 4 is a process yield distribution diagram of double-column continuous flow chromatography in Example 3 of the present invention.
- Fig. 5 is a medium availability distribution diagram of double-column continuous flow chromatography in Example 3 of the present invention.
- FIG. 6 is a schematic diagram of the continuous flow chromatography operation space obtained according to the separation target in Example 3 of the present invention.
- FIG. 7 is a schematic diagram of the steps of a method for implementing multi-column continuous flow chromatography design and analysis based on artificial neural network according to an embodiment of the present invention
- Example 8 is a schematic diagram of the penetration curve of the first artificial neural network training in Example 4 of the present invention.
- Example 9 is a comparison between the characteristic points of the penetration curve and the fitting result of the first artificial neural network in Example 4 of the present invention.
- Example 10 is a schematic diagram of two penetration curves in Example 5 of the present invention.
- Example 11 is a comparison between the penetration curve feature points predicted by the second artificial neural network and the experimental penetration curve in Example 5 of the present invention.
- Figure 12 is a process yield distribution diagram of dual-column continuous flow chromatography in Example 6 of the present invention.
- Fig. 13 is a medium availability distribution diagram of double-column continuous flow chromatography in Example 6 of the present invention.
- Example 14 is a schematic diagram of the continuous flow chromatography operation space obtained according to the separation target in Example 6 of the present invention.
- the embodiment of the present invention discloses a method for realizing multi-column continuous flow chromatography design and analysis based on a chromatography model, which specifically includes the following steps:
- Step 101 Fitting the experimental penetration curve, substituting the experimental penetration curve and the chromatography operation parameters into the chromatography model, fitting the penetration curve, and obtaining the chromatography model parameters;
- Step 102 Predict the penetration curve, limit the chromatographic operation range, and substitute the chromatographic model parameters and chromatographic operation parameters obtained in step 101 into the chromatographic model to obtain one-column and two-column series penetration under different flow rates and different protein concentrations.
- Step 103 process analysis of continuous flow chromatography, substitute the penetration curve and continuous flow basic operating parameters predicted in step 102 into the continuous flow chromatography model to obtain the design parameters and evaluation parameters of the continuous flow chromatography process, and analyze the continuous flow chromatography
- the influence of changes in operating parameters on performance indicators such as process yield and medium utilization of multi-column continuous flow chromatography;
- step 104 the operation space of continuous flow chromatography is optimized. Based on specific separation goals and requirements, an appropriate process yield and medium utilization are determined, and the operation space of optimized continuous flow chromatography design parameters is obtained through the analysis in step 103.
- the tomographic mechanism model is a general rate model considering parallel diffusion, and its equation is as follows:
- the protein adsorption model used is the Langmuir adsorption isotherm model, and its equation is as follows:
- the key operating parameters include the sample loading time in the connected mode and the sample flow rate in the disconnected mode.
- the calculation method is as follows:
- T C is the sample loading time in connected mode, in min;
- U DC represents the sample flow rate in disconnected mode, in mL/min;
- C 0 is the protein loading concentration, in mg;
- T DC is disconnected
- U C is the sample flow rate of the connection mode, the unit is mL/min;
- T 1_1% is the 1% breakthrough time point of one column, the unit is min;
- T 1_s% is one column s % Breakthrough point time, the unit is min;
- T 2_1% is the 2 column 1% breakthrough point time, the unit is min;
- SF is the safety factor.
- T wait T C -T RR if T C >T RR
- T wait 2(T RR -T C ) if T RR >T C
- T CW represents the cleaning time of the connection mode, the unit is min;
- T wait is the waiting time, the unit is min;
- T RR is the total time of elution cleaning and regeneration, the unit is min.
- the sample loading time of the connected mode is the same as the three-column calculation method.
- the waiting time calculation method is as follows:
- T wait 2T C -T RR +T CW if T C >(T RR -T CW )/2
- T wait 2(T RR -2T C -T CW ) if (T RR -T CW )/2>T C
- the key operating parameters include the number of columns, the sample loading time and waiting time of the connected mode, and the sample loading time of the connected mode is the same as the three-column calculation method ,
- the calculation method of the number of bars and waiting time is as follows:
- T wait (N-2)T C +(N-3)T CW -T RR
- the evaluation parameters mainly include process yield and medium utilization.
- the calculation formula of process yield is as follows:
- T 1-95% is the 95% penetration time point of a column, in min.
- step 101 the steps of experimental penetration curve fitting mainly include the following:
- the model parameters include mass transfer related parameters (including axial diffusion coefficient, liquid film mass transfer coefficient, intra-particle solid phase mass transfer coefficient and intra-particle liquid mass transfer coefficient, etc.), adsorption related parameters ( Including saturated adsorption capacity and dissociation equilibrium constant, etc.) and operation related parameters (including empty column flow rate and loading concentration, etc.);
- step 102 the steps of penetration curve prediction are:
- One-column penetration curve prediction Set the range of chromatographic operation parameters, within this range, substitute the chromatographic model parameters and chromatographic operation parameters obtained in step 101 into the chromatographic model to obtain different flow rates and different protein concentrations The next column penetration curve;
- Double-column series penetration curve prediction Set the range of chromatographic operation parameters, within this range, substitute the chromatographic model parameters and chromatographic operation parameters obtained in step 101 into the chromatographic model, and pass the output of one column
- the time-varying protein concentration in the permeation curve is used as the loading concentration of the two columns, and the double-column series breakthrough curves with different flow rates and different protein concentrations are obtained.
- step 103 the process analysis of continuous flow chromatography in step 103 includes the following steps:
- Evaluation parameter calculation step Substitute the design parameters and process arrangement scheme of continuous flow chromatography obtained in the previous step into the evaluation model of continuous flow chromatography above, and calculate the process yield and media of multi-column continuous flow chromatography Utilization.
- step 104 the operation space optimization of continuous flow chromatography includes the following steps:
- Process yield distribution map Based on the design parameter range of continuous flow chromatography, a parameter matrix is generated, and the process yield is calculated for all parameter points in the matrix according to the above method, and the process yield matrix is obtained, and the matrix is linearly interpolated.
- Draw process yield distribution diagrams under different operating conditions including retention time, switching point, and sample protein concentration) for continuous flow chromatography process analysis and optimization.
- Purolite Life Sciences' Praesto Jetted A50 medium was used for the IgG protein penetration experiment.
- the flow rate was 1 mL/min, and the sample protein concentration was 2 mg/mL.
- the sample was stopped and the total volume of the sample protein was loaded. It is 90 times the column volume to obtain the experimental penetration curve.
- the initial values of the chromatographic model parameters are: axial diffusion coefficient 3*10 -7 m 2 /s, liquid film mass transfer coefficient 18*10 -6 m/s, solid phase mass transfer coefficient within particles 4*10 -13 m 2 /s, the liquid phase mass transfer coefficient in the particle is 6*10 -12 m 2 /s, the saturated adsorption capacity is 80 mg/mL, and the dissociation equilibrium constant is 0.2 mg/mL.
- the relevant parameters of the experimental operation are: flow rate 1mL/min, loading protein concentration 2mg/mL.
- the parameters of the fitted model are: axial diffusion coefficient 0.7*10 -7 m 2 /s, liquid film mass transfer coefficient 34*10 -6 m/s, solid phase mass transfer coefficient within particles 0.6*10 -13 m 2 /s, liquid phase mass transfer coefficient within particles 4.5*10 -12 m 2 /s, saturated adsorption capacity 124mg/mL, dissociation equilibrium The constant is 0.13mg/mL.
- Figure 2 shows the comparison between the experimental penetration curve and the fitted penetration curve.
- the mass transfer coefficient is 0.6*10 -13 m 2 /s
- the intra-particle mass transfer coefficient is 4.5*10 -12 m 2 /s
- the saturated adsorption capacity is 124 mg/mL
- the dissociation equilibrium constant is 0.13 mg/mL.
- Example 1 Praesto Jetted A50 medium, a breakthrough curve with a protein concentration of C 0 of 3 mg/mL and a flow rate of 1.5 mL/min for continuous flow chromatography process design, the design process is as follows:
- Dual-column continuous flow chromatography design The sample flow rate U C in the connected mode is the same as that of the protein penetration experiment (1 mL/min), and the sample time T DC in the disconnect mode is the total amount of elution, cleaning and regeneration of the column.
- the time T RR is the same (26 min)
- the column volume of the connection mode cleaning is 4 CV
- the flow rate of the connection mode cleaning is 1.5 mL/min
- the cleaning time T CW of the connection mode can be obtained as 2.6 min.
- the safety factor SF is 0.9
- the switching point s is 80%
- the second column reaches 1% penetration time
- the time T 2_1% is 21.5 min.
- T 1_1% , T 1_s% , T 2_1% , T CW and T RR have the same values as above.
- T C , T CW and T RR are the same as above.
- T C is the same as the three-column continuous flow chromatography process. Since T C >(T RR -T CW )/2,
- N-column continuous flow chromatography design the values of T C , T CW , and T RR are the same as the above.
- the separation target When the separation target is input, such as the process yield is greater than 40g/L/h and the media availability is greater than 80%, the separation target can be met in the above two contour diagrams to meet the process yield greater than 40g/L/h
- the intersection with the media utilization greater than 80% is the appropriate operating space, as shown in Figure 6.
- the description of the above embodiment is a detailed description of the implementation process of multi-column continuous flow chromatography design and analysis method based on the chromatographic mechanism model. Next, the method of multi-column continuous flow chromatography design and analysis based on artificial neural network will be described. The implementation process will be explained in detail.
- the embodiment of the present invention discloses a method for realizing multi-column continuous flow chromatography design and analysis based on artificial neural network, which specifically includes the following steps:
- Step 201 the first artificial neural network training is used to establish a tomographic penetration curve data set and a mechanism model parameter set by using mechanism models and experimental methods, taking the penetration curve data set as input and the mechanism model parameter set as output, and training is obtained The first artificial neural network;
- Step 202 the second artificial neural network training is used to establish a tomographic penetration curve data set and a mechanism model parameter set by using a mechanism model and experimental method, taking the mechanism model parameter set as input and the penetration curve data set as output, and the training is obtained
- the second artificial neural network
- Step 203 penetration curve fitting, which is used to linearly interpolate the penetration curve obtained in the experiment to obtain the characteristic points of the penetration curve, and substitute the characteristic points and experimental operation parameters as input into the first artificial neural network, and the mechanism is obtained by fitting calculation Model parameters;
- Step 204 penetration curve prediction, used to substitute the mechanism model parameters obtained in step 203 into the second artificial neural network, and obtain penetration curves of different flow rates and different protein concentrations according to the prediction range of the tomographic parameters, and compare the penetration curve experiments The data is compared, and if the error is greater than 5%, the first artificial neural network training and the second artificial neural network training are performed again, and steps 203 and 204 are performed again;
- Step 205 the process analysis of continuous flow chromatography is used to substitute the penetration curve and continuous flow basic operating parameters predicted in step 204 into the continuous flow chromatography model to obtain the design parameters and evaluation parameters of the continuous flow chromatography process, and analyze the continuous flow
- the influence of changes in chromatography operation parameters on performance indicators such as process yield and medium utilization of multi-column continuous flow chromatography;
- Step 206 the operation space optimization of continuous flow chromatography is used to determine the appropriate process yield and medium utilization based on specific separation goals and requirements. Through the analysis of step 205, the operation of obtaining optimized continuous flow chromatography design parameters space.
- the tomographic mechanism model is a general rate model considering parallel diffusion, and its equation is as follows:
- the protein adsorption model used is the Langmuir adsorption isotherm model, and its equation is as follows:
- the key operating parameters include the sample loading time in the connected mode and the sample flow rate in the disconnected mode.
- the calculation method is as follows:
- T C is the sample loading time in connected mode, in min;
- U DC represents the sample flow rate in disconnected mode, in mL/min;
- C 0 is the protein loading concentration, in mg;
- T DC is disconnected
- U C is the sample flow rate of the connection mode, the unit is mL/min;
- T 1_1% is the 1% breakthrough time point of one column, the unit is min;
- T 1_s% is one column s % Breakthrough point time, the unit is min;
- T 2_1% is the 2 column 1% breakthrough point time, the unit is min;
- SF is the safety factor.
- T wait T C -T RR if T C >T RR
- T wait 2(T RR -T C ) if T RR >T C
- T CW represents the cleaning time of the connection mode, the unit is min;
- T wait is the waiting time, the unit is min;
- T RR is the total time of elution cleaning and regeneration, the unit is min.
- the sample loading time of the connected mode is the same as the three-column calculation method.
- the waiting time calculation method is as follows:
- T wait 2T C -T RR +T CW if T C >(T RR -T CW )/2
- T wait 2(T RR -2T C -T CW ) if (T RR -T CW )/2>T C
- the key operating parameters include the number of columns, the sample loading time and waiting time of the connected mode, and the sample loading time of the connected mode is the same as the three-column calculation method ,
- the calculation method of the number of bars and waiting time is as follows:
- T wait (N-2)T C +(N-3)T CW -T RR
- the evaluation parameters mainly include process yield and medium utilization.
- the calculation formula of process yield is as follows:
- T 1-95% is the 95% penetration time point of a column, in min.
- step 201 and step 203 the training and application of the first artificial neural network mainly include the following:
- the model parameters include mass transfer related parameters (including axial diffusion coefficient, liquid film mass transfer coefficient, intra-particle solid phase mass transfer coefficient and intra-particle liquid mass transfer coefficient, etc.), Adsorption-related parameters (including saturated adsorption capacity and dissociation equilibrium constant, etc.) and operation-related parameters (including empty tower flow rate and loading concentration, etc.), produce a random number for each of the above parameters within the range of 80%, and convert these random numbers Arrange and generate the mechanism model parameter matrix in a certain order to form 100-10000 groups of mechanism model parameter matrix; substitute each group of parameters into the mechanism model and use the orthogonal configuration method to obtain the corresponding penetration curve; repeat the above steps to obtain the mechanism Model parameter set and corresponding penetration curve data set.
- mass transfer related parameters including axial diffusion coefficient, liquid film mass transfer coefficient, intra-particle solid phase mass transfer coefficient and intra-particle liquid mass transfer coefficient, etc.
- Adsorption-related parameters including saturated adsorption capacity and dissociation equilibrium constant, etc.
- operation-related parameters including empty tower flow rate and loading concentration,
- the experimental method is used to select different media and operating parameters for single-column protein penetration experiments.
- the experimental parameters include the flow rate of the empty column, the loading concentration, the saturated adsorption capacity, and the dissociation equilibrium constant.
- the adsorption capacity and dissociation equilibrium constant are obtained through static adsorption experiments, and 100-10000 experiments are performed to obtain the mechanism model parameter set and the corresponding penetration curve data set.
- the first artificial neural network training select the characteristic points on the penetration curve in the penetration curve data set, that is, the sample loading time that reaches 10%-90% of the penetration point, and normalize the characteristic points to penetrate
- the characteristic points of the penetration curve are the input set, and the corresponding mechanism model parameters are used as the output set to train the artificial neural network, which is recorded as the first artificial neural network.
- the training and application of the second artificial neural network mainly include the following:
- the model parameters include mass transfer related parameters (including axial diffusion coefficient, liquid film mass transfer coefficient, intra-particle solid phase mass transfer coefficient and intra-particle liquid mass transfer coefficient, etc.), Adsorption related parameters (including saturated adsorption capacity and dissociation equilibrium constant, etc.) and operation related parameters (including empty column flow rate and sample concentration, etc.).
- mass transfer related parameters including axial diffusion coefficient, liquid film mass transfer coefficient, intra-particle solid phase mass transfer coefficient and intra-particle liquid mass transfer coefficient, etc.
- Adsorption related parameters including saturated adsorption capacity and dissociation equilibrium constant, etc.
- operation related parameters including empty column flow rate and sample concentration, etc.
- a random number is generated for each of the above parameters in the range of 80%, and these random numbers are arranged in a certain order to generate a mechanism model parameter matrix, thereby forming a 100-10000 sets of mechanism model parameter matrix; substitute each set of parameters into the mechanism model,
- the orthogonal configuration method is used to obtain the corresponding single-column penetration curve and the penetration curve of the two-column loaded with two columns in series; repeat the above steps to obtain the mechanism model parameter set and the corresponding penetration curve data set.
- the experimental method is used to select different media and operating parameters for the protein penetration experiment with double columns in series.
- the experimental parameters include the flow rate of the empty column, the loading concentration, the saturated adsorption capacity, and the dissociation equilibrium constant. , Where the saturated adsorption capacity and dissociation equilibrium constant are obtained through static adsorption experiments, and 100-10000 sets of the above experiments are performed to obtain the mechanism model parameter set and the corresponding penetration curve data set.
- the second artificial neural network training select the characteristic points on the penetration curve in the penetration curve data set, that is, the sample loading time and the sample volume to reach 10%-90% of the penetration point, and two columns are loaded in series. 1% penetration point on the penetration curve.
- the characteristic points on the penetration curve are extracted and normalized, and the mechanism model parameters are used as the input set, and the characteristic points of the corresponding penetration curve are used as the output set.
- the neural network is trained and recorded as the second artificial neural network.
- the steps of re-training the first artificial neural network and the second artificial neural network in step 204 are: using the mechanism model to fit the experimental penetration curve to obtain the mechanism model parameters under the experimental conditions, A new set of mechanism model parameters is randomly generated within the interval of mechanism model parameters ⁇ 30%, and substituted into the mechanism model to obtain a new set of penetration curves, and the new set of penetration curves is merged into the original penetration curve database, using the above methods respectively Perform the first artificial neural network training and the second artificial neural network training again.
- step 205 the process analysis of continuous flow chromatography in step 205 includes the following steps:
- Evaluation parameter calculation step Substitute the design parameters and process arrangement scheme of continuous flow chromatography obtained in the previous step into the evaluation model of continuous flow chromatography above, and calculate the process yield and media of multi-column continuous flow chromatography Utilization.
- step 206 the operation space optimization of continuous flow chromatography includes the following steps:
- Process yield distribution map Based on the design parameter range of continuous flow chromatography, a parameter matrix is generated, and the process yield is calculated for all parameter points in the matrix according to the above method, and the process yield matrix is obtained, and the matrix is linearly interpolated.
- Draw process yield distribution diagrams under different operating conditions including retention time, switching point, and sample protein concentration) for continuous flow chromatography process analysis and optimization.
- Example 4 The first artificial neural network training and penetration curve fitting
- the mechanism model parameter matrix is [5e-7,12e-6,3e-13,9e-12,110,0.12,0.5,2].
- the mechanism model parameter set is composed of 2000 sets of mechanism model parameter matrices, corresponding to 2000 penetration curves, which constitute the penetration curve data set.
- the matrix composed of the characteristic points is [ 63.7,71.6,78.4,84.2,90.9,98.3,105.9,114.7,127.34]; normalize the matrix to obtain a matrix [0.137,0.135,0.136,0.136,0.139,0.142,0.145,0.145,0.143]; All penetration curve data are converted into the above normalized matrix as the input set; the mechanism model parameter set is normalized as the output set; Levenberg-Marquardt is used as the training function, and the root mean square error is used as the objective function for artificial nerve Training, after 116 iterations, the error is 2.84*10 -5 , which is less than 1*10 -3 , which meets the training requirements and obtains the first artificial neural network.
- Example 5 The second artificial neural network training and penetration curve prediction
- a set of mechanism parameters are randomly generated within 80%: axial diffusion coefficient 3*10 -7 m 2 /s, liquid film mass transfer coefficient 18*10 -6 m/s, solid phase mass transfer coefficient within particles 4*10 -13 m 2 /s, the liquid phase mass transfer coefficient in the particle is 6*10 -12 m 2 /s, the saturated adsorption capacity is 80 mg/mL, the dissociation equilibrium constant is 0.2 mg/mL, the flow rate of the empty column is 1 mL/min, and the protein is loaded The concentration is 1mg/mL. Substituting the above parameters into the mechanism model, two solutions are obtained, one is the penetration curve of one column, and the other is the penetration curve of two columns when the two columns are loaded in series, as shown in Figure 10.
- the mechanism model parameter matrix is [3e-7,18e-6,4e-13,6e-12,80,0.2,1,1].
- the mechanism model parameter set is composed of more than 3000 sets of mechanism model parameters, corresponding to more than 3000 sets of penetration curves, which constitute the penetration curve data set.
- the matrix of the characteristic points is [13.0,17.4,21.9, 27.4,34.2,43.2,55.3,72.1,93.7,0.31,1.0,2.1,4.1,7.3,12.3,20.2,32.8,51.2,28.1]; convert all penetration curve data into the above matrix, and then normalize Processing, as the output set.
- the parameter set of the mechanism model is normalized as the input set.
- the neural network calculates the characteristic points of the penetration curve under different protein concentrations and flow rates. For example, the prediction results of a concentration of 1 mg/mL and a flow rate of 0.33 mL/min and 0.5 mL/min are shown in Figure 11.
- the process design of continuous flow chromatography is as follows:
- Dual-column continuous flow chromatography design The sample flow rate U C in the connected mode is the same as that of the protein penetration experiment (1 mL/min), and the sample time T DC in the disconnect mode is the total amount of elution, cleaning and regeneration of the column.
- the time T RR is the same (26 min)
- the column volume of the connection mode cleaning is 4 CV
- the flow rate of the connection mode cleaning is 1 mL/min
- the cleaning time T CW of the connection mode can be obtained as 4 min.
- the safety factor SF is 0.9
- the switching point s is 80%
- the first column reaches 1% penetration time T 1_1% is 6.6min
- the first column reaches s penetration time T 1_s% is 38.1min
- the second column reaches 1% penetration time
- the time T 2_1% is 72.1 min.
- T 1_1% , T 1_s% , T 2_1% , T CW and T RR have the same values as above.
- T C , T CW and T RR are the same as above.
- T C is the same as the three-column continuous flow chromatography process. Since T C >(T RR -T CW )/2,
- N-column continuous flow chromatography design the values of T C , T CW , and T RR are the same as the above.
- the two-column operation cycle time T cycle is 134.5 min, and one column reaches 95% penetration
- the time T 1-95% is 107.8 min, and the column volume CV is 1 mL.
- the separation target When the separation target is input, if the process yield is greater than 17g/L/h and the media utilization is greater than 70%, the separation target can be met in the above two contour diagrams to meet the process yield greater than 17g/L/h The intersection with the media utilization greater than 70%, that is, the appropriate operating space, as shown in Figure 14.
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Chemical & Material Sciences (AREA)
- Theoretical Computer Science (AREA)
- Analytical Chemistry (AREA)
- General Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Biochemistry (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Chemical Kinetics & Catalysis (AREA)
- Databases & Information Systems (AREA)
- Treatment Of Liquids With Adsorbents In General (AREA)
Abstract
基于层析模型实现多柱连续流层析设计、分析的方法，以及一种基于人工神经网络实现多柱连续流层析设计及分析的方法，其中基于层析模型的方法包括以下步骤：步骤101，实验穿透曲线拟合，利用层析模型，拟合得到模型参数；步骤102，穿透曲线预测，将模型参数代入层析模型，得到不同操作条件下的穿透曲线；步骤103，连续流层析的过程分析，将预测的穿透曲线和连续流操作参数代入连续流层析模型，得到过程产率和介质利用度性能指标；步骤104，连续流层析的操作空间优化，基于分离目标，得到连续流层析设计参数的操作空间。基于人工神经网络的方法是使用人工神经网络模型代替层析模型，完成以上各步骤。
Description
本发明涉及生物化工和生物工程领域的蛋白层析分离技术，具体涉及多柱连续流层析设计及分析的方法，包括一种基于层析模型实现多柱连续流层析设计及分析的方法以及一种基于人工神经网络实现多柱连续流层析设计及分析的方法。
连续制造工艺在石油化工、食品及化学工业中已得到广泛的应用，但是在生物技术领域，连续生产技术起步较晚，技术上不够成熟。近年来，一种新型的连续层析分离技术-多柱周期性逆流层析(multi-column periodic counter-current chromatography)，又称连续流层析，被成功应用于蛋白分离，特别是抗体药物生产的蛋白A亲和捕获过程。传统的抗体下游分离过程为基于蛋白A亲和捕获的三步批次层析工艺，其过程效率有限，难以匹配快速增长的上游细胞培养过程产率。另一方面，蛋白A亲和介质价格昂贵，传统批次层析的介质利用度仅60％左右，有必要提高蛋白A亲和介质的利用度，降低介质成本。
连续流层析基本原理是通过双柱串联上样，使用第二根层析柱承接第一根层析柱穿透的蛋白，第一根层析柱在合适的穿透点终止上样，切换至第二根层析柱上样，第一根层析柱则进行洗脱和再生，多柱交替实现连续操作，从而提高过程产率和介质利用度，减少缓冲液消耗和设备规模。专利(US Patent 10099156 B2)描述了一种双柱串联上样的蛋白捕获方式。专利(US Patent 2012/0091063 A1)提出了一种三柱连续流设备，并将其应用于含有单克隆抗体和牛血清白蛋白的混合物分离。专利(US Patent 2017/0016864 A1)提出了多柱连续流蛋白捕获方式，其包括双柱和三柱的串联上样，并对过程进行了实验优化。
整体而言，多柱连续流层析过程复杂，可选择操作参数多，实验优化的工作量极大，若借助数学模型进行合理的过程表征和辅助设计，可以显著提高过程设计和优化的效率，减少实验摸索。虽然已有成熟的数学方法可以针对层析实验的穿透曲线进行拟合和预测，如Baur Daniel等(Biotechnol.J,2016,11:920-931)结合一般性速率模型(General rate model)和缩核模型(Shrinking core model)，对实验穿透曲线进行拟合得到机理参数，然后预测多流速和多浓度下的穿透曲线，从而辅助进行层析过程优化设计。然而，上述专利及论文中使用的层析模型和连续流模型的功能相对单一，缺乏对于不同连续流层析模式和不同操作条件的综合计算、比较和优化，在实际应用中存在许多局限。进一步的，解偏微分方程组需要花费较长的 时间，精度要求越高，所需计算时间就越长。不利于大范围的多参数优化和过程的快速设计和分析。
人工神经网络系统出现于20世纪40年代，由众多可调连接权值的神经元连接而成，具有大规模并行处理、分布式信息存储和自组织自学习等能力。目前，人工神经网络用于层析分离过程分析，已从简单的模式识别发展为基于机理模型的网络预测，如专利(US Patent 5,121,443；EP0395481A2)报道了使用神经网络去除噪声干扰，识别和拆分叠加色谱峰，并对色谱峰各项性质(如保留时间、峰宽等)进行表征。Du等(J.Chromatogr.A,2007,1145:165-174)提出可以使用人工神经网络预测蛋白吸附的穿透曲线。Wang等(J.Chromatogr.A,2017,1487:211-217)提出可以将机理模型参数代入机理模型产生训练集，训练神经网络，以代替计算过程繁琐复杂的机理模型，并对离子交换层析峰型进行拟合。上述专利及论文中使用的神经网络结构简单，功能相对单一，在实际应用中仍然存在许多限制，且难以满足复杂的过程设计和大范围的参数优化要求。鉴于连续流层析过程的复杂性，需要对训练数据集选取进行重新规划，以满足过程设计和分析的需求。
发明内容
鉴于以上存在的技术问题，本发明一方面用于提供一种基于层析模型实现多柱连续流层析设计及分析的方法，旨在对实验产生的穿透曲线进行拟合和精确预测，并基于连续流层析中多个控制参数分析其对过程产率和介质利用度的影响，结合层析机理模型，多种操作模式下的连续流层析模型形成高效综合的系统模型，辅助进行多柱连续流层析的过程分析和优化设计。
本发明的又一方面用于提供一种基于人工神经网络实现多柱连续流层析设计及分析的方法，旨在快速对实验产生的穿透曲线进行拟合和预测，并基于连续流层析中多个控制参数分析其对过程产率和介质利用度的影响，结合机理模型和人工神经网络各自的优点，形成高效综合的系统模型，辅助进行多柱连续流层析的过程分析和优化设计。
为解决上述技术问题，本发明一方面采用如下的技术方案：
一种基于层析机理模型实现多柱连续流层析设计及分析的方法，包括以下步骤：
步骤101，穿透曲线拟合，将实验操作参数代入层析机理模型中，拟合实验所得穿透曲线得到机理模型参数；
步骤102，穿透曲线预测，限定层析操作范围，将步骤101得到的机理模型参数和层析操作参数代入层析机理模型中，得到不同流速与不同蛋白浓度下的一柱和双柱串联穿透曲线。
步骤103，连续流层析的过程分析，将步骤102预测的穿透曲线和连续流基本操作参数 代入连续流层析模型，得到连续流层析过程的设计参数和评估参数，分析连续流层析操作参数变化对多柱连续流层析的过程产率和介质利用度等性能指标的影响；
步骤104，连续流层析的操作空间优化，基于特定的分离目标和要求，确定合适的过程产率和介质利用度，通过步骤103的分析，得到优化的连续流层析设计参数的操作空间。
优选地，步骤101的穿透曲线拟合进一步包括如下步骤：
将实验操作参数以及层析机理模型特征模型参数初值代入层析机理模型中计算穿透曲线，并将计算结果与实验所得穿透曲线进行比较，改变层析模型参数使两者均方根误差达到最小，得到层析模型参数。
优选地，步骤102的穿透曲线预测进一步包括如下步骤：
设定层析操作流速和蛋白质浓度范围，在该范围之内产生层析操作参数矩阵，与步骤101得到的层析模型参数合并，代入层析模型中进行计算，预测得到不同流速和不同蛋白浓度下的一柱和双柱串联穿透曲线。
优选地，所述的连续流层析的过程分析包括如下步骤：
将预测的穿透曲线和连续流层析基本操作参数，代入连续流层析模型中，得到连续流层析的过程设计参数与流程安排方案；
将所得的连续流层析的过程设计参数与流程安排方案代入连续流层析的评估模型，计算得到多柱连续流层析的过程产率和介质利用度。
优选地，所述的连续流层析的操作空间优化包括如下步骤：
基于连续流层析的设计参数范围，生成参数矩阵，对矩阵内的所有参数点计算过程产率，得到过程产率矩阵，对矩阵进行线性插值，绘制在不同操作条件的过程产率分布图，用于连续流层析过程分析和优化；
基于连续流层析的设计参数范围，生成参数矩阵，对矩阵内的所有参数点计算介质利用度，得到介质利用度矩阵，对矩阵进行线性插值后，绘制在不同操作条件的介质利用度分布图，用于连续流层析过程分析和优化；
基于特定的分离目标，在过程产率分布图和介质利用度分布图中分别计算满足分离目标的连续流层析设计参数范围，将两个图的设计参数区域进行叠加，得到同时满足过程产率和介质利用度要求的连续流层析设计参数，并计算连续流层析过程的操作参数与流程安排方案。
优选地，所述的层析机理模型为考虑平行扩散的一般性速率模型。
优选地，所述的基于层析机理模型实现多柱连续流层析设计及分析的方法，其特征在于，所述的连续流层析模型为根据不同操作模式建立的连续流设计模型，连续流层析的评估参数 主要包括过程产率和介质利用度，其中不同操作模式包括二柱、三柱、四柱、N柱，N>4。
以上实现的基于层析机理模型实现多柱连续流层析设计及分析的方法具有如下的有益效果：
(1)采用通用的层析模型对于有限的实验穿透曲线数据进行拟合，得到模型参数可以预测不同流速和不同蛋白浓度下的穿透曲线，基于少量实验数据的基础上对于大范围的层析操作参数变化进行穿透曲线预测，提高优化过程的效率和可靠性；
(2)针对不同模式的连续流层析过程，包括二柱、三柱、四柱和N柱(N>4)，提出各自的过程操作参数和设计参数，可实现不同多柱连续流层析模式的综合分析和比较，从而系统评价不同多柱连续流层析模式，优化多柱连续流层析的分离效果；
(3)基于层析模型的预测功能，可系统分析多个连续流层析的操作参数和设计参数变化对连续流层析分离性能的影响，得到过程产率分布图和介质利用度分布图，从而合理设计连续流层析操作的优化空间，显著提高多柱连续流层析的过程开发效率。
本发明一方面采用如下的技术方案：
一种基于人工神经网络实现多柱连续流层析设计及分析的方法，包括以下步骤：
步骤201，第一人工神经网络训练，采用机理模型和实验方法建立层析穿透曲线数据集和机理模型参数集，以穿透曲线数据集为输入，机理模型参数集为输出，训练得到第一人工神经网络；
步骤202，第二人工神经网络训练，采用机理模型和实验方法建立层析穿透曲线数据集和机理模型参数集，以机理模型参数集为输入，穿透曲线数据集为输出，训练得到第二人工神经网络；
步骤203，穿透曲线拟合，将实验所得穿透曲线进行线性插值，得到穿透曲线的特征点，将特征点和实验操作参数作为输入代入第一人工神经网络，拟合计算得到机理模型参数；
步骤204，穿透曲线预测，将步骤203得到的机理模型参数代入第二人工神经网络，依据层析参数预测范围，得到不同流速与不同蛋白浓度的穿透曲线，并和穿透曲线实验数据进行比较，若误差大于5％，则重新进行第一人工神经网络训练和第二人工神经网络训练，并重新进行步骤203和步骤204；
步骤205，连续流层析的过程分析，将步骤204预测的穿透曲线和连续流基本操作参数代入连续流层析模型，得到连续流层析过程的设计参数和评估参数，分析连续流层析操作参数变化对多柱连续流层析的过程产率和介质利用度等性能指标的影响；
步骤206，连续流层析的操作空间优化，基于特定的分离目标和要求，确定合适的过程产率和介质利用度，通过步骤205的分析，得到优化的连续流层析操作空间。
优选地，步骤201的第一人工神经网络训练和步骤203的穿透曲线拟合进一步包括如下步骤：
产生一定范围内随机分布的若干层析机理模型参数集，代入机理模型方程，利用正交配置法产生穿透曲线数据集，或通过实验得到机理模型参数对应的穿透曲线数据集；
对穿透曲线上的特征点进行提取和归一化处理，确定神经元节点数和网络层数，使用穿透曲线特征点和机理模型参数分别作为输入集和输出集，训练人工神经网络，记为第一人工神经网络；
将实验所得穿透曲线数据进行线性插值，得到穿透曲线特征点，代入已训练好的第一人工神经网络进行计算，得到机理模型中的特征模型参数。
优选地，步骤202的第二人工神经网络训练和步骤204的穿透曲线预测进一步包括如下步骤：
产生一定范围内随机分布的若干层析机理模型参数集，代入机理模型方程，利用正交配置法产生穿透曲线数据集，或通过实验得到机理模型参数对应的穿透曲线数据集；
对穿透曲线上的特征点进行提取和归一化处理，确定神经元节点数和网络层数，使用机理模型参数和穿透曲线特征点分别作为输入集与输出集，训练神经网络，记为第二人工神经网络；
在一定的保留时间和蛋白浓度范围之内产生层析操作参数矩阵，与机理模型参数合并，代入已训练好的第二人工神经网络进行计算，预测得到不同层析操作条件下的穿透曲线。
优选地，步骤204中重新进行第一人工神经网络训练和第二人工神经网络训练过程为：使用机理模型对实验穿透曲线进行拟合，得到该实验条件下的机理模型参数，在该机理模型参数±30％区间内随机生成新的机理模型参数集，代入机理模型中得到新的穿透曲线集，将新的穿透曲线集合并入原先的穿透曲线数据库中，分别对重新进行第一人工神经网络训练和第二人工神经网络训练。
优选地，所述的连续流层析的过程分析包括如下步骤：
将预测的穿透曲线特征点和连续流层析基本操作参数，代入连续流层析模型中，得到连续流层析过程的设计参数与流程安排方案；
评估参数计算步骤，将所得的连续流层析的过程设计参数与流程安排方案代入所述连续流层析的评估模型，计算得到多柱连续流层析的过程产率和介质利用度。
优选地，连续流层析的操作空间优化包括如下步骤：
基于连续流层析的设计参数范围，生成参数矩阵，对矩阵内的所有参数点计算过程产率，得到过程产率矩阵，对矩阵进行线性插值，绘制在不同操作条件的过程产率分布图，用于连 续流层析过程分析和优化；
基于连续流层析的设计参数范围，生成参数矩阵，对矩阵内的所有参数点计算介质利用度，得到介质利用度矩阵，对矩阵进行线性插值后，绘制在不同操作条件的介质利用度分布图，用于连续流层析过程分析和优化；
基于特定的分离目标，在过程产率分布图和介质利用度分布图中分别计算满足分离目标的连续流层析设计参数范围，将两个图的设计参数区域进行叠加，得到同时满足过程产率和介质利用度要求的连续流层析设计参数，并计算连续流层析过程的操作参数与流程安排方案。
优选地，所述的层析机理模型为考虑平行扩散的一般性速率模型。
优选地，所述的连续流层析模型为根据不同操作模式建立的连续流设计模型，连续流层析的评估参数主要包括过程产率和介质利用度，其中不同操作模式包括二柱、三柱、四柱、N柱，N>4。
以上实现的基于人工神经网络实现多柱连续流层析设计及分析的方法具有如下的有益效果：
(1)采用通用的层析机理模型生成人工神经网络的训练集和测试集，大量减少实验工作量，增加人工神经网络的可靠性；
(2)采用人工神经网络代替机理模型进行穿透曲线的拟合和预测，在保持计算精度的情况下，加快了运算速度，有利于大范围的参数分析和优化；
(3)使用机理模型和神经网络相互嵌套的方式进行拟合和预测，降低了神经网络的训练难度，增加了计算结果的可信度；
(4)使用过程中可自行检测拟合效果，发现误差较大则利用机理模型生成新的训练集，重新训练神经网络，形成智能化自学习系统；
(5)针对不同模式的连续流层析过程，包括二柱、三柱、四柱和N柱(N>4)，提出各自的过程操作参数和设计参数，可实现不同多柱连续流层析模式的综合分析；
(6)基于神经网络的计算能力，可系统分析多个操作参数和设计参数变化对连续流层析分离性能的影响，得到过程产率分布图和介质利用度分布图，从而合理设计连续流层析操作的优化空间。
图1为本发明实施例的基于层析模型实现多柱连续流层析设计及分析的方法的步骤示意图；
图2为本发明的实例1中实验穿透曲线和模型拟合穿透曲线的比较
图3为本发明的实例2中两条穿透曲线示意图；
图4为本发明的实例3中双柱连续流层析的过程产率分布图；
图5为本发明的实例3中双柱连续流层析的介质利用度分布图；
图6为本发明的实例3中根据分离目标得到的连续流层析操作空间示意图。
图7为本发明实施例的基于人工神经网络实现多柱连续流层析设计及分析的方法的步骤示意图；
图8为本发明的实例4中第一人工神经网络训练的穿透曲线示意图；
图9为本发明的实例4中穿透曲线特征点和第一人工神经网络拟合结果比较；
图10为本发明的实例5中两条穿透曲线示意图；
图11为本发明的实例5中第二人工神经网络预测的穿透曲线特征点和实验穿透曲线对比；
图12为本发明的实例6中双柱连续流层析的过程产率分布图；
图13为本发明的实例6中双柱连续流层析的介质利用度分布图；
图14为本发明的实例6中根据分离目标得到的连续流层析操作空间示意图。
下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其它实施例，都属于本发明保护的范围。
参照图1，本发明实施例公开了一种基于层析模型实现多柱连续流层析设计及分析的方法，具体包括以下步骤：
步骤101，实验穿透曲线拟合，将实验穿透曲线和层析操作参数代入层析模型中，拟合穿透曲线，得到层析模型参数；
步骤102，穿透曲线预测，限定层析操作范围，将步骤101得到的层析模型参数和层析操作参数代入层析模型中，得到不同流速与不同蛋白浓度下的一柱和双柱串联穿透曲线；
步骤103，连续流层析的过程分析，将步骤102预测的穿透曲线和连续流基本操作参数代入连续流层析模型，得到连续流层析过程的设计参数和评估参数，分析连续流层析操作参数变化对多柱连续流层析的过程产率和介质利用度等性能指标的影响；
步骤104，连续流层析的操作空间优化，基于特定的分离目标和要求，确定合适的过程产率和介质利用度，通过步骤103的分析，得到优化的连续流层析设计参数的操作空间。
为了使本发明实施例的具体实施过程能得到更好的理解，以下将对以上步骤实施的具体过程进行进一步的具体描述。
具体应用实例中，层析机理模型为考虑平行扩散的一般性速率模型，其方程如下：
其中：c为层析柱内蛋白浓度，单位为mg/mL；c
p为介质颗粒内蛋白浓度，单位为mg/mL；c
0为上样蛋白浓度，单位为mg/mL；t为时间，单位为s；D
ax为柱内轴向扩散系数，单位为m
2/s；x为柱内轴向距离，单位为m；u为空塔流速，单位为mL/min；ε为柱内的空隙率；ε
p为颗粒内的孔隙率；k
f为液膜传质系数，单位为m/s；r为颗粒内的径向距离，单位为m；r
p为颗粒半径，单位为m；q为固相蛋白浓度，单位为mg/mL；D
p为颗粒内液相扩散系数，单位为m
2/s；D
s为颗粒内固相扩散系数，单位为m
2/s；L为柱长，单位为m。
上述方程的边界条件为：
在t＝0时，c＝0，c
p＝0；
在x＝0处，
在x＝L处，
在r＝0处，
在r＝r
p处，
所用的蛋白吸附模型为Langmuir吸附等温线模型，其方程如下：
其中：Q
max为饱和吸附量，单位为mg/mL；k
d为解离平衡常数，单位为mg/mL。
具体应用实例中，在连续流层析设计模型的选取上，根据不同的连续流层析操作模式，例如二柱、三柱、四柱、N柱(N>4)柱，建立不同的连续流设计模型，求取过程操作参数与流程安排方案。
(1)当为二柱连续流层析操作模式时，其关键操作参数有连接模式的上样时间和断开模式的上样流速，计算方法如下：
其中：T
C为连接模式的上样时间，单位为min；U
DC代表断开模式的上样流速，单位为mL/min；C
0为蛋白上样浓度，单位为mg；T
DC为断开模式的上样时间，单位为min；U
C为连接模式的上样流速，单位为mL/min；T
1_1％为一柱1％穿透时间点，单位为min；T
1_s％为一柱s％穿透点时间，单位为min；T
2_1％为二柱1％穿透点时间，单位为min；SF为安全因子。
(2)当为三柱连续流层析操作模式时，其关键操作参数有连接模式的上样时间和等待时间，计算方法如下：
T
wait＝T
C-T
RR若T
C＞T
RR
T
wait＝2(T
RR-T
C)若T
RR＞T
C
其中：T
CW代表连接模式的清洗时间，单位为min；T
wait为等待时间，单位为min；T
RR为洗脱清洗再生的总时间，单位为min。
(3)当为三柱连续流层析操作模式时，其关键操作参照有连接模式的上样时间和等待时间，其中连接模式的上样时间与三柱计算方法相同，等待时间计算方法如下：
T
wait＝2T
C-T
RR+T
CW若T
C＞(T
RR-T
CW)/2
T
wait＝2(T
RR-2T
C-T
CW)若(T
RR-T
CW)/2＞T
C
(4)当为N(N>4)柱连续流层析操作模式时，关键操作参数包括柱数、连接模式的上样时间和等待时间，其中连接模式的上样时间与三柱计算方法相同，柱数与等待时间计算方法如下：
T
wait＝(N-2)T
C+(N-3)T
CW-T
RR
其中符号
为向上取整。
具体应用实例中，对于连续流层析评估模型，评估参数主要包括过程产率和介质利用度，过程产率的计算公式如下：
其中P
C为连续流层析的过程产率，单位为g/L/min；U
C为连接模式的上样流速，单位为mL/min；T
DC为断开模式的上样时间，单位为min；CV为柱体积，单位为mL；T
cycle为运行一个循环回到初始状态所需要的总时间，单位为min。
介质利用度计算公式如下：
其中CU
C为连续流层析的介质利用度，单位为％；T
1_95％为一柱95％穿透时间点，单位为min。
通过以上设置的模型和参数，步骤101中，实验穿透曲线拟合的步骤主要包括以下：
(1)依据层析模型，模型参数包括传质相关参数(包括轴向扩散系数、液膜传质系数、颗粒内固相传质系数和颗粒内液相传质系数等)、吸附相关参数(包括饱和吸附量和解离平衡常数等)和操作相关参数(包括空塔流速与上样浓度等)；
(2)将穿透实验中的操作相关参数，赋予初值的传质相关参数和吸附相关参数代入层析机理模型中，采用正交配置法计算产生对应的穿透曲线。
(3)将模型计算所得穿透曲线与实验所得穿透曲线的均方根误差作为目标函数，使用内点法对吸附相关参数和传质相关参数进行拟合，得到目标函数最小时的参数，即拟合的层析模型参数。
具体应用实例中，通过设定的参数和模型，步骤102中，穿透曲线预测的步骤为：
(1)一柱穿透曲线预测：设定层析操作参数范围，在该范围内，将步骤101得到的层析模型参数和层析操作参数代入层析模型中，得到不同流速与不同蛋白浓度下的一柱穿透曲线；
(2)双柱串联穿透曲线预测：设定层析操作参数范围，在该范围内，将步骤101得到的层析模型参数和层析操作参数代入层析模型中，将一柱出口所得穿透曲线中随时间变化的蛋白浓度作为二柱的上样浓度，得到不同流速与不同蛋白浓度的双柱串联穿透曲线。
具体应用实例中，步骤103中连续流层析的过程分析包括如下步骤：
(1)将预测的穿透曲线、连续流层析基本操作参数(包括洗脱清洗再生时间、连接模式清洗的柱体积、安全因子等)和连续流设计参数(切换点选取、保留时间、上样蛋白浓度等)，代入上述连续流层析模型中，得到连续流层析过程的流程安排方案。
(2)评估参数计算步骤：将上一步骤中所得的连续流层析的设计参数与流程安排方案代入上述连续流层析的评估模型，计算得到多柱连续流层析的过程产率和介质利用度。
具体应用实例中，步骤104中，连续流层析的操作空间优化包括如下步骤：
(1)过程产率分布图：基于连续流层析的设计参数范围，生成参数矩阵，对矩阵内的所有参数点根据上述方法计算过程产率，得到过程产率矩阵，对矩阵进行线性插值，绘制在不同操作条件(包括保留时间、切换点、上样蛋白浓度)的过程产率分布图，用于连续流层析过程分析和优化。
(2)介质利用度分布图：基于连续流层析的设计参数范围，生成参数矩阵，对矩阵内的所有参数点根据上述方法计算介质利用度，得到介质利用度矩阵，对矩阵进行线性插值后，绘制在不同操作条件(包括保留时间、切换点、上样蛋白浓度)的介质利用度分布图，用于连续流层析过程分析和优化。
(3)连续流层析的参数优化：基于特定的分离目标(过程产率和介质利用度)，在过程产率分布图和介质利用度分布图中分别计算满足分离目标的连续流设计参数范围，将两个图的设计参数区域进行叠加，得到同时满足过程产率和介质利用度要求的连续流层析设计参数，并计算连续流层析过程的操作参数与流程安排方案。
进一步的，为了使本发明实施例的技术效果更为明显，以下结合数字和图例对发明实施过程列举说明。
实例1实验穿透曲线拟合
(1)实验穿透曲线
采用Purolite Life Sciences公司的Praesto Jetted A50介质进行IgG蛋白的穿透实验，流速为1mL/min，上样蛋白浓度为2mg/mL，达到95％以上穿透浓度后停止上样，上样蛋白总体积为90倍层析柱体积，得到实验穿透曲线。
(2)穿透曲线拟合过程
层析模型参数的初值为：轴向扩散系数3*10
-7m
2/s，液膜传质系数18*10
-6m/s，颗粒内固相传质系数4*10
-13m
2/s，颗粒内液相传质系数6*10
-12m
2/s，饱和吸附量80mg/mL，解离平衡常数0.2mg/mL。实验操作相关参数为：流速1mL/min，上样蛋白浓度2mg/mL。
将模型参数初值和实验操作相关参数代入层析模型中，使用正交配置法产生穿透曲线，并和实验穿透曲线对比，以两者的均方根误差为目标函数，使用内点法求解目标函数的极小值，经过83次迭代后目标函数达到极小值0.011，得到拟合模型参数为：轴向扩散系数0.7*10
-7m
2/s，液膜传质系数34*10
-6m/s，颗粒内固相传质系数0.6*10
-13m
2/s，颗粒内液相传质系数4.5*10
-12m
2/s，饱和吸附量124mg/mL，解离平衡常数0.13mg/mL。
图2为实验穿透曲线和拟合穿透曲线的比较。
实例2穿透曲线预测
(1)一柱穿透曲线预测
使用实例1Praesto Jetted A50介质穿透曲线拟合得到的层析模型参数：轴向扩散系数0.7*10
-7m
2/s，液膜传质系数34*10
-6m/s，颗粒内固相传质系数0.6*10
-13m
2/s，颗粒内液相传质系数4.5*10
-12m
2/s，饱和吸附量124mg/mL，解离平衡常数0.13mg/mL。设定层析操作参数范围为流速0.33mL/min-3mL/min，浓度0.5mg/mL-5mg/mL。在该范围内，层析模型参数和操作参数代入层析模型中，计算得到不同流速和蛋白浓度下的一柱穿透曲线。
(2)双柱串联穿透曲线预测
将上述步骤(1)产生的一柱穿透曲线中随时间变化的蛋白浓度作为二柱的上样蛋白浓度，通过代入与一柱相同的层析模型参数和操作参数，计算得到二柱穿透曲线。例如流速为1.5mL/min、浓度为3mg/mL的一柱和双柱穿透穿透曲线预测结果如图3所示。
实例3连续流层析过程分析和操作空间优化
(1)连续流层析的过程分析
根据实例1 Praesto Jetted A50介质、蛋白浓度C
0为3mg/mL、流速为1.5mL/min的穿透曲线进行连续流层析的过程设计，设计过程如下：
双柱连续流层析设计：连接模式的上样流速U
C与蛋白穿透实验相同(1mL/min),断开模式的上样时间T
DC和层析柱进行洗脱、清洗和再生的总时间T
RR相同(26min)，连接模式清洗的柱体积为4CV，连接模式清洗的流速为1.5mL/min，可求得连接模式的清洗时间T
CW为2.6min。设安全因子SF为0.9，切换点s为80％，一柱达到1％穿透时间T
1_1％为6.5min，一柱达到s穿透时间T
1_s％为25.2min，二柱达到1％穿透时间T
2_1％为21.5min。通过下面两个公式对断开模式的上样流速U
DC和连接模式的上样时间T
C进行求解：
三柱连续流层析设计：T
1_1％，T
1_s％，T
2_1％，T
CW，T
RR的值和前文相同。
由于T
C＜T
RR，故T
wait＝2×(T
RR-T
C)＝26.0(min)
四柱连续流层析设计：T
C，T
CW，T
RR的值和前文相同。
T
C与三柱连续流层析过程相同。由于T
C＞(T
RR-T
CW)/2，故
N柱连续流层析设计：T
C，T
CW，T
RR的值和前文相同。
由公式
可得适用于四柱连续流层析系统。
(2)连续流层析的过程评估和操作空间优化
以双柱连续流层析为例，将上述求得的操作参数代入过程产率和介质利用度的计算公式，其中双柱运行一个周期的时间T
cycle为87.6min，一柱达到95％穿透的时间T
1_95％为33.2min，柱体积CV为1mL。可以得到：
将不同切换点参数(0.1、0.2、…..、0.9)和不同保留时间参数(0.33、0.5、1、1.5、2、2.5、3min)代入上述双柱连续流层析设计和评估方程，得到双柱连续流层析过程的过程产率矩阵和介质利用度矩阵。将矩阵进行线性插值，得到过程产率分布图和介质利用度图，见图4和图5所示。
当输入分离目标，如过程产率大于40g/L/h和介质利用度大于80％后，则可以根据分离目标，在上述两个等高线图中得到满足过程产率大于40g/L/h和介质利用度大于80％的交集，即合适的操作空间，如图6所示。
上述实施例的描述是对基于层析机理模型实现多柱连续流层析设计及分析方法的实施过程的详细说明，接下来将对基于人工神经网络实现多柱连续流层析设计及分析的方法的实施过程进行详细说明。
参照图7，本发明实施例公开了一种基于人工神经网络实现多柱连续流层析设计及分析的方法，具体包括以下步骤：
步骤201，第一人工神经网络训练，用于采用机理模型和实验方法建立层析穿透曲线数 据集和机理模型参数集，以穿透曲线数据集为输入，机理模型参数集为输出，训练得到第一人工神经网络；
步骤202，第二人工神经网络训练，用于采用机理模型和实验方法建立层析穿透曲线数据集和机理模型参数集，以机理模型参数集为输入，穿透曲线数据集为输出，训练得到第二人工神经网络；
步骤203，穿透曲线拟合，用于将实验所得穿透曲线进行线性插值，得到穿透曲线的特征点，将特征点和实验操作参数作为输入代入第一人工神经网络，拟合计算得到机理模型参数；
步骤204，穿透曲线预测，用于将步骤203得到的机理模型参数代入第二人工神经网络，依据层析参数预测范围，得到不同流速与不同蛋白浓度的穿透曲线，并和穿透曲线实验数据进行比较，若误差大于5％，则重新进行第一人工神经网络训练和第二人工神经网络训练，并重新进行步骤203和步骤204；
步骤205，连续流层析的过程分析，用于将步骤204预测的穿透曲线和连续流基本操作参数代入连续流层析模型，得到连续流层析过程的设计参数和评估参数，分析连续流层析操作参数变化对多柱连续流层析的过程产率和介质利用度等性能指标的影响；
步骤206，连续流层析的操作空间优化，用于基于特定的分离目标和要求，确定合适的过程产率和介质利用度，通过步骤205的分析，得到优化的连续流层析设计参数的操作空间。
为了使本发明实施例的具体实施过程能得到更好的理解，以下将对以上步骤实施的具体过程进行进一步的具体描述。
具体应用实例中，层析机理模型为考虑平行扩散的一般性速率模型，其方程如下：
其中：c为层析柱内蛋白浓度，单位为mg/mL；c
p为介质颗粒内蛋白浓度，单位为mg/mL；c
0为上样蛋白浓度，单位为mg/mL；t为时间，单位为s；D
ax为柱内轴向扩散系数，单位为m
2/s；x为柱内轴向距离，单位为m；u为空塔流速，单位为mL/min；ε为柱内的空隙率；ε
p为颗粒内的孔隙率；k
f为液膜传质系数，单位为m/s；r为颗粒内的径向距离，单位为m；r
p为颗粒半径，单位为m；q为固相蛋白浓度，单位为mg/mL；D
p为颗粒内液相扩散系数，单位为m
2/s；D
s为颗粒内固相扩散系数，单位为m
2/s；L为柱长，单位为m。
上述方程的边界条件为：
在t＝0时，c＝0，c
p＝0；
在x＝0处，
在x＝L处，
在r＝0处，
在r＝r
p处，
所用的蛋白吸附模型为Langmuir吸附等温线模型，其方程如下：
其中：Q
max为饱和吸附量，单位为mg/mL；k
d为解离平衡常数，单位为mg/mL。
具体应用实例中，在连续流层析设计模型的选取上，根据不同的连续流层析操作模式，例如二柱、三柱、四柱、N柱(N>4)柱，建立不同的连续流设计模型，求取过程操作参数与流程安排方案。
(1)当为二柱连续流层析操作模式时，其关键操作参数有连接模式的上样时间和断开模式的上样流速，计算方法如下：
其中：T
C为连接模式的上样时间，单位为min；U
DC代表断开模式的上样流速，单位为mL/min；C
0为蛋白上样浓度，单位为mg；T
DC为断开模式的上样时间，单位为min；U
C为连接模式的上样流速，单位为mL/min；T
1_1％为一柱1％穿透时间点，单位为min；T
1_s％为一柱s％穿透点时间，单位为min；T
2_1％为二柱1％穿透点时间，单位为min；SF为安全因子。
(2)当为三柱连续流层析操作模式时，其关键操作参数有连接模式的上样时间和等待时间，计算方法如下：
T
wait＝T
C-T
RR若T
C＞T
RR
T
wait＝2(T
RR-T
C)若T
RR＞T
C
其中：T
CW代表连接模式的清洗时间，单位为min；T
wait为等待时间，单位为min；T
RR为洗脱清洗再生的总时间，单位为min。
(3)当为三柱连续流层析操作模式时，其关键操作参照有连接模式的上样时间和等待时间，其中连接模式的上样时间与三柱计算方法相同，等待时间计算方法如下：
T
wait＝2T
C-T
RR+T
CW若T
C＞(T
RR-T
CW)/2
T
wait＝2(T
RR-2T
C-T
CW)若(T
RR-T
CW)/2＞T
C
(4)当为N(N>4)柱连续流层析操作模式时，关键操作参数包括柱数、连接模式的上样时间和等待时间，其中连接模式的上样时间与三柱计算方法相同，柱数与等待时间计算方法如下：
T
wait＝(N-2)T
C+(N-3)T
CW-T
RR
其中符号
为向上取整。
具体应用实例中，对于连续流层析评估模型，评估参数主要包括过程产率和介质利用度，过程产率的计算公式如下：
其中P
C为连续流层析的过程产率，单位为g/L/min；U
C为连接模式的上样流速，单位为mL/min；T
DC为断开模式的上样时间，单位为min；CV为柱体积，单位为mL；T
cycle为运行一个循环回到初始状态所需要的总时间，单位为min。
介质利用度计算公式如下：
其中CU
C为连续流层析的介质利用度，单位为％；T
1_95％为一柱95％穿透时间点，单位为min。
通过以上设置的模型和参数，步骤201和步骤203中，第一人工神经网络的训练和应用主要包括以下：
(1)机理模型参数集和穿透曲线数据集：
一种实施方式中，依据层析机理模型，模型参数包括传质相关参数(包括轴向扩散系数、液膜传质系数、颗粒内固相传质系数和颗粒内液相传质系数等)、吸附相关参数(包括饱和吸附量和解离平衡常数等)和操作相关参数(包括空塔流速与上样浓度等)，将上述每一个参数均在80％范围内生产一个随机数，将这些随机数按照一定顺序排列产生机理模型参数矩阵，从而形成100-10000组机理模型参数矩阵；将每组参数代入机理模型中，利用正交配置法得到对应的穿透曲线；重复上述步骤，即可得到机理模型参数集和对应的穿透曲线数据集。
另一种实施方式中，利用实验的方法，选择不同的介质和操作参数进行单柱的蛋白穿透实验，实验参数包括空塔流速、上样浓度、饱和吸附量、解离平衡常数，其中饱和吸附量和解离平衡常数通过静态吸附实验得到，进行100-10000实验，得到机理模型参数集和对应的穿透曲线数据集。
(2)第一人工神经网络训练：选取穿透曲线数据集中穿透曲线上的特征点，即达到10％-90％穿透点的上样时间，对特征点进行归一化处理，以穿透曲线的特征点为输入集，对应的机理模型参数作为输出集，训练人工神经网络，记为第一人工神经网络。
(3)第一人工神经网络应用：将实验所得穿透曲线数据进行线性插值，得到穿透曲线的特征点，代入已训练好的第一人工神经网络进行计算，可得到机理模型中对应的模型参数。
具体应用实例中，通过设定的参数和模型，步骤202与步骤204中，对于第二人工神经网络的训练和应用主要包括以下：
(1)机理模型参数集和穿透曲线数据集：
一种实施方式中，依据层析机理模型，模型参数包括传质相关参数(包括轴向扩散系数、液膜传质系数、颗粒内固相传质系数和颗粒内液相传质系数等)、吸附相关参数(包括饱和吸附量和解离平衡常数等)和操作相关参数(包括空塔流速与上样浓度等)。将上述每一个参数均在80％范围内产生一个随机数，将这些随机数按照一定顺序排列产生机理模型参数矩阵，从而形成100-10000组机理模型参数矩阵；将每组参数代入机理模型中，利用正交配置法得到对应的单柱穿透曲线以及双柱串联上样的二柱的穿透曲线；重复上述步骤，即得到机理模型参数集和对应的穿透曲线数据集。
另一种实施方式中，利用实验的方法，选择不同的介质和操作参数进行双柱串联的蛋白穿透实验，此时实验参数包括空塔流速、上样浓度、饱和吸附量、解离平衡常数，其中饱和吸附量和解离平衡常数通过静态吸附实验得到，进行100-10000组上述实验，得到机理模型参数集和对应的穿透曲线数据集。
(2)第二人工神经网络训练：选取穿透曲线数据集中穿透曲线上的特征点，即达到10％-90％穿透点的上样时间和上样量，双柱串联上样二柱的穿透曲线上1％穿透点。对穿透 曲线上的特征点进行提取和归一化处理，以机理模型参数为输入集，对应的穿透曲线的特征点作为输出集，训练神经网络，记为第二人工神经网络。
(3)第二人工神经网络应用：将选择的保留时间和上样蛋白浓度和上述拟合得到的机理模型参数合并，代入已训练好的第二人工神经网络进行计算，可预测得到不同层析操作参数下的穿透曲线的特征点。
具体应用实例中，步骤204中重新进行第一人工神经网络训练和第二人工神经网络训练步骤为：使用机理模型对实验穿透曲线进行拟合，得到该实验条件下的机理模型参数，在该机理模型参数±30％区间内随机生成新的机理模型参数集，代入机理模型中得到新的穿透曲线集，将新的穿透曲线集合并入原先的穿透曲线数据库中，分别使用上述方法重新进行第一人工神经网络训练和第二人工神经网络训练。
具体应用实例中，步骤205中连续流层析的过程分析包括如下步骤：
(1)将预测的穿透曲线特征点、连续流层析基本操作参数(包括洗脱清洗再生时间、连接模式清洗的柱体积、安全因子等)和连续流设计参数(切换点选取、保留时间、上样蛋白浓度等)，代入上述连续流层析模型中，得到连续流层析过程的流程安排方案。
(2)评估参数计算步骤：将上一步骤中所得的连续流层析的设计参数与流程安排方案代入上述连续流层析的评估模型，计算得到多柱连续流层析的过程产率和介质利用度。
具体应用实例中，步骤206中，连续流层析的操作空间优化包括如下步骤：
(1)过程产率分布图：基于连续流层析的设计参数范围，生成参数矩阵，对矩阵内的所有参数点根据上述方法计算过程产率，得到过程产率矩阵，对矩阵进行线性插值，绘制在不同操作条件(包括保留时间、切换点、上样蛋白浓度)的过程产率分布图，用于连续流层析过程分析和优化。
(2)介质利用度分布图：基于连续流层析的设计参数范围，生成参数矩阵，对矩阵内的所有参数点根据上述方法计算介质利用度，得到介质利用度矩阵，对矩阵进行线性插值后，绘制在不同操作条件(包括保留时间、切换点、上样蛋白浓度)的介质利用度分布图，用于连续流层析过程分析和优化。
(3)连续流层析的参数优化：基于特定的分离目标(过程产率和介质利用度)，在过程产率分布图和介质利用度分布图中分别计算满足分离目标的连续流设计参数范围，将两个图的设计参数区域进行叠加，得到同时满足过程产率和介质利用度要求的连续流层析设计参数，并计算连续流层析过程的操作参数与流程安排方案。
进一步的，为了使本发明实施例的技术效果更为明显，以下结合数字和图例对发明实施过程列举说明。
实例4第一人工神经网络训练和穿透曲线拟合
(1)第一人工神经网络训练
在80％范围内随机生成一组机理模型参数：轴向扩散系数5*10
-7m
2/s，液膜传质系数12*10
-6m/s，颗粒内固相传质系数3*10
-13m
2/s，颗粒内液相传质系数9*10
-12m
2/s，饱和吸附量110mg/mL，解离平衡常数0.12mg/mL，空塔流速0.5mL/min，上样蛋白浓度2mg/mL。将上述参数代入机理模型进行计算，得到穿透曲线如图8所示。
机理模型参数矩阵为[5e-7,12e-6,3e-13,9e-12,110,0.12,0.5,2]。
由2000组机理模型参数矩阵构成机理模型参数集，对应于2000条穿透曲线，构成穿透曲线数据集。
选取穿透曲线上的特征点，分别达到10％、20％、30％、40％、50％、60％、70％、80％、90％穿透的时间，由特征点组成的矩阵为[63.7,71.6,78.4,84.2,90.9,98.3,105.9,114.7,127.34]；将该矩阵进行归一化，得到矩阵[0.137,0.135,0.136,0.136,0.139,0.142,0.145,0.145,0.143]；将所有的穿透曲线数据转化为上述归一化矩阵，作为输入集；将机理模型参数集归一化，作为输出集；采用Levenberg-Marquardt为训练函数，以均方根误差为目标函数进行人工神经训练，经过116次迭代后误差为2.84*10
-5，小于1*10
-3，达到训练要求，得到第一人工神经网络。
(2)穿透曲线拟合
采用GE Healthcare公司的Mabselect SuRE介质进行IgG蛋白的穿透实验，流速为0.5mL/min，上样蛋白浓度为1mg/mL；将穿透曲线进行线性插值，得到达到10％-90％穿透的时间点矩阵为[57.4,64.9,71.8,79.3,86.3,93.8,101.2,111.4,125.3]；将该矩阵代入网络1，即可计算得到机理模型参数矩阵为[2e-7,2e-6,4e-13,1.3e-11,90.1,0.3,0.5,1]，图9为实验穿透曲线和拟合穿透曲线的比较。
实例5第二人工神经网络训练和穿透曲线预测
(1)第二人工神经网络训练
在80％范围内随机生成一组机理参数：轴向扩散系数3*10
-7m
2/s，液膜传质系数18*10
-6m/s，颗粒内固相传质系数4*10
-13m
2/s，颗粒内液相传质系数6*10
-12m
2/s，饱和吸附量80mg/mL，解离平衡常数0.2mg/mL，空塔流速1mL/min，上样蛋白浓度1mg/mL。将上述参数代入机理模型中，得到两个解，一个解为一柱穿透曲线，另一个解为双柱串联上样时二柱的穿透曲线，如图10所示。
机理模型参数矩阵为[3e-7,18e-6,4e-13,6e-12,80,0.2,1,1]。
由3000多组机理模型参数构成机理模型参数集，对应于3000多组穿透曲线，构成穿 透曲线数据集。
选取穿透曲线上的特征点，包括达到10％-90％穿透的时间点和上样量，以及二柱1％穿透的时间点，特征点组成的矩阵为[13.0,17.4,21.9,27.4,34.2,43.2,55.3,72.1,93.7,0.31,1.0,2.1,4.1,7.3,12.3,20.2,32.8,51.2,28.1]；将所有的穿透曲线数据转化为上述矩阵，再进行归一化处理，作为输出集。将机理模型参数集归一化作为输入集。采用Levenberg-Marquardt为训练函数，以均方根误差为目标函数进行人工神经训练，经过68次迭代后误差为8.9*10
-4，小于1*10
-3，达到训练要求，得到第二人工神经网络。
(2)穿透曲线预测
使用实例4介质Mabselect SuRE穿透实验拟合计算得到机理模型参数矩阵[2e-7,2e-5,4e-13,1.3e-11,90.1,0.3,0.5,1]，将矩阵代入第二人工神经网络，在不同蛋白浓度和流速下计算得到穿透曲线特征点，例如浓度为1mg/mL、流速分别为0.33mL/min和0.5mL/min的预测结果如图11所示。
实例6连续流层析过程分析和操作空间优化
(1)连续流层析的过程分析
根据Mabselect SuRE介质、蛋白浓度C
0为1mg/mL、流速为1mL/min的穿透曲线进行连续流层析的过程设计，设计过程如下：
双柱连续流层析设计：连接模式的上样流速U
C与蛋白穿透实验相同(1mL/min),断开模式的上样时间T
DC和层析柱进行洗脱、清洗和再生的总时间T
RR相同(26min)，连接模式清洗的柱体积为4CV，连接模式清洗的流速为1mL/min，可求得连接模式的清洗时间T
CW为4min。设安全因子SF为0.9，切换点s为80％，一柱达到1％穿透时间T
1_1％为6.6min，一柱达到s穿透时间T
1_s％为38.1min，二柱达到1％穿透时间T
2_1％为72.1min。通过下面两个公式对断开模式的上样流速U
DC和连接模式的上样时间T
C进行求解：
三柱连续流层析设计：T
1_1％，T
1_s％，T
2_1％，T
CW，T
RR的值和前文相同。
由于T
C＞T
RR，故T
wait＝T
C-T
RR＝9.3(min)
四柱连续流层析设计：T
C，T
CW，T
RR的值和前文相同。
T
C与三柱连续流层析过程相同。由于T
C＞(T
RR-T
CW)/2，故
N柱连续流层析设计：T
C，T
CW，T
RR的值和前文相同。
由公式
可得适用于三柱连续流层析系统。
(2)连续流层析的过程评估和操作空间优化
以双柱连续流层析为例，将上述求得的操作参数代入过程产率和介质利用度的计算公式，其中双柱运行一个周期的时间T
cycle为134.5min，一柱达到95％穿透的时间T
1_95％为107.8min，柱体积CV为1mL。可以得到：
将不同切换点参数(0.1、0.2、…..、0.9)和不同保留时间参数(0.5、1、1.5、2、2.5、3、3.5、4min)代入上述双柱连续流层析设计和评估方程，得到双柱连续流层析过程的过程产率矩阵和介质利用度矩阵。将矩阵进行线性插值，得到过程产率分布图和介质利用度图，见图12和图13所示。
当输入分离目标，如过程产率大于17g/L/h和介质利用度大于70％后，则可以根据分离目标，在上述两个等高线图中得到满足过程产率大于17g/L/h和介质利用度大于70％的交集，即合适的操作空间，如图14所示。
应当理解，本文所述的示例性实施例是说明性的而非限制性的。尽管结合附图描述了本发明的一个或多个实施例，本领域普通技术人员应当理解，在不脱离通过所附权利要求所限定的本发明的精神和范围的情况下，可以做出各种形式和细节的改变。
Claims (15)
- 一种基于层析模型实现多柱连续流层析设计及分析的方法，其特征在于，包括以下步骤：步骤101，实验穿透曲线拟合，将实验穿透曲线和层析操作参数代入层析模型中，拟合穿透曲线，得到层析模型参数；步骤102，穿透曲线预测，限定层析操作范围，将步骤101得到的层析模型参数和层析操作参数代入层析模型中，得到不同流速与不同蛋白浓度下的一柱和双柱串联穿透曲线；步骤103，连续流层析的过程分析，将步骤102预测的穿透曲线和连续流基本操作参数代入连续流层析模型，得到连续流层析过程的设计参数和评估参数，分析连续流层析操作参数变化对多柱连续流层析的过程产率和介质利用度等性能指标的影响；步骤104，连续流层析的操作空间优化，基于特定的分离目标和要求，确定合适的过程产率和介质利用度，通过步骤103的分析，得到优化的连续流层析设计参数的操作空间。
- 如权利要求1所述的基于层析模型实现多柱连续流层析设计及分析的方法，其特征在于，步骤101的实验穿透曲线拟合进一步包括如下步骤：将穿透曲线实验操作参数以及层析模型参数的初始值代入层析模型中，计算穿透曲线，并将计算结果与实验所得穿透曲线进行比较，改变层析模型参数使两者均方根误差达到最小，得到层析模型参数，实现穿透曲线的拟合。
- 如权利要求1所述的基于层析机理模型实现多柱连续流层析设计及分析的方法，其特征在于，步骤102的穿透曲线预测进一步包括如下步骤：设定层析操作流速和蛋白质浓度范围，在该范围之内产生层析操作参数矩阵，与步骤1得到的层析模型参数合并，代入层析模型中进行计算，预测得到不同流速和不同蛋白浓度下的一柱和双柱串联穿透曲线。
- 如权利要求1所述的基于层析模型实现多柱连续流层析设计及分析的方法，其特征在于，所述的连续流层析的过程分析包括如下步骤：将预测的穿透曲线和连续流层析基本操作参数，代入连续流层析模型中，得到连续流层析的过程设计参数与流程安排方案；将所得的连续流层析的过程设计参数与流程安排方案代入连续流层析的评估模型，计算得到多柱连续流层析的过程产率和介质利用度。
- 如权利要求1所述的基于层析模型实现多柱连续流层析设计及分析的方法，其特征在于，所述的连续流层析的操作空间优化包括如下步骤：基于连续流层析的设计参数范围，生成参数矩阵，对矩阵内的所有参数点计算过程产率，得到过程产率矩阵，对矩阵进行线性插值，绘制在不同操作条件的过程产率分布图，用于连续流层析过程分析和优化；基于连续流层析的设计参数范围，生成参数矩阵，对矩阵内的所有参数点计算介质利用度，得到介质利用度矩阵，对矩阵进行线性插值后，绘制在不同操作条件的介质利用度分布图，用于连续流层析过程分析和优化；基于特定的分离目标，在过程产率分布图和介质利用度分布图中分别计算满足分离目标的连续流层析设计参数范围，将两个图的设计参数区域进行叠加，得到同时满足过程产率和介质利用度要求的连续流层析设计参数，并计算连续流层析过程的操作参数与流程安排方案。
- 如权利要求1至5任一所述的基于层析模型实现多柱连续流层析设计及分析的方法，其特征在于，所述的层析模型为考虑平行扩散的一般性速率模型。
- 如权利要求1至5任一所述的基于层析模型实现多柱连续流层析设计及分析的方法，其特征在于，所述的连续流层析模型为根据不同操作模式建立的连续流设计模型，连续流层析的评估参数主要包括过程产率和介质利用度，其中不同操作模式包括二柱、三柱、四柱、N柱，N>4。
- 一种基于人工神经网络实现多柱连续流层析设计及分析的方法，其特征在于，包括以下步骤：步骤201，第一人工神经网络训练，采用机理模型和实验方法建立层析穿透曲线数据集和机理模型参数集，以穿透曲线数据集为输入，机理模型参数集为输出，训练得到第一人工神经网络；步骤202，第二人工神经网络训练，采用机理模型和实验方法建立层析穿透曲线数据集和机理模型参数集，以机理模型参数集为输入，穿透曲线数据集为输出，训练得到第二人工神经网络；步骤203，穿透曲线拟合，将实验所得穿透曲线进行线性插值，得到穿透曲线的特征点，将特征点和实验操作参数作为输入代入第一人工神经网络，拟合计算得到机理模型参数；步骤204，穿透曲线预测，将步骤203得到的机理模型参数代入第二人工神经网络，依据层析参数预测范围，得到不同流速与不同蛋白浓度的穿透曲线特征点，并和穿透曲线实验数据进行比较，若误差大于5％，则重新进行第一人工神经网络训练和第二人工神经网络训练，并重新进行步骤203和步骤204；步骤205，连续流层析的过程分析，将步骤204预测的穿透曲线特征点和连续流基本操 作参数代入连续流层析模型，得到连续流层析过程的设计参数和评估参数，分析连续流层析操作参数变化对多柱连续流层析的过程产率和介质利用度等性能指标的影响；步骤206，连续流层析的操作空间优化，基于特定的分离目标和要求，确定合适的过程产率和介质利用度，通过步骤205的分析，得到优化的连续流层析设计参数的操作空间。
- 如权利要求8所述的基于人工神经网络实现多柱连续流层析设计及分析的方法，其特征在于，步骤201的第一人工神经网络训练和步骤203的穿透曲线拟合进一步包括如下步骤：产生一定范围内随机分布的若干层析机理模型参数集，代入机理模型方程，利用正交配置法产生穿透曲线数据集，或通过实验得到机理模型参数对应的穿透曲线数据集；对穿透曲线上的特征点进行提取和归一化处理，确定神经元节点数和网络层数，使用穿透曲线特征点和机理模型参数分别作为输入集和输出集，训练人工神经网络，记为第一人工神经网络；将实验所得穿透曲线数据进行线性插值，得到穿透曲线特征点，代入已训练好的第一人工神经网络进行计算，得到机理模型中的特征模型参数。
- 如权利要求8所述的基于人工神经网络实现多柱连续流层析设计及分析的方法，其特征在于，步骤202的第二人工神经网络训练和步骤204的穿透曲线预测进一步包括如下步骤：产生一定范围内随机分布的若干层析机理模型参数集，代入机理模型方程，利用正交配置法产生穿透曲线数据集，或通过实验得到机理模型参数对应的穿透曲线数据集；对穿透曲线上的特征点进行提取和归一化处理，确定神经元节点数和网络层数，使用机理模型参数和穿透曲线特征点分别作为输入集与输出集，训练神经网络，记为第二人工神经网络；在一定的保留时间和蛋白浓度范围之内产生层析操作参数矩阵，与机理模型参数合并，代入已训练好的第二人工神经网络进行计算，预测得到不同层析操作条件下的穿透曲线。
- 如权利要求8所述的基于人工神经网络实现多柱连续流层析设计及分析的方法，其特征在于，步骤204中重新进行第一人工神经网络训练和第二人工神经网络训练过程为：使用机理模型对实验穿透曲线进行拟合，得到该实验条件下的机理模型参数，在该机理模型参数±30％区间内随机生成新的机理模型参数集，代入机理模型中得到新的穿透曲线集，将新的穿透曲线集合并入原先的穿透曲线数据库中，分别重新进行第一人工神经网络训练和第二人工神经网络训练。
- 如权利要求8所述的基于人工神经网络实现多柱连续流层析设计及分析的方法，其特征在于，所述的连续流层析的过程分析包括如下步骤：将预测的穿透曲线特征点和连续流层析基本操作参数，代入连续流层析模型中，得到连续流层析的过程设计参数与流程安排方案；将所得的连续流层析的过程设计参数与流程安排方案代入连续流层析的评估模型，计算得到多柱连续流层析的过程产率和介质利用度。
- 如权利要求12所述的基于人工神经网络实现多柱连续流层析设计及分析的方法，其特征在于，所述的连续流层析的操作空间优化包括如下步骤：基于连续流层析的设计参数范围，生成参数矩阵，对矩阵内的所有参数点计算过程产率，得到过程产率矩阵，对矩阵进行线性插值，绘制在不同操作条件的过程产率分布图，用于连续流层析过程分析和优化；基于连续流层析的设计参数范围，生成参数矩阵，对矩阵内的所有参数点计算介质利用度，得到介质利用度矩阵，对矩阵进行线性插值后，绘制在不同操作条件的介质利用度分布图，用于连续流层析过程分析和优化；基于特定的分离目标，在过程产率分布图和介质利用度分布图中分别计算满足分离目标的连续流层析设计参数范围，将两个图的设计参数区域进行叠加，得到同时满足过程产率和介质利用度要求的连续流层析设计参数，并计算连续流层析过程的操作参数与流程安排方案。
- 如权利要求8至13任一所述的基于人工神经网络实现多柱连续流层析设计及分析的方法，其特征在于，所述的层析机理模型为考虑平行扩散的一般性速率模型。
- 如权利要求8至13任一所述的基于人工神经网络实现多柱连续流层析设计及分析的方法，其特征在于，所述的连续流层析模型为根据不同操作模式建立的连续流设计模型，连续流层析的评估参数主要包括过程产率和介质利用度，其中不同操作模式包括二柱、三柱、四柱、N柱，N>4。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US17/623,403 US20220381751A1 (en) | 2019-06-28 | 2019-09-05 | Method for realizing multi-column continuous flow chromatography design and analysis |
Applications Claiming Priority (4)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910573074.7 | 2019-06-28 | ||
| CN201910573074.7A CN110348089B (zh) | 2019-06-28 | 2019-06-28 | 基于层析模型实现多柱连续流层析设计及分析的方法 |
| CN201910573256.4A CN110348090B (zh) | 2019-06-28 | 2019-06-28 | 基于人工神经网络实现多柱连续流层析设计及分析的方法 |
| CN201910573256.4 | 2019-06-28 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2020258515A1 true WO2020258515A1 (zh) | 2020-12-30 |
Family
ID=74060016
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2019/104488 WO2020258515A1 (zh) | 2019-06-28 | 2019-09-05 | 多柱连续流层析设计及分析的方法 |
Country Status (2)
| Country | Link |
|---|---|
| US (1) | US20220381751A1 (zh) |
| WO (1) | WO2020258515A1 (zh) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5729451A (en) * | 1995-12-01 | 1998-03-17 | Coleman Research Corporation | Apparatus and method for fusing diverse data |
| CN101285816A (zh) * | 2008-05-30 | 2008-10-15 | 厦门大学 | 铜锍吹炼的过程参数软测量仪表及其软测量方法 |
| CN101829438A (zh) * | 2010-01-28 | 2010-09-15 | 浙江大学 | 一种模拟移动床色谱操作参数优化方法 |
| CN106596815A (zh) * | 2016-12-07 | 2017-04-26 | 辽宁科技大学 | 用模拟移动床色谱分离复杂混合物中目标物的优化方法 |
| CN107256336A (zh) * | 2017-06-06 | 2017-10-17 | 山东大学 | 幽门螺杆菌对抗生素敏感性和耐药性的量化指标确定方法 |
| CN108717498A (zh) * | 2018-05-19 | 2018-10-30 | 西南石油大学 | 一种油藏相渗曲线模型及相渗曲线计算方法 |
-
2019
- 2019-09-05 WO PCT/CN2019/104488 patent/WO2020258515A1/zh active Application Filing
- 2019-09-05 US US17/623,403 patent/US20220381751A1/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5729451A (en) * | 1995-12-01 | 1998-03-17 | Coleman Research Corporation | Apparatus and method for fusing diverse data |
| CN101285816A (zh) * | 2008-05-30 | 2008-10-15 | 厦门大学 | 铜锍吹炼的过程参数软测量仪表及其软测量方法 |
| CN101829438A (zh) * | 2010-01-28 | 2010-09-15 | 浙江大学 | 一种模拟移动床色谱操作参数优化方法 |
| CN106596815A (zh) * | 2016-12-07 | 2017-04-26 | 辽宁科技大学 | 用模拟移动床色谱分离复杂混合物中目标物的优化方法 |
| CN107256336A (zh) * | 2017-06-06 | 2017-10-17 | 山东大学 | 幽门螺杆菌对抗生素敏感性和耐药性的量化指标确定方法 |
| CN108717498A (zh) * | 2018-05-19 | 2018-10-30 | 西南石油大学 | 一种油藏相渗曲线模型及相渗曲线计算方法 |
Non-Patent Citations (1)
| Title |
|---|
| GAO, ZONGYE, SHI CE, YAO SHAN-JING, LIN DONG-QIANG: "Process Design and Application of Twin-column Continuous Chromatography for Antibody Affinity Separation", JOURNAL OF CHEMICAL ENGINEERING OF CHINESE UNIVERSITIES, ZHEJIANG DAXUE CAILIAO YU HUAXUE GONGCHENG XUEYUAN, CN, vol. 33, no. 1, 28 February 2019 (2019-02-28), CN, pages 117 - 127, XP055769353, ISSN: 1003-9015, DOI: 10.3969/j.issn.1003-9015.2019.00.019 * |
Also Published As
| Publication number | Publication date |
|---|---|
| US20220381751A1 (en) | 2022-12-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112382338B (zh) | 基于自注意力残差网络的dna-蛋白质结合位点预测方法 | |
| Vazquez–Castillo et al. | Design and optimization, using genetic algorithms, of intensified distillation systems for a class of quaternary mixtures | |
| CN107341544A (zh) | 一种基于可分割阵列的可重构加速器及其实现方法 | |
| WO2022100325A1 (zh) | 一种多柱连续流层析捕获蛋白的优化方法 | |
| CN110263925A (zh) | 一种基于fpga的卷积神经网络前向预测的硬件加速实现架构 | |
| CN107103364A (zh) | 一种基于多源域的任务拆分迁移学习预测方法 | |
| CN110348090B (zh) | 基于人工神经网络实现多柱连续流层析设计及分析的方法 | |
| WO2020258515A1 (zh) | 多柱连续流层析设计及分析的方法 | |
| CN112949142A (zh) | 一种基于深度神经网络的ect图像重建方法 | |
| CN114419464A (zh) | 一种基于深度学习的孪生网络变化检测模型 | |
| Sun et al. | Model-based evaluation and model-free strategy for process development of three-column periodic counter-current chromatography | |
| CN109255141A (zh) | 一种汽车车身正向概念设计截面形状优化方法 | |
| CN110348089A (zh) | 基于层析模型实现多柱连续流层析设计及分析的方法 | |
| Behjati et al. | Hierarchical residual attention network for single image super-resolution | |
| Ma et al. | Attentive Enhanced Convolutional Neural Network for Point Cloud Analysis | |
| Fu et al. | Nonlinear wave modeling and dynamic analysis of high-purity heat integrated air separation column | |
| CN112836872B (zh) | 一种基于多gpu的污染物对流扩散方程高性能数值求解方法 | |
| CN114218876A (zh) | 一种基于Apache Spark的分布式流体动力学计算方法 | |
| Liang et al. | A feature selection method based on improved genetic algorithm | |
| CN116072234B (zh) | 基于深度学习的分子动力学轨迹分析方法、存储介质和装置 | |
| Liu et al. | Process design and intensification of multicomponent azeotropes special distillation separation via molecular simulation and system optimization | |
| Yang et al. | Research on composition-process-property prediction of die casting Al alloys via combining feature creation and attention mechanisms | |
| CN110265092A (zh) | 基于人工智能的抗体-抗原分子对接评价方法和系统 | |
| CN113486461B (zh) | 一种基于有限元模拟的球扁钢轧制工艺优化方法 | |
| CN114530217B (zh) | 一种兰炭基多孔碳吸附重金属效率预测方法及相关装置 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 19935067 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 19935067 Country of ref document: EP Kind code of ref document: A1 |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 19935067 Country of ref document: EP Kind code of ref document: A1 |
|
| 32PN | Ep: public notification in the ep bulletin as address of the adressee cannot be established |
Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205 DATED 16/09/2022) |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 19935067 Country of ref document: EP Kind code of ref document: A1 |