CDK9 inhibitor virtual screening method and device based on deep learning, equipment and storage medium

Through a virtual screening method for CDK9 inhibitors based on deep learning, the problems of drug resistance and recurrence in the treatment of acute myeloid leukemia were solved, and an effective CDK9 inhibitor was found, with significant inhibitory activity and potential therapeutic potential, providing Safer and effective treatment options.

CN119007867APending Publication Date: 2024-11-22ZHEJIANG CANCER HOSPITAL
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411066110.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-05
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

In the treatment of acute myeloid leukemia (AML), primary drug resistance and disease recurrence of initial treatment are major obstacles, and existing drug development is difficult to effectively respond to this challenge. The side effects of targeted therapies are greater and lack more effective. and safe treatment options.

Method used

Using the virtual screening method of CDK9 inhibitors based on deep learning, the Morgan fingerprint and JACAD distance of the compound were calculated by obtaining and cleaning CDK9 active compound data, and clustering, selecting the highest pIC50 compound to generate bait molecules, combining deep learning algorithms and The molecular docking program is virtually screened to obtain quasi-target molecules and evaluated in vitro inhibitory activity and biological effects to ultimately obtain potential CDK9 inhibitors.

Benefits of technology

A CDK9 inhibitor with an IC50 value of 31.28 nM was discovered, which significantly reduced the cell viability of hematological tumor cell lines, had high binding affinity and potential for the treatment of myeloid leukemia, and provided a more effective and less side effect treatment regimen. .

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119007867A_ABST
    Figure CN119007867A_ABST
Patent Text Reader

Abstract

The invention provides a CDK9 inhibitor virtual screening method based on deep learning. The CDK9 inhibitor virtual screening method comprises the steps of obtaining CDK9 active compound data, cleaning the data, determining a CDK9 verification data set, calculating Morgan fingerprints and Jaccard distances of all compounds in the CDK9 verification data set, and clustering the compounds to form a plurality of clusters; selecting the compound with the highest pIC50 in the cluster as a representative compound, and generating a bait molecule according to the representative compound; determining a deep learning algorithm according to a quintuple cross validation evaluation result, and determining a molecular docking program according to a molecular docking program evaluation result; performing virtual screening according to the determined deep learning algorithm and the determined molecular docking program, and performing similarity scoring and drug-likeness evaluation on virtual screening results by using RDKit to obtain quasi target molecules; detecting the in-vitro inhibitory activity of the quasi-target molecule to CDK9 to obtain a final target molecule; and performing biological effect evaluation on the final target molecule. The invention further relates to a device, equipment and a storage medium.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of computer-aided drug design, and more specifically to a deep learning-based virtual screening method, device, equipment and storage medium for CDK9 inhibitors. Background Art

[0002] Acute myeloid leukemia (AML) is a diverse hematological malignancy, one of the characteristics of which is the clonal expansion of myeloid mutant cells. AML is the most common type of acute leukemia in adults and is characterized by a wide range of genomic alterations and molecular mutations. These variations provide multiple potential targets for drug development, which may improve the prognosis of the disease. In the related art, the treatment options for AML include traditional therapies (such as cytarabine + anthracycline (7+3) intensive chemotherapy and hypomethylating drugs) and targeted therapies (including immunotherapy and targeted drugs). In addition, allogeneic stem cell transplantation (Allo-SCT) has played a role in improving the survival of eligible AML patients in the past few decades. However, in the treatment of AML, primary resistance to initial treatment and subsequent disease relapse are major obstacles. To improve the survival of patients with AML, it is necessary to explore more effective therapeutic drugs to meet this challenge. Summary of the invention

[0003] To solve the above-mentioned shortcomings, according to a first aspect of the present invention, a method for virtual screening of CDK9 inhibitors based on deep learning is provided, comprising: obtaining CDK9 active compound data, cleaning the data, determining a CDK9 verification data set, calculating the Morgan fingerprint and Jaccard distance of each compound in the CDK9 verification data set, clustering the compounds to form a plurality of clusters; selecting pIC 50 The highest compound is determined as a representative compound, and a bait molecule is generated based on the representative compound; a deep learning algorithm is determined based on the five-fold cross-validation evaluation results, and a molecular docking program is determined based on the molecular docking program evaluation results; virtual screening is performed based on the determined deep learning algorithm and the determined molecular docking program, and the virtual screening results are scored for similarity and evaluated for drug-likeness using RDKit to obtain quasi-target molecules; the in vitro inhibitory activity of the quasi-target molecules on CDK9 is detected to obtain the final target molecules; and the biological effects of the final target molecules are evaluated.

[0004] Optionally, the method further includes: simulating and analyzing the CDK9 inhibitory effect of the final target molecule through molecular dynamics and verifying the same.

[0005] Optionally, the evaluated deep learning algorithms include eight deep learning algorithms: RF, SVR, XGBoost, MLP, Bayesian, Gradient Boosting, LGBM and K-NN, and the evaluated molecular docking programs include two molecular docking programs: Vina-GPU molecular docking program and AutoDock-GPU molecular docking program; the determined deep learning algorithms include at least one of the eight evaluated deep learning algorithms, and the determined molecular docking program includes at least one of the two evaluated molecular docking programs.

[0006] Optionally, the determined deep learning algorithm is the RF deep learning algorithm, and the determined molecular docking program is the Vina-GPU molecular docking program.

[0007] Optionally, the virtual screening includes at least two steps, wherein the first virtual screening step is screening through the RF deep learning algorithm, and the second virtual screening step is screening again through the Vina-GPU molecular docking program.

[0008] Optionally, the first virtual screening step is screening through the RF deep learning algorithm, and also includes: using Hyperopt to search for hyperparameters, and adjusting the RF deep learning algorithm using the following hyperparameter values: n_estimators is set to 380, max_depth is set to 41, min_samples_split is set to 2, min_samples_leaf is set to 1, and max_features is set to 0.1562.

[0009] Optionally, the quasi-target molecules are respectively a first quasi-target molecule and a second quasi-target molecule.

[0010] Optionally, the detecting of the in vitro inhibitory activity of the quasi-target molecule on CDK9 comprises: the IC 50 The value is 31.28nM.

[0011] Optionally, the second quasi-target molecule is the final target molecule, and its chemical formula is:

[0012]

[0013] According to a second aspect of the present invention, a virtual screening device for CDK9 inhibitors based on deep learning is provided, comprising: an acquisition clustering module: used to acquire CDK9 active compound data, clean the data, determine a CDK9 verification data set, calculate the Morgan fingerprint and Jaccard distance of each compound in the CDK9 verification data set, and cluster the compounds to form a plurality of clusters; a determination generation module: used to select pIC50 The highest compound is determined as the representative compound, and a bait molecule is generated based on the representative compound; an evaluation and determination module is used to determine the deep learning algorithm based on the five-fold cross-validation evaluation results and to determine the molecular docking program based on the molecular docking program evaluation results; a virtual screening module is used to perform virtual screening based on the determined deep learning algorithm and the determined molecular docking program, and use RDKit to perform similarity scoring and drug-likeness evaluation on the virtual screening results to obtain quasi-target molecules; and a detection and evaluation module is used to detect the in vitro inhibitory activity of the quasi-target molecule on CDK9 to obtain the final target molecule; and the final target molecule is evaluated for biological effects.

[0014] According to a third aspect of the present invention, there is provided an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the above-mentioned method.

[0015] According to a fourth aspect of the present invention, there is provided a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute the above method.

[0016] According to a fifth aspect of the present invention, there is provided a computer program product, comprising a computer program, wherein the computer program implements the steps of the above method when executed by a processor.

[0017] The embodiment of the present invention establishes an inhibitor discovery method combining deep learning algorithm and molecular docking technology. This method provides a new class of CDK9 inhibitors. The method includes: constructing a comprehensive CDK9 data set, applying various molecular fingerprints, selecting the best deep learning algorithm and molecular docking program for virtual screening of CDK9 inhibitor discovery. Among the identified compounds, compound 2 has inhibitory activity against CDK9, IC 50 The value is 31.28nM. In addition, the compound can also significantly reduce the cell viability of blood tumor cell lines through cell viability assay. Furthermore, the embodiments of the present invention also carried out molecular dynamics simulations to analyze the binding stability and interaction between compound 2 and CDK9 protein. These simulations confirmed the high binding affinity of compound 2, clarified its mechanism of action, and provided detailed insights into the molecular interactions and dynamics of inhibitor-protein complexes. These promising results at both the molecular and cellular levels indicate that compound 2 has great potential for further development as a therapeutic agent. The embodiments of the present invention provide a reference for the development of CDK9 inhibitors with better efficacy and fewer side effects for the treatment of acute myeloid leukemia. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] 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:

[0019] Figure 1 A flowchart of a deep learning-based virtual screening method for CDK9 inhibitors according to an exemplary embodiment of the present invention is shown.

[0020] Figure 2A A schematic diagram showing the distribution of CDK9 active compounds according to an exemplary embodiment of the present invention.

[0021] Figure 2B Schematic diagram showing clustering and t-SNE of CDK9 active compounds according to an exemplary embodiment of the present invention.

[0022] Figure 2C A schematic diagram showing the distribution of CDK9 activity decoy compounds according to an exemplary embodiment of the present invention.

[0023] Figure 2D A schematic diagram showing five-fold cross validation of RDkit molecular fingerprints according to an exemplary embodiment of the present invention is shown.

[0024] Figure 2E A five-fold cross-validation effect diagram of the RF model of an exemplary embodiment of the present invention is shown.

[0025] Figure 2F A diagram showing the test results after hyperparameter optimization based on the RF model of an exemplary embodiment of the present invention is shown.

[0026] Figure 3A A schematic diagram of a ROC curve of the RF model according to an exemplary embodiment of the present invention is shown.

[0027] Figure 3B A schematic diagram of an ROC curve of molecular docking of a CDK9 dataset with a selected protein active pocket based on a molecular docking program according to an exemplary embodiment of the present invention is shown.

[0028] Figure 3C A hybrid virtual screening flow chart of an exemplary embodiment of the present invention is shown.

[0029] Figure 4A A lollipop analysis effect diagram showing CDK9 inhibition rate according to an exemplary embodiment of the present invention.

[0030] Figure 4B A schematic diagram showing dose-response curves of Compound 2 and SNS-032 according to an exemplary embodiment of the present invention is shown.

[0031] Figure 4C A graph showing a comparison of cell viability assay results of Jurkat cell lines treated with the selected compound 1604810-84-5 according to an exemplary embodiment of the present invention.

[0032] Figure 4D A graph showing a comparative effect of cell viability assay of MOLT-4 cell line treated with the selected compound 1604810-84-5 according to an exemplary embodiment of the present invention.

[0033] Figure 5A A schematic diagram showing the RMSD trajectory of Compound 2 according to an exemplary embodiment of the present invention and CDK9 protein over 500 ns is shown.

[0034] Figure 5B A schematic diagram showing the binding mode of compound 2 according to an exemplary embodiment of the present invention and CDK9 protein.

[0035] Figure 5C A schematic diagram showing MM / GBSA contribution analysis of amino acid residues of CDK9 protein according to an exemplary embodiment of the present invention is shown.

[0036] Figure 5D A diagram showing the Gibbs free energy distribution effect of the complex of Compound 2 and CDK9 according to an exemplary embodiment of the present invention.

[0037] Figure 6 A schematic structural diagram of a deep learning-based virtual screening device for CDK9 inhibitors according to an exemplary embodiment of the present invention is shown.

[0038] Figure 7 A schematic structural diagram of an electronic device according to an exemplary embodiment of the present invention is shown. DETAILED DESCRIPTION

[0039] 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.

[0040] 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.

[0041] Figure 1 A flowchart of a deep learning-based virtual screening method for CDK9 inhibitors according to an exemplary embodiment of the present invention is shown.

[0042] S102: Acquire CDK9 active compound data, clean the data, determine a CDK9 verification data set, calculate the Morgan fingerprint and Jaccard distance of each compound in the CDK9 verification data set, and cluster the compounds to form a plurality of clusters.

[0043] In some embodiments, CDK9 active compounds were downloaded from the ChEMBL database (www.ebi.ac.uk / chembl, accessed on January 15, 2024). The IC 50 values ​​and the corresponding SMILES format. The open source cheminformatics toolkit RDKit was used to remove invalid SMILES, standardize the remaining SMILES (including desalting and structure normalization), and eliminate duplicates. Subsequently, Morgan fingerprints and Jaccard distances were calculated to cluster the compounds into 50 groups. In order to visualize the distribution of these clusters, the t-distributed stochastic neighbor embedding (t-SNE) method was used. In addition, the deep learning-based method DeepCoy was used to generate property-matching bait molecules. In order to ensure the diversity of the bait compound skeleton, bait compounds were generated based on the active molecular structures in the 50 clusters.

[0044] Furthermore, combined with FIG. 2A to FIG. 2C As shown, an embodiment of the present invention obtained 1002 CDK9 active compounds from the ChEMBL database. These data were cleaned and processed using RDKit so that a machine learning model could be subsequently constructed and performance verified. The spatial distribution characteristics of the molecules were calculated using t-SNE. The results show that the obtained CDK9 dataset has good spatial distribution characteristics. In order to subsequently verify the machine learning algorithm and molecular docking program, the molecules were clustered and representative compounds were selected to obtain bait molecules with similar physicochemical properties but different structures. The molecules were then divided into 50 categories based on Morgan fingerprints and Jaccard distances, and pIC 50 The most representative molecules. The deep learning generation algorithm DeepCoy was used to generate 2,500 decoy molecules based on these representative compounds.

[0045] S104: Select pIC in the cluster 50 The highest ranking compound was determined as a representative compound, based on which the decoy molecule was generated.

[0046] S106: Determine the deep learning algorithm based on the five-fold cross-validation evaluation results and determine the molecular docking program based on the molecular docking program evaluation results.

[0047] In order to convert the molecular structure from SMILES string to a format suitable for machine learning analysis, the embodiment of the present invention divides the data set into two parts, training and testing, at a ratio of 9:1. Various molecular fingerprints were generated using the RDKit chemical informatics library, including RDKit, Atom pairs, Morgan, and MACCS fingerprints. Each fingerprint type has a different characterization of the molecular structure. The performance of these fingerprints under the Random Forest (RF) model was evaluated using a five-fold cross-validation method, and the RDKit fingerprint was selected for further verification based on the superior performance of the RDKit fingerprint. Then, the ability of several models to predict the affinity of CDK9 inhibitors using RDKit fingerprints was evaluated. All models were built using the Scikit-learn library, and molecular characterization and data set classification were performed using RDKit. According to the results of the five-fold cross-validation, the RF model was selected for further virtual screening. In order to improve the predictive performance of the RF model, we used Hyperopt to search for hyperparameters and adjusted the RF model using the following hyperparameters: n_estimators, max_depth, min_samples_split, min_samples_leaf, and max_features.

[0048] The parameter ranges are as follows:

[0049] n_estimators (the number of trees in the forest) is between 50 and 500 with a step size of 10;

[0050] max_depth (maximum depth of each tree) is between 5 and 50, with a step size of 1;

[0051] min_samples_split (the minimum number of samples required to split an internal node) is between 2 and 10, in steps of 1; min_samples_leaf (the minimum number of samples required for a leaf node) is between 1 and 10, in steps of 1; and

[0052] max_features (the number of features to consider for each split) is between 0.1 and 1.0.

[0053] In some embodiments, the hyperparameters are determined as:

[0054] n_estimators: 380;

[0055] max_depth: 41;

[0056] min_samples_split: 2;

[0057] min_samples_leaf: 1; and

[0058] max_features: 0.1562.

[0059] The embodiment of the present invention downloaded proteins with a resolution less than All CDK9 protein structures were collected. These structures included: 3BLH, 3BLR, 3BLQ, 3TN8, 3MI9, 3MY1, 4BCH, 4IMY, 4RO5, and 7NWK for molecular docking evaluation. Only CDK9 and its co-crystal ligands were retained, and the protein was prepared using MGLTools 1.5.7. The protein preparation included checking hydrogen atoms and removing water molecules. Then, the performance of two docking software tools, AutoDock-GPU and Vina-GPU, was evaluated, and the evaluation results were analyzed using ROC curves, area under the curve (AUC), t-tests, box plots, and score distribution histograms.

[0060] Specifically, combined FIG. 2D to FIG. 2F As shown, in order to determine the most suitable machine learning algorithm for the CDK9 dataset, a five-fold cross validation was used to determine the best molecular fingerprint. Cross validation was performed using the CDK9 active compound dataset and the RDKit, Atom pairs, Morgan, and MACCS fingerprints in the RF model (Table 1). According to the average R 2 and RMSE results, RDKit was selected as the molecular characterization tool for the model. Then, eight machine learning algorithms were constructed, namely random forest (RF), support vector regression (SVR), extreme gradient boosting (XGBoost), multilayer perceptron (MLP), Bayesian, gradient boosting, light gradient boosting machine (LGBM) and k-nearest neighbor (K-NN). Based on the results, we selected the RF model as one of the virtual screening tools for this study (Table 2). In order to enhance the robustness of the model, we performed a hyperparameter search for the RF model and determined the optimal parameters. The training and testing results based on these optimal parameters are as follows: training RMSE: 0.253, training R 2 : 0.936; Test RMSE: 0.609, Test R 2 :0.611.

[0061] Table 1. Selection of fingerprints

[0062]

[0063] Table 2. Selection of deep learning algorithms

[0064]

[0065] Combination FIG. 3A to FIG. 3B As shown, in order to validate the deep learning model and molecular docking program, the CDK9 validation dataset was used. The performance of the RF model was evaluated by predicting the CDK9 validation dataset, and its AUC-ROC was 0.80. For molecular docking evaluation, the RCSB database was downloaded with a resolution of less than All CDK9 protein structures were analyzed, and docking simulations were performed on different protein structures using AutoDock-GPU and Vina-GPU. The analysis showed that the 7NWK structure evaluated by Vina-GPU performed best, with an AUC of 0.72, and the t-test result was statistically significant (t-statistic = 5.36670124, p = 8.76E-08; Table 3). Therefore, the 7NWK structure and Vina-GPU method were selected for the second round of virtual screening.

[0066] Table 3 Double T test results of different screening combinations

[0067]

[0068] S108: Perform virtual screening according to the determined deep learning algorithm and the determined molecular docking program, and use RDKit to perform similarity scoring and drug-likeness evaluation on the virtual screening results to obtain quasi-target molecules.

[0069] The embodiment of the present invention uses RDKit to prepare a MCE-like drug compound library. The preparation process includes: removing invalid SMILES, standardizing SMILES (including desalting and structural normalization), and removing duplicate SMILES. Reported CDK9 active molecules are excluded from the compound library. The final MCE-like drug compound library includes 2 compounds and is saved in SMILES format.

[0070] The virtual screening workflow consists of two steps: RF model and Vina-GPU.

[0071] In the first round of screening, the RF model was selected based on the results of five-fold cross validation. Compounds were characterized using RDKit molecular fingerprinting. The optimal parameters were determined through hyperparameter search, and the trained pt weights were saved for use in the screening phase. Based on the results, the prediction of pIC 50 280 compounds with values ​​greater than 7 were subjected to the second round of screening.

[0072] In the second round of screening, 7NWK protein and Vina-GPU software (both of which showed the best evaluation results) were used to dock the 280 compounds obtained in the first round of screening. The settings of Vina-GPU are as follows: the length, width and height dimensions of the docking box are The number of threads was set to 8000; Energy_range was set to 0. Subsequently, we selected the top 50 compounds for further processing based on their Vina scores less than or equal to -9.9 kcal / mol.

[0073] The embodiment of the present invention uses RDKit to perform similarity scoring on 50 compounds obtained through virtual screening and evaluates whether they meet the five principles of drug similarity. First, we perform similarity scoring on these 50 compounds based on the existing CDK9 active molecule dataset and select 34 compounds with scores less than or equal to 0.35 for further processing.

[0074] RDKit was further used to match these 34 compounds with Lipinski's Rule of Five, with the following criteria: molecular weight less than or equal to 500; logP less than or equal to 5; number of hydrogen bond donors less than or equal to 10; number of hydrogen bond acceptors less than or equal to 5; number of rotatable bonds less than 10. Ultimately, 22 compounds that met the "Lipinski's Rule of Five" criteria were identified, and two of them were selected for further bioactivity evaluation and molecular dynamics simulation.

[0075] Specifically, Figure 3C As shown, in order to find potential CDK9 inhibitors, in an embodiment of the present invention, a hybrid virtual screening workflow was constructed, and virtual screening was performed using the MCE drug compound library (https: / / www.medchemexpress.com / ). The MCE drug compound library was prepared using RDkit, and the reported CDK9 active molecules were deleted from the library. After preparation, a total of 8602 molecules entered the virtual screening workflow. The virtual screening workflow includes two stages:

[0076] In the first round of virtual screening, an RF model was selected based on the results of five-fold cross-validation. The model was optimized using hyperparameter tuning to determine the best parameters and then trained for this screening. 50 280 molecules with values ​​greater than 7 were advanced to the second round of screening. In order to fully utilize the advantages of structure-based virtual screening, Vina-GPU was used in the second round of screening. Compounds were ranked according to their binding affinity scores to the CDK9 active pocket, with lower scores indicating stronger binding affinity. The top 50 molecules were selected for further screening.

[0077] To ensure the chemical diversity and drug-likeness of the selected compounds, we used RDKit to perform two rounds of screening on the 50 compounds obtained: the first round was similarity scoring, and the second round was whether they met the five principles of Lipinski. Ultimately, 21 compounds with similarity scores below 0.35 and meeting the drug-likeness criteria were identified. Subsequently, compounds 1 and 2 were selected for further bioactivity evaluation.

[0078] The first quasi-target molecule (Compound 1) has the structural formula:

[0079]

[0080] The second quasi-target molecule (Compound 2) has the structural formula:

[0081]

[0082] Table 4. Scores at each stage and inhibition rates of the two compounds finally selected

[0083]

[0084] S110: detecting the in vitro inhibitory activity of the quasi-target molecule on CDK9 to obtain the final target molecule; and evaluating the biological effect of the final target molecule.

[0085] Combination FIG. 4A to FIG. 4D As shown, selected compounds were biologically evaluated using the ADP-GLO assay to measure their inhibition of CDK9 at a concentration of 25 μM. Figure 4A The inhibition rate of the compounds is shown. Among the tested compounds, compound 2 had the highest inhibition rate of 102.57%, followed by compound 1 with an inhibition rate of 44.75%. The inhibition rate of the positive control drug Staurosporine was 104.31%. To further evaluate the efficacy of these compounds, the IC 50 The dose-response curves of compound 2 and positive control SNS-032 are shown in Figure 4B IC of compound 2 50 The value was 31.28 nM, indicating that it has significant inhibitory activity against CDK9, while the IC 50 The value is 2.938nM.

[0086] In addition, if Figure 4C and Figure 4DAs shown, the cell viability of two blood tumor cell lines treated with compound 2 and the laboratory's existing CDK9 positive control 1604810-84-5 was also evaluated. The cell viability assay showed that compound 2 significantly reduced the cell viability of the tested blood tumor cell lines, with an efficacy comparable to 1604810-84-5. Together, these findings indicate that the selected compounds inhibit CDK9 at the molecular level and exhibit potent biological activity in cell-based assays, highlighting their potential as therapeutic agents for acute myeloid leukemia.

[0087] About In Vitro Kinase Assays

[0088] First, the compounds were dissolved in dimethyl sulfoxide (DMSO) to make a 10 mM stock solution, which was then diluted 50-fold to obtain the desired test concentrations. The concentration range of 0.1 to 10,000 nM was determined by 10-fold serial dilutions. Two solutions were prepared using an assay buffer consisting of 5× buffer, 5 mM MgCl2, 1 mM DTT, and ddH2O:

[0089] (1) 2× mixed solution of ATP and substrate;

[0090] (2) 2× mixed solution of kinase and metal.

[0091] The Echo 655 robot transferred 25 nL of compound solution to a 384-well plate. Subsequently, 2.5 μL of 2× kinase and metal mixture was added and incubated at 25°C for 10 minutes. Subsequently, 2.5 μL of 2× substrate and ATP solution was added and incubated at the same temperature for another 50 minutes. At the same time, a 2× mixture of XL665 and antibody was prepared. Then 5 μL of kinase detection reagent was added and incubated at 25°C for another 60 minutes. Fluorescence was measured at 620 nm and 665 nm wavelengths using a microplate reader, and the dose response of the data was analyzed using GraphPad 7.0 software, and the IC was calculated using the following formula 50 value:

[0092] Y=Bottom+(Top-Bottom) / (1+10^((LogIC50-X)*hillslope))

[0093] About cell culture and CCK8 assay

[0094] Human AML cell lines (including HL-60 and MV4-11) were cultured in RPMI-1640 (Gibco) supplemented with 10% FBS (Hyclone), 1% penicillin / streptomycin (Hyclone), 1% non-essential amino acids (Gibco), 2mM L-glutamine (Sigma), 1mM sodium pyruvate (Sigma) and 55μL β-mercaptoethanol (Sigma). Cells were cultured in complete medium at 37°C with 5% CO2.

[0095] Cell growth inhibition was assessed using a cell counting kit-8 (CCK-8, Dojindo Laboratories). First, Jurkat-3 and MOLT-4 cells were counted and seeded into 96-well culture plates at 10,000 cells per well. The cells were cultured at 37°C, 5% CO2 for 24 hours, and then the culture medium was replaced with different concentrations of compound 1 diluted in the corresponding culture medium. After 48 hours of co-culture, 10 μL of CCK-8 reagent was added to each well. After one hour of incubation at 37°C, the optical density (OD) at a wavelength of 450 nanometers was measured using a Sunrise multi-function microplate reader (Teken Life Sciences). The percentage of cell viability at each concentration relative to the control group was calculated, and the half-maximum effective concentration (EC50) was determined using GraphPad Prism 7.0.

[0096] About Biological Assessment

[0097] The selected compounds were biologically evaluated using the ADP-GLO assay to measure their inhibition of CDK9 at a concentration of 25 μM. Among the compounds tested, compound 2 showed the highest inhibition of 102.57%, followed by compound 1 with an inhibition of 44.75%. The positive control drug Staurosporine showed an inhibition of 104.31%. To further evaluate the potency of these compounds, the IC 50 Values, dose response curves of compound 2 and positive control SNS-032 (reference Figure 4B ). IC of compound 2 50 The value was 31.28 nM, indicating that it has significant inhibitory activity against CDK9, while the IC 50 The value is 2.938nM.

[0098] In addition, the cell viability of two blood tumor cell lines treated with compound 2 and the laboratory's existing CDK9 positive control 1604810-84-5 was evaluated. The cell viability assay showed that compound 2 significantly reduced the cell viability of the tested blood tumor cell lines, with an efficacy comparable to that of 1604810-84-5. Together, these findings indicate that the selected compounds inhibit CDK9 at the molecular level and exhibit potent biological activity in cell-based assays, highlighting their potential as therapeutic agents for acute myeloid leukemia.

[0099] About Molecular Dynamics

[0100] The initial structure of the complex between CDK9 and compound 2 was obtained from the highest docking score conformation of Vina-GPU. Classical molecular dynamics simulations were performed using the GROMACS2020.7 software package to study the stability and interaction mechanism of the complex. The protein was parameterized using the AMBER99SB-ILDN force field, and the ligand was parameterized using the AMBER99SB-ILDN force field. The complex was placed in a cubic box, ensuring a minimum distance of 1.5 nm from the edge of the box, and was dissolved using TIP3P water molecules. Sodium ions (Na + ) and chloride ions (Cl - ) to neutralize the system and reach a physiological ionic strength of 0.15 M. Energy minimization was performed using a steepest descent algorithm to eliminate steric collisions and poor contacts. This process was complete when the maximum force dropped below 1000 kJ / mol / nm. The system went through a two-step equilibrium process:

[0101] (1) NVT balancing: 500 ps at 300 K using a constant number, volume, and temperature (NVT) combination;

[0102] (2) NPT equilibrium: A calibration of 500 ps was performed at 300 K and 1 bar using a constant number of particles, pressure and temperature (NPT) combination. A Parrinello-Rachman barometer was used to maintain the pressure. Production molecular dynamics (MD) simulations were performed for 500 ns at 300 K and 1 bar NPT conditions. The LINCS algorithm constrained all bond lengths with a time step of 2 fs. Long-range electrostatic interactions were calculated using the Particle Mesh Ewald (PME) method. The MD trajectories were analyzed using GROMACS built-in tools to calculate the root mean square deviation (RMSD), hydrogen bonds and gyration radius (Rg). The calculation formula for Rg is as follows:

[0103]

[0104] Among them, m i is the mass of the ith particle, R i (y) is the coordinate of the ith particle in the y direction, and R i (z) is the coordinate of the ith particle in the Z direction.

[0105] The binding free energy of the complex was estimated using the molecular mechanics / Poisson–Boltzmann surface area (MM / PBSA) method, and the free energy was decomposed for each residue by APBS.

[0106] ΔG binding =ΔE MM +ΔG solv -TΔS

[0107] Specifically, combined FIG. 5A to FIG. 5D As shown, in order to further analyze the binding mode of compound 2 with CDK9, a 500ns molecular dynamics simulation was performed using GROMACS 2020.7. After 300ns of simulation, the binding of compound 2 to CDK9 tended to be stable. The conformation of the complex was extracted from the last frame of the simulation for further interaction analysis. It was found that compound 2 formed a hydrogen bond with GLY25 in the active pocket of CDK9 and two hydrogen bonds with CYS106, indicating that compound 2 has a high binding affinity with CDK9. In addition, compound 2 was observed to form strong hydrophobic interactions with ILE25, VAL33, and LEU156 in the active pocket of CDK9, providing detailed structural information for subsequent compound optimization. This comprehensive MD simulation scheme provides detailed insights into the dynamic behavior and interaction mechanism of the CDK9 inhibitor complex, which helps to understand its binding stability and potential as a therapeutic agent.

[0108] According to an embodiment of the present invention, Figure 6 As shown, an embodiment of the present invention further provides a CDK9 inhibitor virtual screening device based on deep learning, and the CDK9 inhibitor virtual screening device based on deep learning 600 comprises:

[0109] Acquisition clustering module 602: used to acquire CDK9 active compound data, clean the data, determine a CDK9 verification data set, calculate the Morgan fingerprint and Jaccard distance of each compound in the CDK9 verification data set, and cluster the compounds to form multiple clusters;

[0110] Determine generation module 604: for selecting pIC in the cluster 50 The highest compound is determined as a representative compound, and a decoy molecule is generated based on the representative compound;

[0111] Evaluation and determination module 606: used to determine the deep learning algorithm according to the five-fold cross-validation evaluation result and determine the molecular docking program according to the molecular docking program evaluation result;

[0112] Virtual screening module 608: used to perform virtual screening according to the determined deep learning algorithm and the determined molecular docking program, and use RDKit to perform similarity scoring and drug-likeness evaluation on the virtual screening results to obtain quasi-target molecules; and

[0113] Detection and evaluation module 610: used to detect the in vitro inhibitory activity of the quasi-target molecule on CDK9 to obtain the final target molecule; and to evaluate the biological effect of the final target molecule.

[0114] 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.

[0115] According to an embodiment of the present invention, the present invention also provides an electronic device, a readable storage medium and a computer program product.

[0116] According to an embodiment of the present invention, the present invention provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute any one of the above methods.

[0117] According to an embodiment of the present invention, the present invention provides a computer program product, which includes: a computer program, the computer program is stored in a readable storage medium, at least one processor of an electronic device can read the computer program from the readable storage medium, and at least one processor executes the computer program so that the electronic device executes the solution provided by any of the above embodiments.

[0118] According to an embodiment of the present invention, the present invention further provides an electronic device, Figure 7 A schematic block diagram of an example electronic device 900 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0119] like Figure 7As shown, the device 900 includes a computing unit 901, which can perform various appropriate actions and processes according to a computer program stored in a ROM 902 or a computer program loaded from a storage unit into a RAM 903. In the RAM 903, various programs and data required for the operation of the device 900 can also be stored. The computing unit 901, the ROM 902, and the RAM 903 are connected to each other via a bus 904. An I / O interface 905 is also connected to the bus 904.

[0120] A number of components in the device 900 are connected to the I / O interface 905, including: an input unit 906, such as a keyboard, a mouse, etc.; an output unit 907, such as various types of displays, speakers, etc.; a storage unit 908, such as a disk, an optical disk, etc.; and a communication unit 909, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 909 allows the device 900 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0121] The computing unit 901 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 901 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 901 performs the various methods and processes described above, such as the problem handling method. For example, in some embodiments, the problem handling method may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 908. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 900 via the ROM 902 and / or the communication unit 909. When the computer program is loaded into the RAM 903 and executed by the computing unit 901, one or more steps of the problem handling method described above may be performed. Alternatively, in other embodiments, the computing unit 901 may be configured to perform the problem handling method in any other appropriate manner (e.g., by means of firmware).

[0122] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0123] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.

[0124] In the context of the present invention, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0125] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube)) or an LCD for displaying information to the user; and a keyboard and pointing device through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback; and input from the user can be received in any form.

[0126] The systems and techniques described herein can be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components, or a computing system that includes frontend components, or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by digital data communication (e.g., a communication network) in any form or medium. Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0127] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services. The server may also be a server of a distributed system, or a server combined with a blockchain.

[0128] The embodiments of the present invention establish an inhibitor discovery method that combines deep learning algorithms and molecular docking technology. This method provides a new class of CDK9 inhibitors. The method includes: constructing a comprehensive CDK9 data set, applying various molecular fingerprints, selecting the best deep learning algorithm and molecular docking program for virtual screening of CDK9 inhibitor discovery. In addition, the compound can also significantly reduce the cell viability of blood tumor cell lines through cell viability assays. Furthermore, the embodiments of the present invention also perform molecular dynamics simulations to analyze the binding stability and interaction of compound 2 with CDK9 protein. These simulations confirm the high binding affinity of compound 2. These promising results at both the molecular and cellular levels indicate that compound 2 has great potential for further development as a therapeutic agent. The embodiments of the present invention provide a reference for the development of CDK9 inhibitors with better efficacy and fewer side effects for the treatment of acute myeloid leukemia.

[0129] 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 virtual screening method for CDK9 inhibitors based on deep learning, characterized in that: include: acquiring CDK9 active compound data, cleaning the data, determining a CDK9 validation data set, calculating a Morgan fingerprint and a Jaccard distance for each compound in the CDK9 validation data set, and clustering the compounds to form a plurality of clusters; Select the pIC in the cluster 50 The highest compound is determined as a representative compound, and a bait molecule is generated based on the representative compound; Determine the deep learning algorithm based on the five-fold cross-validation evaluation results and determine the molecular docking program based on the molecular docking program evaluation results; Perform virtual screening based on a determined deep learning algorithm and a determined molecular docking program, and use RDKit to perform similarity scoring and drug-likeness evaluation on the virtual screening results to obtain quasi-target molecules; detecting the in vitro inhibitory activity of the quasi-target molecule on CDK9 to obtain the final target molecule; The biological effects of the final target molecule are evaluated.

2. The deep learning-based virtual screening method for CDK9 inhibitors according to claim 1, wherein: Also includes: The CDK9 inhibitory effect of the final target molecule was simulated and verified by molecular dynamics.

3. The deep learning-based virtual screening method for CDK9 inhibitors according to claim 2, wherein: The evaluated deep learning algorithms include eight deep learning algorithms: RF, SVR, XGBoost, MLP, Bayesian, GradientBoosting, LGBM and K-NN; the evaluated molecular docking programs include two molecular docking programs: Vina-GPU molecular docking program and AutoDock-GPU molecular docking program; the determined deep learning algorithms include at least one of the eight evaluated deep learning algorithms, and the determined molecular docking programs include at least one of the two evaluated molecular docking programs.

4. The deep learning-based virtual screening method for CDK9 inhibitors according to claim 3, wherein: The determined deep learning algorithm is the RF deep learning algorithm, and the determined molecular docking program is the Vina-GPU molecular docking program.

5. The deep learning-based virtual screening method for CDK9 inhibitors according to claim 4, wherein: The virtual screening includes at least two steps, wherein the first virtual screening step is screening through the RF deep learning algorithm, and the second virtual screening step is screening again through the Vina-GPU molecular docking program.

6. The deep learning-based virtual screening method for CDK9 inhibitors according to claim 5, wherein: The first virtual screening step is screening through the RF deep learning algorithm, and also includes: using Hyperopt to perform hyperparameter search, and adjusting the RF deep learning algorithm using the following hyperparameter values: n_estimators is set to 380, max_depth is set to 41, min_samples_split is set to 2, min_samples_leaf is set to 1, and max_features is set to 0.1562.

7. The deep learning-based virtual screening method for CDK9 inhibitors according to claim 6, wherein: The quasi-target molecules are respectively a first quasi-target molecule and a second quasi-target molecule.

8. The deep learning-based virtual screening method for CDK9 inhibitors according to claim 7, wherein: The detection of the in vitro inhibitory activity of the quasi-target molecule on CDK9 comprises: the IC 50 The value is 31.28nM.

9. The deep learning-based virtual screening method for CDK9 inhibitors according to claim 8, wherein: The second quasi-target molecule is the final target molecule, and its chemical formula is:

10. A virtual screening device for CDK9 inhibitors based on deep learning, characterized in that: include: A clustering module is used to obtain CDK9 active compound data, clean the data, determine a CDK9 validation data set, calculate the Morgan fingerprint and Jaccard distance of each compound in the CDK9 validation data set, and cluster the compounds to form multiple clusters; Determine the generation module: used to select the pIC in the cluster 50 The highest compound is determined as a representative compound, and a bait molecule is generated based on the representative compound; Evaluation and determination module: used to determine the deep learning algorithm based on the five-fold cross-validation evaluation results and determine the molecular docking program based on the molecular docking program evaluation results; Virtual screening module: used to perform virtual screening based on a determined deep learning algorithm and a determined molecular docking program, and use RDKit to perform similarity scoring and drug-likeness evaluation on the virtual screening results to obtain quasi-target molecules; as well as Detection and evaluation module: used to detect the in vitro inhibitory activity of the quasi-target molecule on CDK9 and obtain the final target molecule; The biological effects of the final target molecule are evaluated.

11. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claim 19.

12. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 19.

13. A computer program product comprising a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 19.

Citation Information

Cited By

  • Compound binding stability evaluation model based on Gaussian mixture and method thereof

    CN121393588A