AXL inhibitor high-throughput virtual screening method and device based on deep learning

Abstract

The invention relates to a deep learning-based AXL inhibitor high-throughput virtual screening method, which is characterized by comprising the following steps: respectively obtaining a ligand and a receptor protein, and performing format conversion on the structures of the ligand and the receptor protein by using the same program; a database is established and used for high-throughput virtual screening, and the database at least comprises a first database and a second database; evaluating and determining a deep learning algorithm and a molecular docking program; performing high-throughput virtual screening according to the determined deep learning algorithm and the molecular docking program, and obtaining quasi target molecules through the high-throughput virtual screening; detecting the in-vitro inhibition activity of the quasi target molecule on AXL to obtain a final target molecule; and performing biological effect evaluation on the final target molecule. The invention also relates to a device.

Description

Technical Field

[0001] The present invention relates to the field of computer-aided drug design, and more specifically to a high-throughput virtual screening method and device for AXL inhibitors based on deep learning. Background Art

[0002] AXL (also known as Ark, Ufo, and Tyro-7) is a member of the TAM receptor tyrosine kinase family, which also includes Mer and Tyro-3. Upon binding to growth arrest-specific protein 6 (Gas6), AXL dimerizes and becomes autophosphorylated, thereby activating downstream mitogen-activated protein kinase (MAPK) and phosphatidylinositol 3-kinase / protein kinase B (PI3K / AKT) pathways.

[0003] Computer-aided drug design (CADD) combined with sophisticated computational modeling techniques has greatly accelerated the drug discovery process. This approach, particularly through virtual screening, has become a key tool for increasing the efficiency of drug discovery and development. Summary of the invention

[0004] To address the above-mentioned shortcomings, according to a first aspect of the present invention, a high-throughput virtual screening method for AXL inhibitors based on deep learning is provided, comprising: obtaining ligands and receptor proteins respectively, and converting the structures of the ligands and the receptor proteins into formats using the same program; establishing a database for high-throughput virtual screening, the database comprising at least a first database and a second database; evaluating and determining a deep learning algorithm and a molecular docking program; performing high-throughput virtual screening according to the determined deep learning algorithm and molecular docking program, and obtaining a quasi-target molecule through the high-throughput virtual screening; detecting the in vitro inhibitory activity of the quasi-target molecule on AXL to obtain a final target molecule; and evaluating the biological effect of the final target molecule.

[0005] Optionally, the method further comprises: simulating and analyzing and verifying the AXL inhibitory effect of the final target molecule by molecular dynamics.

[0006] Optionally, the first database is the Maybridge drug database, and the second database is the MCE drug database.

[0007] Optionally, the evaluated deep learning algorithms include at least: a first deep learning algorithm and a second deep learning algorithm, and the evaluated molecular docking programs include three molecular docking programs: AutoDock-GPU molecular docking program, Autodock Vina molecular docking program, and LeDock molecular docking program.

[0008] Optionally, the determined deep learning algorithms include: PLANET deep learning algorithm and DeepDock deep learning algorithm.

[0009] Optionally, the order of the high-throughput virtual screening is set as: first perform the first step of screening through the PLANET deep learning algorithm, then perform the second step of screening through the DeepDock deep learning algorithm, and finally perform the third step of screening.

[0010] Optionally, the affinity of the molecules screened in the first step by the PLANET deep learning algorithm is greater than or equal to 6.782.

[0011] Optionally, the quasi-target molecules are respectively a first quasi-target molecule, a second quasi-target molecule, a third quasi-target molecule, a fourth quasi-target molecule, a fifth quasi-target molecule, a sixth quasi-target molecule, a seventh quasi-target molecule, an eighth quasi-target molecule, and a ninth quasi-target molecule.

[0012] Optionally, the ninth quasi-target molecule is the final target molecule, and its chemical formula is:

[0013]

[0014] Optionally, the detecting the in vitro inhibitory activity of the quasi-target molecule on AXL comprises: detecting the IC 50 value, and the IC 50 The value is 9.38 nM, and the IC 50 The value is 112.0 nM, and the IC 50 The value is 22.3nM.

[0015] According to a second aspect of the present invention, a high-throughput virtual screening device for AXL inhibitors based on deep learning is provided, comprising: an acquisition conversion module: used to respectively acquire ligands and receptor proteins, the structures of the ligands and the receptor proteins being format converted using the same program; a database establishment module: used for high-throughput virtual screening, the database comprising at least a first database and a second database; an evaluation and determination module: used to evaluate and determine a deep learning algorithm and a molecular docking program; a virtual screening module: used to perform high-throughput virtual screening according to the determined deep learning algorithm and molecular docking program, and obtain a quasi-target molecule through the high-throughput virtual screening; and a detection and evaluation module: used to detect the in vitro inhibitory activity of the quasi-target molecule on AXL to obtain a final target molecule; and perform a biological effect evaluation on the final target molecule.

[0016] Embodiments of the present invention Embodiments of the present invention provide a high-throughput virtual screening method to identify a potent novel inhibitor of AXL. By using advanced algorithms such as PLANET and DeepDock, the final target molecule was determined to be a highly effective AXL inhibitor with significant inhibitory biological activity (IC 50 =9.378 nM). Further exploration by molecular dynamics (MD) simulation revealed the specific binding mode of the final target molecule (compound 9) to AXL. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The following are drawings of exemplary embodiments of the present invention, and the same or similar reference numerals are used in the drawings to represent the same or similar elements. In the drawings:

[0018] Figure 1 A flowchart of a deep learning-based high-throughput virtual screening method for AXL inhibitors according to an exemplary embodiment of the present invention is shown.

[0019] Figure 2A ROC curves for molecular docking of the DUD-E dataset with selected receptor conformations using selected programs according to an exemplary embodiment of the present invention are shown.

[0020] Figure 2B A schematic diagram showing EF1% and NEF1% evaluation of a data set by selected software according to an exemplary embodiment of the present invention.

[0021] Figure 2C T-tests and block diagrams showing selected software validations of exemplary embodiments of the present invention.

[0022] Figure 3 A schematic diagram showing a process for screening AXL inhibitors according to an exemplary embodiment of the present invention.

[0023] Figure 4A A schematic diagram showing the inhibition rate (%) analysis of AXL by 9 selected compounds according to an exemplary embodiment of the present invention (LDC-1267 is a positive control).

[0024] Figure 4B The IC values ​​of selected compounds (3, 5, 6 and 9) of exemplary embodiments of the present invention are shown. 50 Evaluation diagram.

[0025] Figure 5A A graph showing the RMSD value of compound 9 and protein (AXL) at 100 ns according to an exemplary embodiment of the present invention.

[0026] Figure 5B A schematic diagram showing the residue contributions of the compound 9-AXL complex according to an exemplary embodiment of the present invention.

[0027] Figure 5C A schematic diagram showing the binding mode of compound 9 to AXL after MD simulation according to an exemplary embodiment of the present invention.

[0028] Figure 5D A schematic diagram of a two-dimensional plot of compound 9 and AXL is shown according to an exemplary embodiment of the present invention.

[0029] Figure 6 A schematic structural diagram of a deep learning-based high-throughput virtual screening device for AXL inhibitors according to an exemplary embodiment of the present invention is shown. DETAILED DESCRIPTION

[0030] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation modes in conjunction with the accompanying drawings.

[0031] In the present invention, the term "and / or" is intended to cover all possible combinations and sub-combinations of the listed elements, including any one, any sub-combination or all of the elements listed separately, without excluding other elements. Unless otherwise specified, the terms "first", "second", etc. are used to describe various elements without intending to limit the positional relationship, timing relationship or importance relationship of these elements. Such terms are only used to distinguish one element from another. Unless otherwise specified, the orientation or positional relationship indicated by the terms "front, back, up, down, left, right", etc. is usually based on the orientation or positional relationship shown in the accompanying drawings, which is only for the convenience and simplification of description, and cannot be understood as limiting the scope of protection of the present invention.

[0032] AXL (also known as Ark, Ufo, and Tyro-7) is a member of the TAM receptor tyrosine kinase family, which also includes Mer and Tyro-3. Upon binding to growth arrest-specific protein 6 (Gas6), AXL dimerizes and autophosphorylates, thereby activating downstream mitogen-activated protein kinase (MAPK) and phosphatidylinositol 3-kinase / protein kinase B (PI3K / AKT) pathways. The extracellular membrane region of AXL contains two immunoglobulin repeats (Ig and Ig) and two fibronectin type III domains (FNIII), which are binding sites for endogenous ligands. AXL is a key regulator of cellular function and is associated with the growth of various tissues as well as the survival, proliferation, and invasion of tumor cells. The prominent role of AXL in the development of various cancers makes it a drug target for the treatment of cancers such as glioblastoma multiforme and acute myeloid leukemia. A large amount of experimental data shows that downregulation of AXL can inhibit the growth of malignant tumors such as mesothelioma and colorectal adenocarcinoma. At the same time, overexpression of AXL in breast cancer and pancreatic cancer is also significantly associated with an increased incidence of metastasis. In addition, the activation of AXL is also closely related to the expression of phosphorylated-AKT and matrix metalloproteinase-9 (MMP9), which can promote cell migration and invasion. According to the different binding modes, small molecule inhibitors of AXL can be divided into two categories. Currently reported type I inhibitors include Bemcentinib (BGB324), Amuvatinib (MP-470), Dubermatinib (TP-0903), etc. Among them, Bemcentinib is a highly specific and selective AXL inhibitor, which has been widely studied in various cancer models such as breast cancer, prostate cancer, lung cancer, pancreatic cancer and ovarian cancer. In addition, Foretinib (GSK1363089) and Sitravatinib (MGCD516) as specific type II AXL inhibitors are also in different stages of research.

[0033] Computer-aided drug design (CADD) combined with sophisticated computational modeling techniques has greatly accelerated the drug discovery process. This approach, especially through virtual screening, has become a key tool to improve the efficiency of drug discovery and development. By learning the intricate interactions between proteins and ligands, deep learning models are able to skillfully predict the binding affinity and conformation of molecules, which not only simplifies the screening process for candidate drugs, but also helps discover potential drugs with new structures and significant inhibitory activity.

[0034] The embodiment of the present invention provides a high-throughput virtual screening method for AXL inhibitors based on deep learning, which is used to discover new compounds that inhibit AXL. First, an "internal" database is established, which contains various compounds from Maybridge and MCE drug compound libraries. Secondly, PLANET is used as the main tool for the first virtual screening stage. After the first virtual screening, candidate compounds are further screened according to DeepDock, and nine compounds with good inhibitory properties against AXL are finally determined. Among them, compound 9 showed a fairly high inhibitory activity, and its IC 50 The value is 9.378nM. In order to further understand the dynamic interaction between compound 9 and AXL, molecular dynamics simulations were performed in the embodiments of the present invention. Through these simulation analyses, the stability of the compound 9-AXL complex was further studied, its conformational changes over time were observed, and key interactions affecting binding affinity were identified. This comprehensive analysis not only highlights the unique binding characteristics of compound 9 to AXL, but also provides important inspiration for the development and enhancement of AXL inhibitors.

[0035] Figure 1 A flowchart of a deep learning-based high-throughput virtual screening method for AXL inhibitors according to an exemplary embodiment of the present invention is shown.

[0036] S102: respectively obtaining the ligand and the receptor protein, and converting the structures of the ligand and the receptor protein into different formats using the same program.

[0037] In some embodiments, an X-ray co-crystal protein structure (PDBID: 5U6B) is obtained from the RCSB protein database for further protein preparation. The embodiments of the present invention use the Rosetta 3.13 suite to preprocess the protein structure. The preprocessing step includes: removing all non-protein atoms, such as ligands, ions, and water molecules from the PDB file. Then, the residues in the protein structure are renumbered and normalized, and the resulting structure is converted into a PDB format compatible with Rosetta. The preprocessed structure was verified by SAVES v6.0, and the process included Ramachandran conformational analysis, geometric and physical property evaluation of the structure, evaluation of the reliability of the protein structure model based on statistical data, display of error distribution, and environmental analysis-based methods to evaluate the quality of the protein structure model. Through these rigorous preprocessing and verification procedures, the protein structure used in the embodiments of the present invention can ensure compatibility with subsequent Rosetta protocols and workflows, and ensure its reliability in the following analysis. Finally, the protein structure is converted to PDBQT format using AutoDockTools (version 1.5.7), which includes atomic partial charges and specific atomic types, such as AutoDock atom types. The ligand structures were prepared using the same AutoDockTools program as the protein.

[0038] S104: Establishing a database for high-throughput virtual screening, wherein the database includes at least a first database and a second database.

[0039] Molecular diversity is a key factor in virtual screening because it has been shown to significantly improve hit rates. In order to enhance the molecular diversity in the compound library, the embodiments of the present invention combine the Maybridge drug database with the MCE drug database to establish an "in-house" database. To further improve the "in-house" database, AutoDock and Open Babel software are used. These software are used to generate the three-dimensional structure of the compound library, verify whether its proton state is correct, and confirm whether its molecular format is appropriate.

[0040] In some embodiments, high-throughput virtual screening is applied using the Maybridge drug database as the first database and the MCE drug database as the second database. In some embodiments, the ligand structure is prepared using the same AutoDockTools program as the protein, and the prepared ligand structure is converted to the PDBQT format, which includes adding Gasteiger charges and assigning rotatable bonds. In addition, Open Babel (version 2.4.0) is used to convert the ligand into MOL2 format for LeDock-based screening. Converted into MOL2 format for LeDock-based screening. After converting the ligand structure into multiple formats, it can be used in various screening schemes, and it is also convenient for compatibility with different software suites. In summary, these rigorous procedures ensure the reliable and efficient use of the "in-house" drug database in ligand screening.

[0041] S106: Evaluate and finalize deep learning algorithms and molecular docking programs.

[0042] The main function of molecular docking is to determine the most favorable conformation of the ligand within the active site of the target protein and to estimate the strength of the interaction between the protein and the ligand. With the emergence of artificial intelligence, prediction tools based on artificial intelligence have become increasingly important. These tools are good at analyzing large amounts of molecular and protein interaction data sets and can more accurately predict the binding mode and affinity of ligands to proteins. In addition, choosing the right molecular docking software is a key factor in effective virtual screening.

[0043] First, an AXL dataset containing active substances and decoys was prepared for AXL. The active ligands were from the ChEMBL database (ebi.ac.uk / chembl / ), and the number of decoys was 2900, which were generated based on the chemical similarity with the active ligands, using the molecular weight, logP, number of rotatable bonds, number of hydrogen bond acceptors / donors, and net charge criteria from the DUD-E database (dude.docking.org). Next, the molecules were converted to MOL2 format using Open Babel 2.4.0.

[0044] The embodiments of the present invention use five different programs to dock all molecules into their respective target structures: AutoDock-GPU, AutoDock-Vina, LeDock, PLANET, and DeepDock. The docking scores obtained were used to construct receiver operating characteristic curves (ROC curves) and the performance of these programs was evaluated by T-test. In addition, we also evaluated the performance of these programs using the EF1% metric and the NEF1% metric, the former being used to measure the ability of the model to detect active instances in the top-ranked molecules, and the latter being the ratio of EF1% to the highest EF1% (maximum EF1%) achievable on a given data set.

[0045] The results of the evaluation are presented in full detail in Figure 2, with the focus on identifying the software that guarantees the highest accuracy and reliability in the virtual screening process. The area under the ROC curve is an important indicator for evaluating the discriminative ability of the model, especially when distinguishing between active and inactive (or decoy) data. Figure 2A In the results, DeepDock and PLANET showed excellent performance, which was reflected in their large areas under the ROC curves, with AUC values ​​of 0.90 and 0.86, respectively. These results indicate that DeepDock and PLANET have the most effective discrimination ability in distinguishing AXL-specific active datasets from decoy datasets ( FIG. 2A to FIG. 2C ). Based on these findings, embodiments of the present invention integrate DeepDock and PLANET into a high-throughput virtual screening method.

[0046] S108: Perform high-throughput virtual screening according to the determined deep learning algorithm and molecular docking program, and obtain quasi-target molecules through the high-throughput virtual screening.

[0047] In the present embodiment, the first step of screening: the "internal" database screening was performed according to the PLANET affinity, and the best 2,000 molecules were selected for scoring. Then, the second step of screening evaluated the molecules by a combination of the DeepDock method and the third step of screening method. As a result, nine molecules were screened for further biological activity evaluation.

[0048] In some embodiments, the third screening method is a special manual screening.

[0049] The DeepDock program on GitHub is a deep learning framework for protein-ligand docking, a computational technique for predicting binding affinities between proteins and small molecules. The program uses a convolutional neural network (CNN) to learn the relationship between the three-dimensional structures of proteins and ligands and their corresponding binding affinities. First, the massif algorithm is applied to determine the interface properties when AXL binds to molecules. Then, the software is used to dock and score various molecules.

[0050] Using the "in-house" database and the AXL protein structure, high throughput virtual screening (HTVS) was used to identify potential target molecules. In the embodiment of the present invention, the database has a total of 160,000 small molecule compounds, providing sufficient choices for virtual screening. The virtual screening workflow is as follows Figure 3As shown. First, the molecules in the "internal" database were filtered through the PLANET module, and finally 2,000 molecules with a PLANET affinity of 6.782 or more were screened. These molecules were then further screened using the DeepDock algorithm, which evaluates the impact of various atoms on target binding and facilitates the manual screening process. Subsequently, experienced medicinal chemists evaluated these molecules, focusing on high binding affinity and potential drugability, which may have led to the exclusion of some molecules. The screening process ultimately identified the following nine compounds, see Table 1.

[0051] The first quasi-target molecule (Compound 1) has the structural formula:

[0052]

[0053] The second quasi-target molecule (Compound 2) has the structural formula:

[0054]

[0055] The third quasi-target molecule (Compound 3) has the structural formula:

[0056]

[0057] The fourth quasi-target molecule (Compound 4) has the structural formula:

[0058]

[0059] The fifth quasi-target molecule (Compound 5) has the structural formula:

[0060]

[0061] The sixth quasi-target molecule (Compound 6) has the structural formula:

[0062]

[0063] The seventh quasi-target molecule (Compound 7) has the structural formula:

[0064]

[0065] The eighth quasi-target molecule (Compound 8) has the structural formula:

[0066]

[0067] The ninth quasi-target molecule (Compound 9) has the structural formula:

[0068]

[0069] S110: detecting the in vitro inhibitory activity of the quasi-target molecule on AXL to obtain the final target molecule; and evaluating the biological effect of the final target molecule.

[0070] In some embodiments, the compound is first dissolved in dimethyl sulfoxide (DMSO) to prepare a 10 mM stock solution, and then diluted to 50 times the concentration of the test drug solution before use. The test concentrations are 0.1 nM, 1 nM, 10 nM, 100 nM, 1000 nM and 10,000 nM, respectively, obtained by 10-fold serial dilution. In kinase reaction buffer (1 volume of 5X kinase buffer and 4 volumes of water; 5 mM MgCl 2 ; 1mM DTT). Transfer 25nL of the complex dilution containing Echo655 to a 384test plate. After centrifugation, add 2.5μL of 2x kinase / metal ion solution to the 384test plate and incubate at 25°C for 10 minutes. Then, add 2.5μL of 2x substrate / ATP solution to the detection plate and incubate at 25°C for 60 minutes. Prepare a 2x Sa-XL 665 / TK&STK antibody-Cryptate mixture with HTRF analysis buffer, add 5μL Sa-XL 665 / TK&STK antibody-Cryptate to each well, centrifuge at 1000rpm for 1 minute, and incubate at 25°C for 1 hour. Finally, calculate the IC of AXL kinase using the formula 50 value.

[0071] Y=Bottom+(Top-Bottom) / (1+10^((LogIC 50 -X)*Hill Slope))

[0072] Based on the HTVS results, the inhibition rate of AXL by nine selected compounds (each at a concentration of 25 μM) was further evaluated. Figure 4A As shown, compared with other compounds, compound 3, compound 5, compound 6 and compound 9 showed higher inhibition rates. Therefore, the IC values ​​of these four compounds on AXL were further determined. 50 It is worth noting that compound 9 has a stronger inhibitory activity against AXL (IC 50 =9.38 nM). In addition, compound 3 (IC 50 =112.0 nM) and compound 5 (IC 50 =22.3 nM) also showed commendable inhibitory activity ( Figure 4B ). These comprehensive analyses not only validated the effectiveness of the HTVS process, but also laid a solid foundation for the development of innovative AXL inhibitors.

[0073] Table 1. Docking scores and inhibition rates of nine compounds

[0074]

[0075] In addition, the embodiments of the present invention also perform molecular dynamics simulations. The embodiments of the present invention use the Amber software package (version 2022) for molecular dynamics (MD) simulations. Amber is a widely used and rigorously validated MD simulation and analysis software package.

[0076] The protein and ligand structures were initially prepared using the AmberTools package (version 23). Hydrogen atoms were added to the protein structure using the LEaP module and assigned appropriate ionization states based on the pKa values ​​of the titratable residues. The protein structure was solvated in a rectangular box of TIP3P water molecules. The minimum distance between the protein and the box border was 30 Counterions were added to neutralize the overall charge of the system. The ligand structures were prepared using the Leap module to generate the necessary Amber compatible force field parameters. The General Amber Force Field (GAFF) was assigned to the ligand atoms, while the partial atomic charges were calculated using the AM1-BCC method.

[0077] The prepared system was energy minimized in three stages. First, the steepest descent algorithm was applied twice for 5000 steps each, followed by the conjugate gradient algorithm for another 5000 steps. The system was then gradually heated from 0 K to the target temperature of 300 K over a period of 100 ps using a Langewinter thermostat. Positional restraints were applied to the protein and ligand heavy atoms. Subsequently, the system was equilibrated in the NPT ensemble at 300 K and 1 atm for 1 ns, and the restraints were gradually reduced to zero.

[0078] Production MD simulations were performed at 300 K and 1 atm in the NPT ensemble using the pmemd.cuda module for computational efficiency. The time step was 2 fs and the SHAKE algorithm was used to constrain bonds involving hydrogen atoms. Long-range electrostatic interactions were handled using the Particle Mesh Ewald (PME) method. The total simulation time for each system was 100 ns and the coordinates were saved every 150 ps for subsequent analysis.

[0079] The MD simulation trajectories were analyzed using the CPPTRAJ module in the AmberTools software package. The MD simulation trajectories were analyzed. Key properties such as root mean square deviation (RMSD), root mean square fluctuation (RMSF), and radius of gyration (Rg) were calculated to assess the stability, flexibility, and overall conformational changes of the protein-ligand complex. In addition, hydrogen bonds and other relevant interactions were analyzed to gain insight into the binding mechanism of the protein-ligand. By applying these rigorous methods, we ensured the reliability and insight of the MD simulations of protein-ligand interactions in this study.

[0080] Based on the evaluation of inhibitory activity, molecular dynamics (MD) simulations of compound 9 and AXL were performed to explain their interaction and provide a basis for further structural optimization work. Figure 5A As shown, the simulation results show that after 100 nanoseconds, the binding of the protein to compound 9 tends to be stable, indicating that this is a reliable complex that can be subjected to in-depth rational analysis. The last conformation was extracted for detailed study. In particular, compound 9 can generate stable H-bond interactions with Pro621, Met623 and Asp627, which makes it highly potent in inhibiting AXL ( Figure 5B and 5C ). At the same time, Figure 5B and 5D As shown, compound 9 can also form strong hydrophobic interactions with Val550, Phe622, Gly626, and Met679. The corresponding key interactions between compound 9 and AXL may provide valuable insights for the discovery and development of highly active AXL inhibitors.

[0081] In some embodiments, the above results were statistically analyzed. The results are presented as mean ± standard deviation of three independent experiments. Data were analyzed using GraphPad Prism 8.0 software (San Diego, California, USA). The significance of the difference between the two groups was determined by independent sample t-test and ANOVA multiple comparison test. Statistically significant p values ​​are marked as: *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001.

[0082] According to an embodiment of the present invention, Figure 6 is a schematic diagram of a framework of a high-throughput virtual screening device for AXL inhibitors based on deep learning according to an exemplary embodiment of the present invention, such as Figure 6 As shown, an embodiment of the present invention further provides a high-throughput virtual screening device for AXL inhibitors based on deep learning. The high-throughput virtual screening device 600 for AXL inhibitors based on deep learning comprises:

[0083] The acquisition conversion module 602 is used to respectively acquire the ligand and the receptor protein, and the structures of the ligand and the receptor protein are converted into formats using the same program.

[0084] Database establishment module 604: used for high-throughput virtual screening, the database includes at least a first database and a second database.

[0085] Evaluation and determination module 606: used to evaluate and determine the deep learning algorithm and molecular docking program.

[0086] Virtual screening module 608: used to perform high-throughput virtual screening according to the determined deep learning algorithm and molecular docking program, and obtain quasi-target molecules through the high-throughput virtual screening.

[0087] Detection and evaluation module 610: used to detect the in vitro inhibitory activity of the quasi-target molecule on AXL to obtain the final target molecule; and to evaluate the biological effect of the final target molecule.

[0088] It should be noted that the device embodiment in the embodiment of the present invention can refer to the above method embodiment, which will not be described in detail here.

[0089] The embodiments of the present invention provide a high-throughput virtual screening method to identify potent novel inhibitors of AXL. By using advanced algorithms such as PLANET and DeepDock, compound 9 was determined to be a highly effective AXL inhibitor with significant inhibitory biological activity (IC 50 =9.378nM). Further exploration by molecular dynamics (MD) simulation revealed the specific binding mode of compound 9 to AXL. Notably, compound 9 formed stable hydrogen bond interactions with Pro621, Met623, and Asp627 in AXL, resulting in high inhibitory potency. In addition, compound 9 also established stable hydrophobic interactions with Val550, Phe622, Gly626, and Met679. These key interactions between compound 9 and AXL provide profound inspiration for the discovery and development of highly active AXL inhibitors, and lay the foundation for future structural optimization and drug development work.

[0090] It is to be understood that the present invention is described by some embodiments, and it is known to those skilled in the art that various changes or equivalent substitutions may be made to these features and embodiments without departing from the spirit and scope of the present invention. In addition, under the teachings of the present invention, these features and embodiments may be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the scope of protection of the present invention.

Claims

1. A high-throughput virtual screening method for AXL inhibitors based on deep learning, characterized in that: include: The ligand and receptor proteins are obtained respectively, and the structures of the ligand and the receptor proteins are converted into formats using the same program; Establishing a database for high-throughput virtual screening, wherein the database comprises at least a first database and a second database; Evaluate and finalize deep learning algorithms and molecular docking procedures; According to the determined deep learning algorithm and molecular docking program, high-throughput virtual screening is performed to obtain quasi-target molecules through the high-throughput virtual screening; detecting the in vitro inhibitory activity of the quasi-target molecule on AXL to obtain the final target molecule; The biological effect of the final target molecule is evaluated.

2. The high-throughput virtual screening method for AXL inhibitors based on deep learning according to claim 1, characterized in that: Also includes: The AXL inhibitory effect of the final target molecule was simulated and verified by molecular dynamics.

3. The high-throughput virtual screening method for AXL inhibitors based on deep learning according to claim 2, characterized in that: The first database is the Maybridge drug database, and the second database is the MCE drug database.

4. The high-throughput virtual screening method for AXL inhibitors based on deep learning according to claim 3, characterized in that: The evaluated deep learning algorithms include at least: a first deep learning algorithm and a second deep learning algorithm, and the evaluated molecular docking programs include three molecular docking programs: AutoDock-GPU molecular docking program, Autodock Vina molecular docking program, and LeDock molecular docking program.

5. The high-throughput virtual screening method for AXL inhibitors based on deep learning according to claim 4, characterized in that: The confirmed deep learning algorithms include: PLANET deep learning algorithm and DeepDock deep learning algorithm.

6. The high-throughput virtual screening method for AXL inhibitors based on deep learning according to claim 5, characterized in that: The order of the high-throughput virtual screening is set as: firstly, the first step of screening is performed by the PLANET deep learning algorithm, then the second step of screening is performed by the DeepDock deep learning algorithm, and finally the third step of screening is performed.

7. The high-throughput virtual screening method for AXL inhibitors based on deep learning according to claim 6, characterized in that: The affinity of the molecules screened in the first step by the PLANET deep learning algorithm is greater than or equal to 6.

782.

8. The deep learning-based high-throughput virtual screening method for AXL inhibitors according to claim 7, characterized in that: The quasi-target molecules are respectively a first quasi-target molecule, a second quasi-target molecule, a third quasi-target molecule, a fourth quasi-target molecule, a fifth quasi-target molecule, a sixth quasi-target molecule, a seventh quasi-target molecule, an eighth quasi-target molecule, and a ninth quasi-target molecule.

9. The deep learning-based high-throughput virtual screening method for AXL inhibitors according to claim 8, characterized in that: The ninth quasi-target molecule is the final target molecule, and its chemical formula is:

10. The deep learning-based high-throughput virtual screening method for AXL inhibitors according to claim 9, characterized in that: The detecting of the in vitro inhibitory activity of the quasi-target molecule on AXL comprises: detecting the IC 50 value, and the IC 50 The value is 9.38 nM, and the IC 50 The value is 112.0 nM, and the IC 50 The value is 22.3nM.

11. A high-throughput virtual screening device for AXL inhibitors based on deep learning, characterized in that: include: Acquisition conversion module: used to obtain ligand and receptor proteins respectively, and the structures of the ligand and the receptor proteins are converted into formats using the same program; Database establishment module: used for high-throughput virtual screening, the database includes at least a first database and a second database; Evaluation and determination module: used to evaluate and determine deep learning algorithms and molecular docking programs; Virtual screening module: used to perform high-throughput virtual screening according to the determined deep learning algorithm and molecular docking program, and obtain quasi-target molecules through the high-throughput virtual screening; as well as Detection and evaluation module: used to detect the in vitro inhibitory activity of the quasi-target molecule on AXL and obtain the final target molecule; The biological effect of the final target molecule is evaluated.

Images