Radionuclide ligand, screening method thereof and application of radionuclide ligand in preparation of radiodiagnosis drugs

Abstract

The invention discloses a radionuclide ligand, a screening method of the radionuclide ligand and application of the radionuclide ligand in preparation of radiodiagnosis drugs, and belongs to the field of biomedical detection materials. The radionuclide ligand disclosed by the invention has a structure as shown in formulas 1-400. The invention further discloses a screening method of the radionuclide ligand and application of the radionuclide ligand in preparation of radiodiagnosis drugs. The radionuclide ligand disclosed by the invention can be better bonded with radionuclide, is good in stability, rich in type and large in quantity, contributes to promoting the stability improvement and application range expansion of a PET imaging agent, provides a powerful technical support for the development of novel radiodiagnosis drugs, and has important scientific significance and clinical application prospects.

Description

Technical Field

[0001] This invention belongs to the field of biomedical detection materials, specifically relating to radionuclide ligands and their screening methods and their application in the preparation of radiodiagnostic drugs. Background Technology

[0002] Radiopharmaceutical chemistry, an important branch of radiochemistry, focuses primarily on radioisotope preparations and labeling agents used for clinical diagnosis or treatment, and is also one of the foundations of nuclear medicine. Radiopharmaceuticals are divided into diagnostic drugs and therapeutic drugs. Diagnostic radiopharmaceuticals can be further divided into positron-emitting radiopharmaceuticals and single-photon radiopharmaceuticals, corresponding to positron emission tomography (PET) and single-photon emission computed tomography (SPECT) techniques, respectively. In PET imaging, radiopharmaceuticals containing positron-emitting nuclides decay and release positrons. These positrons quickly (approximately 10⁻⁶ s) annihilate with ordinary electrons in the environment, converting into two gamma photons (energy 511 keV) emitted at 180° angles and detected by the PET detector. Clearly, PET images the distribution concentration (i.e., radiation intensity) of radiopharmaceuticals in various tissues and organs within the body; therefore, the design and control of the structure and properties of these drugs directly affect the quality of PET imaging and the key distribution areas.

[0003] In practical applications, most radioactive metal nuclides exist in cation form, therefore labeling methods often involve coordination (especially chelation) with suitable organic ligands. These organic ligands often have a dual function: on the one hand, they need to stably chelate metal ions, and on the other hand, they need to be covalently linked to target molecules (such as monoclonal antibodies). Organic ligands can be synthesized in advance, have no time requirements, and can be transported over long distances. They can be bound to radioactive metal ions through a rapid coordination reaction before use, often without the need for separation and purification. However, for most non-metallic light elements, labeling methods are very time-sensitive, often involving multiple steps such as organic synthesis and column chromatography separation, and are not suitable for long-distance transportation. In addition, some radioactive metal nuclides with relatively short half-lives, such as... 43 Sc (t1 / 2 = 3.891 h), 68 Ga (t1 / 2 = 67.71 min), 82m Rb (t1 / 2 = 6.472 h), etc., can also be used for PET imaging and related medical and physiological research by binding with appropriate ligands.

[0004] Therefore, designing and developing more ligands for positron-emitting radioactive metals and synthesizing labeled compounds is of significant scientific importance; finding the mechanisms to improve the stability of PET imaging agents will further expand the application scenarios (spatiotemporal) of PET technology, and broaden and deepen people's understanding of positron-emitting radionuclides, especially novel radionuclides (such as...). 43 Sc、 68 Ga、 89 Understanding the coordination chemistry of Zr (e.g., Zr) is currently one of the research hotspots in this field. Summary of the Invention

[0005] The problem to be solved by this invention is to provide radionuclide ligands and their screening methods and their applications in the preparation of radiodiagnostic drugs, so as to expand the types of ligands for radiodiagnostic drugs and improve the stability of radiodiagnostic drugs.

[0006] The first aspect of this invention provides radionuclide ligands, the specific structures of which are shown in formulas 1 to 400 below:

[0007]

[0008] .

[0009] A second aspect of the present invention provides a method for screening radionuclide ligands, characterized by comprising the following steps: (1) Extract coordination compound data containing metal ions from the database, and classify and obtain information on their coordination bond length, coordination atom type, ligand functional group and overall molecular structure. (2) Using external data and density functional theory, a machine learning model that correlates coordination bond length and coordination bond energy data is trained. (3) Using machine learning models, the coordination bond length is correlated with the coordination bond energy obtained through theoretical calculation, and a quantitative structure-activity relationship model between the ligand molecular structure and the thermodynamic stability of the labeled compound is established. (4) Generate candidate ligand molecular structures based on the established structure-activity relationship model, and use the structure-activity relationship model to predict their coordination stability with the target radionuclide, and screen out a predetermined number of stable candidate ligands. (5) Stable candidate ligands and their labeled compounds with target radionuclides are prepared by a high-throughput synthesis and characterization platform, and their stability is experimentally determined to obtain radionuclide ligands.

[0010] Furthermore, in step (1), the database is the Cambridge Structure Database.

[0011] Furthermore, the metal ions in step (1) include Sc, Cu, Ga, Sr, Zr, Y, lanthanides, In, Ca, Ba and Hf ions.

[0012] Furthermore, the training in step (2) includes the following steps: using a graph neural network, the coordinate atoms and bonds are represented as nodes and edges in the graph; using stochastic gradient descent to minimize the calculation of coordinate bond energy and the prediction of the graph neural network model to achieve training; monitoring the prediction error of the model for the validation set samples during training, and stopping training when the prediction error no longer decreases.

[0013] Furthermore, in step (5), the high-throughput synthesis is carried out on an automated synthesis platform. The automated synthesis platform uses a Bayesian optimization algorithm and a Gaussian process as a surrogate model to automatically iteratively optimize the conditions for ligand synthesis or labeling reaction.

[0014] A third aspect of the invention provides the use of radionuclide ligands in the preparation of radiodiagnostic drugs.

[0015] A fourth aspect of the present invention provides a radiodiagnostic drug comprising the aforementioned radionuclide ligand and radionuclide.

[0016] Furthermore, radioactive nuclides are 43 Sc、 64 Cu、 68 Ga、 83 Sr or 89 Zr.

[0017] The present invention has the following beneficial effects: The radionuclide ligands of this invention exhibit good bonding with radionuclides and demonstrate excellent stability. By integrating coordination compound data from the Cambridge Structure Database and machine learning modeling, a quantitative structure-activity relationship (QSAR) between the ligand molecular structure and the thermodynamic stability of the labeled compound was established. This enables efficient and rational design of radionuclide ligands. This screening method significantly improves the efficiency and accuracy of ligand screening, overcoming the limitations of traditional trial-and-error methods that rely on experience. It can rapidly identify ligand structures with high stability to the target radionuclide. A high-throughput synthesis and validation platform accelerates the experimental preparation and performance evaluation of candidate ligands, significantly shortening the R&D cycle and reducing development costs. The diverse and abundant types of radionuclide ligands presented in this invention contribute to improving the stability and expanding the application range of PET imaging agents, providing strong technical support for the development of novel radiodiagnostic drugs, and possessing significant scientific value and clinical application prospects. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall technical roadmap for the screening method of radionuclide ligands of the present invention. Figure 2 A flowchart of a homogeneous network architecture; Figure 3 A 64-well reaction tray and control software interface for an automated high-throughput synthesis experimental reaction platform; Figure 4 A schematic diagram of the algorithm for determining the maximum value of the objective function for Bayesian optimization is shown below; (a) is the first batch of iterations; (b) is the visualization result of the expected improvement of the harvest function in the first batch of iterations; (c) is the second batch of iterations; (d) is the visualization result of the expected improvement of the harvest function in the second batch of iterations; (e) is the third batch of iterations; and (f) is the visualization result of the expected improvement of the harvest function in the third batch of iterations. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention; that is, the described embodiments are only a part of the embodiments of this invention, and not all of them.

[0020] Therefore, the following detailed description of the embodiments of the present invention is not intended to limit the scope of the claimed invention, but only to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0021] The features and performance of the present invention will be further described in detail below with reference to embodiments.

[0022] Example The overall technical route of the screening method for radionuclide ligands in this invention is as follows: Figure 1 As shown.

[0023] 1. Data Collection and Preprocessing 1.1 Data Source Using the Cambridge Structure Database (CSD, version 5.42+, data updated to March 2022), data was retrieved and downloaded via its provided Python API to generate a local structured sub-database. All coordination compound structure data for target metallic elements (Sc, Cu, Ga, Sr, Zr) and their similar elements (Y, lanthanides, In, Ca, Ba, Hf) were retrieved, with a total number of coordination compounds less than 10. 5 The order of magnitude of the coordination compounds of the relevant metals is shown in Table 1.

[0024] Table 1. Statistical table of coordination compounds of relevant metals in the structural database.

[0025] 1.2 Data Extraction and Structuring The coordination compounds collected in 1.1 were separated and extracted according to the following information: Information on the type of coordinating atom and its coordination bond length with the metal ion, ligand functional groups, and overall molecular structure.

[0026] During the data extraction phase, common coordinating atoms in (Sc, Cu, Ga, Sr, Zr) complexes, including N, O, C, P, S, and B, were primarily selected. Considering the subjective and diverse nature of the "functional group" concept, no specific structural constraints were imposed on the functional groups. Subsequently, AI learning combined with local atomic clustering was used to autonomously determine the functional group structures. However, a statistical analysis of common functional groups and molecular structures was conducted, including Hydroxyl, Phosphate, Sulfhydryl, Carbonyl, Carboxyl, Hydroxamic acid, and macrocyclic structures. This statistical analysis will aid in the interpretation of the AI's learning results.

[0027] We used a self-written Python script (which is publicly available at https: / / github.com / liuchonggroup / cation) to clean, deduplicatize, and format the data, and then built a local structured sub-database.

[0028] 2. Modeling of Coordination Bond Energy and Stability 2.1 Bond Energy Data Generation Bond lengths and bond energies of known external coordination compounds were calculated using density functional theory (DFT) to construct a training set. During training set construction, a graph neural network (GNN) was used to model the bond energy, representing the atoms and bonds of the complex as nodes and edges in a graph. Stochastic gradient descent was used to minimize the calculated coordination bond energies and the GNN model predictions to achieve training. The prediction error of the model for the validation set samples was monitored during training; training was stopped when the prediction error no longer decreased.

[0029] The Transformer Conv. algorithm (which is publicly available at https: / / github.com / liuchonggroup / BL2E / tree / main) maps bond lengths to coordination bond energies, and is used to predict the bond energies of compounds in this project's database.

[0030] 2.2 Structure-Property Relationship Modeling Different model structures were explored and hyperparameters were optimized. The dataset (molecular structure information and thermodynamic stability, i.e., bond energy information) was divided into training and testing sets. Cross-validation was used to determine the correlation coefficient of the final model, and the optimal model architecture was selected for determining the structure-activity relationship. Furthermore, since positron-emitting radioactive metal complexes are mostly chelate coordinations, this model will consider the significant enhancement effect of multidentate coordination on coordination bond energy.

[0031] Specifically, the input includes features such as coordinating atoms, functional groups, bond lengths, and bond energies. A graph isomorphism network (GIN) is used to establish structure-property relationships. Support vector machine (SVM) and random forest (RF) are used as baseline models for comparison to highlight the advantages of GIN in processing graph-structured data.

[0032] GIN comprises nine graph isomorphic layers. Local graph clustering is implemented in layers 4, 7, and 9 to integrate local information and derive the integrated complex vector representation. The vector representations from layers 4, 7, and 9 are then concatenated and fed into a fully connected neural network predictor to predict the coordination bond energy of the complex. The graph isomorphic network architecture flowchart is shown below. Figure 2 As shown.

[0033] SVM and RF modeling methods differ from GIN. SVM and RF modeling do not involve compiling graph representations; instead, they utilize more traditional molecular fingerprints (MFP) and molecular descriptors (MDp). Molecular descriptors and molecular fingerprints are extracted using Rdkit and Babel methods, simultaneously extracting information such as the atomic number of the central metal element, electron configuration, electronegativity, and formal charges. This information is then assembled into feature vectors and input into the SVM or RF model for training. Model training and evaluation utilize scikit-learn methods.

[0034] 3. Ligand structure generation and evaluation The trained GIN model is used to generate candidate ligand molecular structures, and the generated ligand structures are predicted to correlate with the target metal ion (e.g., ...). 68 Ga 3+ The stability of the complexes formed was assessed. Natural Language Processing (NLP) tools (BERT) were used to mine ligand stability data reported in the literature, and the results were cross-validated with the model predictions.

[0035] 4 High-throughput synthesis and validation 4.1 Automated Synthesis Platform Use automated high-throughput synthesis platforms such as Figure 3 As shown, the sample tray (made of PTFE) that is matched with the automatic sample feeding mechanism of the platform has 8 × 8 = 64 holes. Each hole has a groove at the bottom and can fix reaction bottles with a capacity of 4 mL, 8 mL and 20 mL. There are 5 trays in total, so 320 sets of experiments can be carried out in each batch. Figure 3The interface of the host computer control software specifically designed for this platform is shown, allowing for the systematic setting and adjustment of experimental conditions for the same batch using both automatic optimization (left) and exhaustive methods (right). The platform currently has four sample loading channels and three additional channels that can be expanded, all of which can withstand high-ionic-strength aqueous solutions, volatile organic solvents, strong acids, and strong bases. It is particularly noteworthy that this high-throughput automated platform is especially suitable for systematic reaction studies, minimizing human error in both large-scale repeated experiments and fine-tuning of conditions. Furthermore, this automated platform is also suitable for the reactivity studies of radionuclides, as it can minimize human contact with raw materials and reactors.

[0036] 4.2 Bayesian optimization of reaction conditions Bayesian optimization was performed using a Gaussian process (GP) surrogate model, with the reaction yield as the objective function.

[0037] In each iteration, 3–5 sample points are selected for experiments to update the surrogate model and gradually approach the optimal reaction conditions.

[0038] Figure 4 This diagram illustrates the execution of the algorithm for finding the global optimum of a one-dimensional non-convex objective function for a reaction system within the framework of this project. Each row represents a batch iteration of the model. The left column includes the objective function (solid black line), the Gaussian surrogate model (dashed blue line, yellow filled area), the current sample points (blue pentagrams), and the next sample points (green pentagrams). The right column visualizes the expected improvement of the harvest function. It is particularly noteworthy that for the expected value of the objective function predicted by GP (dashed blue line), the yellow filled area nearby represents the region where the objective function has a 95% probability of appearing; therefore, a larger vertical span indicates greater variance and uncertainty. It is also important to note that the globally optimal algorithm is an open-source package developed at https: / / github.com / Zhang-Zhiyuan-zzy / hotpot.git, which is publicly available on GitHub.

[0039] 4.3 Coordination reaction and stability test The synthesized ligands underwent coordination reactions with target metal ions, and the structures of the complexes were characterized using UV-Vis spectroscopy and mass spectrometry. The thermodynamic stability and decomposition kinetics of the complexes under physiological conditions were determined to validate the AI ​​prediction results.

[0040] 5 Results and Analysis The above method was used to screen and synthesize 400 radionuclide ligand structures, and the radionuclide ligand pairs were specific to certain radionuclides ( 43 Sc、 64 Cu、 68 Ga、 83Sr and 89 The binding energy of Zr) E The values ​​are shown in Table 2 below.

[0041] Table 2 Binding energies of radionuclide ligands to specific radionuclides

[0042] As shown in Table 2, the radionuclide ligands synthesized by the above method have good binding ability with radionuclides, and the resulting radiodiagnostic drugs have strong stability.

[0043] The present invention has been described according to the above embodiments. It should be understood that the above embodiments do not limit the present invention in any way. All technical solutions obtained by equivalent substitution or equivalent transformation fall within the scope of the present invention.

Claims

1. A radionuclide ligand, the specific structure of which is shown in formulas 1 to 400 below: 。 2. The method of screening for a radionuclide ligand according to claim 1, characterized in that, Includes the following steps: (1) Extract coordination compound data containing metal ions from the database, and classify and obtain information on their coordination bond length, coordination atom type, ligand functional group and overall molecular structure. (2) Using external data and density functional theory, a machine learning model that correlates coordination bond length and coordination bond energy data is trained. (3) Using a machine learning model, the coordination bond length is correlated with the coordination bond energy obtained through theoretical calculation, and a quantitative structure-activity relationship model between the ligand molecular structure and the thermodynamic stability of the labeled compound is established. (4) Generate candidate ligand molecular structures based on the established structure-activity relationship model, and use the structure-activity relationship model to predict the stability of their coordination with the target radionuclide, thereby screening out a predetermined number of stable candidate ligands. (5) The stable candidate ligands and their labeled compounds with the target radionuclide are prepared by a high-throughput synthesis and characterization platform, and their stability is experimentally determined to obtain the radionuclide ligands.

3. The method of screening for a radionuclide ligand according to claim 2, wherein, The database used in step (1) is the Cambridge Structure Database.

4. The method of screening for a radionuclide ligand according to claim 2, wherein, The metal ions in step (1) include Sc, Cu, Ga, Sr, Zr, Y, lanthanides, In, Ca, Ba and Hf ions.

5. The method of screening for a radionuclide ligand according to claim 2, wherein, The training in step (2) includes the following steps: using a graph neural network, with coordinate atoms and bonds represented as nodes and edges in the graph; using stochastic gradient descent to minimize the calculation of coordinate bond energy and the prediction of the graph neural network model to achieve training; monitoring the prediction error of the model for the validation set samples during training, and stopping training when the prediction error no longer decreases.

6. The method of screening for a radionuclide ligand according to claim 2, wherein, In step (5), high-throughput synthesis is carried out on an automated synthesis platform. The automated synthesis platform uses a Bayesian optimization algorithm and a Gaussian process as a surrogate model to automatically iteratively optimize the conditions for ligand synthesis or labeling reaction.

7. The use of the radionuclide ligand according to claim 1 in the preparation of radiodiagnostic drugs.

8. A radiodiagnostic drug, characterized by, Includes the radionuclide ligand and radionuclide as described in claim 1.

9. Use according to claim 8, wherein the compound is ###0002### The radionuclide is 43 Sc, 64 Cu, 68 Ga, 83 Sr or 89 Zr.

Images