Ionizable lipid, screening method therefor and application thereof

Abstract

Provided are an ionizable lipid, a screening method therefor and application thereof. The molecular structure of the ionizable lipid can be predicted on the basis of a model algorithm, such that fundamental properties such as in-vivo mRNA delivery efficiency and apparent pKa can be predicted, and thus the present invention can be applied to formulation screening of mRNA lipid nanoparticles. Moreover, a physiologically-based pharmacokinetic (PBPK) model is developed for calculating a rate constant for a key step in the pharmacokinetics of lipid nanoparticles. The method and the model provide guidance for the research on and development of lipid nanoparticles, thereby facilitating the rapid development of mRNA-LNP drugs. By using a model simulation technique, a variety of new ionizable lipids, which can be used for an mRNA-LNP delivery system, can be obtained by means of screening, and the high delivery efficiency thereof has also been effectively verified.

Description

Ionizable lipids, screening methods and uses thereof TECHNICAL FIELD

[0001] The present application relates to the technical field of nucleic acid delivery, in particular to ionizable lipids, screening methods and uses thereof. BACKGROUND

[0002] Messenger ribonucleic acid (mRNA) can be used as an effective pharmaceutical active ingredient in modern medicine to deliver into the body to prevent and control various diseases.

[0003] Lipid nanoparticles (LNP) is an important mRNA delivery system, which is characterized by using ionizable lipids as its main component to complete the construction of the mRNA delivery system. Ionizable lipids contain nitrogen atoms in their head groups, which can ionize (positively charged) in acidic environments and remain uncharged in neutral environments. The ionizability of ionizable lipids allows LNP to effectively encapsulate mRNA during preparation and promotes mRNA escape from endosomes after LNP enters cells, thereby exerting its function. Moreover, compared to lipids that are always positively charged, LNP prepared from ionizable lipids is safer.

[0004] In the prior art, the structure screening of ionizable lipids is carried out by trial and error method, that is, a large number of candidate lipids are synthesized and used to prepare RNA (mRNA or small interfering RNA) loaded LNP (RNA-LNP) and test their delivery efficiency, thereby selecting the ionizable lipid with the highest delivery efficiency. However, it is obvious that this research method has many limitations. For example, the success rate of finding satisfactory lipids is low, the research period is long, and the consumption of experimental materials and animals is large. Moreover, the chemical structure space of the obtained lipids is often too large, and the screening efficiency by single experimental trial and error is too low. In addition, general trial and error experiments mainly investigate the final effect of candidate compounds, ignoring quantitative research on key steps in the mechanism of action, and lack sufficient understanding of these candidate compounds.

[0005] Therefore, developing a rapid, efficient and low-cost ionizable lipid screening system is of great significance for the in-depth mining of ionizable lipids, the development of lipid nanoparticle systems, and mRNA delivery, etc.

[0006] SUMMARY

[0007] The present application aims to at least solve one of the above technical problems existing in the prior art. To this end, the present application aims to provide ionizable lipids and screening methods and applications thereof. Based on the many deficiencies existing in the prior art in which ionizable lipids are screened by trial and error experiments, the present application proposes a method of designing or assisting in the design of ionizable lipids using a model algorithm. In the present application, the designed model can be effectively used to quickly screen ionizable lipids for mRNA delivery, and the ionizable lipid molecules screened based on the model all have the expected effect, so it can be shown that the model in the present application can be used for the design and screening of ionizable lipid molecules. In addition, the present application also provides a PBPK model which can be used to calculate the rate constant of the key step in the pharmacokinetics of lipid nanoparticles, thereby effectively analyzing and predicting the effect of existing lipid nanoparticles.

[0008] In a first aspect of the present application, a method for constructing an LNP property prediction model is provided, comprising the following steps:

[0009] (1) Preprocessing the collected LNP data, and extracting the ionizable lipid molecular structure from the preprocessed data;

[0010] (2) Encoding the extracted ionizable lipid molecular structure;

[0011] (3) Combining the encoded information with other necessary information representing the LNP composition to obtain input information;

[0012] (4) Modeling the input information to obtain an LNP property prediction model.

[0013] In some embodiments of the present application, in step (1), the LNP data includes: LNP formulation, apparent pKa value, particle size, type of protein encoded by the included mRNA, mRNA encapsulation rate, experimental animal species, administration route, administration dose, and mRNA expression level. Of course, those skilled in the art can further add other LNP data according to actual use requirements to further improve the model, including but not limited to the above LNP data.

[0014] In some embodiments of the present application, the LNP formulation includes the chemical structure of the ionizable lipid and the type of the auxiliary lipid. Of course, those skilled in the art can further add other LNP formulation information according to actual use requirements to further improve the model, including but not limited to the above LNP formulation information.

[0015] In some embodiments of the present application, the LNP data can be based on existing databases or experimental data.

[0016] In some embodiments of the present application, in step (1), the pretreatment is to improve the uniformity of LNP data.

[0017] In some embodiments of the present application, the pretreatment should maintain as much data as possible while improving the uniformity of LNP data.

[0018] In some embodiments of the present application, the improvement of LNP data uniformity includes: retaining data with consistent administration routes, consistent measurement indicators, comparable mRNA expression levels to standard LNP, and the same or equivalent formulation composition as standard LNP.

[0019] In some embodiments of the present application, the improvement of LNP data uniformity specifically includes: removing data that is not measured; removing data that is not administered by intravenous or intramuscular injection; removing data that is not measured by mRNA expression level using fluorescence signal, luciferase concentration, or human erythropoietin (hEPO) concentration caused by mRNA delivery; removing data that is not measured by luciferase fluorescence signal in the whole body or liver of the test animal; and retaining data that can convert the mRNA expression level of LNP into a fold change based on the standard LNP formulation.

[0020] Of course, those skilled in the art can adjust the standards for improving the uniformity of LNP data according to the species selection of the subject and other actual needs, so that it is not limited to the above requirements.

[0021] In some embodiments of the present application, the standard LNP formulation is an LNP formulation composed of an ionizable lipid, a helper lipid, cholesterol, and a PEG lipid.

[0022] In some embodiments of the present application, the ionizable lipid includes MC3 or its derivative esters.

[0023] In some embodiments of the present application, the helper lipid includes DSPC.

[0024] In some embodiments of the present application, the PEG lipid includes PEG2000-DMG or its derivative esters.

[0025] In some embodiments of the present application, the standard LNP formulation is composed of MC3, DSPC, cholesterol, and PEG2000-DMG, which is the LNP lipid formulation of the first approved siRNA drug. Of course, those skilled in the art can also select other well-known LNP formulations as standard formulations for comparison, and the selection of the standard formulation does not change the compound itself obtained by screening, but only serves as a baseline.

[0026] In some embodiments of the present application, the standard LNP formulation consists of MC3, DSPC, cholesterol and PEG2000-DMG in a molar ratio of 50:10:38.5:1.5. The known standard expression levels of this standard LNP formulation include: luciferase concentration of 198 ng / g in liver tissue at 4 hours post administration of mRNA at a concentration of 0.3 mg / kg; luciferase flux of 2.57E+9 p / s in liver at 6 hours post administration of mRNA at a concentration of 0.5 mg / kg (data from Moderna); luciferase flux of 8.66E+8 p / s in whole body at 6 hours post administration of mRNA at a concentration of 0.5 mg / kg (data from Tufts University); plasma hEPO concentration of 1570, 1830, 810 ng / mL at 3, 6, 24 hours post administration of mRNA at a concentration of 0.5 mg / kg, respectively. Among them, the expression protein concentration indices obtained from the above different data sources are comparable, but the flux indices are not comparable because they are measured by photomultiplier signal amplification, and the measurement itself depends on the experimental instruments used.

[0027] In some embodiments of the present application, the encoding method comprises using simplified molecular-input line-entry system (SMILES), extended connectivity fingerprints (ECFP) sequences, molecular physicochemical property descriptors, 2D molecular descriptors, 3D molecular descriptors, molecular structure pictures, molecular graphs and molecular coordinates, etc.

[0028] In some embodiments of the present application, the feature information is an extended connectivity fingerprint (ECFP) sequence.

[0029] In some embodiments of the present application, the ionizable lipid molecule structure is converted into an extended connectivity fingerprint (ECFP) by the RDKit software package in Python. Of course, those skilled in the art can also use other equivalent technical means to obtain the extended connectivity fingerprint, including but not limited to ChemAxon.

[0030] In some embodiments of the present application, the conversion parameters of the extended connectivity fingerprint include the radius and bit number of the ECFP.

[0031] In some embodiments of the present application, the radius of the ECFP is set to 8-10, and the bit number is set to 512-2048.

[0032] In some embodiments of the present application, the radius of the ECFP is set to 9, and the bit number is set to 1024.

[0033] In some embodiments of the present application, the other information characterizing the LNP composition includes, but is not limited to, the types and proportions of helper lipids, cholesterol, PEG lipids, and mRNA.

[0034] In some embodiments of the present application, the LNP properties include apparent pKa and mRNA delivery efficiency.

[0035] In some embodiments of the present application, the apparent pKa and mRNA delivery efficiency are obtained based on different LNP property prediction models.

[0036] In some embodiments of the present application, the property prediction model is composed of a modeling algorithm, hyperparameters matched with the algorithm, and weight parameters obtained by training the model with data.

[0037] In some embodiments of the present application, the modeling algorithm used can include, but is not limited to, LightGBM, random forest, XGBoost, decision tree, support vector machine, artificial neural network, deep neural network, residual network, recurrent neural network, long short-term memory network, convolutional neural network, and Transformer, etc.

[0038] In some embodiments of the present application, the modeling algorithm used is LightGBM.

[0039] In some embodiments of the present application, the hyperparameters of the mRNA delivery efficiency model include: colsample_bynode = 0.8, colsample_bytree = 0.5, learning_rate = 0.1, max_depth = 3, num_leaves = 4, reg_alpha = 1, reg_lambda = 1, subsample = 0.7, subsample_freq = 3. The hyperparameters of the apparent pKa model include: colsample_bytree = 1, learning_rate = 0.01, max_depth = 40, n_estimators = 700, num_leaves = 45, objective ='regression', subsample = 0.8. Of course, the selection of hyperparameters is not unique, and the purpose of determining hyperparameters is to help the model to train good performance on the data set. As long as the model can predict accurately, without underfitting and overfitting, any combination of hyperparameters is acceptable.

[0040] In some embodiments of the present application, the weight parameters obtained by training the model with data refer to parameters automatically obtained by the model in the training process according to the defined algorithm and data after the hyperparameters are determined. The weight parameters describe how the model calculates the value of the property step by step according to the input variable parameters. Because of the randomness in the training process, the weight parameters are not unique.

[0041] In some embodiments of the present application, the LNP property prediction model can be reconstructed by researchers in the field based on conventional operations in the field after the modeling algorithm and the training data are known.

[0042] In a second aspect of the present application, the LNP property prediction model constructed by the construction method of the first aspect of the present application is provided.

[0043] In the present application, the LNP property prediction model can be used to predict different LNP properties such as apparent pKa and mRNA delivery efficiency, and the prediction results are accurate.

[0044] In a third aspect of the present application, the application of the LNP property prediction model of the second aspect of the present application in LNP evaluation is provided.

[0045] In some embodiments of the present application, the LNP evaluation includes apparent pKa evaluation and mRNA delivery efficiency evaluation.

[0046] In a fourth aspect of the present application, a screening method of ionizable lipids is provided, which comprises the following steps:

[0047] (1) identifying the structural features affecting the properties of LNP in the ionizable lipids using the LNP property prediction model of the second aspect of the present application to obtain structural fragments, and splicing the structural fragments to construct a molecular library;

[0048] (2) screening the compounds with an apparent pKa range of 6-7 and a mRNA delivery efficiency higher than that of the control prescription from the molecular library using the LNP property prediction model of the second aspect of the present application, and obtaining the ionizable lipids.

[0049] In some embodiments of the present application, the screening method specifically comprises the following steps:

[0050] (1) scoring the molecular structure of the ionizable lipids using the LNP property prediction model of the second aspect of the present application, obtaining high-scored ionizable lipid feature structural fragments, splicing different feature structural fragments to construct an ionizable lipid molecular library, and combining the ionizable lipid molecular structure information with other LNP information to obtain an LNP prescription containing the ionizable lipids to be evaluated.

[0051] (2) Using the LNP property prediction model of the second aspect of the present application, the LNP constructed in step (1) is screened to find a scheme with an apparent pKa range of 6-7 and a mRNA delivery efficiency higher than that of the control prescription, and the corresponding ionizable lipid is found, i.e.

[0052] In some embodiments of the present application, in step (1), the molecular structure score of the ionizable lipid is achieved by analyzing the LNP property prediction model of the second aspect of the present application and the collected LNP data. The structure score includes: classification of the extended connectivity fingerprint sites of the ionizable lipid, and calculation of the extended connectivity fingerprint score of the ionizable lipid.

[0053] In some embodiments of the present application, the classification of the extended connectivity fingerprint sites includes: calculating the importance of each input variable through the weight parameters of the property prediction model, which includes whether the substructure of the ionizable lipid represented by each ECFP site helps to improve the mRNA delivery efficiency, if it helps to improve the mRNA delivery efficiency, it is classified as positive, otherwise, it is classified as negative.

[0054] In some embodiments of the present application, the prediction of mRNA delivery efficiency refers to predicting that the mRNA delivery efficiency of the LNP is higher or lower than that of the standard LNP.

[0055] In some embodiments of the present application, the judgment of whether it helps to improve the mRNA delivery efficiency is that when the ECFP value is 1 (i.e. the site exists), it helps to make the mRNA delivery efficiency higher than that of the standard LNP, otherwise, it is not conducive to improving the mRNA delivery efficiency.

[0056] In some embodiments of the present application, the importance of each input variable of the property prediction model, or the classification of the extended connectivity fingerprint sites of the ionizable lipid, can be calculated by the SHAP package in Python. Of course, researchers in the field can also obtain this information through other machine learning model interpretability algorithms.

[0057] In some embodiments of the present application, the extended connectivity fingerprint score of the ionizable lipid is obtained according to the following formula: ECFP score = N positive -d x N negative

[0058] In the formula, N positive is the number of positive ECFP sites, N negativeis the number of negative ECFP sites. Here, the parameter d is 0.7. This scoring formula is not unique, as long as the formula can reward positive ECFP sites and penalize negative ECFP sites according to the ionizable lipids, and the ECFP score has a consistent trend with the mRNA delivery efficiency.

[0059] In some embodiments of the present application, in step (1), the high-scored ionizable lipid feature structure fragments are obtained by directly referring to the structure fragments represented by the high-scored ionizable lipid positive ECFP sites, or extracting characteristic structure fragments from the high-scored ionizable lipids as a whole.

[0060] In some embodiments of the present application, in step (1), the splicing of different characteristic structure fragments to construct the ionizable lipid molecule library is performed according to the rules of ionizable lipid molecules.

[0061] In some embodiments of the present application, the control prescription is a standard LNP formula, and the standard LNP formula is an LNP formula composed of ionizable lipids, auxiliary lipids, cholesterol, and PEG lipids.

[0062] In some embodiments of the present application, the standard LNP formula is composed of MC3, DSPC, cholesterol, and PEG2000-DMG.

[0063] In some embodiments of the present application, the standard LNP formula is composed of MC3, DSPC, cholesterol, and PEG2000-DMG at a molar ratio of 50:10:38.5:1.5.

[0064] Of course, other LNP formulas known to have effects can also be selected by those skilled in the art as standard formulas for comparison, and the selection of the standard formula does not change the compounds screened, but is only used as a control.

[0065] In some embodiments of the present application, the structural features include:

[0066] wherein,

[0067] x is selected from 0 or 1,

[0068] y is selected from an integer from 0 to 8,

[0069] z is selected from an integer from 1 to 4,

[0070] R4 is selected from -H or -CH3.

[0071] In a fifth aspect of the present application, there is provided an ionizable lipid, which is screened by the screening method of the fourth aspect of the present application.

[0072] In some embodiments of the present application, the ionizable lipid has at least one of the following structural features:

[0073] wherein,

[0074] x is selected from 0 or 1,

[0075] y is selected from an integer from 0 to 8,

[0076] z is selected from an integer from 1 to 4,

[0077] R4 is selected from -H or -CH3.

[0078] In some embodiments of the present application, the ionizable lipid has at least two of the above structural features, and other parameters are as defined above.

[0079] In some embodiments of the present application, the ionizable lipid has at least three of the above structural features, and other parameters are as defined above.

[0080] In some embodiments of the present application, the ionizable lipid comprises:

[0081] In a sixth aspect of the present application, there is provided a use of the ionizable lipid of the fifth aspect of the present application in drug delivery.

[0082] In some embodiments of the present application, the drug delivery comprises preparation of a drug delivery carrier or related products.

[0083] In a seventh aspect of the present application, there is provided an LNP, which comprises the ionizable lipid of the fifth aspect of the present application and a nucleic acid molecule.

[0084] In some embodiments of the present application, the nucleic acid molecule comprises modified and unmodified nucleic acid molecules.

[0085] In some embodiments of the present application, the modification comprises chemical modification and enzymatic modification.

[0086] In the present application, the term "chemical modification" refers to the modification of the chemical activity and biological function of a nucleic acid molecule by introducing specific chemical groups thereon.

[0087] In some embodiments of the present application, the modification refers to improvement or amelioration.

[0088] In some embodiments of the present application, the chemical modification comprises backbone modification, base modification and nucleotide modification (modification of sugar group or phosphate group).

[0089] In the present application, the term "enzymatic modification" refers to modification of nucleic acid molecules by using specific enzymatic catalytic reactions.

[0090] In some embodiments of the present application, the enzymatic modification comprises methylation modification, phosphorylation modification and acetylation modification.

[0091] In some embodiments of the present application, the nucleic acid molecules comprise RNA and DNA.

[0092] In some embodiments of the present application, the RNA comprises mRNA and siRNA.

[0093] In some embodiments of the present application, the ionizable lipids are chemically modified.

[0094] In some embodiments of the present application, the chemical modification comprises targeting molecule modification, reporter group modification and functional group modification.

[0095] In some embodiments of the present application, the targeting molecule comprises saccharides and derivatives thereof, peptides, antibodies and biotin.

[0096] In some embodiments of the present application, the reporter group comprises fluorescent groups, tag sequences and nuclides.

[0097] In some embodiments of the present application, the functional group comprises polymers having at least one of the effects of improving stability, improving particle size, improving flocculation effect, improving encapsulation efficiency or drug loading, improving biocompatibility.

[0098] In some embodiments of the present application, the functional group comprises, but is not limited to, polyethylene glycol, poloxamer, chitosan and polyethylene.

[0099] In some embodiments of the present application, the LNP further carries a second active pharmaceutical molecule and / or a pharmaceutically acceptable adjuvant.

[0100] In some embodiments of the present application, the second active pharmaceutical molecule has the same or different function as the nucleic acid molecule.

[0101] The eighth aspect of the present application provides the use of the LNP of the seventh aspect of the present application in drug delivery.

[0102] In some embodiments of the present application, the drug delivery comprises preparation of drug delivery carriers or related products.

[0103] In a ninth aspect of the present application, a drug delivery carrier is provided, which comprises the ionizable lipid of the fifth aspect of the present application or the LNP of the seventh aspect of the present application.

[0104] In a tenth aspect of the present application, a method for constructing a PBPK model is provided, comprising the following steps:

[0105] (1) classifying the in vivo metabolic kinetics data and the cellular metabolic kinetics data of the LNP, respectively;

[0106] (2) according to the transport mechanism of the LNP in vivo or in cells, writing differential equations of the mass change rate of each transport pathway required, combining all the differential equations into a differential equation system, i.e. obtaining the PBPK model.

[0107] In some embodiments of the present application, the in vivo transport pathways that need to describe the mass change rate include:

[0108] lipids distributed in the blood circulation system and organs along with the blood flow, lipids permeated from the blood part of the organs to the interstitium, lipids transported from the interstitium to the cells mediated by receptors, decomposition of the LNP in the cells to form free ionizable lipids, and hydrolytic metabolism of the ionizable lipids. Of course, other organs can be added as needed by those skilled in the art, in which case the distribution of the LNP components in the organ needs to be deducted from the "other organs" compartment.

[0109] In some embodiments of the present application, the blood circulation system and organs include but are not limited to veins, arteries, pulmonary vessels, liver, spleen and other organs.

[0110] In some embodiments of the present application, the obtained in vivo PBPK model is:

[0111] Blood flow:

[0112] Permeation of blood to interstitium:

[0113] Receptor-mediated phagocytosis of LNP from interstitium to cells:

[0114] Decomposition of LNP in cells to ionizable lipids:

[0115] Hydrolysis of ionizable lipids:

[0116] In the formula, M represents the mass of the lipid, represents the mass change rate in a small time unit. Q represents the blood flow. In addition to C receptorAll symbols C with subscripts denote the concentration of lipids in each compartment. C receptor is the receptor concentration mediating LNP phagocytosis. Similarly, all V symbols denote the volume of various organs or sub-organs. P denotes the permeability of LNP, S denotes the organ endothelial area, and their product is the parameter representing the rate of lipid exchange between blood and interstitial space of the organ. k in denotes the phagocytosis rate of cells, which depends on the concentrations of lipids and receptors according to the law of chemical kinetics. k dis denotes the LNP disintegration rate to release free ionizable lipids. k el denotes the metabolic rate, which represents the hydrolysis rate of ionizable lipids.

[0117] In some embodiments of the present application, the rate of change in compartment i is calculated as the sum of the changes of all processes (assuming there are n processes involved) related to it, i.e.,

[0118] or converting the rate of mass change in all i compartments to the rate of concentration change multiplied by volume, i.e.,

[0119] In some embodiments of the present application, the in vivo metabolic kinetic data is data from single species or multiple species.

[0120] In some embodiments of the present application, the species include humans and non-human animals.

[0121] In some embodiments of the present application, the non-human animals include murine, canine, feline, equine, bovine, avian, and primate animals.

[0122] In some embodiments of the present application, the non-human animals include rats and mice.

[0123] In some embodiments of the present application, the cellular metabolic kinetic data includes the transport processes of LNP components between endosomes, macropinosomes, lysosomes, autophagosomes in cells, and the disintegration, efflux, and RNA release processes of LNP.

[0124] In some embodiments of the present application, the pathways of the described rate of mass change in cells include:

[0125] lipids and RNA from early endosomes to late endosomes, lipids and RNA from late endosomes to lysosomes, LNP disintegration to form free lipids and RNA, efflux of lipids and RNA, RNA release to the cell plasma after release events, lipids and RNA encapsulated in autophagosomes after release events, lipids and RNA from autophagosomes to autolysosomes.

[0126] In some embodiments of the application, the resulting cellular PBPK model is:

[0127] Lipid and RNA from early to late endosome:

[0128] Lipid and RNA from late endosome to lysosome:

[0129] Free lipid and RNA dissociated from the LNP:

[0130] Excretion of lipid and RNA:

[0131] RNA released to the cellular plasma after the release event:

[0132] RNA encapsulated in autophagosome after the release event:

[0133] Lipid encapsulated in autophagosome after the release event:

[0134] Lipid and RNA from autophagosome to autolysosome:

[0135] where M represents the mass of the lipid or RNA, and the rate of change. The subscript of the symbol indicates the location, situation or process experienced by the corresponding mass (specifically, for location, EE represents early endosome; LE represents late endosome; LY represents lysosome; AP represents autophagosome; AL, represents autolysosome. For situation and process, "ass" represents assembly; "dis" represents dissociation; "lip" represents lipid; "eg" represents excretion; "rel" represents release). Furthermore, it is assumed that k LE_LY is equal to k AP_AL . The parameter f rel represents the proportion of RNA released into the cytoplasm at the time of the release event to the RNA experiencing the release event.

[0136] In some embodiments of the application, the rate of change in compartment i is calculated as the sum of the changes of all processes associated with it (assuming there are n processes associated), i.e.,

[0137] In some embodiments of the present application, the subject information includes weight, and volume of liver, lung, other organs, spleen, vein, artery, hepatic blood vessels, hepatic cells, hepatic interstitium, splenic blood vessels, splenic interstitium, splenic cells, blood vessels of other organs, cells of other organs, blood vessels of lung, hepatic arterial blood flow, hepatic venous blood flow, pulmonary blood flow, blood flow of other organs, splenic blood flow, hepatic endothelial surface area, splenic endothelial surface area, endothelial surface area of other organs, hematocrit, receptor concentration in liver, receptor dilution from liver to spleen.

[0138] In some embodiments of the present application, the physiological parameters such as organ volume, blood flow, etc. are obtained from public data.

[0139] In some embodiments of the present application, the value of hepatic arterial blood flow is calculated by the sum of blood flow of hepatic artery, large intestine, small intestine, and pancreas.

[0140] In some embodiments of the present application, the receptor concentration in liver is assumed to be 1.

[0141] In some embodiments of the present application, the receptor dilution from liver to spleen is obtained by fitting the observed data.

[0142] In some embodiments of the present application, the model parameters other than the physiological parameters are obtained by fitting the observed data.

[0143] In some embodiments of the present application, the fitting of the observed data uses error models including “constant”, “proportional”, “exponential”, and “combined”. These four error models are standard methods defined in MALTAB SimBiology. Other modeling tools can also include error models that can be used for similar fitting.

[0144] In an eleventh aspect of the present application, there is provided a PBPK model constructed by the construction method of the tenth aspect of the present application.

[0145] The PBPK model is used to calculate the rate constants of key steps in the pharmacokinetics of lipid nanoparticles. The PBPK model is constructed by collecting and organizing the in vivo or in cell metabolism kinetics data of RNA-LNP, obtaining differential equations by classifying and organizing the data, and further organizing the differential equations into a differential equation system. As described above, when constructing, the physiological related parameters in the differential equation system can be obtained from public data sets, and the parameters related to the LNP formula in the differential equation system are calculated by fitting the collected in vivo or in cell metabolism kinetics data. The fitting is to find the appropriate parameter values by calculation software so that the results calculated by the differential equation system are consistent with the collected data, and the parameter values are the calculated parameter results.

[0146] In some embodiments of the present application, the contained information includes concentration-time curves of each component of the drug in each vesicle in each tissue or cell in vivo; wherein each component of the drug includes ionizable lipids, RNA, etc.; the tissues in vivo include blood in large blood vessels, plasma, and various organs in vivo, which can be further subdivided into secondary tissues, including blood, plasma, extracellular fluid, cells, etc. in organs; each vesicle in the cell includes early endosomes, late endosomes, lysosomes, autophagosomes, etc. Each transport process is represented by a differential equation to represent the rate of transport of each component or all of the LNP from the previous tissue or vesicle to the next tissue or vesicle, with the unit of amount of substance / time; wherein the amount of substance can be in weight or molar units; the parameters involved in the differential equation include parameters related to the LNP formulation such as the rate constant of component transport, clearance rate, etc., and physiological-related parameters such as tissue volume, blood flow rate, etc.

[0147] In a twelfth aspect of the present application, the PBPK model of the eleventh aspect of the present application is provided for use in the evaluation of drug pharmacokinetics.

[0148] In some embodiments of the present application, the drug includes a drug delivery carrier loaded with a drug active molecule.

[0149] In some embodiments of the present application, the drug delivery carrier includes LNP.

[0150] The beneficial effects of the present application are:

[0151] 1. The present application provides an ionizable lipid and its screening method and application, thereby solving the many deficiencies in the prior art of using trial and error experiments to screen ionizable lipids, and realizing efficient and accurate screening and identification of ionizable lipids.

[0152] 2. The screening method in the present application is obtained by relying on a big data model algorithm, and the obtained model can be effectively used for rapid screening of ionizable lipids for mRNA delivery, and the ionizable lipid molecules screened based on the model all have the expected effect, with small error and high accuracy, providing a favorable means for the development of ionizable lipid molecules.

[0153] 3. The ionizable lipid molecules provided in the present application have suitable apparent pKa values and high mRNA delivery efficiency, and have equivalent or better LNP properties compared to the standard LNP formulations known in the prior art, thereby producing equivalent or better drug delivery effects, providing more selectivity for the development of drug delivery carriers.

[0154] 4. The PBPK model provided in the present application can be used to calculate the rate constant of the key step in the pharmacokinetics of lipid nanoparticles, thereby effectively analyzing and predicting the effects of existing lipid nanoparticles. BRIEF DESCRIPTION OF DRAWINGS

[0155] Figure 1. Predicted apparent pKa of 9 ionizable lipids on validation set.

[0156] Figure 2A. Effect of structural features (ECFP) on mRNA delivery efficiency of ionizable lipids.

[0157] Figure 2B. Linear relationship between positive ECFP scores and mRNA delivery efficiency of ionizable lipids.

[0158] Figures 3A-3D. mRNA delivery efficiency of lipid nanoparticles of different ionizable lipids (whole body fluorescence imaging in mice); where Figures 3A and 3C are time curves of total luminescence; 3B and 3D are area under the curve (AUC) of total luminescence **, <0.01; ***, <0.001, ****, <0.0001.

[0159] Figure 4. Schematic diagram of PBPK model in vivo.

[0160] Figure 5. Schematic diagram of PBPK model of cells (complex version). Where solid lines represent processes included in actual simulation; dashed lines represent processes omitted in simulation due to limitations of data.

[0161] Figure 6. Schematic diagram of PBPK model of cells (simplified version). Where solid lines represent processes included in actual simulation; dashed lines represent processes omitted in simulation due to limitations of data.

[0162] Figure 7. PBPK model fitting results of metabolic data of three mRNA-LNPs in rats and fitting parameters; solid lines represent simulation results, and dots represent original data. Model conditions are according to original experimental design. Results are expressed as concentrations of ionizable lipids in tissues.

[0163] Figures 8A-8B. PBPK model fitting results of metabolic data of siRNA-LNPs in mice. Solid lines represent simulation results, and dots represent original data. Model conditions are according to original experimental design. Results are expressed as concentrations of lipids in tissues. Where Figure 8A is a mouse intravenously injected with siRNA-LNP at a lipid dose of 11.1 mg / kg. Figures 8B and 8C are mice intravenously injected with siRNA-LNP at a dose of 0.3 mg / kg of different particle sizes (equivalent to about 3.42 mg / kg DMAP-DLP).

[0164] Figure 9. PBPK model fitting results of metabolic data of siRNA-LNPs at different doses in humans. Solid lines represent simulation results, and dots represent original data. Model conditions are according to original experimental design. Results are expressed as concentrations of ionizable lipids (MC3) in plasma.

[0165] FIGS. 10A-10C are the PBPK model fitting results of LNP cellular transport data of ionizable lipids C12-200 and MC3. The solid lines represent the simulation results, and the dots represent the original data; wherein, FIG. 10A is the cellular uptake, LNP dissociation, and cellular efflux results of C12-200 LNP; FIG. 10B is the cellular uptake and siRNA distribution at each level of vesicle of MC3 LNP; the model conditions are according to the original experimental design; FIG. 10C is the corresponding parameters of FIGS. 10A and 10B.

[0166] FIGS. 11A-11B are the PBPK model fitting and simulation results of cellular transport data of ionizable lipids L319, MC3, and the comparison of three ionizable lipids RNA release related parameters. The solid lines represent the simulation results, and the dots and the dashed lines represent the original data. The results are represented as the amount or proportion or probability of siRNA in the cell. The model conditions are according to the original experimental design. Wherein, FIG. 11A is the cellular uptake data fitting of L319 LNP. FIG. 11B and FIG. 11C are the LNP RNA release process fitting and simulation calculation. FIG. 11D is the comparison of three ionizable lipids RNA release related parameters. DETAILED DESCRIPTION

[0167] Unless otherwise indicated, the following terms and phrases used herein are intended to have the following meanings. A particular term or phrase should not be construed as indefinite or unclear unless specifically defined, but should be understood according to the ordinary meaning. When a trade name appears herein, it is intended to refer to its corresponding product or active ingredient thereof.

[0168] Unless otherwise specified, when a group has one or more available sites for linkage, any one or more sites of the group can be linked to other groups by a chemical bond. When the connection mode of the chemical bond is not fixed, and there is an H atom at the available site, the number of H atoms at the site will be reduced to the corresponding valence number of groups corresponding to the number of connected chemical bonds when the chemical bond is connected. The chemical bond connecting the site to other groups can be represented by a straight solid line bond a straight dashed line bond or a wavy line . For example, the straight solid line bond in -OCH3 represents the connection to other groups through the oxygen atom in the group; the straight dashed line bond in -NH2 represents the connection to other groups through both ends of the nitrogen atom in the group; the wavy line in -Ph represents the connection to other groups through the 1 and 2 carbon atoms in the phenyl group.

[0169] The compounds of the present application can be confirmed by conventional methods well known to those skilled in the art, such as nuclear magnetic resonance techniques.

[0170] The compounds of the present application can be prepared by a variety of synthetic methods well known to those skilled in the art, including the specific embodiments set forth below, embodiments formed by combining other synthetic methods with the embodiments set forth below, and equivalents thereof as appreciated by those skilled in the art. Preferred embodiments include, but are not limited to, the examples of the present application.

[0171] The materials and solvents used in the present application are commercially available.

[0172] Unless otherwise indicated, the test or test method is a conventional method in the art.

[0173] The compounds are named according to the conventional naming principles in the art or using the software naming, and the commercially available compounds are named using the supplier catalog name.

[0174] In the present application, the compound groups, reagents or techniques well known to those skilled in the art can be represented by abbreviations, including but not limited to: messenger RNA (mRNA), lipid nanoparticle (LNP), polyethylene glycol (PEG), extended connectivity fingerprint (ECFP), human erythropoietin (hEPO), distearoylphosphatidylcholine (DSPC), dimyristoyl (DMG), simplified molecular-input line-entry system (SMILES), accuracy (ACC), mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE), 1,3-dicyclohexyl carbodiimide (DCC), 4-dimethylaminopyridine (DMAP), dichloromethane (DCM), thin layer chromatography (TLC), petroleum ether (PE), ethyl acrylate (EA), methanol (MeOH), physiologically-based pharmacokinetics (PBPK), 5-fold cross-validation (5CV).

[0175] Example 1 A method of screening ionizable lipids for use in mRNA delivery

[0176] In this example, the method of screening ionizable lipids for use in mRNA delivery comprises the following steps:

[0177] (1) Collect and organize existing LNP experimental data (training set) for screening ionizable lipids from databases and / or literature, and require the data to include information such as the ionizable lipids, auxiliary lipids, cholesterol, PEG lipids, types and proportions of mRNA, dosing regimen, subjects, measurement results and measurement time, etc. contained in the LNP. Among them, the data of ionizable lipids, auxiliary lipids, cholesterol, PEG lipids can be represented by categorical variables, chemical descriptors or ECFP, and other data can be represented by conventional representation methods.

[0178] In this example, LNP experimental data information is derived from 9 patent documents (US20160151284A1, WO2017112865A1, US9868691B2, WO2020061367A1, WO2017004143A1, WO2019036028A1, WO2018200943A1, WO2015199952A1, WO2017075531A1) and the following academic articles:

[0179] Sabnis, S.; Kumarasinghe, E. S.; Salerno, T.; Mihai, C; Ketova, T.; Senn, J. J.; Lynn, A.; Bulychev, A.; McFadyen, I.; Chan, J.; Almarsson, Stanton, M. G.; Benenato, K. E. A Novel Amino Lipid Series for mRNA Delivery: Improved Endosomal Escape and Sustained Pharmacology and Safety in Non-Human Primates. Mol. Ther. 2018, 26 (6), 1509-1519. https: / / doi.org / 10.1016 / j.ymthe.2018.03.010.

[0180] Fenton, O. S.; Kauffman, K. J.; McClellan, R. L.; Appel, E. A.; Dorkin, J. R.; Tibbitt, M. W.; Heartlein, M. W.; DeRosa, F.; Langer, R.; Anderson, D. G. Bioinspired Alkenyl Amino Alcohol Ionizable Lipid Materials for Highly Potent In Vivo mRNA Delivery. Adv. Mater. 2016, 28 (15), 2939-2943. https: / / doi.org / 10.1002 / adma.201505822.

[0181] Hajj, K. A.; Ball, R. L.; Deluty, S. B.; Singh, S. R.; Strelkova, D.; Knapp, C. M.; Whitehead, K. A. Branched-Tail Lipid Nanoparticles Potently Deliver mRNA In Vivo Due to Enhanced Ionization at Endosomal pH. Small 2019, 15 (6), 1805097. https: / / doi.org / 10.1002 / smll.201805097.;

[0182] Miao, L.; Li, L.; Huang, Y.; Delcassian, D.; Chahal, J.; Han, J.; Shi, Y.; Sadtler, K.; Gao, W.; Lin, J.; Doloff, J. C.; Langer, R.; Anderson, D. G. Delivery of mRNA Vaccines with Heterocyclic Lipids Increases Anti-Tumor Efficacy by STING-Mediated Immune Cell Activation. Nat. Biotechnol. 2019, 37 (10), 1174-1185. https: / / doi.org / 10.1038 / s41587-019-0247-3.

[0183] Cornebise, M.; Narayanan, E.; Xia, Y.; Acosta, E.; Ci, L.; Koch, H.; Milton, J.; Sabnis, S.; Salerno, T.; Benenato, K. E. Discovery of a Novel Amino Lipid That Improves Lipid Nanoparticle Performance through Specific Interactions with mRNA. Adv. Funct. Mater. 2022, 32 (8), 2106727. https: / / doi.org / 10.1002 / adfm.202106727.

[0184] Zhao, X.; Chen, J.; Qiu, M.; Li, Y.; Glass, Z.; Xu, Q. Imidazole-Based Synthetic Lipidoids for In Vivo mRNA Delivery into Primary T Lymphocytes. Angew. Chem. Int. Ed Engl. 2020, 59 (45), 20083-20089. https: / / doi.org / 10.1002 / anie.202008082.

[0185] Qiu, M.; Glass, Z.; Chen, J.; Haas, M.; Jin, X.; Zhao, X.; Rui, X.; Ye, Z.; Li, Y.; Zhang, F.; Xu, Q. Lipid Nanoparticle-Mediated Codelivery of Cas9 mRNA and Single-Guide RNA Achieves Liver-Specific in Vivo Genome Editing of Angptl3. Proc. Natl. Acad. Sci. 2021, 118 (10), e2020401118. https: / / doi.org / 10.1073 / pnas.2020401118.

[0186] Kauffman, K. J.; Dorkin, J. R.; Yang, J. H.; Heartlein, M. W.; DeRosa, F.; Mir, F. F.; Fenton, O. S.; Anderson, D. G. Optimization of Lipid Nanoparticle Formulations for mRNA Delivery in Vivo with Fractional Factorial and Definitive Screening Designs. Nano Lett. 2015, 15 (11), 7300-7306. https: / / doi.org / 10.1021 / acs.nanolett.5b02497.

[0187] Miao, L.; Lin, J.; Huang, Y.; Li, L.; Delcassian, D.; Ge, Y.; Shi, Y.; Anderson, D. G. Synergistic Lipid Compositions for Albumin Receptor Mediated Delivery of mRNA to the Liver. Nat. Commun. 2020, 11 (1), 2424. https: / / doi.org / 10.1038 / s41467-020-16248-y.

[0188] Among them, after the collection of the data is completed, subsequent data processing work is also carried out to improve the uniformity of the data while maintaining as much data as possible to the greatest extent. The standard for data processing is: retaining data with consistent administration routes, consistent measurement indicators, comparable mRNA expression levels with standard LNP, and the same or equivalent formulation composition as the standard LNP. Therefore, in this embodiment, the specific data processing method is: removing data that is not measured in mice; removing data that is not administered by intravenous or intramuscular injection; removing data that does not measure mRNA expression levels by fluorescence signal, luciferase concentration or human erythropoietin (hEPO) concentration; removing data that does not measure luciferase fluorescence signal in the whole body or liver of the test animal; retaining data that can convert the mRNA expression level of LNP into the expression level fold based on the standard LNP formula. Samples with mRNA expression levels greater than the standard LNP are labeled positive, samples less than the standard formula are labeled negative, and in addition, the standard LNP formula is also labeled positive. Finally, the collected data set contains a total of 397 LNP formulations, of which there are 382 different ionizable lipids.

[0189] Among them, the standard LNP formula is used as a control. The standard LNP formula consists of MC3 (ionizable lipid), DSPC (helper lipid), cholesterol and PEG2000-DMG (PEG lipid) with a molar ratio of 50:10:38.5:1.5. In the field, the standard LNP formula is usually taken as a control because it is actually the LNP lipid formula of the first approved siRNA drug, which has good effect known. The known standard expression levels of the standard LNP formula include: 4 hours after administration of mRNA at a concentration of 0.3 mg / kg, the luciferase concentration in liver tissue is 198 ng / g; 6 hours after administration of mRNA at a concentration of 0.5 mg / kg, the luciferase fluorescence flux in the liver is 2.57E+9 p / s (data from Moderna); 6 hours after administration of mRNA at a concentration of 0.5 mg / kg, the luciferase fluorescence flux in the whole body is 8.66E+8 p / s (data from Tufts University); 3, 6, 24 hours after administration of mRNA at a concentration of 0.5 mg / kg, the plasma hEPO concentration is 1570, 1830, 810 ng / mL, respectively. The expression protein concentrations obtained from the above different data sources are comparable, but the fluorescence flux is not comparable because it is measured by a photomultiplier signal amplifier, and the measurement itself depends on the experimental instrument used.

[0190] (2) Extract all ionizable lipid molecular structures from the compounds in the training set above, and convert these molecular structures into their simplified molecular linear input specification (SMILES), thereby obtaining the ECFP of these ionizable lipid molecular structures.

[0191] In this embodiment, the extracted information is specifically: the formula of the liposome nanoparticle (LNP) (including the chemical structure of the ionizable lipid, the type of the helper lipid), the apparent pKa value, the particle size, the type of the mRNA-encodes protein, the mRNA encapsulation rate, the experimental animal species, the administration route, the administration dose, and the mRNA expression level.

[0192] In this embodiment, the ionizable lipid molecular structure is converted into an extended connectivity fingerprint (ECFP) by the RDKit software package in Python. The radius of the ECFP is set to 9, and the bit number is set to 1024. Each ionizable lipid molecule has a unique ECFP sequence (a string sequence composed of "0" and "1"). The type of the helper lipid, cholesterol, and PEG lipid is represented by a categorical variable, and the molar proportion of each type of lipid is represented by a numerical variable.

[0193] (3) Based on the input information in step (2), a model is established to correlate the input information with the corresponding LNP prescription, administration scheme, and resulting drug delivery result, which is represented by the model-predicted characteristics of the corresponding LNP, such as mRNA delivery efficiency, apparent pKa, etc. The construction algorithm for establishing the model can be LightGBM, random forest, XGBoost, deep neural network, support vector machine, etc.

[0194] In this embodiment, the construction algorithm for establishing the model is LightGBM.

[0195] In this embodiment, the structural fragments of ionizable lipid molecules that have a positive effect on improving mRNA delivery efficiency can be determined by the model and the comparison data, and these structural fragments can be arranged and combined according to the structural rules of ionizable lipid molecules, resulting in nearly 20 million molecules. These molecules are combined with the prescription information of standard LNP, and then the apparent pKa and mRNA delivery efficiency are predicted by the established model.

[0196] In the field, due to the huge volume of AI models, it is difficult to realize the mathematical sense of instantiation, therefore, only the parameters in the model are saved after the construction is completed and used. When it is needed to be used again, the weights are loaded to reinitialize a model, thereby completing new detection.

[0197] After the data collection is completed, a complete data set is divided into a training set and a test set. The training set is used to train the model, and the test set is used to evaluate the prediction performance of the model. The splitting of the data set adopts stratified sampling method to maintain the same classification proportion (positive and negative samples) in the separated data set as the original data. Based on the consideration of potential bias that may be generated by different data sources (original literature), stratified sampling strategy is necessary to ensure the consistency of data source distribution between data sets.

[0198] Specifically, in this embodiment, the stratified sampling strategy is realized by the scikit-learn (sklearn) software package. In this embodiment, the training set and the test set are obtained in a ratio of 4:1, and the data distribution is shown in Table 1.

[0199] Table 1 Prediction results of the model on mRNA delivery efficiency in a data set

[0200] During the model training process, the hyperparameters of the model are randomly searched to adjust. Briefly, 1000 combinations of hyperparameters are randomly selected from the hyperparameter space to train on the training set, and the results of 5-fold cross-validation (5_CV) are used to select the best model.

[0201] In the present embodiment, the important hyperparameters of the optimal model obtained include: colsample_bynode = 0.8, colsample_bytree = 0.5, learning_rate = 0.1, max_depth = 3, num_leaves = 4, reg_alpha = 1, reg_lambda = 1, subsample = 0.7, subsample_freq = 3. As described above, with the above parameters, the model can be reconstructed based on new data.

[0202] In predicting the apparent pKa of LNP, no further special data processing work is needed. When predicting the apparent pKa of LNP, the data set used contains 352 LNP formulations, of which 350 are different ionizable lipids. The data set is divided into three subsets, namely the training set, the validation set and the test set, when used, the size ratio of the data set is about 8:1:1 (ignoring decimal places). Among them, the infrequent molecules that appear less than three times in the data set are forced to be included in the training set in order to train the model in the most comprehensive molecular structure space as possible. The rest of the data is randomly stratified sampled according to the apparent pKa (<6, 6-7, 7-8 and >8 four specifications). Finally, the training set, the validation set and the test set contain 278, 37 and 37 samples respectively. The model is trained on the training set, while the hyperparameters are adjusted on the validation set to obtain the best configuration.

[0203] Finally, the performance of the model is evaluated using the test set to evaluate its generalization ability.

[0204] The apparent pKa is predicted based only on the LNP formulation, and the model is also trained using the LightGBM algorithm to establish a regression model. At the same time, the random search method is used to fine-tune the hyperparameters of the model, and a total of 800 different parameter combinations are tested. The final model hyperparameters are: colsample_bytree = 1, learning_rate = 0.01, max_depth = 40, n_estimators = 700, num_leaves = 45, objective ='regression', subsample = 0.8. As above, with the above parameters, the model can be reconstructed based on new data.

[0205] In evaluating the prediction performance of the above standard regression model, MAE, MSE, RMSE and R 2 are used for evaluation. These indicators are calculated as shown by the following formula.

[0206] where RMSE is simply a measure of the size of the error between predicted and actual values. MAE is another metric used to evaluate the prediction error. MAE is similar to RMSE, measuring the average absolute difference between predicted and actual values. But differently, MAE focuses more on the absolute error rather than their squared values. R-squared (R 2 ) is a statistical measure that indicates the degree of fit of the model to the data, the closer the score is to 1, the better the fit.

[0207] The prediction performance of a classification model is evaluated by four metrics, including accuracy (ACC), recall, precision, and F1_score (F1). ACC represents the proportion of samples that are correctly predicted. Recall represents the proportion of correctly predicted positive samples to all actual positive samples. Precision is the proportion of correctly predicted positive samples to all samples predicted as positive. F1 takes into account both recall and precision, and considers the class distribution to evaluate the accuracy of the classification model.

[0208] The definitions of these metrics are shown in the following formulas:

[0209] where TP represents true positive, TN represents true negative, FP represents false positive, and FN represents false negative.

[0210] The performance of the constructed model was tested using the above test set, and the results are as follows:

[0211] Table 2 is the performance of the model for predicting mRNA delivery efficiency on the collected data set. Table 3 is the validation result of the model for predicting mRNA delivery efficiency using an external validation set including 14 ionizable lipids (not included in the original data set). Table 4 is the performance of the model for predicting apparent pKa on the collected data set. Figure 1 is the result of the model for predicting apparent pKa on an external validation set including 9 ionizable lipids (not included in the original data set).

[0212] Table 2 Performance of the model for predicting mRNA delivery efficiency on the collected data set

[0213] Table 3 Performance of the model for predicting mRNA delivery efficiency on the external validation set

[0214] *,Prediction represents the predicted value of the model trained based on the training data set.

[0215] Prediction al represents the prediction value of the model trained based on the entire data set after determining the hyperparameter adjustment method by the 5_CV method.

[0216] Table 4 Performance of the model for predicting apparent pKa on the collected data set

[0217] The results of Table 2 show that the model for predicting mRNA delivery efficiency has good performance on the training set, 5-fold cross-validation (5CV), and test set, with all indicators above 0.75. The results of Table 3 show that after fixing the hyperparameters, the results of the model trained on the entire data set can be generalized to new lipid structures, with a prediction accuracy of mRNA delivery efficiency of more than 0.8. The results of Table 4 further show that the model for predicting apparent pKa has good prediction performance on the training set, validation set, and test set, with R 2 The results of Figure 1 also confirm the above conclusion that in 9 external tests, the predicted apparent pKa of 7 samples is close to the experimental value, and the model has good generalization effect.

[0218] (4) In this embodiment, the inventors calculated the feature importance of input information (parameters) on the model output after training the model by SHAP algorithm, especially the importance of sequence encoding of each ECFP site on mRNA delivery efficiency (as shown in Figure 2A). Each ECFP site represents a substructure of an ionizable lipid molecule, and if the presence of the site helps to classify the lipid as a lipid with high mRNA delivery efficiency, the site is also considered positive (and vice versa). Then, the lipids are scored according to the number of positive and negative ECFP sites they contain using the following formula: ECFP score = N positive -d x N negative

[0219] In the formula, N positive is the number of positive ECFP sites, and N negative is the number of negative ECFP sites; when the parameter d is 0.7, the correlation coefficient between the ECFP score and the mRNA delivery efficiency can be maximized (as shown in Figure 2B).

[0220] In this embodiment, evaluating the ECFP score of ionizable lipids helps to provide interpretability for the machine learning model and to identify which structural features of ionizable lipids are beneficial or detrimental to mRNA delivery. The ECFP score data (positive and negative) can be used to classify subsequent lipids or LNPs (by building a classification model), so that it can be used to qualitatively predict whether the mRNA delivery efficiency of the LNP is better than the standard formulation.

[0221] (5) Construction of LNP virtual prescription library:

[0222] The structural features of the ionizable lipids that affect the properties of LNP are identified by the ECFP score described above, and the structural features that significantly positively or negatively affect the properties of LNP are distinguished. The ionizable lipids in the dataset are sorted by ECFP score from high to low, and ionizable lipids containing as many positive structural features as possible (such as the top 50 lipids) are induced, and characteristic substructure fragments (such as cycloalkanes, branched alkanes, ester bonds, and combinations thereof containing a certain number of carbon atoms) are extracted therefrom. Then, the molecular library is generated by molecular fragment splicing according to the regularity of the ionizable lipid molecules, and the LNP virtual prescription library is constructed by combining the ionizable lipid molecular structure and other information of LNP.

[0223] The general regularity of the target structure for molecular fragment splicing includes but is not limited to: the ionizable lipid can be divided into a head and a tail, wherein the head is a small group containing a nitrogen atom, and the tail is a long carbon chain (which can contain ester bonds, disulfide bonds, unsaturated bonds, and the like). The head can be obtained by statistical data collection, and the tail can be obtained by adding a certain length of carbon chain to the above-mentioned extracted fragments. Directly connecting the head and the tail can obtain new lipid molecular structures. A total of nearly 20 million molecules are obtained by this method.

[0224] (6) Compound screening in the LNP virtual prescription library using the model:

[0225] The model established in step (4) is used to predict the apparent pKa and mRNA delivery efficiency of the LNP generated in the above prescription library, and compounds with pKa in the range of 6-7 and mRNA delivery efficiency higher than that of the control prescription (positive LNP prescription) are selected. Because each of the above-mentioned molecular fragments actually participates in the construction of multiple lipid molecules, the positive prescription rate can be calculated for each fragment; select molecular fragments with high positive prescription rate for further generation of ionizable lipid molecules and corresponding LNP prescriptions, use the established model to predict, select the LNP with the highest probability of pKa in the range of 6-7 and mRNA delivery efficiency higher than that of the control prescription as the recommended prescription, and the ionizable lipid corresponding to the recommended prescription is the ionizable lipid molecule screened by the model.

[0226] Example 2 Ionizable lipids obtained based on the model

[0227] In this example, the inventors set the prescription with apparent pKa in the range of 6-7 and mRNA delivery efficiency as positive as the positive prescription. The positive rate of each substructure fragment is calculated and the fragments are sorted, and the final fragments include:

[0228] wherein, in the above structure, x is selected from 0 or 1, y is selected from an integer from 0 to 8, z is selected from an integer from 1 to 4, and R4 is selected from -H or -CH3.

[0229] According to the method in the above examples, the above fragments are recombined into new ionizable lipids, and other necessary information is combined into the LNP prescription, and the effect is predicted by the model in the examples of the present application, and the prescription with the highest probability of apparent pKa in 6-7 and mRNA delivery efficiency is selected, and the corresponding ionizable lipid structure is the ionizable lipid (LQ085-LQ091, LQ093 and LQ094) screened by the model.

[0230] In the following examples, a plurality of ionizable lipid molecules based on the model screening scheme in the above examples are listed, and their corresponding synthesis and characterization information are provided.

[0231] (1) LQ085

[0232] The synthetic route of LQ085 is as follows:

[0233] wherein, the specific preparation method is:

[0234] Step 1. Preparation of LQ013-1

[0235] According to the feeding amount shown in the following table, 6-bromohexanoic acid, 2-ethylhexanol, DCC, DMAP and DCM were added to the reaction bottle, and stirred at room temperature for 12 h. TLC (PE:EA = 20:1) showed that the reaction was complete (the Rf value of LQ013 was 0.6), then the reaction liquid was filtered with diatomite and rotary evaporated, and 2 g of colorless oil was obtained after purification, with a yield of 75%. (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0236] Raw material ratio table for preparation of LQ013-1

[0237] Step 2. Preparation of LQ085-1

[0238] To a reaction flask was added SM102-1, 1-(3-aminopropyl)-4-methylpiperazine, potassium carbonate, and acetonitrile in the amounts shown in the table below, and the reaction was stirred at 30 °C for 12 h. TLC (DCM:MeOH = 10:1) showed the reaction to be complete (Rf value of LQ085-1 was 0.4) and the reaction was then concentrated, diluted with 100 mL of ethyl acetate, washed twice with 100 mL of water, dried over anhydrous sodium sulfate, and concentrated. Purification gave 2.5 g of colorless oil in 46% yield (flash chromatography, dichloromethane / methanol 100:1 to 20:1).

[0239] Table of raw material proportions for the preparation of LQ085-1

[0240] Step 3. Preparation of LQ085

[0241] To a reaction flask was added LQ085-1, LQ013-1, potassium carbonate, potassium iodide, and acetonitrile in the amounts shown in the table below, and the reaction was stirred at 80 °C for 12 h. TLC (DCM:MeOH = 10:1) showed the reaction to be complete (Rf value of LQ085 was 0.6) and the reaction was then concentrated and purified to give 1.5 g of colorless oil in 42% yield (flash chromatography, dichloromethane / methanol 100:1 to 20:1).

[0242] Table of raw material proportions for the preparation of LQ085

[0243] The proton nuclear magnetic resonance spectrum of the resulting end product is as follows:

[0244] 1 H NMR (600 MHz, Chloroform-d) δ 4.86 - 4.81 (m, 1H), 3.99 - 3.93 (m, 2H), 2.78 (d, J = 53.1 Hz, 6H), 2.42 (t, J = 6.9 Hz, 2H), 2.32 (s, 3H), 2.30 (t, J = 7.4 Hz, 2H), 2.26 (t, J = 7.5 Hz, 2H), 1.90 - 1.81 (m, 2H), 1.71 - 1.52 (m, 9H), 1.48 (q, J = 7.1, 6.5 Hz, 4H), 1.37 - 1.19 (m, 42H), 0.90 - 0.84 (m, 12H).

[0245] (2) LQ086

[0246] The synthetic route for LQ086 is shown below:

[0247] The specific preparation method is as follows:

[0248] Step 1. Preparation of LQ086-1

[0249] To a reaction flask was added SM102-1, 1-(3-aminopropyl)pyrrolidine, potassium carbonate and acetonitrile in the amounts shown in the table below, and heated to 60°C and stirred for 12h. TLC (DCM:MeOH = 10:1) showed the reaction to be complete (Rf value of LQ085-1 was 0.15) and the reaction was then spun down. The reaction was then diluted with 100 mL of ethyl acetate and washed twice with 100 mL of water, the organic phase was dried over anhydrous sodium sulfate and spun down, and purified to give 1.7 g of colorless oil in 37% yield (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0250] Table of raw material proportions for preparation of LQ086-1

[0251] Step 2. Preparation of LQ086

[0252] To a reaction flask was added LQ086-1, LQ013-1, potassium carbonate, potassium iodide and acetonitrile in the amounts shown in the table below, and stirred at 80°C for 12h. TLC (DCM:MeOH = 10:1) showed the reaction to be complete (Rf value of LQ085 was 0.6) and the reaction was then spun down. Purification gave 1.1 g of colorless oil in 45% yield (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0253] Table of raw material proportions for preparation of LQ086

[0254] The proton nuclear magnetic resonance spectrum of the resulting end product was:

[0255] 1 H NMR (600 MHz, Chloroform-d) δ 4.84 (p, J = 6.2 Hz, 1H), 4.00 - 3.93 (m, 2H), 3.10 (d, J = 6.6 Hz, 6H), 3.02 - 2.97 (m, 2H), 2.65 (t, J = 6.7 Hz, 2H), 2.52 (q, J = 8.4 Hz, 4H), 2.29 (dt, J = 22.6, 7.4 Hz, 4H), 2.10 - 1.99 (m, 6H), 1.66 - 1.57 (m, 4H), 1.55 - 1.45 (m, 8H), 1.35 - 1.22 (m, 39H), 0.91 - 0.84 (m, 12H).

[0256] (3) LQ087

[0257] The synthetic route for LQ087 is shown below:

[0258] wherein the specific preparation method is:

[0259] Step 1. Preparation of LQ087-1

[0260] According to the amount of raw materials shown in the table below, N-methylbenzylamine, tert-butyl N-(4-bromobutyl)carbamate, potassium iodide, potassium carbonate and acetonitrile were added to the reaction bottle, and stirred at 60°C for 12h. After TLC (DCM:MeOH=10:1) showed that the reaction was complete (the Rf value of LQ087-1 was 0.8), the reaction solution was spin-dried. Then diluted with 100mL of ethyl acetate, washed with 100mL of water twice, and the organic phase was dried with anhydrous sodium sulfate and spin-dried, and then purified to obtain 1.8g of colorless oil with a yield of 61% (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0261] Raw material ratio table for preparation of LQ087-1

[0262] Step 2. Preparation of LQ087-2

[0263] According to the amount of raw materials shown in the table below, LQ087-1 and hydrogen chloride-dioxane solution (4M) were added to the reaction bottle, and stirred at room temperature for 2h. After TLC (DCM:MeOH=10:1) showed that the reaction was complete (the Rf value of LQ087-2 was 0.1), the reaction solution was spin-dried. Purification obtained 880mg of colorless oil with a yield of 75% (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0264] Raw material ratio table for preparation of LQ087-2

[0265] Step 3. Preparation of LQ087-3

[0266] According to the amount of raw materials shown in the table below, SM102-1, LQ087-2, potassium carbonate and acetonitrile were added to the reaction bottle, and stirred at 30°C for 12h. After TLC (DCM:MeOH=10:1) showed that the reaction was complete (the Rf value of LQ087-3 was 0.3), the reaction solution was spin-dried. Then diluted with 100mL of ethyl acetate, washed with 100mL of water twice. The organic phase was dried with anhydrous sodium sulfate and spin-dried, and then purified to obtain 760mg of colorless oil with a yield of 51% (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0267] Raw material ratio table for preparation of LQ087-3

[0268] Step 4. Preparation of LQ087-4

[0269] To a reaction flask was added 8-bromooctanoic acid, 2-cyclohexylethanol, DCC, DMAP and DCM in the amounts shown in the table below and stirred at room temperature for 12 h. TLC (PE:EA = 20:1) showed the reaction to be complete (Rf value of LQ087-4 was 0.6) after which the reaction was filtered over celite and spun dry. Purification gave 2.2 g of colorless oil in 66.7% yield (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0270] Table of raw material ratios for preparation of LQ087-4

[0271] Step 5. Preparation of LQ087

[0272] To a reaction flask was added LQ087-3, LQ087-4, potassium iodide, potassium carbonate and acetonitrile in the amounts shown in the table below and stirred at 60 °C for 12 h. TLC (DCM:MeOH = 10:1) showed the reaction to be complete (Rf value of LQ087 was 0.5) after which the reaction was spun dry. It was then diluted with 100 mL of ethyl acetate and washed with 100 ml of water twice. The organic phase was dried over anhydrous sodium sulfate and spun dry. Purification gave 420 mg of colorless oil in 38% yield (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0273] Table of raw material ratios for preparation of LQ087-4

[0274] The proton nuclear magnetic resonance spectrum of the resulting end product was:

[0275] 1 H NMR (600 MHz, Chloroform-d) δ 7.30 (d, J = 4.4 Hz, 4H), 7.25 - 7.21 (m, 1H), 4.88 - 4.83 (m, 1H), 4.09 (t, J = 6.9 Hz, 2H), 3.47 (s, 2H), 2.51 (s, 5H), 2.38 (t, J = 6.8 Hz, 2H), 2.27 (td, J = 7.5, 5.3 Hz, 4H), 2.18 (s, 3H), 1.70 (tdd, J = 11.8, 5.9, 2.7 Hz, 4H), 1.66 - 1.56 (m, 6H), 1.51 (dd, J = 11.9, 5.1 Hz, 13H), 1.35 - 1.12 (m, 41H), 0.93 (td, J = 11.7, 3.0 Hz, 2H), 0.87 (t, J = 7.0 Hz, 6H).

[0276] (4) LQ089

[0277] The synthetic route of LQ089 is shown below:

[0278] The specific preparation method is:

[0279] Step 1. Preparation of LQ089-1

[0280] According to the amount of raw materials shown in the table below, 8-bromooctanoic acid, cyclohexanemethanol, DCC, DMAP and DCM were added to the reaction bottle, and stirred at room temperature for 16 h. TLC (PE:EA=20:1) showed that the reaction was complete (the product Rf value was 0.6), then the reaction solution was filtered with diatomite and rotary dried, and 2.6 g of colorless oil was obtained after purification, with a yield of 93% (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0281] Raw material ratio table for preparation of LQ089-1

[0282] Step 2. Preparation of 9-octadecenol

[0283] According to the amount of raw materials shown in the table below, magnesium turnings and anhydrous tetrahydrofuran were added to the reaction bottle, and heated to 45°C under nitrogen protection. 1-bromononane was slowly added, and the temperature rose and began to reflux after the Grignard reaction was triggered. After complete addition, the reaction was incubated at 45°C for 2 hours, then cooled to 2-10°C, and n-nonyl aldehyde was slowly added. After the addition was completed, the reaction was stirred at room temperature for 12 h. TLC (PE:EA=20:1) showed that the reaction was complete (the product Rf value was 0.3). The reaction solution was cooled to 2-10°C, and 100 mL of 1N dilute hydrochloric acid was slowly added. After the addition was completed, it was stirred for 30 minutes, then 600 mL of ethyl acetate and 1 L of water were added for extraction. The organic phase was dried over anhydrous sodium sulfate and rotary evaporated. After purification, 81 g of white solid was obtained, with a yield of 85% (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0284] Raw material ratio table for preparation of 9-octadecenol

[0285] Step 3. Preparation of LQ089-2

[0286] According to the amount of raw materials shown in the table below, 9-octadecenol, 8-bromooctanoic acid, DCC, DMAP and DCM were added to the reaction bottle, and stirred at room temperature for 16 h. TLC (PE:EA=20:1) showed that the reaction was complete (the product Rf value was 0.6), then the reaction solution was filtered with diatomite and rotary dried, and 3.8 g of colorless oil was obtained after purification, with a yield of 80% (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0287] Raw material ratio table for preparation of LQ089-2

[0288] Step 4. Preparation of LQ089-3

[0289] To a reaction flask was added LQ089-2, ethanolamine and acetonitrile in the amounts shown in the table below and heated to 30°C and stirred for 16h. TLC (DCM:MeOH = 10:1) showed the reaction to be complete (product Rf value 0.4) and the reaction was then spun down. The reaction was then diluted with 100 mL of ethyl acetate and washed twice with 100 mL of water and the organic phase was dried over anhydrous sodium sulfate, spun down and purified to give 2.9g of colorless oil in 80% yield (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0290] Table of raw material ratios for preparation of LQ089-3

[0291] Step 4. Preparation of LQ089

[0292] To a reaction flask was added LQ089-3, LQ089-1, potassium carbonate, potassium iodide and acetonitrile in the amounts shown in the table below and heated to 80°C and stirred for 16h. TLC (DCM:MeOH = 10:1) showed the reaction to be complete (product Rf value 0.6) and the reaction was then filtered. The filter cake was then washed twice with 50 mL of ethyl acetate and the filtrate was spun down. Purification gave 3g of colorless oil in 52% yield (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0293] Table of raw material ratios for preparation of LQ089

[0294] The proton nuclear magnetic resonance spectrum of the resulting end product was:

[0295] 1 H NMR (600 MHz, Chloroform-d) δ 4.84 (p, J = 6.3 Hz, 1H), 3.86 (d, J = 6.6 Hz, 2H), 3.74 (t, J = 5.1 Hz, 2H), 2.85 (t, J = 5.1 Hz, 2H), 2.73 (t, J = 7.9 Hz, 4H), 2.27 (dt, J = 12.2, 7.5 Hz, 4H), 1.74 - 1.68 (m, 4H), 1.64 - 1.56 (m, 10H), 1.48 (d, J = 13.1 Hz, 4H), 1.35 - 1.19 (m, 42H), 0.99 - 0.91 (m, 2H), 0.86 (t, J = 7.0 Hz, 6H).

[0296] (5) LQ090

[0297] The synthetic route of LQ090 is shown below:

[0298] Wherein, the specific preparation method is:

[0299] Step 1. Preparation of LQ090-1

[0300] According to the amount of raw materials shown in the table below, 8-bromooctanoic acid, tetrahydrocumin alcohol, DCC, DMAP and DCM were added to the reaction bottle, and stirred at room temperature for 16h. TLC (PE:EA=20:1) showed that the reaction was complete (the product Rf value was 0.65), then the reaction solution was filtered with diatomite, and rotary evaporation. After purification, 2g of colorless oil was obtained, with a yield of 87% (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0301] Raw material ratio table for preparation of LQ090-1

[0302] Step 2. Preparation of LQ090-2

[0303] According to the amount of raw materials shown in the table below, 9-heptadecanol, 8-bromooctanoic acid, DCC, DMAP and DCM were added to the reaction bottle, and stirred at room temperature for 16h. TLC (PE:EA=20:1) showed that the reaction was complete (the product Rf value was 0.6), then the reaction solution was filtered with diatomite, and rotary evaporation. After purification, 3.6g of colorless oil was obtained (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0304] Raw material ratio table for preparation of LQ090-2

[0305] Step 3. Preparation of LQ090-3

[0306] According to the amount of raw materials shown in the table below, LQ090-2, ethanolamine and acetonitrile were added to the reaction bottle, and heated to 30℃ and stirred for 16h. TLC (DCM:MeOH=10:1) showed that the reaction was complete (the product Rf value was 0.4), then the reaction solution was rotary evaporation. Then diluted with 100mL of ethyl acetate, washed with 100mL of water twice, and the organic phase was dried with anhydrous sodium sulfate, rotary evaporation, and purification to obtain 2.8g of colorless oil (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0307] Raw material ratio table for preparation of LQ090-3

[0308] Step 4. Preparation of LQ090

[0309] To a reaction flask was added LQ090-3, LQ090-1, potassium carbonate, potassium iodide and acetonitrile in the amounts shown in the table below, and heated to 80°C and stirred for 16h. After TLC (DCM:MeOH = 10:1) showed the reaction was complete (product Rf value of 0.6), the reaction was filtered, and the filter cake was washed twice with 50 mL of ethyl acetate. The filtrate was spun dry and purified to give 800 mg of colorless oil (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0310] Table of raw material ratios for LQ090 preparation

[0311] The proton nuclear magnetic resonance spectrum of the resulting end product is:

[0312] 1 H NMR (600 MHz, Chloroform-d) δ 4.88 - 4.83 (m, 1H), 4.03 (dd, J = 11.1, 6.2 Hz, 1H), 3.98 (dd, J = 11.1, 5.8 Hz, 1H), 3.57 (t, J = 5.3 Hz, 2H), 2.63 (t, J = 5.3 Hz, 2H), 2.50 (t, J = 7.6 Hz, 4H), 2.27 (dt, J = 9.2, 7.5 Hz, 4H), 1.75 (hd, J = 6.8, 4.7 Hz, 1H), 1.68 (dq, J = 12.1, 3.9 Hz, 1H), 1.61 (p, J = 7.3 Hz, 4H), 1.47 (tdd, J = 18.9, 9.1, 4.6 Hz, 10H), 1.37 - 1.19 (m, 40H), 0.91 - 0.84 (m, 18H).

[0313] (6) LQ091

[0314] The synthetic route of LQ091 is shown below:

[0315] The specific preparation method is:

[0316] Step 1. Preparation of LQ091-1

[0317] To a reaction flask was added 8-bromooctanoic acid, tetrahydrocuminyl alcohol, DCC, DMAP and DCM in the amounts shown in the table below, and stirred at room temperature for 16h. After TLC (PE:EA = 20:1) showed the reaction was complete (product Rf value of 0.65), the reaction was filtered with celite, spun dry, and purified to give 2g of colorless oil, with a yield of 87% (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0318] Table of raw material ratios for LQ091-1 preparation

[0319] Step 2. Preparation of 9-octadecenol

[0320] According to the table below, magnesium chips and anhydrous tetrahydrofuran were added to a reaction bottle, heated to 45°C under nitrogen protection, and 1-bromononane was slowly added dropwise. The temperature rose and began to reflux after the Grignard reaction was triggered. After the addition was complete, the reaction was incubated at 45°C for 2 hours, then cooled to 2-10°C, and n-nonyl aldehyde was slowly added dropwise. After the addition was complete, the reaction was stirred at room temperature for 12 hours. TLC (PE:EA=20:1) showed that the reaction was complete (the product had an Rf value of 0.3). The reaction was cooled to 2-10°C, and 100 mL of 1N dilute hydrochloric acid was slowly added dropwise. After the addition was complete, the mixture was stirred for 30 minutes, then 600 mL of ethyl acetate and 1 L of water were added for extraction. The organic phase was dried over anhydrous sodium sulfate and then rotary evaporated. Purification yielded 81 g of white solid with a yield of 85% (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0321] Table of raw material ratios for preparation of 9-octadecenol

[0322] Step 3. Preparation of LQ091-2

[0323] According to the table below, 9-octadecenol, 8-bromooctanoic acid, DCC, DMAP, and DCM were added to a reaction bottle, which was stirred at room temperature for 16 hours. TLC (PE:EA=20:1) showed that the reaction was complete (the product had an Rf value of 0.6). The reaction mixture was then filtered with diatomite, rotary evaporated, and purified to obtain 3.8 g of colorless oil with a yield of 80% (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0324] Table of raw material ratios for preparation of LQ091-2

[0325] Step 3. Preparation of LQ091-3

[0326] According to the table below, LQ091-2, ethanolamine, and acetonitrile were added to a reaction bottle, which was heated to 30°C and stirred for 16 hours. TLC (DCM:MeOH=10:1) showed that the reaction was complete (the product had an Rf value of 0.4). The reaction mixture was then rotary evaporated. It was then diluted with 100 mL of ethyl acetate, washed twice with 100 mL of water, and the organic phase was dried over anhydrous sodium sulfate, rotary evaporated, and purified to obtain 2.9 g of colorless oil with a yield of 80% (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0327] Table of raw material ratios for preparation of LQ091-3

[0328] Step 4. Preparation of LQ091

[0329] To a reaction flask was added LQ091-3, LQ091-1, potassium carbonate, potassium iodide and acetonitrile in the amounts shown in the table below, and heated to 80°C and stirred for 16h. After TLC (DCM:MeOH = 10:1) showed the reaction was complete (product Rf value of 0.6), the reaction was filtered, the filter cake was washed twice with 50 mL of ethyl acetate, the filtrate was spun dry and purified to give 900 mg of colorless oil (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0330] Table of raw material ratios for LQ091 preparation

[0331] The proton nuclear magnetic resonance spectrum of the resulting end product is:

[0332] 1 H NMR (600 MHz, Chloroform-d) δ 4.88 - 4.83 (m, 1H), 4.04 - 3.97 (m, 2H), 3.57 (t, J = 5.1 Hz, 2H), 3.51 - 3.44 (m, 1H), 2.63 (s, 2H), 2.50 (t, J = 7.8 Hz, 4H), 2.28 (dt, J = 9.3, 7.5 Hz, 4H), 1.93 (dq, J = 12.7, 4.0 Hz, 2H), 1.76 (pd, J = 6.9, 4.7 Hz, 1H), 1.68 (dq, J = 12.0, 3.9 Hz, 2H), 1.65 - 1.56 (m, 5H), 1.54 - 1.41 (m, 10H), 1.31 - 1.23 (m, 36H), 0.90 - 0.85 (m, 18H).

[0333] (7) LQ093

[0334] The synthetic route for LQ093 is shown below:

[0335] Wherein, the specific preparation method is:

[0336] Step 1. Preparation of LQ093-1

[0337] To a reaction flask was added 9-octadecenol, 8-bromooctanoic acid, DCC, DMAP and DCM in the amounts shown in the table below, and stirred at room temperature for 16h. After TLC (PE:EA = 20:1) showed the reaction was complete (product Rf value of 0.6), the reaction was filtered with celite, spun dry, and purified to give 3.8 g of colorless oil (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0338] Table of raw materials for the preparation of LQ093-1

[0339] Step 2. Preparation of LQ093-2

[0340] To a reaction flask was added 9-heptadecanol, 8-bromooctanoic acid, DCC, DMAP and DCM in the amounts shown in the table below and stirred at room temperature for 16 h. TLC (PE:EA = 20:1) showed the reaction to be complete (product Rf = 0.6) and the reaction was filtered through celite, spun down and purified to give 3.6 g of a colourless oil (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0341] Table of raw materials for the preparation of LQ093-2

[0342] Step 3. Preparation of LQ093-3

[0343] To a reaction flask was added LQ093-2, ethanolamine and acetonitrile in the amounts shown in the table below and stirred at 30 °C for 16 h. TLC (DCM:MeOH = 10:1) showed the reaction to be complete (product Rf = 0.4) and the reaction was spun down, diluted with 100 mL of ethyl acetate and washed twice with 100 mL of water, the organic phase was dried over anhydrous sodium sulphate, spun down and purified to give 2.8 g of a colourless oil (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0344] Table of raw materials for the preparation of LQ093-3

[0345] Step 4. Preparation of LQ093

[0346] To a reaction flask was added LQ093-3, LQ093-1, potassium carbonate, potassium iodide and acetonitrile in the amounts shown in the table below and stirred at 80 °C for 16 h. TLC (DCM:MeOH = 10:1) showed the reaction to be complete (product Rf = 0.6) and the reaction was filtered, the filter cake was washed twice with 50 mL of ethyl acetate, the filtrate was spun down and purified to give 900 mg of a colourless oil (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0347] Table of raw materials for the preparation of LQ093

[0348] The proton nuclear magnetic resonance spectrum of the final product was:

[0349] 1H NMR (600 MHz, Chloroform-d) δ 4.86 (p, J = 6.2 Hz, 2H), 3.61 (t, J = 5.3 Hz, 2H), 2.68 (s, 2H), 2.55 (t, J = 7.8 Hz, 4H), 2.27 (t, J = 7.5 Hz, 4H), 1.64 - 1.58 (m, 4H), 1.54 - 1.46 (m, 12H), 1.33 - 1.22 (m, 62H), 0.87 (t, J = 7.0 Hz, 12H).

[0350] (8) LQ094

[0351] The synthetic route of LQ094 is shown below:

[0352] The specific preparation method is:

[0353] Step 1. Preparation of LQ094-1

[0354] According to the amount of raw materials shown in the table below, 9-octadecanol, 8-bromooctanoic acid, DCC, DMAP and DCM were added to the reaction bottle, and stirred at room temperature for 16 h. TLC (PE:EA = 20:1) showed that the reaction was complete (the Rf value of the product was 0.6), then the reaction solution was filtered with diatomite, and dried by rotary evaporation. After purification, 3.8 g of colorless oil was obtained with a yield of 80% (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0355] Raw material ratio table for preparation of LQ094-1

[0356] Step 2. Preparation of LQ094

[0357] According to the amount of raw materials shown in the table below, LQ94-1, ethanolamine, K2CO3, KI and acetonitrile were added to the reaction bottle, and stirred at 80°C for 12 h. TLC (DCM:MeOH = 10:1) showed that the reaction was complete (the Rf value of the product was 0.6), then the reaction solution was filtered, the filter cake was washed twice with 50 ml of ethyl acetate, and the filtrate was dried by rotary evaporation. After purification, 700 mg of colorless oil was obtained (flash chromatography, dichloromethane / methanol from 100:1 to 20:1).

[0358] Raw material ratio table for preparation of LQ094

[0359] The nuclear magnetic resonance hydrogen spectrum of the obtained final product is:

[0360] 1H NMR (600 MHz, Chloroform-d) δ 4.85 (p, J = 6.3 Hz, 2H), 3.68 (t, J = 5.3 Hz, 2H), 2.76 (s, 2H), 2.64 (s, 4H), 2.27 (t, J = 7.5 Hz, 4H), 1.61 (q, J = 7.2 Hz, 4H), 1.54 (s, 4H), 1.50 (d, J = 6.3 Hz, 8H), 1.33 - 1.22 (m, 64H), 0.87 (t, J = 7.0 Hz, 12H).

[0361] Example 3 Experimental validation of new ionizable lipids screened by model

[0362] In this example, when testing the in vivo / in vitro effects of the above LNP of LQ085-091, LQ093, LQ094, ionizable lipids DLin-MC3-DMA (abbreviated as MC3) and SM102 were selected as controls (as shown in Table 5).

[0363] The specific experimental method is as follows:

[0364] In vitro experiment:

[0365] An appropriate amount of the above ionizable lipid to be tested, cholesterol, DSPC and DMG-PEG2000 were dissolved in ethanol to prepare a mother liquor of each lipid. Then, the ionizable lipid to be tested: DSPC: cholesterol: DMG-PEG2000 was mixed in a molar ratio of 50:10:38.5:1.5 to prepare a mixed lipid solution. The final concentration of the ionizable lipid to be tested in the solution was 12.5 mM. Luciferase mRNA was dispersed in citrate buffer to prepare an acidic mRNA solution. Using a PNI microfluidic device, under the condition of a nitrogen-phosphorus ratio of 6:1, at a flow rate of 12 mL / min, the mRNA solution and the lipid ethanol solution were mixed in a volume ratio of 3:1 (mRNA solution: lipid ethanol solution). The mixture was dialyzed (12-24 hours) into 0.01 M PBS to remove ethanol. After dialysis, the LNP solution was concentrated by ultrafiltration (Amicon-Ultra, MWCO 10KDa). The LNP solution was sterilized by a 0.22 μm sterile filter. The particle size and polydispersity index (PDI) of the LNP were measured using a Malvern particle size analyzer. The encapsulation efficiency (EE) of the LNP was determined using a Quant-it Ribogreen RNA detection kit, and the mRNA concentration was determined using a Stunner high-throughput concentration and particle size analyzer. The apparent pKa of the LNP was detected by TNS (2-(p-tolylamino)-6-naphthalenesulfonic acid) co-incubation method.

[0366] Table 5. Basic characterization indicators of lipid nanoparticles for different ionizable lipids.

[0367] Table 5 shows that the predicted apparent pKa of LNPs containing newly synthesized ionizable lipids were all between 6 and 7, which is close to the measured values. The mRNA encapsulation efficiency of all LNPs was above 90%, and the particle size was uniform.

[0368] In vivo experiments:

[0369] Female Balb / C mice (purchased from Vital River) aged 6-8 weeks were used as experimental subjects. Each group of mice was injected intravenously with the aforementioned LNP loaded with luciferase mRNA, at a dose of 5 μg per mouse. At 3, 10, 24, and 48 hours post-administration, D-luciferin potassium salt was injected intraperitoneally, and the total luminescence in the mice was detected using the IVIS Spectrum small animal imaging system.

[0370] Figures 3A-3D show the measurement results of fluorescence emission signals throughout the mouse body.

[0371] Experimental results showed that the mRNA delivery efficiency of LQ087, LQ089, LQ090, LQ093, and LQ094 was not inferior to that of the standard MC3 formulation, and the delivery efficiency of LQ089 and LQ091 was superior to that of the standard formulation, with statistically significant differences.

[0372] Example 4: A PBPK model and its application in evaluating the pharmacokinetics of lipid nanoparticles

[0373] In this embodiment, a PBPK model and its construction method are provided. This model can be used to evaluate the pharmacokinetics of lipid nanoparticles, especially to calculate the rate constants of key steps in the pharmacokinetics of lipid nanoparticles.

[0374] In this embodiment, the allocation ratio and method of the training set and test set are the same as in the above embodiments.

[0375] The specific construction method is as follows:

[0376] In vivo LNP metabolic kinetic data were collected (rat data were obtained from US9868691B2 and Sabnis, S.; Kumarasinghe, ES; Salerno, T.; Mihai, C.; Ketova, T.; Senn, JJ; Lynn, A.; Bulychev, A.; McFadyen, I.; Chan, J.; Almarsson, Stanton, M. G.; Benenato, K. E. A Novel Amino Lipid Series for mRNA Delivery: Improved Endosomal Escape and Sustained Pharmacology and Safety in Non-Human Primates. Mol. Ther. 2018, 26 (6), 1509-1519. https: / / doi.org / 10.1016 / j.ymthe.2018.03.010.; Mouse data from Hen, S.; Tam, Y. Y. C.; Lin, P. J. C.; Sung, M. M. H.; Tam, Y. K.; Cullis, P. R. Influence of Particle Size on the in Vivo Potency of Lipid Nanoparticle Formulations of siRNA. J. Controlled Release 2016, 235, 236-244. https: / / doi.org / 10.1016 / j.jconrel.2016.05.059. and Mui, B. L.; Tam, Y. K.; Jayaraman, M.; Ansell, S. M.; Du, X.; Tam, Y. Y. C.; Lin, P. J.; Chen, S.; Narayanannair, J. K.; Rajeev, K. G.; Manoharan, M.; Akinc, A.; Maier, M. A.; Cullis, P.; Madden, T. D.; Hope, M. J. Influence of Polyethylene Glycol Lipid Desorption Rates on Pharmacokinetics and Pharmacodynamics of siRNA Lipid Nanoparticles. Mol. Ther. Nucleic Acids 2013, 2 (12), e139. https: / / doi.org / 10.1038 / mtna.2013.66.; Human data from U.S. Food and Drug Administration. NDA 210922 - Patisiran - Cross-Discipline Team Leader Review. 2018.) to establish the PBPK model in vivo.For rat data, SD rats weighing 225-250 g were injected intravenously with a dose of 0.2 mg / kg of human erythropoietin (hEPO) encoding mRNA-LNPs. The LNPs were composed of ionizable lipid, DSPC, cholesterol, and PEG-lipid in a molar ratio of 50:10:38.5:1.5. The ratio of nitrogen in the ionizable lipid and the ratio of phosphates in the mRNA backbone (N / P ratio) was estimated to be 5.67 (cf. Hassett, K. J.; Benenato, K. E.; Jacquinet, E.; Lee, A.; Woods, A.; Yuzhakov, O.; Himansu, S.; Deterling, J.; Geilich, B. M.; Ketova, T.; Mihai, C.; Lynn, A.; McFadyen, I.; Moore, M. J.; Senn, J. J.; Stanton, M. G.; Almarsson, C. J. Pharm. Sci. 2017, 106, 1173-1181. https: / / doi.org / 10.1016 / j. phas.2017.01.015). Ciaramella, G.; Brito, L. A. Optimization of Lipid Nanoparticles for Intramuscular Administration of mRNA Vaccines. Mol. Ther. -Nucleic Acids 2019, 15, 1-11. https: / / doi.org / 10.1016 / j.omtn.2019.01.013.). The ionizable lipids used included MC3, SM-102, and Lipid 18 (referred to as Lipid 18 in the patent literature and as Lipid5 in the article). For mouse data, 6-8 week old C57B1 / 6 mice were injected intravenously with 0.3 mg / kg of siRNA, and it was loaded in LNPs of similar composition. The ionizable lipid used was DMAP-BLP, which is an analog of MC3. The PEG-lipid fraction ranged from 0.25% to 5% to produce LNPs with different particle sizes. The N / P ratio was 3 and 6. For human data, the data came from the phase 1 clinical trial (ALN-TTR02-001 and ALN-TTR02-005) of the first approved siRNA drug (patisiran or Onpattro®). In the trial, healthy volunteers were injected intravenously with 0.5, 0.3, 0.15, 0.05, and 0.01 mg / kg of patisiran, and the LNP composition was very similar to the one used in the mice described above. Ciaramella, G.; Brito, L. A. Optimization of Lipid Nanoparticles for Intramuscular Administration of mRNA Vaccines. Mol. Ther. -Nucleic Acids 2019, 15, 1-11. https: / / doi.org / 10.1016 / j.omtn.2019.01.013.). The ionizable lipids used included MC3, SM-102, and Lipid 18 (referred to as Lipid 18 in the patent literature and as Lipid5 in the article). For mouse data, 6-8 week old C57B1 / 6 mice were injected intravenously with 0.3 mg / kg of siRNA, and it was loaded in LNPs of similar composition. The ionizable lipid used was DMAP-BLP, which is an analog of MC3. The PEG-lipid fraction ranged from 0.25% to 5% to produce LNPs with different particle sizes. The N / P ratio was 3 and 6. For human data, the data came from the phase 1 clinical trial (ALN-TTR02-001 and ALN-TTR02-005) of the first approved siRNA drug (patisiran or Onpattro®). In the trial, healthy volunteers were injected intravenously with 0.5, 0.3, 0.15, 0.05, and 0.01 mg / kg of patisiran, and the LNP composition was very similar to the one used in the mice described above.

[0377] Cellular LNP metabolism kinetic data (from HeLa cell experiments) were collected, which included data for delivery and RNA release for three ionizable lipids C12-200, MC3, and L319 LNPs (see Gilleron, J.; Querbes, W.; Zeigerer, A.; Borodovsky, A.; Marsico, G.; Schubert, U.; Manygoats, K.; Seifert, S.; Andree, C.; M.; Epstein-Barash, H.; Zhang, L.; Koteliansky, V.; Fitzgerald, K.; Fava, E.; Bickle, M.; Kalaidzidis, Y.; Akinc, A.; Maier, M.; Zerial, M. Image-Based Analysis of Lipid Nanoparticle-Mediated siRNA Delivery, Intracellular Trafficking and Endosomal Escape. Nat. Biotechnol. 2013, 31 (7), 638-646. https: / / doi.org / 10.1038 / nbt.2612.; Sahay, G.; Querbes, W.; Alabi, C; Eltoukhy, A.; Sarkar, S.; Zurenko, C; Karagiannis, E.; Love, K.; Chen, D.; Zoncu, R.; Buganim, Y.; Schroeder, A.; Langer, R.; Anderson, D. G. Efficiency of siRNA Delivery by Lipid Nanoparticles Is Limited by Endocytic Recycling. Nat. Biotechnol. 2013, 31 (7), 653-658. https: / / doi.org / 10.1038 / nbt.2614.; and Wittrup, A.; Ai, A.; Liu, X.; Hamar, P.; Trifonova, R.; Charisse, K.; Manoharan, M.; Kirchhausen, T.; Lieberman, J. Visualizing Lipid-Formulated siRNA Release from Endosomes and Target Gene Knockdown. Nat. Biotechnol. 2015, 33 (8), 870-876. https: / / doi.org / 10.1038 / nbt.3298.). These data include the transport processes of LNP components between endosomes, macropinosomes, lysosomes, autophagosomes in cells, and the processes of LNP decomposition, efflux, and RNA release. Among them, in the data of C12-200, the phagocytosis, decomposition, and efflux of intracellular LNPs are also involved.In the data of MC3, the time course of LNP phagocytosis, the fraction of two endosomes and lysosomes, and the fraction of siRNA release associated with the amount of phagocytosis were involved. In the data of L319, the data about LNP phagocytosis, the proportion of phagosomae that triggered release events, and the proportion of siRNA released from endosomes and retained were provided.

[0378] The significance of the PBPK model construction is that it highlights the PK behavior of ionizable lipids through the in vivo RNA-LNP PBPK. The structure of the in vivo PBPK model is shown in Figure 4. It can be found that in the model, the whole body is divided into six main compartments, including veins, arteries, pulmonary vessels, liver, spleen, and “other organs”. LNP will be distributed in the circulatory system and all organs along with the blood flow. Since the LNP formulation involved in the data is mainly targeted at the liver and spleen, and the targeting effect on other tissues is minimal, only the liver and spleen models are mechanically established. The relevant processes include the penetration of LNP between the blood vessels and the interstitial space of the organ, the phagocytosis of LNP from the interstitium into the space inside a cell (wrapped in endosomes) by receptor-mediated LNP, the dissociation of LNP to release free ionizable lipids, and the hydrolysis of ionizable lipids. In order to simplify the model, only the blood volume of the lung is selected as a reference to establish a simple organ model to represent all other organs, in which the processes of phagocytosis, decomposition, and hydrolysis are combined into one process.

[0379] The differential equations of the in vivo PBPK model thus obtained can be summarized as follows:

[0380] Distribution through blood flow:

[0381] Penetration of blood to interstitium:

[0382] Receptor-mediated LNP phagocytosis:

[0383] LNP decomposition into ionizable lipids:

[0384] Hydrolysis of ionizable lipids:

[0385] In the formula, M represents the mass of lipids, and Q represents the blood flow. All symbols C other than C receptor represent the concentration of lipids in various compartments, and the subscript thereof indicates the specific location. C receptoris the receptor concentration mediating the phagocytosis of LNP. Likewise, all V symbols represent the volume of various organs or sub- organs. P represents the permeability of LNP, S represents the endothelial area of organs, and their product is a parameter representing the rate of exchange of lipid mass between blood and interstitial space. k in represents the rate of phagocytosis of cells, which depends on the concentration of lipid and receptor according to the law of chemical kinetics. dis represents the rate of decomposition of LNP to release free ionizable lipid. k el represents the rate of metabolism, which represents the rate of hydrolysis of ionizable lipid.

[0386] The change of lipid mass in any compartment is affected by multiple factors. For example, the interstitial lipid mass in the liver will change due to permeation from blood to interstitial space, as well as cell phagocytosis, etc. This process will result in an increase in mass, i.e. dM > 0, otherwise dM < 0. Therefore, the rate of change in compartment i is calculated as the sum of the changes of all processes related to it (assuming there are n processes related), i.e.

[0387] The rate of change of mass of all i compartments above can be converted to the rate of change of concentration of mass and volume, i.e.

[0388] When fitting pharmacokinetic data with a system of differential equations, it is necessary to first give the initial mass or concentration of each compartment. Take the vein as an example, when fitting the pharmacokinetic data of intravenous injection, the mass concentration of the vein blood is the corresponding injection amount, and the mass of the rest of the compartments is 0.

[0389] In this embodiment, the physiological related parameters required by the collected model are shown in Table 6. The collection of parameters combines the basic information of the subject animal or human, such as age, weight, etc.

[0390] Table 6 Physiological related parameters of PBPK model

[0391] a, The value of "hepatic arterial blood flow" is calculated by the sum of blood flow of hepatic artery, large intestine, small intestine and pancreas. B, "Intrahepatic receptor concentration" is assumed to be 1, and "receptor extension ratio from liver to spleen" is determined by fitting the ratio of receptor concentration of liver and spleen.

[0392] Based on the transport processes of LNP in cells, the resulting cell PBPK model structure is shown in FIG. 5 and FIG. 6. It can be found that LNP is first taken up by various receptors and phagocytosed into early endosomes or macropinosomes. Then, a series of protein exchanges between endosomes or macrophage phagosomes and cells occur, promoting their maturation to form late endosomes and further form lysosomes. After phagocytosis, LNP begins to decompose. At the late endosome stage, a significant proportion of LNP will be secreted outside the cell through mobile vesicles. At the same time, a small amount of LNP contained in the late endosome will trigger an RNA release event. However, even if the release occurs, only part of the RNA in the LNP will be released into the cytoplasm. After the release event, the late endosome will be wrapped by autophagosomes and then integrated into autolysosomes. FIG. 5 and FIG. 6 show two versions of the model, respectively. The complex version is the most mechanistic model, which presents the mechanism of LNP transport in detail, while the simple version does not explicitly simulate the dissociation of LNP because it is more compatible with the existing data. In these models, the transport processes indicated by dashed lines are simplified because they are not the main focus of the modeling target or the experimental methods in the prior art do not support their simulation. The omitted processes include LNP dissociation in the culture medium, complete LNP excretion from the cell, reuptake of dissociated lipids and RNA, and metabolism of lipids and RNA.

[0393] In the simulation of the cell PBPK model, the process of LNP uptake directly references the uptake curve of the original data. In addition, a weight ratio parameter is introduced to describe the difference in the amount of lipids and RNA in the LNP taken up by the cells. The mass changes of substances caused by other processes are modeled as a series of differential equations, which are as follows:

[0394] Lipids and RNA from early to late endosomes:

[0395] Lipids and RNA from late endosomes to lysosomes:

[0396] Free lipids and RNA dissociated from LNP:

[0397] Excretion of lipids and RNA:

[0398] RNA released into the cell plasma after the release event:

[0399] RNA wrapped in autophagosomes after the release event:

[0400] Lipids wrapped in autophagosomes after the release event:

[0401] Lipid and RNA from autophagosome to autolysosome:

[0402] where M represents the mass of lipid or RNA, and k is the kinetic rate. The subscript of the symbol indicates the location, state or process experienced by the corresponding mass (specifically, for location, EE represents early endosome; LE represents late endosome; LY represents lysosome; AP represents autophagosome; AL represents autolysosome. For state and process, "ass" represents assembly; "dis" represents dissociation; "lip" represents lipid; "eg" represents efflux; "rel" represents release). In addition, it is assumed that k LE_LY is equal to k AP_AL . The parameter f rel represents the ratio between the RNA released into the cytoplasm at the time of the release event and the RNA experiencing the release event.

[0403] Likewise, the change in mass of lipid or RNA in any compartment is actually the result of multiple processes. Therefore, the ODE (ordinary differential equation) for any mass change takes the form provided in equation The cell flow model does not rely on prior knowledge of physiological parameters, such as the total volume of cells in culture. One reason for this is that the uptake process will be the same as what is presented in the data, so the effect of cell number on uptake does not need to be taken into account in this case. Another reason is that all experimental data collected is presented in percentage terms, which can be easily modeled with a first order kinetic rate.

[0404] In this example, all of the above equations were written into the SimBiology APP of MATLAB to construct the PBPK model. In the model fitting, the default estimation method "Isqnonlin" was chosen to fit the parameters. Four error models, "constant", "proportional", "exponential" and "combined", were compared to obtain the best performance results. The fitting results were mainly verified by visual inspection and Bayesian Information Criterion (BIC) and Akaike Information Criterion (AIC) and log-likelihood.

[0405] The fitting results of the model are shown in FIG. 7, FIGS. 8A-8C, FIG. 9, FIGS. 10A-10C, FIGS. 11A-11D. FIG. 7 is the data fitting result of rats, and the involved LNP includes three ionizable lipids, MC3, SM-102 and Lipid 18, respectively. Under the condition of intravenous injection of 0.2 mg / kg mRNA (equivalent to about 2.23 mg / kg MC3, about 2.46 mg / kg SM-102 and Lipid 18) in rats, kin k dis k el The magnification relative to MC3 was used to describe the differences in metabolic processes among the three LNPs. This demonstrates that the type of ionizable lipid significantly affects the various transport rates of LNPs. All model simulations were performed under conditions consistent with the original experimental design (see US9868691B2 and Sabnis, S.; Kumarasinghe, ES; Salerno, T.; Mihai, C.; Ketova, T.; Senn, JJ; Lynn, A.; Bulychev, A.; McFadyen, I.; Chan, J.; Almarsson, Stanton, M. G.; Benenato, K. E. A Novel Amino Lipid Series for mRNA Delivery: Improved Endosomal Escape and Sustained Pharmacology and Safety in Non-Human Primates. Mol. Ther. 2018, 26 (6), 1509-1519. https: / / doi.org / 10.1016 / j.ymthe.2018.03.010). FIGS. 8A-8C are the results of PBPK model fitting of siRNA-LNP metabolism data in mice, with some simplifications in the model structure due to limitations of the data. The ionizable lipids involved are MC3 and DMAP-BLP. The particle sizes of the LNPs are 80, 78, 45 nm, respectively. Mice were injected intravenously with siRNA-LNP at a lipid dose of 11.1 mg / kg or 0.3 mg / kg siRNA-LNP (equivalent to about 3.42 mg / kg DMAP-DLP). The results demonstrate that the difference in LNP particle size can significantly affect the uptake rate in organs.The simulation conditions for all models were according to the original experimental design (see Hen, S.; Tam, Y. Y. C; Lin, P. J. C; Sung, M. M. H.; Tam, Y. K.; Cullis, P. R. Influence of Particle Size on the in Vivo Potency of Lipid Nanoparticle Formulations of siRNA. J. Controlled Release 2016, 235, 236-244. https: / / doi.org / 10.1016 / j.jconrel.2016.05.059. and Mui, B. L.; Tam, Y. K.; Jayaraman, M.; Ansell, S. M.; Du, X.; Tam, Y. Y. C; Lin, P. J.; Chen, S.; Narayanannair, J. K.; Rajeev, K. G.; Manoharan, M.; Akinc, A.; Maier, M. A.; Cullis, P.; Madden, T. D.; Hope, M. J. Influence of Polyethylene Glycol Lipid Desorption Rates on Pharmacokinetics and Pharmacodynamics of siRNA Lipid Nanoparticles. Mol. Ther. Nucleic Acids 2013, 2 (12), e139. https: / / doi.org / 10.1038 / mtna.2013.66). Figure 9 is the PBPK model fitting results for human data of siRNA-LNP at different doses, the model structure was simplified due to the limitation of data. The ionizable lipid is MC3. The data is from marketed drug. The results of the Phase 1 clinical trial of Givlaari® (ALN-TTR02-001 and ALN-TTR02-005), all model simulation conditions were performed according to the original experimental design (see U.S. Food and Drug Administration. NDA 210922 - Patisiran - Cross-Discipline Team Leader Review. 2018). FIGS. 10A-10C, FIGS. 11A-11D are the results of PBPK model fitting of LNP cellular uptake data, probability calculations of RNA release into cells, and the ability of LNP of the three lipids to release their encapsulated RNA for the ionizable lipids C12-200, L319, MC3, illustrating the differences between the three ionizable lipids. All model simulation conditions were performed according to the original experimental design (see Gilleron, J.; Querbes, W.; Zeigerer, A.; Borodovsky, A.; Marsico, G.; Schubert, U.; Manygoats, K.; Seifert, S.; Andree, C.; M.; Epstein-Barash, H.; Zhang, L.; Koteliansky, V.; Fitzgerald, K.; Fava, E.; Bickle, M.; Kalaidzidis, Y.; Akinc, A.; Maier, M.; Zerial, M. Image-Based Analysis of Lipid Nanoparticle-Mediated siRNA Delivery, Intracellular Trafficking and Endosomal Escape. Nat. Biotechnol. 2013, 31 (7), 638-646. https: / / doi.org / 10.1038 / nbt.2612.; Sahay, G.; Querbes, W.; Alabi, C; Eltoukhy, A.; Sarkar, S.; Zurenko, C; Karagiannis, E.; Love, K.; Chen, D.; Zoncu, R.; Buganim, Y.; Schroeder, A.; Langer, R.; Anderson, D. G. Efficiency of siRNA Delivery by Lipid Nanoparticles Is Limited by Endocytic Recycling. Nat. Biotechnol. 2013, 31 (7), 653-658. https: / / doi.org / 10.1038 / nbt.2614., and Wittrup, A.; Ai, A.; Liu, X.; Hamar, P.; Trifonova, R.; Charisse, K.; Manoharan, M.; Kirchhausen, T.; Lieberman, J. Visualizing Lipid-Formulated siRNA Release from Endosomes and Target Gene Knockdown. Nat. Biotechnol. 2015, 33 (8), 870-876. https: / / doi.org / 10.1038 / nbt.3298.).

[0406] The above embodiments are the preferred embodiments of the present application, but the embodiments of the present application are not limited to the above embodiments, and any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present application should be equivalent replacement methods, and are included in the protection scope of the present application.

Claims

1. A method for constructing a LNP property prediction model, comprising the following steps: (1) preprocessing collected LNP data, and extracting ionizable lipid molecular structures from the preprocessed data; (2) encoding the extracted ionizable lipid molecular structures; (3) combining the encoded information with other necessary information representing LNP composition to obtain input information; (4) modeling the input information to obtain a LNP property prediction model.

2. The construction method according to claim 1, characterized in that, In step (1), the LNP data includes: LNP formulation, apparent pKa value, particle size, mRNA-encoded protein type contained, mRNA encapsulation rate, experimental animal species, administration route, administration dose, and mRNA expression level; wherein the LNP formulation includes the chemical structure of ionizable lipid, the type of auxiliary lipid.

3. The construction method of claim 1, wherein, In step (1), the preprocessing is to improve the uniformity of LNP data, and the improvement of LNP data uniformity preferably includes: retaining data with consistent administration route, consistent measurement index, mRNA expression level comparable to that of a standard LNP, and formulation composition identical or equivalent to that of a standard LNP.

4. The construction method according to claim 3, characterized in that, The standard LNP formulation is a LNP formulation composed of ionizable lipid, auxiliary lipid, cholesterol, and PEG lipid.

5. The construction method according to claim 4, characterized in that, The ionizable lipid includes MC3 or its derivative esters; the auxiliary lipid includes DSPC; and the PEG lipid includes PEG2000-DMG or its derivative esters.

6. The construction method of claim 1, wherein, The encoding method includes using simplified molecular-input line-entry system (SMILES), extended connectivity fingerprint (ECFP) sequence, molecular physical and chemical property descriptor, 2D molecular descriptor, 3D molecular descriptor, molecular structure picture, molecular graph, and molecular coordinates for encoding.

7. The construction method of claim 1, wherein, The algorithms used in the modeling include LightGBM, random forest, XGBoost, decision tree, support vector machine, artificial neural network, deep neural network, residual network, recurrent neural network, long short-term memory network, convolutional neural network, and Transformer. 8.A LNP property prediction model constructed by the method of any one of claims 1-7. 9.Use of the LNP property prediction model of claim 8 in LNP evaluation.

10. Use according to claim 9, characterized in that, The LNP evaluation includes apparent pKa evaluation and mRNA delivery efficiency evaluation. 11.A method for screening ionizable lipids, comprising the following steps: (1) identifying structural features affecting LNP properties in ionizable lipids using the LNP property prediction model of claim 8 to obtain structural fragments, and splicing the structural fragments to construct a molecular library; (2) screening compounds with an apparent pKa range of 6-7 and mRNA delivery efficiency higher than that of a control formulation from the molecular library using the LNP property prediction model of claim 8. In step (2), the compounds are obtained by splicing the structural features in the molecular library, and the splicing is performed according to the rules of ionizable lipid molecules.

12. The screening method according to claim 11, wherein, wherein, 13. The screening method according to claim 12, characterized in that, The structural features include: x is selected from 0 or 1, y is an integer from 0 to 8, z is an integer from 1 to 4, R4 is selected from -H or -CH3. ​ 14. An ionizable lipid characterized in that, The ionizable lipid is screened by the screening method of any one of claims 11-13, and the ionizable lipid has at least one structural feature of claim 13.

15. The ionizable lipid of claim 14, wherein, The ionizable lipid comprises:

16. Use of the ionizable lipid of any one of claims 14-15 in drug delivery.

17. A lipid nanoparticle (LNP) characterized in that, The LNP comprises the ionizable lipid of any one of claims 14-15 and a nucleic acid molecule.

18. The LNP of claim 17, wherein, The nucleic acid molecule comprises RNA and DNA, and the RNA comprises mRNA and siRNA.

19. The LNP of claim 17, wherein, The LNP is further loaded with a second active pharmaceutical molecule and / or a pharmaceutically acceptable adjuvant.

20. Use of the LNP of any one of claims 17-19 in drug delivery.

21. A drug delivery vehicle, characterized in that, The drug delivery carrier comprises the ionizable lipid of any one of claims 14-15 or the LNP of any one of claims 17-19.

22. A method for constructing a physiologically-based pharmacokinetic (PBPK) model, comprising the following steps: (1) classifying the in vivo metabolic kinetic data and the cellular metabolic kinetic data of the LNP, respectively; (2) according to the transport mechanism of the LNP in vivo or in cells, writing the differential equations of the mass change rate of each transport pathway required, and combining all the differential equations into a differential equation system, i.e. obtaining the PBPK model.

23. The method of construction of claim 22, wherein, The in vivo transport pathways that need to describe the mass change rate include: lipids distributed in the blood circulation system and between organs with blood flow; lipids permeating from the blood part of the organ to the interstitium; lipids transported from the interstitium to the cell mediated by receptors; LNP decomposition in the cell to form free ionizable lipids; hydrolytic metabolism of ionizable lipids.

24. The method of construction of claim 22, wherein, The intracellular transport pathways that need to describe the mass change rate include: lipids and RNA from early endosomes to late endosomes, lipids and RNA from late endosomes to lysosomes, LNP dissociation to form free lipids and RNA, discharge of lipids and RNA, release of RNA to the cell plasma after the release event, lipids and RNA wrapped in autophagosomes after the release event, lipids and RNA from autophagosomes to autophagic lysosomes.

25. The PBPK model constructed by the construction method of any one of claims 22-24.

26. Use of the PBPK model of claim 25 in drug pharmacokinetic evaluation.

Images