Nanocrystal development system based on AI and API-auxiliary material assembling and pairing, working method and application

Abstract

The invention discloses a nanocrystal development system based on AI and API-auxiliary material assembling and pairing and a working method and application thereof, and belongs to the technical field of computer-aided preparation preparation. A structure-effect relationship is used as a common feature in a nanocrystal research and development process, and an attention-based graph neural network and transfer learning are adopted, so that PANDA can coordinately process a multi-target task, effective auxiliary materials can be quickly identified to stabilize active pharmaceutical ingredients (APIs), and the research and development efficiency of the nanocrystal is improved. The production of the nanocrystal with the optimal treatment effect and the minimum side effect is facilitated. In addition, PANDA shows huge potential in developing high-efficiency and low-toxicity colorectal cancer targeted nanocrystals and balancing API-auxiliary materials from multiple aspects of design, manufacturing, evaluation and the like. In a word, the appearance of the PANDA thoroughly changes the research and development process of the whole nanocrystal, and the PANDA has a wide prospect in the aspect of clinical transformation.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of AI and biomedicine, specifically relating to an AI-supported nanocrystal development platform (PANDA), which aims to improve the development efficiency of nanocrystals and achieve the required functions. PANDA shows great potential in developing highly efficient and low-toxicity colorectal cancer-targeting nanocrystals, balancing API-excipient design, manufacturing, and evaluation. Background Technology

[0002] Poor water solubility of active pharmaceutical ingredients (APIs) is a significant challenge in drug development. Nanocrystals, due to their superior solubility and carrier-free advantages, have been approved by the FDA as an important strategy to address the limitations of hydrophobic APIs in bioavailability, particularly in solving the problem of poor water solubility in oral formulations. Nanocrystals offer advantages such as convenient preparation and excellent synergistic therapeutic effects; however, the selection of APIs and excipients is crucial in their preparation. The entire nanocrystal development process involves multiple stages, including API-excipient design, manufacturing, preclinical testing, and clinical research. The entire process is fraught with unpredictable challenges in screening and optimization, leading to a high dependence on the expertise of developers and manufacturers, resulting in inefficiency. To overcome these difficulties, scientists have been actively exploring AI-driven methods for nanocrystal development, aiming to improve the accuracy and efficiency of nanocrystal development.

[0003] Given AI's strength in capturing structure-function relationships, it holds immense potential for creating specific data-driven modules and guiding nanocrystal development. Beyond single-objective tasks, AI can leverage schedule optimization paradigms to design rational platforms that manage the nanocrystal development process. Such platforms can balance multiple objectives with both complex commonalities and unique characteristics; however, this potential remains largely untapped. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides an AI-supported nanocrystal development platform (PANDA), which combines AI, API, and auxiliary material assembly and pairing experimental big data. It utilizes a large amount of real experimental data to build, train, and optimize AI models, and rapidly expands the scale of auxiliary material screening through AI models. This greatly improves experimental efficiency while reducing screening costs. At the same time, AI is used to optimize the preparation conditions of nanocrystals, accelerating the mass production and industrialization of nanocrystals.

[0005] The technical solution of the present invention:

[0006] In a first aspect, the present invention provides a nanocrystal development system based on an AI and API-auxiliary material assembly pairing database, comprising:

[0007] The data acquisition module is used to obtain structure-effect relationship analysis data of APIs and excipients from the API-excipient pairing database, experimental parameters of API-excipient pairing co-assembly of nanocrystals by transient nanocomposite method (FNC) and evaluation data of nanocrystals after co-assembly treatment;

[0008] The data processing module is used to process the structure-effect relationship analysis data of APIs and excipients, the experimental parameters of API-excipient pairing FNC co-assembly, and the evaluation data of nanocrystals after co-assembly treatment.

[0009] The model training module is used to train and validate the initial model based on the structure-effect relationship analysis data of APIs and excipients, the FNC experimental parameters of API-excipient pairing co-assembly, and the evaluation data of nanocrystals after co-assembly treatment, so as to obtain the prediction model.

[0010] The model execution module is used to input API structure information and excipient structure information into the prediction model, predict the assembly pairing strength and FNC preparation conditions, and output the results.

[0011] Secondly, the present invention provides a method for operating the system as follows:

[0012] Data were obtained from the API-excipient pairing database, including analysis of the structure-effect interaction between APIs and excipients, preparation conditions and parameters for API-excipient pairing co-assembly, and evaluation experiments on nanocrystals after co-assembly treatment.

[0013] The experimental data on the pairing and assembly of the APIs and excipients were processed;

[0014] The initial model was trained and validated based on data from the structure-effect interaction analysis of the APIs and excipients, the preparation conditions parameters for API-excipient pairing and co-assembly, and the evaluation experiments of nanocrystals after co-assembly treatment, in order to obtain a predictive model.

[0015] The structural information of the excipients and APIs to be screened is input into the prediction model to perform nano-assembly prediction, thereby completing the screening of API-excipient pairs.

[0016] Optionally, the step of processing the assembly experimental data of the APIs and excipients includes:

[0017] The experimental parameters of the structure and assembly of APIs and excipients are converted into feature data that can be recognized by the initial model to obtain the feature matrix;

[0018] The assembly experimental data of APIs and excipients were combined and separated.

[0019] Optionally, the steps of training and validating the initial model to obtain the prediction model based on data such as the structure-effect interaction analysis of the APIs and excipients, experimental parameters of API-excipient pairing co-assembly, and evaluation experiments of nanocrystals after co-assembly treatment include:

[0020] Training datasets were obtained from the structure-effect interaction analysis of the APIs and excipients, experimental parameters of API-excipient pairing co-assembly, and evaluation experiments of nanocrystals after co-assembly treatment to train and validate the initial model.

[0021] Test datasets were obtained from the structure-effect interaction analysis of the APIs and excipients, the preparation conditions and parameters of API-excipient pairing co-assembly, and the evaluation experiments of nanocrystals after co-assembly treatment. The trained model was then used to make predictions and evaluations to obtain a predictive model.

[0022] Optionally, the step of obtaining a test dataset from data such as the structure-effect interaction analysis of the APIs and excipients, the preparation condition parameters for API-excipient pairing co-assembly, and the evaluation experiments of nanocrystals after co-assembly treatment, and then using this dataset to predict and evaluate the trained model to obtain a predictive model, includes:

[0023] Test datasets were obtained from the structure-effect interaction analysis of the APIs and excipients, the preparation conditions and parameters of API-excipient pairing co-assembly, and the evaluation experiments of nanocrystals after co-assembly treatment to predict the trained model.

[0024] The prediction results are evaluated based on preset evaluation indicators.

[0025] Optionally, the preset evaluation indicators are as follows:

[0026] To further evaluate the stabilizing effect of excipients in each API-excipient combination on API, the size-based pre-defined evaluation index is considered reasonable when the particle size of the API-excipient nanocrystals is less than three-quarters of that of the free API assembled formulation. The predicted parameters are considered acceptable when the nanocrystal particle size is less than 250 nm and the PDI is less than 0.3 under the predicted preparation conditions.

[0027] Optionally, before the step of training and validating the initial model based on data such as the structure-effect interaction analysis between the APIs and excipients, the preparation condition parameters for API-excipient co-assembly, and the evaluation experiments of the co-assembled nanocrystals, to obtain the predictive model, the following steps are included:

[0028] The algorithm of the initial model is modified based on the physicochemical properties and structural information of APIs and excipients;

[0029] Local and global attention mechanisms were introduced to construct a message-passing network, which learned the structural features of functional groups and the overall molecular structure, respectively. To further enhance the predictive power of this module, API and excipient spectral features were combined with the attention mechanism to capture the molecular interactions between them.

[0030] Thirdly, this application provides a computer program that can run on a processor, the processor executing the computer program to implement the steps of the nanocrystal development platform based on the AI ​​and API-auxiliary assembly pairing database support described above.

[0031] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the nanocrystal development platform based on the AI ​​and API-auxiliary assembly pairing database supported as described above.

[0032] Finally, this invention provides an application of PANDA, in which PANDA-designed nanocrystals significantly improve the therapeutic efficacy and biocompatibility of targeted APIs for colorectal cancer.

[0033] First, 7694 possible pairings of 2436 APIs and 165 water-soluble excipients were identified from the FDA-approved list. These were then screened according to PANDA and formed into nanocrystals using flash nanocomposite technology (FNC). The nanocrystals produced by FNC were demonstrated to have smaller size, more uniform distribution, and consistent morphology compared to pure API aggregates and nanocrystals produced through batch mixing. Notably, FNC technology provides additional benefits in terms of reproducibility and stable manufacturing without batch-to-batch variability.

[0034] GSK3326595 (GSK) is a small molecule inhibitor of arginine methyltransferase 5 (PRMT5). The main obstacle to its clinical translation is its poor water solubility after oral administration. Therefore, using the PANDA platform, a library of FDA-approved excipients was screened, and it was found that the food additive Alizarin Red S (ARS) can effectively stabilize GSK3326595, forming uniform and stable nanocrystals (GANC), thereby maximizing therapeutic efficacy and minimizing systemic toxicity.

[0035] Mice carrying CT26 tumors were administered oral nanocrystals selected through PANDA screening, and the growth of colorectal cancer was monitored. Compared with GSK or batch-mixed GANCs, FNC-based GANC administration resulted in significant tumor inhibition in mice.

[0036] Free GSK significantly inhibited PRMT5 phosphorylation in the liver and tumors within 4 hours, but both reverted to normal levels after 24 hours. In contrast, GANC induced sustained and minimal PRMT5 inhibition in both tumors and the liver after 24 hours.

[0037] A solution mixing apparatus includes an injection pump or peristaltic pump, a multi-inlet vortex mixer, and a product collection device, wherein the multi-inlet mixer is configured to perform the steps of the working method described in any of the preceding claims.

[0038] This invention takes the structure-effect relationship as a common feature in the nanocrystal development process. Employing attention-based graph neural networks and multi-task learning, PANDA can coordinate multiple target tasks and rapidly identify effective excipients to stabilize active pharmaceutical ingredients (APIs), facilitating the production of nanocrystals with optimal therapeutic efficacy and minimal side effects. Furthermore, PANDA demonstrates significant potential in developing highly effective and low-toxicity colorectal cancer-targeting nanocrystals, balancing API-excipient relationships across design, manufacturing, and evaluation. In conclusion, the emergence of PANDA has revolutionized the entire nanocrystal development process and holds broad promise for clinical translation.

[0039] The nanocrystal development platform based on AI and API-excipient assembly pairing database described in this invention offers the following significant advantages when used to improve the therapeutic efficacy and biocompatibility of targeted APIs:

[0040] (1) High sensitivity and specificity: PANDA achieved 91.8% sensitivity and 93.2% specificity on the validation set;

[0041] (2) High accuracy: The accuracy of the PANDA prediction module is above 0.8;

[0042] (3) Excellent nano properties: Nanocrystals produced by FNC under the guidance of PANDA have smaller size and more uniform morphology compared with pure API aggregates and nanocrystals produced by batch mixing;

[0043] (4) Good repeatability: PANDA-guided FNC technology offers the additional advantage of repeatability and stability of the manufacturing process with no batch-to-batch differences.

[0044] (5) Good therapeutic effect: Nanocrystals produced by FNC under the guidance of PANDA have longer drug release time and better tumor treatment effect. Attached Figure Description

[0045] Figure 1 It includes PANDA's workflow diagram and thermal visualizations of the 30 highest-scoring APIs and excipients.

[0046] Figure 2 This is a graph showing the predicted and experimental results of whether GSK and 10 selected excipients can self-assemble into nanocrystals. AR: Allura Red, TT: Tartrazine Yellow, ARS: Alizarin Red S, CM: Carmine, MEG: Methylglucosamine, EGCG: (-)-Epigallocatechin Gallate, CAP: Capecitabine, RY: Saffron Yellow, CGA: Chlorogenic Acid, MBB: 4-(Methoxycarbonyl)phenylboronic Acid;

[0047] Figure 3 This is a drug release curve of GANC in simulated body fluids;

[0048] Figure 4 This is a tumor growth curve of T26 tumor-bearing mice;

[0049] Figure 5 These are images showing the body weights of tumor-bearing mice in each group;

[0050] Figure 6 This is a quantitative analysis of the biodistribution of GANC in CT26 tumor-bearing mice. Detailed Implementation

[0051] The technical solution of the present invention will be further described below with reference to specific embodiments.

[0052] Example 1: This invention provides a nanocrystal development system based on an AI and API-auxiliary material assembly pairing database, comprising:

[0053] Data were obtained from the API-excipient assembly pairing database, including analysis of the structure-effect interaction between APIs and excipients, experimental parameters of API-excipient pairing co-assembly, and evaluation experiments of nanocrystals after co-assembly treatment.

[0054] The experimental data on the pairing and assembly of the APIs and excipients were processed;

[0055] The initial model was trained and validated based on data obtained from the API-excipient assembly pairing database, including analysis of the structure-effect interaction between APIs and excipients, experimental parameters of API-excipient pairing co-assembly, and evaluation experiments of nanocrystals after co-assembly treatment, in order to obtain a predictive model.

[0056] The structural information of the excipients and APIs to be screened is input into the prediction model to perform nano-assembly prediction, thereby completing the screening of API-excipient pairs.

[0057] In a specific example of the present invention, 2,100 combinations containing 50 active ingredients and 42 excipients were first screened from FDA-approved drugs through experiments. An API-excipient assembly pairing database containing 129,318 drug-drug interactions was constructed using the DrugBank dataset.

[0058] Then, data such as the structure-effect interaction between APIs and excipients, physicochemical properties such as solubility and stability, preparation conditions for API-excipient co-assembly, and evaluation experiments of nanocrystals after co-assembly treatment were obtained from the pre-constructed API-excipient assembly pairing database.

[0059] Furthermore, the experimental data on the assembly of APIs and excipients were processed.

[0060] To further evaluate the stabilizing effect of excipients on corresponding APIs, a size considered reasonable was defined as the size of the API-excipient nanocrystals being less than three-quarters the size of the free APIs assembled formulation. A screening of 2082 measured API-excipient combinations revealed that 734 (35.3%) showed the expected size reduction. Furthermore, the dataset's balanced class composition contributes to improved prediction accuracy. Therefore, PANDA can eliminate the need for tedious high-throughput screening, rapidly identifying and inferring more API-excipient pairs in these assembled formulations, and representing their features using molecular graphs. Simultaneously, local and global attention mechanisms were introduced, constructing a message-passing network that learned functional group and overall molecular structure features, respectively.

[0061] Furthermore, data such as the structure-effect interaction analysis of APIs and excipients, experimental parameters of API-excipient pairing co-assembly, and evaluation experiments of co-assembled nanocrystals were input into the initial model for training and validation. At the same time, the algorithm hyperparameters were adjusted according to the model performance. After the initial model training was completed, the final model performance was obtained using test set data.

[0062] To further improve PANDA's predictive power, the spectral features of APIs and excipients were combined with an attention mechanism to capture their molecular interactions. In a retrospective evaluation based on 10-fold cross-validation, PANDA demonstrated superior performance compared to other machine learning modules such as random forests and k-nearest neighbors. The average area under the ROC curve reached 0.884, indicating that PANDA accurately captured structure-assembly interactions. After further parameter optimization, PANDA ultimately achieved a sensitivity of 91.8% and a specificity of 93.25% on the validation set.

[0063] Furthermore, after obtaining the final model, the structures of the APIs and excipients to be screened are input into the model to predict the assembly probability of the APIs and excipients, thereby completing the screening of APIs-excipient pairs.

[0064] Given its outstanding performance, PANDA was applied to another experimental test set containing 627 API-excipient combinations recorded under different nanoprecipitation conditions, as well as a comprehensive test set containing 121 API-excipient combinations from the literature. Unsurprisingly, PANDA also demonstrated a high degree of sensitivity and specificity in identifying nanocrystals, which means that PANDA can identify effective API-excipient combinations that can form stable nanocrystals under various conditions.

[0065] Example 2

[0066] To verify the potential of PANDA in guiding the development of highly efficient and low-toxicity nanocrystals, excipient pairing prediction was performed for colorectal cancer-targeting APIs, and the selected API-excipient co-assembled nanomedicines were validated for therapeutic efficacy and biocompatibility in a colorectal cancer model. The specific steps are as follows.

[0067] GSK is currently the most advanced molecular inhibitor in clinical trials targeting high PRMT5 expression. However, adverse reactions occurred in 89% of patients after use, which seriously hinders its further translation into clinical applications. The adverse reactions of GSK3326595 mainly stem from poor water solubility and bioavailability after oral administration. Therefore, using the PANDA platform, an effective excipient was identified, and a highly efficient and low-toxicity GSK nanocrystal was developed.

[0068] A screening of the FDA-approved excipient library revealed that the food additive ARS can effectively stabilize GSK3326595, forming uniform and stable nanocrystals. Therefore, GSK / ARS nanocrystals (GANC) were successfully prepared using PANDA-guided FNC technology. Figure 3 This is a drug release curve of GANC in simulated body fluids. As can be seen from the figure, compared with GSK aggregates, GANC has a smaller size, more uniform distribution, and a longer drug release time.

[0069] The specific preparation method of GSK / ARS nanocrystals (GANC) is as follows: GSK3326595 and Alizarin Red S are dissolved in dimethyl sulfoxide (9 mM) and water (1 mM), respectively. They are then introduced into a multi-inlet vortex mixer through different channels using a micro-injection pump (the flow rates of the dimethyl sulfoxide and aqueous phases are 29 mL / min and 30 mL / min, respectively). The mixed product is collected and purified to remove organic solvents and unassembled free molecules. After lyophilization, it is stored.

[0070] Mice carrying CT26 tumors were administered GANC orally, and the therapeutic efficacy and biosafety of GANC in colorectal cancer were monitored. Compared with GSK or batch-mixed GANC, administration of GANC based on FNC significantly inhibited tumor growth in mice (e.g., Figure 4 As shown). In daily treatment with free GSK, a significant decrease in mouse body weight was observed; however, in daily treatment with GANC, the decrease in body weight was not significant (as shown). Figure 5 (As shown).

[0071] Next, the phosphorylation status of NLRC5, a downstream pathway of PRMT5, was monitored as a marker of GSK activity and toxicity. Free GSK significantly inhibited PRMT5 phosphorylation in both the liver and tumors within 4 hours, but both reverted after 24 hours. In contrast, GANCs continued to inhibit PRMT5 phosphorylation in both tumors and the liver after 24 hours, suggesting that GANCs may reduce the side effects of PRMT5 inhibitors. Consistent with specific PRMT5 pathway regulatory mechanisms, the distribution trends of GANCs in tumors and organs showed a different pattern than those of free GSKs, with a difference of up to 4.57-fold between tumors and the liver. These results indicate that nanocrystalline GANCs co-assembled with ARS can improve therapeutic efficacy and reduce side effects by altering the biodistribution of GSK3326595.

[0072] In summary, an AI-powered platform has been developed that simplifies nanocrystal development, offering advantages such as convenience, time-saving, and cost-effectiveness, while improving the accuracy of drug efficacy prediction and facilitating clinical translation. This end-to-end platform underscores the necessity for improved reporting standards in biomedical research and the pharmaceutical industry. By using FDA-approved excipients (dyes, nutrients, and food compounds, etc.), PANDA can accelerate the research process for highly efficient and low-toxicity nanocrystals of various APIs.

[0073] Specific examples are used in this paper to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application.

Claims

1. A nanocrystal development system based on AI and API-auxiliary material assembly pairing database support, characterized in that, The system includes: The data acquisition module is used to obtain structure-effect relationship analysis data of APIs and excipients from the API-excipient pairing database, experimental parameters of API-excipient pairing co-assembly of nanocrystals by transient nanocomposite method (FNC) and evaluation data of nanocrystals after co-assembly treatment; The data processing module is used to process the structure-effect relationship analysis data of APIs and excipients, the experimental parameters of API-excipient pairing FNC co-assembly, and the evaluation data of nanocrystals after co-assembly treatment. The model training module is used to train and validate the initial model based on the structure-effect relationship analysis data of APIs and excipients, the FNC experimental parameters of API-excipient pairing co-assembly, and the evaluation data of nanocrystals after co-assembly treatment, so as to obtain the prediction model. The model execution module is used to input API structure information and excipient structure information into the prediction model, predict the assembly pairing strength and FNC preparation conditions, and output the results.

2. The working method of the development system according to claim 1, characterized in that, Includes the following steps: S1. Obtain experimental parameters for API-excipient pairing and assembly, and structure-effect relationship analysis data of APIs and excipients from a variety of combinations of different APIs and excipients screened from the FDA-approved drug database. S2. API-excipients are co-assembled into nanocrystals by FNC using intermolecular non-covalent interactions, and evaluation data of the co-assembled nanocrystals are obtained. The data are then integrated and analyzed to establish a database of API-excipient assembly pairing and preparation conditions. S3. Based on the structure-effect relationship analysis data of APIs and excipients, the preparation condition parameters of API-excipient pairing assembly, and the evaluation data of nanocrystals after co-assembly, the initial model is trained and validated to obtain the prediction model. S4. Input the APIs and the FDA-approved excipients to be screened into the prediction model to perform co-assembly pairing prediction, in order to obtain the excipients paired with the APIs and their suitable preparation conditions.

3. The working method of the development system according to claim 2, characterized in that: In step S1, the preparation conditions parameters for API-excipient pairing assembly include the flow rate of the organic phase, the flow rate of the aqueous phase, the concentration of API, the concentration of excipients, and the ratio of API to excipients.

4. The working method according to claim 2, characterized in that, Test datasets were obtained from the structure-effect interaction analysis of APIs and excipients, the preparation conditions parameters of API-excipient pairing co-assembly, and the evaluation data of nanocrystals after co-assembly treatment to make predictions for the trained model. The prediction results are evaluated based on preset evaluation indicators; The preset evaluation index is that the particle size of API-excipient nanocrystals is less than three-quarters of that of freely assembled API formulations. For the task of optimizing preparation conditions, the preset evaluation index is that the particle size of self-assembled nanocrystals is less than 250 nanometers and the polydispersity index is less than 0.

3.

5. The working method according to claim 2, characterized in that, Local and global attention mechanisms were introduced to construct a message passing network, which learns the structural features of functional groups and the overall molecular structure features, respectively. API and excipient spectral features were combined with the attention mechanism to capture the molecular interactions between them.

6. A nanocrystal, characterized in that, The nanocrystals include GSK3326595 and Alizarin Red S, an excipient screened using the method described in claims 2-5.

7. The method for preparing nanocrystals according to claim 6, characterized in that, Includes the following steps: GSK3326595 and Alizarin Red S were dissolved in dimethyl sulfoxide and water, respectively, and then fed into a multi-inlet vortex mixer through different channels using a micro-injection pump. The mixture was collected and purified to remove organic solvents and unassembled free molecules, and then lyophilized for storage.

8. The application of the nanocrystal described in claim 6 as a small molecule inhibitor of arginine methyltransferase 5.

9. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps of the working method according to any one of claims 2 to 5.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the working method as described in any one of claims 2 to 5.

Images