An automated synthesis testing system and method for rapid optimization of formulations

Abstract

The present invention discloses an automated synthesis test system and method for realizing rapid optimization of a formulation. By combining automated chemical synthesis, automated testing, and machine learning, a method for realizing rapid optimization of a formulation is proposed, and a class of automated synthesis test systems is presented. The system includes an automated synthesis device, an automated testing device, and a machine learning algorithm module. The above modules can achieve: automatically synthesizing an olfactory sensor, automatically testing to obtain experimental data, automatically predicting a formulation with better performance, and guiding the closed-loop of iterative automated synthesis. This system can achieve rapid optimization of a target formulation in a high-dimensional variable space and obtain a locally optimal formulation composition in a relatively short time.

Description

Technical Field

[0001] The present invention belongs to the field of biochemical synthesis, and particularly relates to an automated synthesis test system and method for realizing rapid formulation optimization. Background Art

[0002] Precise monitoring of leaked gases is crucial for safe operation and pollution prevention in chemical industrial parks. Existing point-type olfactory sensors can only monitor single chemical components, and have problems such as low cost performance, sparse networking, and poor anti-interference ability, and cannot meet the monitoring requirements for large-scale hazardous chemical leakage gases. Although the developing infrared hyperspectral cameras can cover large detection areas, due to large systematic errors and high prices, etc., they cannot be deployed on a large scale for the time being, and there are still uncovered field-of-view dead angles.

[0003] Regarding how to efficiently develop gas-sensitive element formulations, this project proposes closed-loop iterative solutions of "R & D - screening - machine learning" and "machine learning - R & D". Traditional manual synthesis experimental methods have problems such as long development cycles and difficulty in achieving optimization in a multi-dimensional variable space. Through the construction of a high-throughput automated synthesis and screening platform and the combination of machine learning algorithms, this research realizes the functions of automatic synthesis of gas-sensitive elements, automatic testing to obtain data, and iterative optimization by machine learning, thereby breaking through the constraints of the multi-dimensional variable space and finding the global optimization solution.

[0004] In machine learning, generally all or part of the labeled training data is required to train the model, but in actual usage scenarios or production environments, obtaining labeled samples faces problems such as high costs and poor consistency. For the gas-sensitive unit development field faced by this research, traditional manual synthesis methods are inefficient and difficult to operate, and it is difficult to obtain a large number of labeled samples with good consistency. At the same time, there are many factors affecting the performance of gas-sensitive units, and a small amount of data cannot achieve global optimization. The research on the optimization algorithm of the sensitive film formulation based on active learning proposed by this project can, on the one hand, quickly perform experimental labeling of formulations using a high-throughput automated synthesis and screening platform to provide a large number of training sets for machine learning, and on the other hand, through the algorithm of active machine learning, enable the system to actively select samples with potential high performance and high value for the next round of synthesis screening tests during the iterative process, accelerating the system convergence speed, so that after fewer iterative cycles, a gas-sensitive unit formulation with performance close to the theoretical optimum can be obtained.

[0005] Bayesian optimization has important applications in determining the parameters of neural networks. Facing a large number of hyperparameters in complex neural networks, it is difficult to determine the hyperparameter values through functional relationships to obtain the optimal performance of neural networks. The Bayesian optimization algorithm is used to solve the optimization problem of models with implicit relationships from parameters to results. In the present invention, the components of various chemical synthesis raw materials in the formula are regarded as parameters to be optimized, and the performance of the synthesized samples is regarded as the result, which is actually measured by a high-throughput automated synthesis test system. Through the high-throughput automated synthesis test system, this method can obtain the actual performance of a large number of formulas, and combined with the Bayesian optimization algorithm, it can guide the system to quickly optimize the formula parameters. Summary of the Invention

[0006] The object of the present invention is to provide an automated synthesis test system and method for realizing rapid optimization of formulas in view of the deficiencies of the prior art.

[0007] The object of the present invention is achieved by the following technical solutions:

[0008] In the first aspect, the present invention provides an automated synthesis test system for realizing rapid optimization of formulas, including an automated synthesis device, an automated test device, and a machine learning algorithm module;

[0009] The automated synthesis device includes an automated pipetting platform and a sample transfer device; the automated pipetting platform includes an automated pipette, one or more pipette tips, one or more multi-well dropping plates, and a chemical waste bucket; the sample transfer device includes a substrate and motors provided at both ends of the substrate.

[0010] The automated test device includes a gas supply device, a high-throughput test gas chamber, a lead screw motor, and a detection device; the gas supply device is connected to the high-throughput test gas chamber, and a gas rapid mixing grid plate is provided at the connection; the opening and closing of the high-throughput test gas chamber are driven by the lead screw motor; the high-throughput test gas chamber also includes a number of air change valves.

[0011] The gas supply device includes: a number of gas cylinders, each gas cylinder is connected to a gas circuit component through a trachea, and each trachea is connected with a mass flow meter and a main valve.

[0012] The detection device includes a visible light detection device and a hyperspectral detection device for detecting the high-throughput test gas chamber.

[0013] The machine learning algorithm module includes:

[0014] Database: recording existing formulas and the experimental results corresponding to the existing formulas;

[0015] Prediction module: Based on the existing formulations in the database and the experimental results corresponding to the existing formulations, using the Bayesian optimization principle and the Gaussian method to give one or more predicted formulations that are more likely to achieve higher performance;

[0016] Compilation module: Based on the predicted formulations given by the prediction module, compile the predicted formulations into execution commands and send them to the automated synthesis device.

[0017] Further, the pipette tips are compatible with an automated pipette and are disposable.

[0018] Further, the visible light detection device includes a camera, an observation window, and an illumination device; the observation window is opened at the top of the high-throughput test chamber and is composed of a transparent resin or glass; the camera is located above the observation window; the illumination device is a ring of LED light sources surrounding the observation window.

[0019] Further, the hyperspectral detection device includes a detection platform, an optical fiber, and a single-channel spectrometer; the detection platform is placed inside the high-throughput test chamber, an optical fiber is installed at the bottom of the detection platform, one end of the optical fiber is connected to the detection platform, and the other end is connected to the single-channel spectrometer.

[0020] Further, the gas rapid mixing grid plate includes several layers of porous plates and rings, and each two layers of porous plates and a ring form a diffusion gas chamber, and the holes of each layer of porous plates are not aligned with each other.

[0021] In a second aspect, the present invention provides an automated synthesis test method for rapid formulation optimization, including the following steps:

[0022] (1) The automated synthesis device synthesizes an olfactory sensor according to the existing formulation given by the machine learning algorithm module, and the existing formulation includes the types and dosages of each raw material reagent; the automated pipette moves each raw material reagent in the existing formulation into the porous dropping plate to be mixed into the olfactory sensor stock solution; then the automated pipette moves and drops the synthesized olfactory sensor stock solution onto the surface of the substrate; after the olfactory sensor stock solution dries, a gas-sensitive unit is formed; the gas-sensitive unit and the substrate form an olfactory sensor; the waste liquid generated during the above process and the used disposable pipette tips are transferred to a chemical waste bucket;

[0023] (2) Start the motor, drive the movement of the substrate by the rolling of the motors at both ends, so as to transport the olfactory sensor from the automated synthesis device to the high-throughput test chamber in the automated test device; then close the high-throughput test chamber;

[0024] (3) Introduce a mixed gas composed of multiple gases in different proportions into the high-throughput test chamber, and the olfactory sensors in the high-throughput test chamber react with the mixed gas; subsequently, detect the olfactory sensors after the reaction through a visible light detection device or the hyperspectral detection device;

[0025] (3.1) Detection process of the visible light detection device: The olfactory sensors react with the introduced mixed gas in the high-throughput test chamber and produce color changes. The camera takes pictures of the gas-sensitive unit array on the surface of the olfactory sensors and compares the color change amounts of the gas-sensitive unit array. The color change amounts are read through the R, G, B channels of the pictures; the color change amount data is used to input an evaluation function to quantitatively analyze the performance of the formulation of the olfactory sensors;

[0026] (3.2) Detection process of the hyperspectral detection device: The olfactory sensors react with the introduced mixed gas in the high-throughput test chamber and produce color changes. The single-channel spectrometer reads the spectral responses of each gas-sensitive unit of the olfactory sensors on the detection platform through optical fibers to obtain spectral response curves; subsequently, the spectral response curves are input into an evaluation function to quantitatively analyze the performance of the formulation of the olfactory sensors;

[0027] (4) According to the detection results obtained in the steps, quantify the performance of the olfactory sensors with an evaluation function, and store the quantification results and the existing formulations in a database;

[0028] (5) The prediction module is based on the existing formulations in the database and the experimental results corresponding to the existing formulations.

[0029] The beneficial effects of the present invention are as follows:

[0030] 1. In the field of sensor development, an innovative method of using high-throughput synthesis and high-throughput testing to optimize the sensor formulation in a high-dimensional variable space through machine learning iteration;

[0031] 2. An innovative method is proposed for flexible sensitive film type olfactory sensors, using high-throughput synthesis and high-throughput testing to optimize the sensor formulation in a high-dimensional variable space through machine learning iteration;

[0032] 3. A method is proposed to shorten the gas diffusion time in the high-throughput test chamber, shorten the single test time, and improve the high-throughput detection efficiency of the system by quickly mixing one or more gases introduced into the high-throughput test chamber;

[0033] 4. A method is proposed to observe the olfactory sensors using hyperspectral and obtain the spectral responses of the olfactory sensors;

[0034] 5. A method is proposed to import the spectral response of olfactory sensors into a machine learning algorithm module to increase the data dimension of machine learning and improve the convergence speed of machine learning.

[0035] 6. Machine learning is applied to the development of olfactory sensor formulations.

[0036] 7. An innovative method is proposed to use the results of high-throughput experiments to replace traditional data sets for training machine learning models.

[0037] 8. An innovative method is proposed to apply Bayesian optimization and Gaussian methods to the development of olfactory sensors for optimizing the olfactory sensor formulation. Description of the Drawings

[0038] Figure 1 is the "Automated Synthesis Test System" device;

[0039] Figure 2 is the "Automated Synthesis Test System" system;

[0040] Figure 3 is the olfactory sensor;

[0041] Figure 4 is the sample transfer device;

[0042] Figure 5 is the automated test device;

[0043] Figure 6 is the system operation process;

[0044] Figure 7 is the gas rapid mixing grid;

[0045] Figure 8 is the gas rapid mixing grid;

[0046] Figure 9 is the gas rapid grid;

[0047] Figure 10 is the machine learning algorithm module;

[0048] In the figure, 1 - automated pipette, 2 - multi - well dropping plate, 3 - lead screw motor, 4 - automated pipetting platform, 5 - chemical waste barrel, 6 - sample transfer device, 7 - visible light detection device, 8 - high - throughput test gas chamber, 9 - hyperspectral detection device, 10 - olfactory sensor stock solution, 11 - substrate, 12 - gas - sensitive unit, 13 - gas - sensitive unit array, 14 - olfactory sensor, 15 - motor, 16 - camera, 17 - observation window, 18 - lighting device, 19 - gas rapid mixing grid plate, 20 - detection platform, 21 - optical fiber, 22 - single - channel spectrometer, 23 - automated test device, 24 - gas circuit element, 25 - mass flowmeter, 26 - gas cylinder, 27 - main valve, 28 - gas supply device, 29 - air change valve. Specific Embodiment

[0049] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0050] As Figure 1 and Figure 2 shown, the present invention provides an automated synthesis and test system for realizing rapid formula optimization, including an automated synthesis device, an automated test device, and a machine learning algorithm module.

[0051] The automated synthesis device includes an automated pipetting platform 4 and a sample transfer device 6; the automated pipetting platform 4 includes an automated pipette 1, one or more pipette tips, one or more multi - well dropping plates 2, and a chemical waste barrel 5; the pipette tips are disposable and are matched with the automated pipette 1, and the automated pipette 1 and the pipette tips are used for pipetting and mixing chemical substances during the synthesis process; the multi - well dropping plate 2 is used for holding chemical substances during the synthesis process; the chemical waste barrel 5 is used for holding the disposable pipette tips used during the synthesis process; the functions of the automated pipetting platform 4 are: 1) multiple chemical raw materials are pipetted and mixed according to the formula ratio given by the machine learning algorithm module to synthesize the olfactory sensor stock solution; 2) the olfactory sensor stock solution is dropped onto the surface of the substrate 11 carrying chemical substances, and after the olfactory sensor stock solution dries, a gas - sensitive unit is formed. The substrate and the gas - sensitive unit together constitute an olfactory sensor 14 that can react with the gas to be measured, which is used for subsequent tests.

[0052] The described automated synthesis and testing system has the characteristic of high throughput: The automated pipetting platform 4 can independently synthesize different olfactory sensor stock solutions 10 in each well of the porous dropper plate 2 according to different prediction formulas, improving the efficiency of the automated pipetting platform 4; A variety of olfactory sensor stock solutions 10 independently synthesized in the porous dropper plate 2 are dropped onto the surface of the substrate 11 in the form of a matrix of MxN points by the automated pipetting gun 1; After drying, a gas-sensitive unit array arranged in a matrix form is formed on the surface of the substrate 11.

[0053] Definition of olfactory sensor: The entire synthesis process of the olfactory sensor is detailed in Figure 3 .. The olfactory sensor stock solution 10 is dropped onto the surface of the substrate 11, and each gas-sensitive colorimetric point formed after the sensor stock solution dries is called a gas-sensitive unit 12, and multiple gas-sensitive units form a gas-sensitive unit array 13 arranged in a matrix form. The gas-sensitive unit array 13 is located on the surface of the substrate 11, and the two together constitute the olfactory sensor 14.

[0054] The sample transfer device 6 is used to transport the olfactory sensor 14 from the automated pipetting platform 4 to the high-throughput test gas chamber 8. The sample transfer device 6 is set as a pair of motors 15 configured at both ends of the entire system. The sample is placed on the conveyor belt connecting the two ends of the motors, and through the rolling of the two ends of the motors 15, the transportation from the automated synthesis device to the automated testing device is realized, as Figure 4 shown. The sample transfer device 6 is used to transfer the olfactory sensor 14 between the automated pipetting platform 4 and the high-throughput test gas chamber 8; The olfactory sensor 14 includes the gas-sensitive unit array 13 and the substrate 11 carrying the gas-sensitive unit array 13, and the composition of the olfactory sensor 14 is as Figure 3 shown.

[0055] The substrate 11 is a flexible membrane carrying chemical reagents, and the material is porous PTFE material.

[0056] As Figure 5 shown, the automated testing device includes a gas supply device 28, a high-throughput test gas chamber 8, a lead screw motor 3, and a detection device; The gas supply device 28 is connected to the high-throughput test gas chamber 8, and a gas rapid mixing grid plate 19 is provided at the connection; The opening and closing of the high-throughput test gas chamber 8 is driven by the lead screw motor 3; The high-throughput test gas chamber 8 also includes a number of ventilation valves 29, and the ventilation valves are used to discharge the test gas after the reaction.

[0057] The detection device includes a visible light detection device 7 and a hyperspectral detection device 9 for detecting the high-throughput test chamber 8; the function of the automated test device is as follows: in the test process, the olfactory sensor 14 is placed in the high-throughput test chamber 8, and the gas supply device 28 supplies the test gas into the high-throughput test chamber 8. The olfactory sensor 14 will react with the test gas, resulting in a color change. The visible light detection device 7 is used to detect the visible light response of the olfactory sensor 14, and the hyperspectral detection device 9 is used to detect the spectral response of the olfactory sensor 14.

[0058] The gas supply device 28 is used to supply the detection target gas with a specified concentration into the high-throughput test chamber 8. The concentration of the detection target gas can be adjusted within 0% - 100% according to the R & D requirements of the sensor, and the type of the detection target gas is determined according to the R & D requirements of the olfactory sensor. For example, when developing the olfactory sensor formula for carbon dioxide, carbon dioxide gas is supplied.

[0059] The carbon dioxide sensor is a type of olfactory sensor. If the sensor developed here is used to detect carbon dioxide, carbon dioxide gas will be supplied into the test chamber to test the response of the sensor. If it is an olfactory sensor for detecting gases such as ammonia and methane, ammonia, methane and other gases will be supplied into the test chamber.

[0060] Our system is not used to develop olfactory sensors for a specific gas, but is general-purpose. For example, if a customer needs a methane sensor, this system can be used to develop a methane sensor.

[0061] The gas supply device 28 includes: a number of gas cylinders 26 for storing the test gas; each gas cylinder 26 is connected to a gas path component 24 through a gas pipe, and a mass flow meter 25 and a main valve 27 are connected to each gas pipe; the mass flow meter is used to control the mass flow rate of the test gas flowing out of the gas cylinder and can modulate the test gas with the required concentration; the main valve is used to control the opening and closing of the gas cylinder.

[0062] The high-throughput test chamber 8 is used to detect the response of the olfactory sensor in the environment of the detection target gas with a specified concentration; the high-throughput test chamber is composed of an airtight chamber and a detection device for reading the optical and spectral responses of the reagent. The detection device is used to detect the response of the olfactory sensor 14 carried on the surface of the substrate in the high-throughput test chamber 8, identify it and convert it into an electrical signal. The detection device includes a visible light detection device 7 and a hyperspectral detection device 9 for detecting the high-throughput test chamber 8.

[0063] The visible light detection device 7 includes a camera 16, an observation window 17, and an illumination device 18; the observation window 17 is opened at the top of the high-throughput test chamber 8 and is composed of a piece of transparent resin or glass; the camera 16 is located above the observation window 17; the illumination device 18 is a ring of LED light sources surrounding the observation window 17.

[0064] The visible light detection device 7 can convert the observed color change of the olfactory sensor 14 into a change in RGB values. The visible light detection device 7 measures the performance of the olfactory sensor 14 by comparing the color change of the olfactory sensor 14 through colorimetry.

[0065] The hyperspectral detection device 9 includes a detection platform 20, an optical fiber 21, and a single-channel spectrometer 22; the detection platform 20 is placed inside the high-throughput test chamber 8, and the olfactory sensor 14 is placed on the detection platform 20. An optical fiber 21 is installed at the bottom of the detection platform 20. One end of the optical fiber 21 is connected to the detection platform 20, and the other end is connected to the single-channel spectrometer 22.

[0066] The inlet of the high-throughput test chamber 8 is located on the side wall of the high-throughput test chamber 8. A gas rapid mixing grid plate 19 is provided at the inlet; the inlet of the high-throughput test chamber is connected to the gas supply device 28; the test gas stored in the gas supply device 28 is input into the high-throughput test chamber 8 through the gas rapid mixing grid plate 19. The gas rapid mixing grid plate 19 adopts a multi-layer porous structure and includes one or more porous plates and one or more diffusion chambers; the porous plates and the diffusion chambers are arranged at intervals.

[0067] The gas rapid mixing grid plate 19 includes:

[0068] Several layers of porous plates. Each two layers of porous plates and a circular ring form a layer of diffusion chamber, and the holes of each layer of porous plates are not aligned with each other. As Figure 7 shown, the gas rapid mixing grid plate 19 includes 3 layers of porous plates and 2 circular rings, forming 2 layers of diffusion chambers; the holes of the 3 layers of porous plates are not aligned with each other. After the gas is ejected from the first layer of porous plate, it is blocked and refluxed by the wall surface of the second layer of porous plate, forming a turbulent flow in the first diffusion chamber, enhancing the mixing effect of the gas; similarly, after the gas is ejected from the second layer of porous plate, it is blocked and refluxed by the wall surface of the third layer of porous plate, forming a turbulent flow in the second diffusion chamber, enhancing the mixing effect of the gas, as Figure 8 and Figure 9 shown. The number of layers of the porous plate can be modified according to the specific situation.

[0069] The automated test device has the characteristic of high throughput: 1) The automated test device can simultaneously detect and read the responses of multiple gas sensing units 12 arranged in a matrix. The simultaneous detection of multiple gas sensing units 12 can improve the test efficiency of the automated test device 23; 2) During the detection process, when ventilating, the gas concentration in the high-throughput test gas chamber can reach equilibrium within a short time (1 s, the lowest relative concentration is 90%), thus shortening the inflation time required for a single detection and improving the detection efficiency; 3) In the present invention, a method of setting a gas rapid mixing grid plate is also proposed to rapidly mix the gas.

[0070] As Figure 10 shown, the machine learning algorithm module includes:

[0071] Database: Records existing formulations and the experimental results corresponding to the existing formulations;

[0072] Prediction module: Based on the existing formulations in the database and the experimental results corresponding to the existing formulations, use the method of Bayesian optimization to obtain one or more predicted formulations that are more likely to achieve higher performance; specifically, regard the formulation as an input point in the multi-variable parameter space, and regard the performance of the formulation test that the user is interested in (such as sensitivity, response time, etc.) as the output. There will be an unknown black-box function relationship between the input and the output, and it is necessary to find the extreme value of this function at the lowest cost; use Bayesian optimization and its derivative algorithms to handle this task, use the observed values of the parameter space that have been collected and recorded through high-throughput experiments as prior data points, use models such as Gaussian processes and random forests to fit these prior data points, and then use the selected acquisition function to evaluate the fitted hypothetical model, weigh the extreme value positions given by the prior data points and the positions with less information, predict the parameter points in the parameter space where the extreme value of the black-box function is most likely to appear under the observed prior conditions, and use these parameter points as the observed points for the next round of experimental synthesis and testing, that is, the formulations to be explored in the next round of experiments.

[0073] Compilation module: Based on the predicted formulations given by the prediction module, compile the predicted formulations into execution commands and send them to the automated synthesis device.

[0074] Corresponding to the above system embodiment, the present invention also provides an automated synthesis test method for realizing rapid optimization of formulations, which is realized based on an automated synthesis test system for realizing rapid optimization of formulations, and includes the following steps:

[0075] (1) The automated synthesis device synthesizes the olfactory sensor 14 according to the existing formula given by the machine learning algorithm module. The existing formula includes the types and dosages of various raw material reagents. The automated pipette 1 moves each raw material reagent in the existing formula into the porous dropping plate 2 according to the existing formula and mixes them into the olfactory sensor stock solution 10. Subsequently, the automated pipette 1 moves and drops the synthesized olfactory sensor stock solution 10 onto the surface of the substrate 11, i.e., the porous PTFE film. After the olfactory sensor stock solution 10 dries, a gas-sensitive unit 12 is formed. At this time, the gas-sensitive unit 12 and the substrate 11 constitute the olfactory sensor 14. The waste liquid generated during the above process and the used disposable pipette tips are transferred into the chemical waste bucket 5. As Figure 3 shown.

[0076] The gas sensor, the visible light detection system capable of detecting the response of the gas-sensitive unit in the gas environment to be measured, and the hyperspectral detection system together constitute an olfactory sensing system.

[0077] The automated pipetting platform is used to pipette and mix multiple chemical substance raw materials according to the formula ratio given by the machine learning algorithm module, synthesize the olfactory sensor stock solution, and drop the olfactory sensor stock solution onto the surface of the substrate carrying the chemical substances. After the olfactory sensor stock solution dries, a gas-sensitive unit is formed. The substrate and the gas-sensitive unit together constitute the olfactory sensor 14 that can react with the gas to be measured.

[0078] (2) Start the motor 15. The movement of the substrate 11 is driven by the winding of the two-end motor 15, so as to transport the olfactory sensor 14 from the automated synthesis device to the high-throughput test gas chamber 8 in the automated test device. Subsequently, close the high-throughput test gas chamber 8. The opening and closing of the high-throughput test gas chamber 8 are driven by the lead screw motor 3. When the high-throughput test gas chamber 8 is opened, the lead screw motor 3 rotates forward, driving the upper cover of the high-throughput test gas chamber 8 to open. When the high-throughput test gas chamber 8 is closed, the lead screw motor 3 rotates reversely, driving the upper cover of the high-throughput test gas chamber 8 to fall and closely contact the bottom cover of the high-throughput test gas chamber 8 to achieve airtight sealing of the gas chamber.

[0079] (3) Introduce a mixed gas composed of multiple gases in different proportions into the high-throughput test gas chamber 8. The olfactory sensor 14 in the high-throughput test gas chamber 8 reacts with the mixed gas. Subsequently, the reacted olfactory sensor 14 is detected by the visible light detection device 7 or the hyperspectral detection device 9.

[0080] (3.1) Detection process of the visible light detection device 7: The olfactory sensor 14 reacts with the mixed gas in the high-throughput test chamber 8 and produces a color change. The camera 16 in the visible light detection device 7 takes a photo of the gas-sensitive unit array 13 on the surface of the olfactory sensor 14 through the observation window 17 located at the top of the high-throughput test chamber, and compares the color change amount of the gas-sensitive unit array 13. The color change amount is read through the R, G, B channels of the photo. The color change amount data is used to input an evaluation function to quantitatively analyze the performance of the formulation of the olfactory sensor 14.

[0081] The evaluation function is a function used to convert the R, G, B color change amounts into the performance of the olfactory sensor. The specific design is determined by actual requirements. For example: The values of the R, G, B channels are real numbers between 0 and 255. Suppose after ventilation for a period of time, the total value of the numerical changes of the R, G, B channels is E, and the ventilation concentration is h. Then the resolution of the sensor to the gas concentration can be evaluated by k = E / h. It means the color change amount caused by the change of the unit concentration of the gas to be measured. If we want the gas sensor to be more sensitive to the concentration of the gas to be measured, then the value of k should be as large as possible. More complex evaluation functions can also be used, involving some mathematical algorithms, which are difficult to explain in detail here. Moreover, the evaluation function here needs to be customized according to project requirements, such as the evaluation function of the color change amount, the evaluation function of the change speed, etc.

[0082] (3.2) Detection process of the hyperspectral detection device 9: In the process of (3), the olfactory sensor 14 reacts with the mixed gas in the high-throughput test chamber 8 and produces a color change. Multiple optical fibers 21 in the hyperspectral detection device are located below the substrate 11 of the olfactory sensor 14, and the end of each optical fiber is aligned with each gas-sensitive unit 12 of the gas-sensitive unit array 13 of the olfactory sensor 14. The other end of the optical fiber is connected to a single-channel spectrometer 22, and the single-channel spectrometer reads the spectral response of each gas-sensitive unit 12. The spectral response curve is used to input an evaluation function to quantitatively analyze the performance of the formulation of the olfactory sensor 14.

[0083] (4) According to the RGB color change values obtained in step (3) and the detection results of the spectral response curve, the performance of the olfactory sensor 14 is quantified by an evaluation function, and the quantified results and the existing formulations are stored in the database; specifically, step three will obtain the original data of the test module, and different quantified indicators (such as the sensitivity, response time, etc. of the sensor in a certain environment) can be extracted according to different user requirements, and these quantified indicators need to be processed including normalization according to the needs of the optimization algorithm and recorded as the target observation values for algorithm optimization.

[0084] (5) The prediction module is based on the existing formulations in the database and the experimental results corresponding to the existing formulations.

Claims

1. An automated synthesis test system for realizing rapid optimization of a formula, characterized in that, It includes an automated synthesis device, an automated testing device, and a machine learning algorithm module; The automated synthesis device includes an automated pipetting platform (4) and a sample transfer device (6); the automated pipetting platform (4) includes an automated pipette (1), one or more pipette tips, one or more multi-well dropping plates (2), and a chemical waste bin (5); the sample transfer device (6) includes a substrate (11) and motors (15) provided at both ends of the substrate (11); The automated testing device is located on the substrate (11); The automated testing device includes a gas supply device (28), a high-throughput test gas chamber (8), a lead screw motor (3), and a detection device; the gas supply device (28) is connected to the high-throughput test gas chamber (8), and a gas rapid mixing grid plate (19) is provided at the connection; the opening and closing of the high-throughput test gas chamber (8) is driven by the lead screw motor (3); the high-throughput test gas chamber (8) further includes a number of ventilation valves (29); The gas supply device (28) includes: a number of gas cylinders (26), each gas cylinder (26) is connected to a gas path component (24) through a trachea, and a mass flow meter (25) and a main valve (27) are connected to each trachea; The detection device includes a visible light detection device (7) and a hyperspectral detection device (9) for detecting the high-throughput test gas chamber (8); The machine learning algorithm module includes: Database: Records existing formulations and the experimental results corresponding to the existing formulations; Prediction module: Based on the existing formulations in the database and the experimental results corresponding to the existing formulations, use the Bayesian optimization principle and the Gaussian method to give one or more predicted formulations that are more likely to achieve higher performance; Compilation module: Based on the predicted formulations given by the prediction module, compile the predicted formulations into execution commands and send them to the automated synthesis device.

2. The automated synthesis test system for achieving rapid optimization of a formulation according to claim 1, wherein The pipette tip is matched with the automated pipette and is disposable.

3. An automated synthesis test system for rapid optimization of a formulation according to claim 1, characterized in that The substrate (11) is a flexible film for carrying chemical reagents.

4. An automated synthesis test system for rapid optimization of a formulation according to claim 1, characterized in that, The visible light detection device (7) includes a camera (16), an observation window (17), and an illumination device (18); the observation window (17) is opened at the top of the high-throughput test gas chamber (8) and is composed of a transparent resin or glass; the camera (16) is located above the observation window (17); the illumination device (18) is a ring of LED light sources surrounding the observation window (17).

5. An automated synthesis test system for rapid optimization of a formulation according to claim 1, characterized in that, The hyperspectral detection device (9) includes a detection platform (20), an optical fiber (21), and a single-channel spectrometer (22); the detection platform (20) is placed in the high-throughput test gas chamber (8), an optical fiber (21) is installed at the bottom of the detection platform (20), one end of the optical fiber (21) is connected to the detection platform (20), and the other end is connected to the single-channel spectrometer (22).

6. An automated synthesis test system for rapid optimization of a formulation according to claim 1, characterized in that, The gas rapid mixing grid plate (19) includes several layers of perforated plates and rings, and each two layers of perforated plates and a ring form a diffusion gas chamber, and the holes of each layer of perforated plates are not aligned with each other.

7. An automated synthesis test method for rapid formulation optimization of the automated synthesis test system for rapid formulation optimization according to any one of claims 1 to 6, characterized in that, It includes the following steps: (1)The automated synthesis device synthesizes the olfactory sensor (14) according to the existing formula given by the machine learning algorithm module. The existing formula includes the types and dosages of various raw material reagents. The automated pipette (1) moves each raw material reagent in the existing formula into the multi-well dropping plate (2) according to the existing formula and mixes them into the stock solution (10) of the olfactory sensor. Subsequently, the automated pipette (1) moves and drops the synthesized stock solution (10) of the olfactory sensor onto the surface of the substrate (11). After the stock solution (10) of the olfactory sensor dries, a gas-sensitive unit (12) is formed. The gas-sensitive unit (12) and the substrate (11) constitute the olfactory sensor (14). The generated waste liquid and the used disposable pipette tips are transferred into the chemical waste bucket (5). (2)Start the motor (15). The movement of the substrate (11) is driven by the winding of the two motors (15) at both ends, so as to transport the olfactory sensor (14) from the automated synthesis device to the high-throughput test gas chamber (8) in the automated test device. Subsequently, close the high-throughput test gas chamber (8). (3)Introduce a mixed gas composed of multiple gases in different proportions into the high-throughput test gas chamber (8). The olfactory sensor (14) in the high-throughput test gas chamber (8) reacts with the mixed gas. Subsequently, the reacted olfactory sensor (14) is detected by the visible light detection device (7) or the hyperspectral detection device (9). (3.1)Detection process of the visible light detection device (7): The olfactory sensor (14) reacts with the introduced mixed gas in the high-throughput test gas chamber (8) and produces a color change. The camera (16) takes a photo of the gas-sensitive unit array (13) on the surface of the olfactory sensor (14), and compares the color change amount of the gas-sensitive unit array (13). The color change amount is read through the R, G, B channels of the photo. The color change amount data is used to input into the evaluation function to quantitatively analyze the performance of the formula of the olfactory sensor (14). (3.2)Detection process of the hyperspectral detection device (9): The olfactory sensor (14) reacts with the introduced mixed gas in the high-throughput test gas chamber (8) and produces a color change. The single-channel spectrometer (22) reads the spectral response of each gas-sensitive unit of the olfactory sensor (14) on the detection platform (20) through the optical fiber (21) to obtain the spectral response curve. Subsequently, the spectral response curve is input into the evaluation function to quantitatively analyze the performance of the formula of the olfactory sensor (14). (4)According to the detection results obtained in step (3), the performance of the olfactory sensor (14) is quantified by the evaluation function, and the quantified results and the existing formula are stored in the database. (5)The prediction module gives one or more predicted formulas that are more likely to achieve higher performance based on the existing formulas in the database and the corresponding experimental results, using the Bayesian optimization principle and the Gaussian method.

Images