Prediction device, particle size distribution prediction method and particle size distribution prediction program

By using a predictive device and a learned model, the challenge of controlling the size distribution of lipid particles in microfluidic devices was solved, achieving accurate prediction of sizes from 100 to 1000 nm, thus improving prediction accuracy and production stability.

CN123125655APending Publication Date: 2026-11-17KONICA MINOLTA INC +1
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202580027020.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2024-04-09
Filing Date
2025-02-07
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing microfluidic devices struggle to accurately control the particle size distribution when fabricating lipid particles around 1000 nm. While gradient boosting methods are beneficial, they require a large amount of experimental data, resulting in minimal improvement in prediction accuracy.

Method used

A predictive device is used to obtain the production conditions and constituent lipids of lipid particles. Then, machine learning is performed using a learned model to predict the particle size distribution. This includes pretreatment to remove experimental conditions that affect accuracy, adding lipid ratio and diversity index, and controlling the particle size distribution.

Benefits of technology

It enables accurate size distribution prediction of lipid particles ranging from 100 to 1000 nm, improving prediction accuracy and reliability, and promoting the utilization of microfluidic devices and stable customization for large-scale production.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123125655A_ABST
    Figure CN123125655A_ABST
Patent Text Reader

Abstract

The prediction device (100) includes: an information acquisition unit (102) for acquiring the manufacturing conditions of lipid particles composed of any one of phospholipids, anionic lipids and cationic lipids and the constituent lipids of the lipid particles; a learned model (103) for predicting the particle size distribution of lipid particles based on the acquired manufacturing conditions of lipid particles and the constituent lipids of the lipid particles; and an output unit (104) for outputting the particle size distribution of lipid particles predicted by the learned model (103).
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to a prediction device, a method for predicting particle size distribution using the prediction device, and a particle size distribution prediction program. Background Technology

[0002] Lipid particles, modulated by mixing phospholipids or other lipids (e.g., ionized lipids, PEGylated lipids, cholesterol), are primarily used as nanomedicines due to their relatively high safety and biocompatibility. Formulations encapsulated and loaded into lipid particles are bio-inspired nanocarriers that can protect drugs from degradation, control drug release, enable targeted drug delivery, alter in vivo distribution, and improve bioavailability.

[0003] Furthermore, lipid particles are among the most popular cell-mimicking compartments for creating artificial cells. Therefore, lipid particles are used not only in drug delivery systems (DDS) to encapsulate and carry drugs, cosmetics, and pesticides, but also play a central role in the bottom-up creation of bacterial-sized artificial cells.

[0004] The cellular mimicry compartments of lipid particles include substrate uptake, protein movement, phospholipid biosynthesis, cell division and related processes, and the evolution of membrane proteins. The main parameters of lipid particles are their size, lipid composition, and laminar structure, which determine their physicochemical properties and alter compound encapsulation efficiency, cellular uptake efficiency, and targeted tissue transferability.

[0005] Furthermore, in the bottom-up construction of artificial cells, the morphology and size of lipid particles can cause significant changes in the behavior of inward-oriented cytoskeletal proteins.

[0006] Therefore, the control of lipid particle size and lipid composition plays an important role in a wide range of fields, such as constructing artificial cells using bacterial-sized lipid particles.

[0007] Currently, there are two main strategies for fabricating lipid particles. One is a top-down approach that involves size reduction processes following large vesicles (extrusion, high shear stress, ultrasonic treatment). The other is a bottom-up approach that promotes the self-assembly of individual lipid monomers to form vesicles.

[0008] In recent years, the use of microfluidic devices as a bottom-up approach to fabrication has become widespread. A major advantage of using microfluidics to fabricate lipid particles is the ability to stably scale up production, thanks to the high reproducibility of the resulting lipid particle size distribution and the scalability provided by microfluidic parallelization. Here, the inventors have discovered that by making the lipids have the same composition as *E. coli*, liposomes as small as 1 μm can be fabricated even in microfluidics.

[0009] In the assembly of lipid particles via microfluidics, the size of the lipid particles can be controlled by adjusting the total flow rate and ratio of the injection rates of the organic and aqueous phases. Furthermore, the inherent properties of the lipids used (charge, phase transition temperature, molecular weight, etc.), the chip structure, and the mixing temperature are important factors determining the particle size and polydispersity of the resulting lipid particles.

[0010] Here, existing microfluidic devices are able to control the particle size distribution of lipid particles in the range of approximately 100 nm to 300 nm.

[0011] However, due to the interaction of multiple factors such as the flow rate ratio of buffer to lipid, lipid composition, and pH of buffer, especially when fabricating lipid particles of around 1000 nm, it is difficult to achieve accurate distribution control in microfluidic devices.

[0012] To analyze such complex factors, machine learning-based predictions are being made, such as in gradient boosting methods like LightGBM (Light Gradient Boosting Machine), which can accurately predict the target variable and assess the contribution of explanatory variables (see, for example, Patent Documents 1 and 2).

[0013] Patent Document 1: Japanese Patent Application Publication No. 2023-022170

[0014] Patent Document 2: Japanese Patent Application Publication No. 2021-085849

[0015] While gradient boosting is a useful method, the accuracy improvement gained by adjusting the hyperparameters is negligible, requiring the addition of additional features. Therefore, it has been indicated that a considerable amount of experimental data is needed for gradient boosting. Consequently, predicting lipid particle size is challenging using gradient boosting. Summary of the Invention

[0016] Therefore, the objective of this invention is to provide a prediction device capable of predicting the accurate size distribution of lipid particles with a size of 100 to 1000 [nm], a particle size distribution prediction method for the prediction device, and a particle size distribution prediction program.

[0017] That is, the above-mentioned problems of the present invention are solved by the following configuration.

[0018] (1) A prediction device, wherein:

[0019] The information acquisition department acquires the manufacturing conditions of lipid particles composed of any one of phospholipids, anionic lipids, and cationic lipids, as well as the constituent lipids of the aforementioned lipid particles.

[0020] The learned model, based on the obtained lipid particle fabrication conditions and the constituent lipids of the lipid particles, predicts the particle size distribution of the lipid particles; and

[0021] The output section outputs the particle size distribution of lipid particles predicted by the learned model.

[0022] (2) According to the prediction device described in (1), wherein,

[0023] The learned model described above was obtained by machine learning using the production conditions of the lipid particles and the composition of the lipid particles as explanatory variables and the particle size distribution of the lipid particles as the target variable.

[0024] (3) The prediction device according to (1) or (2), wherein,

[0025] The above-mentioned lipid particle size distribution is a curve representing the proportion of lipid particles at each particle size.

[0026] (4) According to the prediction device described in (3), wherein,

[0027] The particle sizes mentioned above are in the range of 0 [nm] to 10000 [nm].

[0028] (5) According to the prediction device described in (2), wherein,

[0029] The learned model described above was obtained through machine learning using the preprocessed explanatory or target variables described above.

[0030] (6) According to the prediction device described in (5), wherein,

[0031] For the explanatory variables mentioned above, the physical properties of lipids were obtained or predicted in the preprocessing, and a weighted average was added in conjunction with the lipid proportion of each lipid particle.

[0032] (7) According to the prediction device described in (5), wherein,

[0033] For the explanatory variables mentioned above, a lipid diversity index was added to the pretreatment process.

[0034] (8) According to the prediction device described in (5), wherein,

[0035] For the aforementioned target variable, the experimental data after removing lipid particle samples that had aggregated were used as variables in the above preprocessing.

[0036] (9) According to the prediction device described in (8), wherein,

[0037] Based on the above experimental data, experimental conditions were defined as follows: the removal of phosphatidylethanolamine relative to the weight of the lipids constituting the lipid particles exceeded 60%, or the lipid flow rate was defined as the sum of the lipid and buffer flow rates exceeded 0.6.

[0038] (10) According to the prediction device described in (1), wherein,

[0039] The conditions for preparing the lipid particles include at least one or a combination thereof selected from the flow rate ratio of the composition solution used for preparing lipid particles, pH, and lipid composition.

[0040] (11) A method for predicting particle size distribution using a prediction device, comprising:

[0041] The information acquisition department acquires the manufacturing conditions of lipid particles composed of any one of phospholipids, anionic lipids, and cationic lipids, as well as the steps for the formation of the lipid particles.

[0042] The learned model predicts the particle size distribution of lipid particles based on the obtained lipid particle fabrication conditions and the lipid particle composition; and

[0043] The output section outputs the particle size distribution of lipid particles predicted by the learned model.

[0044] (12) A particle size distribution prediction program, wherein a computer is made to perform:

[0045] The conditions for obtaining lipid particles composed of any one of phospholipids, anionic lipids, and cationic lipids, and the steps for forming the lipid particles.

[0046] The steps involve enabling the learned model to predict the particle size distribution of lipid particles based on the obtained lipid particle fabrication conditions and the lipid particle composition; and

[0047] The step is to output the particle size distribution of lipid particles predicted by the learned model.

[0048] According to the present invention, it is possible to predict the accurate size distribution of lipid particles with a size of 100 to 1000 nm. Attached Figure Description

[0049] Figure 1 This is a diagram illustrating an example of the structure of the prediction device according to this embodiment.

[0050] Figure 2 This is a flowchart illustrating the particle size distribution prediction process performed by the prediction device involved in this embodiment.

[0051] Figure 3A This is a graph representing the predicted accuracy of the particle size distribution as a preprocessing step, prior to excluding steps S003 and S005.

[0052] Figure 3B It means from generation Figure 3A The graph shows the prediction accuracy of the particle size distribution after excluding the processing steps S003 and S005 in the results processing.

[0053] Figure 4 This is an explanatory diagram illustrating the concepts of obtaining the physical property values ​​shown in step S007 and predicting the physical property values ​​shown in step S009.

[0054] Figure 5A It is a chart (one of the charts) showing the particle size distribution of a specified lipid predicted by the predictive device.

[0055] Figure 5B This is a chart of measured values ​​of lipid particles actually produced using a microfluidic system (Part 1).

[0056] Figure 6A It is a chart (Part Two) showing the particle size distribution of the specified lipids predicted by the prediction device.

[0057] Figure 6B This is a chart showing the measured values ​​of lipid particles actually produced using a microfluidic system (Part Two). Detailed Implementation

[0058] The following is a detailed description of the methods for implementing the present invention. Furthermore, the embodiments described below are merely examples for implementing the present invention and should be appropriately modified or changed according to the structure of the device to which the present invention is applied and various conditions. Therefore, the present invention is not limited to the embodiments described below. In addition, a portion of the embodiments described below may be appropriately combined to form the invention. Furthermore, the same reference numerals are used for the same components, and descriptions are appropriately omitted.

[0059] <This implementation method>

[0060] [Overall structure of the image processing device]

[0061] Figure 1This diagram illustrates a structural example of the prediction apparatus according to this embodiment. The prediction apparatus 100 according to this embodiment is configured with a CPU 101, an information acquisition unit 102, a learned model 103, an output unit 104, a display unit 105, a ROM (Read-Only Memory) 106, a RAM (Random Access Memory) 107, and an external storage device 108. Furthermore, the external storage device 108 is configured to include a particle size distribution prediction program 109.

[0062] like Figure 1 As shown, CPU 101 is a processing unit that oversees the entire control prediction device 100. CPU 101 performs various controls by executing programs stored in ROM 106 or external storage device 108. For example, CPU 101 executes the particle size distribution prediction program 109 stored in external storage device 108 to perform the following... Figure 2 The flowchart shown illustrates the particle size distribution prediction process.

[0063] Information acquisition unit 102 acquires the manufacturing conditions and constituent lipids of lipid particles composed of any one of phospholipids, anionic lipids, and cationic lipids. Information acquisition unit 110 is, for example, composed of operating components such as a keyboard, mouse, various switches, buttons, and touch panels that receive various operations from the user. Furthermore, lipid particles refer to vesicles composed of at least one or a combination of ionized (pH-sensitive) lipids, PEGylated lipids, cholesterol, or phospholipids.

[0064] The learned model 103 predicts the particle size distribution of lipid particles based on the obtained lipid particle fabrication conditions and the constituent lipids of the lipid particles. The lipid particle fabrication conditions include at least one or a combination thereof selected from the flow rate ratio of the composition solution used for lipid particle fabrication, pH, and lipid composition. The constituent lipids of the lipid particles consist of ionized lipids, PEGylated lipids, phospholipids, and cholesterol. The predicted lipid particle size distribution is a curve representing the proportion of lipid particles present at each particle size. Furthermore, the particle sizes of the lipid particles range from 0 [nm] to 10000 [nm].

[0065] The learned model 103 is constructed by machine learning using lipid particle manufacturing conditions and lipid particle composition as explanatory variables and lipid particle size distribution as the target variable. The learned model 103 can also be obtained by machine learning using pre-processed explanatory or target variables. For example, for the explanatory variables, the physical properties of lipids can be obtained or predicted during pre-processing, and a weighted average can be added along with the lipid proportions of each lipid particle. Furthermore, for the explanatory variables, a lipid diversity index can be added during pre-processing.

[0066] Alternatively, for the target variable, experimental data after removing (excluding) samples of lipid particles that have agglomerated can be used as variables during preprocessing. In this case, experimental data can also exclude experimental conditions where phosphatidylethanolamine accounts for more than 60% of the weight of the lipids constituting the lipid particles. Furthermore, experimental data can also exclude actual conditions where the lipid flow rate exceeds 0.6 relative to the sum of the lipid and buffer flow rates. These values ​​are examples and are not limited to these. For example, experimental conditions where the ratio of phosphatidylethanolamine to lipid weight exceeds 75% can also be excluded.

[0067] The learned model 103 is constructed by including a pre-learned recognizer and stores and saves various parameters. In addition, the learned model 103 only needs to be able to predict the particle size distribution of lipid particles based on the obtained lipid particle manufacturing conditions and the lipid composition of the lipid particles. For example, different predefined learning models can also be applied.

[0068] The output unit 104 outputs the particle size distribution of lipid particles predicted by the learned model 103. The output unit 104 outputs the data to a printer, display, or information processing device, for example.

[0069] The display unit 105 may be composed of display devices such as liquid crystal displays, organic EL displays, and printers, or software (viewers) used for display.

[0070] ROM106 stores, for example, the control program of control prediction device 100.

[0071] RAM 107, for example, is composed of DRAM (Dynamic Random Access Memory) and functions as working memory that temporarily stores the data required by CPU 101 to execute the particle size distribution prediction program 109.

[0072] External storage device 108, for example, is a hard disk drive (HDD) that stores necessary data in an interconnected manner.

[0073] [Particle size distribution prediction processing of the prediction device]

[0074] Next, regarding the particle size distribution prediction process of lipid particles performed by the prediction device 100 containing the above-described structure, refer to... Figure 1 And use a flowchart to illustrate.

[0075] Figure 2 This is a flowchart illustrating the particle size distribution prediction process performed by the prediction device according to this embodiment. Furthermore, steps S001 to S011 correspond to preprocessing.

[0076] First, the prediction device 100 accepts the user's operation and obtains data from the information acquisition unit 102 on the production conditions of lipid particles composed of any one of phospholipids, anionic lipids, and cationic lipids, as well as data on the lipids that make up the lipid particles (step S001).

[0077] Next, the prediction device 100 sets experimental conditions in which phosphatidylethanolamine is removed (excluded) at a ratio of more than 75% of the weight of the lipids constituting the lipid particles (step S003). Furthermore, the experimental conditions in the exclusion setting are not limited to the condition where phosphatidylethanolamine exceeds 75%, and phosphatidylethanolamine may be composed of other proportions.

[0078] Furthermore, the prediction device 100 sets experimental conditions where the flow rate of lipid removal (exclusion) exceeds 0.6 relative to the sum of the flow rates of lipid and buffer (step S005). Additionally, the experimental conditions in the exclusion setting are not limited to the condition where the flow rate of lipid exceeds 0.6 relative to the sum of the flow rates of lipid and buffer.

[0079] Figure 3A This is a graph representing the predicted accuracy of the particle size distribution as preprocessing, prior to excluding steps S003 and S005. Figure 3A In the chart, the horizontal axis represents the prediction, and the vertical axis represents the correct answer.

[0080] Figure 3A The sample showing agglomeration deviates significantly from the predicted accuracy of 300 indicated by the straight line, suggesting a decrease in accuracy. Therefore, in steps S003 and S005, the prediction device 100 excludes samples with lower predicted accuracy due to particle size distribution.

[0081] Figure 3B This indicates that the measurement is from Figure 3A The graph shows the results of the particle size distribution measurement after excluding the processing steps S003 and S005. Figure 3B The charts are also related Figure 3ASimilarly, the horizontal axis represents the prediction, and the vertical axis represents the correct answer.

[0082] like Figure 3B As shown, the plot, which is closer to the straight line indicating a prediction accuracy of 300, shows reduced bias. This suggests that by excluding samples from experimental conditions that decrease accuracy, it is possible to achieve a significant improvement in data accuracy while simultaneously predicting the particle size distribution of lipid particles.

[0083] Return to Figure 2 Let's continue with the explanation.

[0084] The prediction device 100 receives the compound name of the lipid used in the experiment as input via the information acquisition unit 102. The prediction device 100 acquires the physical property values ​​of the lipid from external data (e.g., a compound information database) (acquisition of physical property values) (step S007). Then, the prediction device 100 adds a weighted average of the physical property values ​​corresponding to the proportion of lipid contained in the particles as an explanatory variable (step S008).

[0085] Furthermore, the prediction device 100 receives input from the SMILES notation (Simplified Molecular Input line Entry System) of the lipids used in the experiment via the information acquisition unit 102. SMILES notation is a method of representing the chemical structure of a molecule using ASCII alphanumeric strings (represented in text form) without structural ambiguity. The prediction device 100 applies a natural language processing model to the string represented in SMILES notation to predict the physical properties of the lipids (property value prediction) (step S009). Then, the prediction device 100 adds a weighted average of the predicted lipid properties, which corresponds to the proportion of lipids contained in the particles, to the explanatory variables (step S010).

[0086] Figure 4 This is an explanatory diagram illustrating the concepts of obtaining the physical property values ​​shown in step S007 and predicting the physical property values ​​shown in step S009.

[0087] In step S007, the prediction device 100 as follows Figure 4 As shown in the previous paragraph, the lipid property values ​​are obtained from the compound information database by accepting the compound name of the lipid as input in the information acquisition unit 102. Then, in step S008, the prediction device 100 calculates a weighted average based on the obtained lipid property values ​​and adds it as an explanatory variable.

[0088] In this case, for example, the physical properties of the lipids obtained can be logP, mass, number of hydrogen bond donors, number of hydrogen bond acceptors, etc.

[0089] On the other hand, in step S009, the prediction device 100 as Figure 4 As shown in the next paragraph, by accepting the SMILES notation of lipids as input in the information acquisition unit 102, the physical property values ​​of lipids are predicted according to a natural language processing model. Then, in step S010, the prediction device 100 adds a weighted average of the predicted lipid physical property values ​​(lipid physical property prediction values) to the explanatory variables based on the predicted physical property values ​​(lipid physical property prediction values).

[0090] In this case, for example, as predicted physical properties, those that meet the criteria include Balaban J, Labute ASA, solubility, and number of valence electrons.

[0091] Return to Figure 2 Let's continue with the explanation.

[0092] Next, the prediction device 100 adds the diversity index of each particle as an explanatory variable based on the proportion of lipids that make up the lipid particles (step S011).

[0093] Here, as an example of a diversity index, Equation (1) represents the Simpson diversity index to be applied.

[0094] [Formula 1]

[0095]

[0096] Where, n i The proportion (mass%) of the i-th lipid relative to the whole.

[0097] The prediction device 100 predicts the particle size distribution of lipid particles based on the production conditions of the lipid particles and the constituent lipids of the lipid particles by performing (setting) preprocessing from step S001 to step S011 (step S013).

[0098] Figure 5A This is a graph showing the particle size distribution of the specified lipids predicted by the prediction device 100. On the other hand, Figure 5B This is a graph showing the measured values ​​of lipid particles actually fabricated using microfluidic devices. Figure 5A , Figure 5B In the figure, the horizontal axis represents the size [nm], and the vertical axis represents the intensity [%).

[0099] like Figure 5A As shown in the graphs, prediction models 501-503 represent predictions for each flow rate ratio of lipid flow rate to buffer flow rate. The prediction device 100 can adjust the flow rate of the microfluidic system and control the size of the lipid particles by controlling the flow rate ratio. Furthermore, the lipid particle size distribution is a curve representing the proportion of lipid particles at each size.

[0100] Prediction model 501 represents the case where the lipid-buffered fluid flow rate is 0.3. Prediction model 502 represents the case where the lipid-buffered fluid flow rate is 0.4. Prediction model 503 represents the case where the lipid-buffered fluid flow rate is 0.5.

[0101] Prediction models 501-503 all encompass and predict ranges from 0 [nm] to approximately 10 [nm]. 4 The range of [nm] is specifically predicted to be around 10. 2 [nm] ~ Approximately 10 3 Size distribution in [nm].

[0102] On the other hand, such as Figure 5B As shown in the charts, measured value models 551-553 represent the measured values ​​of each flow rate ratio of lipid flow rate to buffer flow rate. That is, these measured value models 551-553 are the actual fabrication results of lipid particles fabricated using microfluidic devices.

[0103] Measured value model 551 represents the case where the lipid-buffered fluid flow rate is 0.3. Measured value model 552 represents the case where the lipid-buffered fluid flow rate is 0.4. Measured value model 553 represents the case where the lipid-buffered fluid flow rate is 0.5.

[0104] The measured values ​​of models 551-553 all exhibit curves with the same characteristics as the predicted models 501-503. This indicates that the predicted models 501-503 of the prediction device 100, by inputting specified experimental conditions, can predict the results in the production of lipid particles. In particular, the predicted models 501-503 of the prediction device 100 encompass approximately 10... 2 [nm] ~ Approximately 10 3 The size of [nm].

[0105] Similarly, by changing the experimental conditions, we can verify whether the size of lipid particles can also be predicted under other experimental conditions.

[0106] Figure 6A This is a graph showing the particle size distribution of the specified lipids predicted by the prediction device 100. On the other hand, Figure 6B This is a graph showing the measured values ​​of lipid particles actually fabricated using microfluidic devices. Figure 6A , Figure 6B In the figure, the horizontal axis represents the size [nm], and the vertical axis represents the intensity [%).

[0107] like Figure 6AAs shown in the charts, prediction models 601–603 represent the predictions for each flow rate ratio of lipid to buffer. Prediction model 601 represents the case where the lipid to buffer flow rate is 0.1. Prediction model 602 represents the case where the lipid to buffer flow rate is 0.2. Prediction model 603 represents the case where the lipid to buffer flow rate is 0.4.

[0108] Prediction models 601-603 all encompass and predict ranges from 0 [nm] to approximately 10 [nm]. 4 The range of [nm] is specifically predicted to be around 10. 2 [nm] ~ Approximately 10 3 Size distribution in [nm].

[0109] On the other hand, such as Figure 6B As shown in the charts, measured value models 661-663 represent the measured values ​​of each flow rate ratio of lipid flow rate to buffer flow rate. These measured value models 661-663, like measured value models 5511-553, represent the actual fabrication results of lipid particles using microfluidic devices.

[0110] Measured value model 661 represents the case where the flow rate of lipid to buffer is 0.1. Measured value model 662 represents the case where the flow rate of lipid to buffer is 0.2. Measured value model 663 represents the case where the flow rate of lipid to buffer is 0.4.

[0111] The measured values ​​of models 661-663 all exhibit curves with the same characteristics as the predicted models 601-603. This indicates that, by inputting specified experimental conditions, the predicted models 601-603 of the prediction device 100, like the predicted models 501-503, can predict the results in the production of lipid particles. In particular, the predicted models 501-503 of the prediction device 100 encompass approximately 10... 2 [nm] ~ Approximately 10 3 The size of [nm].

[0112] As described above, the prediction device 100 according to this embodiment is configured with an information acquisition unit 102, a learned model 103, and an output unit 104. The information acquisition unit 102 acquires the manufacturing conditions and constituent lipids of lipid particles composed of any one of phospholipids, anionic lipids, and cationic lipids. The learned model 103 predicts the particle size distribution of the lipid particles based on the acquired manufacturing conditions and constituent lipids. The output unit 104 outputs the particle size distribution of the lipid particles predicted by the learned model 103.

[0113] Therefore, the prediction device 100 according to this embodiment can predict ranges from 100 to 1000 [nm] (10 3The accurate size distribution of lipid particles with a size of [nm].

[0114] Furthermore, the prediction device 100 according to this embodiment is configured to include a learned model 103. The learned model 103 can also be a learned model obtained by machine learning, using the production conditions of lipid particles and the composition of lipid particles as explanatory variables and the particle size distribution of lipid particles as the target variable.

[0115] Based on the above structure, the learned model 103 can perform data analysis called supervised learning and classify explanatory variables according to the target variable. Therefore, it can classify the production conditions of lipid particles and the lipids that make up the lipid particles with high accuracy.

[0116] Furthermore, in the prediction device 100 of this embodiment, the particle size distribution of lipid particles may be a curve representing the proportion of lipid particles at each particle size.

[0117] Based on the above structure, the prediction models 501-503 and 601-603 in the prediction device 100 can adjust the flow rate of the microfluidic system, for example, by controlling the flow rate ratio, and thus control the size of the lipid particles. Therefore, the prediction models 501-503 and 601-603 can serve as curves representing the presence ratio of lipid particles at each particle size.

[0118] Furthermore, in the prediction device 100 of this embodiment, each particle size may be set to a range of 0 [nm] to 10000 [nm].

[0119] Based on the above structure, since it is possible to predict larger particle sizes than lipid particles produced by existing microfluidic devices, it is possible to promote the utilization of microfluidic devices and achieve stable customization at the production scale.

[0120] Furthermore, in the prediction device 100 of this embodiment, the learned model 103 may also be obtained by machine learning through preprocessed explanatory variables or target variables.

[0121] According to the above structure, the prediction device 100 uses the learned model 103 to perform machine learning on the target variable or explanatory variable, thereby improving accuracy and reliability, and enhancing versatility.

[0122] Furthermore, in the prediction device 100 according to this embodiment, the physical properties of lipids can be acquired or predicted during preprocessing, and a weighted average can be added in conjunction with the lipid proportion of each lipid particle. In this case, a lipid diversity index can also be added during preprocessing for the explanatory variable. Alternatively, experimental data after removing samples of lipid particles that have agglomerated can be used as a variable for the target variable during preprocessing.

[0123] In particular, experimental data can also exclude experimental conditions where phosphatidylethanolamine accounts for more than 60% of the weight of the lipids constituting the lipid particles or where the lipid flow rate is more than 0.6 times the sum of the lipid and buffer flow rates.

[0124] Based on the above structure, the prediction device 100 can achieve a significant improvement in the accuracy of the data by performing such preprocessing, and at the same time be able to predict the size distribution with high precision.

[0125] In particular, the prediction device 100 can dramatically improve accuracy by setting experimental conditions to be excluded based on the experimental data. For example, in Figure 3A In the middle, the prediction accuracy is set at a coefficient of 0.678, but... Figure 3B In this study, by excluding the prescribed experimental conditions, the prediction accuracy can be increased to 0.884 by the prescribed coefficient.

[0126] Furthermore, the prediction device 100 according to this embodiment may also include at least one or a combination thereof selected from the flow rate ratio, pH, and lipid composition of the composition solution used for lipid particle production as lipid particle production conditions.

[0127] Based on the above structure, the prediction device 100 can predict what size lipid particles can be produced under specified production conditions (experimental conditions).

[0128] Explanation of reference numerals in the attached figures: 100…Prediction device; 101…CPU; 102…Information acquisition unit; 103…Learned model; 104…Output unit; 105…Display unit; 106…ROM; 107…RAM; 108…External storage device; 109…Particle size distribution prediction program.

Claims

1. A prediction device, wherein, have: The information acquisition unit acquires the manufacturing conditions of lipid particles composed of any one of phospholipids, anionic lipids, and cationic lipids, as well as the constituent lipids of the lipid particles. The learned model predicts the particle size distribution of lipid particles based on the obtained manufacturing conditions and the constituent lipids of the lipid particles; and The output section outputs the particle size distribution of lipid particles predicted by the learned model.

2. The prediction device according to claim 1, wherein, The learned model is obtained by machine learning using the production conditions of the lipid particles and the lipid composition as explanatory variables, and the particle size distribution of the lipid particles as the target variable.

3. The prediction device according to claim 1 or 2, wherein, The particle size distribution of the lipid particles is a curve representing the proportion of lipid particles at each particle size.

4. The prediction device according to claim 3, wherein, The particle size ranges from 0 nm to 10,000 nm.

5. The prediction device according to claim 2, wherein, The learned model is obtained by performing machine learning on the preprocessed explanatory variables or the target variables.

6. The prediction device according to claim 5, wherein, For the explanatory variables, the physical properties of lipids are obtained or predicted in the preprocessing, and a weighted average is added in conjunction with the lipid proportion of each lipid particle.

7. The prediction device according to claim 5, wherein, For the explanatory variables, a lipid diversity index was added during the pretreatment.

8. The prediction device according to claim 5, wherein, For the target variable, in the preprocessing, experimental data after removing lipid particle samples that have agglomerated are used as variables.

9. The prediction device according to claim 8, wherein, Based on the experimental data, experimental conditions in which phosphatidylethanolamine is removed relative to the weight of the lipids constituting the lipid particles by more than 60% or the lipid flow rate is greater than 0.6 relative to the sum of the lipid and buffer flow rates are used.

10. The prediction device according to claim 1, wherein, The conditions for preparing the lipid particles include at least one or a combination thereof selected from the flow rate ratio, pH, and lipid composition of the composition solution used for preparing lipid particles.

11. A method for predicting particle size distribution using a prediction device, wherein, include: The information acquisition unit acquires the manufacturing conditions of lipid particles composed of any one of phospholipids, anionic lipids, and cationic lipids, as well as the steps for the formation of the lipid particles. The learned model predicts the particle size distribution of lipid particles based on the obtained lipid particle fabrication conditions and the lipid particle composition; and The step of outputting the lipid particle size distribution predicted by the learned model.

12. A particle size distribution prediction program, wherein, To make the computer perform: The conditions for obtaining lipid particles composed of any one of phospholipids, anionic lipids, and cationic lipids, and the steps for the composition of the lipid particles; The step of enabling the learned model to predict the particle size distribution of lipid particles based on the obtained lipid particle manufacturing conditions and the lipid particle composition; and The step of outputting the particle size distribution of lipid particles predicted by the learned model.

Citation Information

Patent Citations

  • Method, device, system, and program for cell analysis, and method, device, and program for generating trained artificial intelligence algorithm

    JP2021085849A

  • Characterization and identification of biological structures

    JP2023022170A