Drug affinity prediction method, apparatus, processing device, program product, and medium

Abstract

This application provides a method, apparatus, processing device, program product, and medium for predicting drug affinity, applied in the field of drug development technology. The method includes: acquiring first information, including the three-dimensional structure information of a target molecule and the three-dimensional structure information of a ligand; determining the quantum effect priority of each atom within the binding pocket of the target molecule based on the first information, whereby the quantum effect priority characterizes the degree of contribution of an atom to the quantum effect in the interaction between the target molecule and the ligand; dividing each atom into at least two computational domains based on the quantum effect priority of each atom, whereby the computational domains include a quantum computing region and a classical computing region; and allocating computational resources according to the at least two computational domains to obtain a predicted value of the drug affinity between the target molecule and the ligand. This method solves the problem of the difficulty in accurately predicting drug affinity in existing technologies.

Description

Technical Field

[0001] This application relates to the field of drug development technology, specifically to a method, apparatus, processing equipment, program product, and medium for predicting drug affinity. Background Technology

[0002] In the field of new drug development, tens of thousands of new drugs are developed every year. If traditional experimental techniques are used to find and determine the target, it is necessary to find the corresponding target for each new drug and determine the affinity, which is a huge challenge for researchers.

[0003] However, existing technologies struggle to accurately predict drug affinity. Summary of the Invention

[0004] The purpose of this application is to provide a method, apparatus, processing equipment, program product and medium for predicting drug affinity, so as to solve the problem that the prior art is difficult to achieve accurate prediction of drug affinity.

[0005] One embodiment of this application provides a method for predicting drug affinity, the method comprising: Obtain first information, which includes: the three-dimensional structural information of the target molecule and the three-dimensional structural information of the ligand; Based on the first information, the quantum effect priority of each atom in the binding pocket of the target molecule is determined, and the quantum effect priority is used to characterize the degree to which the atom contributes to the quantum effect in the interaction between the target molecule and the ligand. Based on the quantum effect priority of each atom, each atom is divided into at least two computational domains, the computational domains including: a quantum computational domain and a classical computational domain; The predicted drug affinity between the target molecule and the ligand is obtained by scheduling computational resources according to the at least two computational domains.

[0006] Optionally, determining the quantum effect priority of each atom within the binding pocket of the target molecule based on the first information includes: Based on the first information, obtain the binding pocket of the target molecule; Using each atom within the binding pocket as a node and the chemical bonds between the atoms as edges, a molecular graph corresponding to the binding pocket is constructed, and an initial feature vector is assigned to each node in the molecular graph. The molecular diagram is input into a quantum effect priority prediction network to obtain the quantum effect priority of each atom; In the quantum effect priority prediction network, the first... The node at the th The feature vector of layer +1 is obtained by aggregating a first feature vector and at least one second feature vector, wherein the first feature vector is the... The node at the th The feature vector of the layer, the second feature vector is the first layer. The node at the th Feature vectors of neighboring nodes of the layer and All are integers greater than or equal to 1.

[0007] Optionally, dividing each atom into at least two computational domains based on the quantum effect priority of each atom includes at least one of the following: Atoms with a quantum effect priority greater than or equal to a first threshold are assigned to the quantum computing region; Atoms with a quantum effect priority less than a first threshold are assigned to the classical computing region; The first threshold is set according to the distribution of quantum effect priority of each atom.

[0008] Optionally, the step of scheduling computing resources according to the at least two computing domains to obtain the predicted drug affinity between the target molecule and the ligand includes: Based on the characteristic information of the quantum computing region, molecular dynamics simulation parameters are generated, wherein the characteristic information includes: scale information and / or chemical complexity information; Based on the molecular dynamics simulation parameters, conformational sampling is performed to obtain at least two conformational snapshots, wherein the conformational sampling includes enhanced sampling of the quantum computing region; Based on the second information of the at least two conformational snapshots, N target conformational snapshots are selected from the at least two conformational snapshots, where N is a positive integer, and the second information includes: energy information and structural information; Based on the N target conformation snapshots, determine the enthalpy change and entropy change of the binding process, wherein the binding process is the process by which the target molecule and the ligand bind from the free state to form a complex; Based on the enthalpy change, a convergence determination is made; If the convergence condition is met, the predicted value of the drug affinity is determined based on the enthalpy change and the entropy change. The convergence condition includes: the standard deviation of the enthalpy change is less than a second threshold.

[0009] Optionally, determining the enthalpy change and entropy change of the binding process based on the N target conformation snapshots includes: A Hamiltonian is constructed for each of the target conformation snapshots, the Hamiltonian including: a quantum mechanical Hamiltonian and a molecular mechanical Hamiltonian; The total Hamiltonian is determined based on the Hamiltonian of the N target conformation snapshots. Based on the variable quantum algorithm, the ground state electron energy of each target conformation snapshot is determined according to the total Hamiltonian; Based on the Boltzmann weighted ensemble average algorithm, the ensemble average enthalpy of the bound state and the ensemble average enthalpy of the unbound state are determined. The enthalpy change of the bonding process is determined based on the ensemble average enthalpy of the bonded state and the ensemble average enthalpy of the unbonded state. The entropy change of the combination process is determined based on the interaction entropy method.

[0010] Optionally, the method further includes: If the convergence condition is not met, execute the target operation and re-execute the target steps until the convergence condition is met. The target step is: to schedule computing resources according to the at least two computing domains to obtain the predicted value of drug affinity between the target molecule and the ligand; The target operation includes at least one of the following: Expanding the boundaries of the quantum computing domain; Increase the sampling duration.

[0011] One embodiment of this application also provides a drug affinity prediction device, the device comprising: The information acquisition module is used to acquire first information, which includes: the three-dimensional structural information of the target molecule and the three-dimensional structural information of the ligand; A first processing module is configured to determine the quantum effect priority of each atom within the binding pocket of the target molecule based on the first information, wherein the quantum effect priority is used to characterize the degree to which the atom contributes to the quantum effect in the interaction between the target molecule and the ligand. The second processing module is used to divide each of the atoms into at least two computational domains according to the quantum effect priority of each atom, wherein the computational domains include: a quantum computing region and a classical computing region; The third processing module is used to schedule computing resources according to the at least two computing domains to obtain the predicted value of drug affinity between the target molecule and the ligand.

[0012] One embodiment of this application also provides a processing device, which includes a processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the drug affinity prediction method as described in any of the preceding claims.

[0013] One embodiment of this application also provides a computer program product, including computer instructions that, when executed by a processor, implement the steps of the drug affinity prediction method as described in any of the preceding claims.

[0014] One embodiment of this application also provides a readable storage medium, wherein the readable storage medium stores a program that, when executed by a processor, implements the steps in the drug affinity prediction method as described in any of the preceding claims.

[0015] At least one of the above-mentioned technical solutions of this application has the following beneficial effects: In the drug affinity prediction method of this application embodiment, by calculating the priority of quantum effects, key atoms that contribute significantly to quantum effects can be accurately identified, overcoming the problems of wasted computing power or omission of key regions in traditional schemes; based on the priority of quantum effects, atoms are dynamically divided into quantum computing regions and classical computing regions, which can realize the on-demand allocation and scheduling of computing resources, thereby reducing the computing dimension and time cost while ensuring prediction accuracy, and achieving the optimal balance between accuracy and computing power consumption in drug affinity prediction. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a drug affinity prediction method according to one embodiment of this application. Figure 2 This is a flowchart illustrating a drug affinity prediction method according to another embodiment of this application; Figure 3 This is a structural diagram of a drug affinity prediction system according to one embodiment of this application; Figure 4 This is a schematic diagram of the priority prediction of the Abl kinase-imatinib complex by the Quantum Effect Priority Prediction Network (QEP-Net) according to one embodiment of this application. Figure 5 This is a schematic diagram of the structure of a drug affinity prediction device according to one embodiment of this application. Detailed Implementation

[0017] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and are not used to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same class, and the number of objects is not limited; for example, the first object can be one or more.

[0018] In various embodiments of the present invention, it should be understood that the sequence number of each process described below does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0019] In the embodiments of this application, the term "and / or" describes the relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following associated objects have an "or" relationship.

[0020] It should be understood that "B corresponding to A" means that B is associated with A, and B can be determined based on A. However, it should also be understood that determining B based on A does not mean determining B solely based on A; B can also be determined based on A and / or other information.

[0021] In the embodiments of this application, the term "multiple" refers to two or more, and other quantifiers are similar.

[0022] In addition, the terms "system" and "network" are often used interchangeably in this article.

[0023] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0024] This application addresses the problem that existing technologies struggle to accurately predict drug affinity by providing a method, apparatus, processing equipment, program product, and medium for predicting drug affinity.

[0025] like Figure 1 As shown, one embodiment of this application provides a method for predicting drug affinity, including: Step 101: Obtain first information, which includes: three-dimensional structural information of the target molecule and three-dimensional structural information of the ligand.

[0026] Here, the target molecule is a biomolecule or functional molecule with a recognizable binding pocket or binding region. For example, the target molecule can be a protein, enzyme, receptor, ion channel, nucleic acid, viral protein, artificial enzyme, or other biomolecule with a binding pocket; the ligand can be a small molecule drug, candidate compound, peptide, nucleotide analog, or other molecule capable of binding to the target molecule.

[0027] Step 102: Based on the first information, determine the quantum effect priority of each atom within the binding pocket of the target molecule. The quantum effect priority characterizes the degree to which the atom contributes to the quantum effect in the interaction between the target molecule and the ligand. In other words, the quantum effect priority characterizes the contribution of the atom to the quantum mechanical effect in the binding energy during the binding process of the target molecule and the ligand, or it characterizes the priority of the atom in requiring description using quantum mechanical methods. That is, the larger the quantum effect priority value, the more accurately the atom needs to be described using high-precision quantum mechanical methods to more accurately calculate its contribution to the overall binding energy.

[0028] Here, the strength of the interaction between the target molecule and the ligand is quantitatively assessed through quantum effect priority.

[0029] Step 103: Based on the quantum effect priority of each atom, divide each atom into at least two computational domains, the computational domains including: a quantum computing region and a classical computing region.

[0030] It should be noted that the quantum mechanics (QM) region is a critical core region (e.g., the region where chemical reactions or key interactions occur), and computational resources can be allocated to it preferentially, while molecular mechanics (MM) is a peripheral environment region with relatively low computational priority.

[0031] Step 104: Schedule computing resources according to the at least two computing domains to obtain the predicted drug affinity between the target molecule and the ligand.

[0032] In other words, by dividing computing domains, intelligent allocation of computing resources can be achieved for different computing domains, thereby achieving an effective balance between computing accuracy and cost.

[0033] In this embodiment, by calculating the priority of quantum effects, key atoms that contribute significantly to quantum effects can be accurately identified, overcoming the problems of wasted computing power or omission of key regions in traditional schemes. Based on the priority of quantum effects, atoms are dynamically divided into quantum computing regions and classical computing regions, which can realize the on-demand allocation and scheduling of computing resources. This reduces the computational dimension and time cost while ensuring prediction accuracy, achieving the optimal balance between accuracy and computing power consumption in drug affinity prediction.

[0034] like Figure 2 As shown, in some embodiments, step 102, namely determining the quantum effect priority of each atom within the binding pocket of the target molecule based on the first information, includes the following steps: Step 1021: Obtain the binding pocket of the target molecule based on the first information.

[0035] Here, the binding pocket (also known as the active pocket or active binding pocket) can be identified based on the initial information input by the user (including the three-dimensional structure information of the target molecule and the three-dimensional structure information of the ligand). For example, a composite algorithm based on distance and solvent accessibility can be used to automatically identify and precisely define the binding pocket region of the target molecule.

[0036] Step 1022: Using each atom in the binding pocket as a node and the chemical bonds between the atoms as edges, construct a molecular graph corresponding to the binding pocket, and assign an initial feature vector to each node in the molecular graph.

[0037] For example, the atomic connections and chemical properties (atom type, charge, hybridization, etc.) of the binding pocket can be used to construct a molecular graph: G=(V,E), where V represents the set of atomic nodes and E represents the set of chemical bonds. Furthermore, based on the atomic type, charge, hybridization, and other chemical properties of the binding pocket, each node (i∈V) can be assigned an initial feature vector. .

[0038] Step 1023: Input the molecular diagram into a quantum effect priority prediction network to obtain the quantum effect priority of each atom; wherein, in the quantum effect priority prediction network, the first... The node at the th The feature vector of layer +1 is obtained by aggregating a first feature vector and at least one second feature vector, wherein the first feature vector is the... The node at the th The feature vector of the layer, the second feature vector is the first layer. The node at the th Feature vectors of neighboring nodes of the layer and All are integers greater than or equal to 1.

[0039] It's important to note that QEP-Net is a specially designed Graph Neural Network (GNN) model trained to predict the priority of each atom's contribution to the overall binding energy in the binding pocket (i.e., predicting the quantum effect priority of atoms, or priority scores). QEP-Net not only identifies key interactions but also quantitatively assesses which chemical environments (such as aromatic rings, halogen-containing, or metal-ion-containing regions) most require high-precision quantum mechanical descriptions. The quantum effect priorities predicted by QEP-Net can then be used to guide a multi-step physics simulation process, essentially applying GNNs as a "smart navigation module" in physics simulations.

[0040] For example, the quantum effect priority prediction network QEP-Net can update node features through multiple rounds of message passing when predicting quantum effect priority. Its core iterative expression can be represented as: ; in, Represents atoms (nodes) (that is, the first) (node) in the The feature vector of the layer, Indicates the first The node at the th The feature vector of the layer, Indicates the first The neighboring nodes of the nth node. Among them, the (nth)th node's neighboring nodes. +1) layer, the Feature vectors of each node It is due to the inherent characteristics of its superior layer. (i.e., the first feature vector) and all its neighboring nodes The information obtained after aggregating the features (i.e., the second feature vector) is determined by a learnable update function.

[0041] It should also be noted that in QEP-Net, after L layers of iteration, the final node features The input is fed into a fully connected layer, and QEP-Net outputs the quantum effect priority of each atom, which is a quantum effect priority scalar fraction. .

[0042] In some embodiments, dividing each atom into at least two computational domains based on the quantum effect priority of each atom includes at least one of the following: (1) Assign atoms with a quantum effect priority greater than or equal to a first threshold to the quantum computing region; (2) Assign atoms with a quantum effect priority less than the first threshold to the classical computing region; The first threshold is set according to the distribution of quantum effect priority of each atom.

[0043] In this embodiment, based on the priority of quantum effects of atoms { }, automatically set a first threshold To satisfy all > The atoms are divided into QM regions (quantum computing regions), and the boundaries of the QM regions / MM regions are adaptively and dynamically defined.

[0044] like Figure 2 As shown, in some embodiments, step 104, which involves scheduling computing resources according to the at least two computing domains to obtain a predicted drug affinity value between the target molecule and the ligand, includes: Step 1041: Generate molecular dynamics simulation parameters based on the feature information of the quantum computing region, wherein the feature information includes: scale information and / or chemical complexity information.

[0045] Here, molecular dynamics (MD) simulation parameters can be dynamically generated based on the size and chemical complexity of the quantum computing domain. For example, specific molecular dynamics simulation parameters may include at least one of the following: sampling duration (e.g., conformation sampling duration), frequency of conformation snapshot acquisition, temperature, time step, enhancement sampling parameters, bias potential energy parameters, bias region (region where bias potential energy is applied), and number of conformations screened.

[0046] Step 1042: Based on the molecular dynamics simulation parameters, perform conformational sampling to obtain at least two conformational snapshots, wherein the conformational sampling includes enhanced sampling of the quantum computing region.

[0047] Here, Gaussian accelerated molecular dynamics (GaMD) can be used to perform more powerful enhanced sampling of regions with significant quantum effects (quantum computing regions).

[0048] Step 1043: Based on the second information of the at least two conformational snapshots, select N target conformational snapshots from the at least two conformational snapshots, where N is a positive integer. The second information includes energy information and structural information.

[0049] After the molecular dynamics simulation is completed, N target conformational snapshots can be selected from the at least two conformational snapshots based on the energy and structural information of each conformational snapshot. The energy information includes at least one of conformational potential energy, relative energy, estimated binding energy, and estimated free energy; the structural information includes at least one of structural similarity between conformational snapshots, root mean square deviation (RMSD), critical atom distance, and critical dihedral angle.

[0050] Specifically, N conformational snapshots that satisfy preset energy conditions and whose structural differences (judged based on structural information) satisfy preset difference conditions can be preferentially selected as target conformational snapshots. Among them, satisfying the preset difference conditions for structural differences includes at least one of the following: the structural similarity between conformational snapshots is lower than a preset similarity threshold; the root mean square deviation (RMSD) between conformational snapshots is greater than or equal to a preset RMSD threshold.

[0051] In other words, after the molecular dynamics simulation is completed, the energy and structural information can be combined to preferentially select N conformational snapshots with lower energy and greater structural differences as the target conformational snapshots.

[0052] Step 1044: Based on the N target conformation snapshots, determine the enthalpy change and entropy change of the binding process, wherein the binding process is the process by which the target molecule and the ligand bind from the free state to form a complex.

[0053] Step 1045: Based on the enthalpy change, perform a convergence determination; Here, we can analyze the enthalpy change. The statistical error is analyzed to determine convergence. For example, convergence can be determined by judging enthalpy change. Standard deviation Is it below the preset accuracy threshold? (i.e., the second threshold) is achieved; if it is lower than the second threshold, it means that the convergence condition is met.

[0054] Step 1046: If the convergence condition is met, determine the predicted value of drug affinity based on the enthalpy change and the entropy change. The convergence condition includes: the standard deviation of the enthalpy change is less than a second threshold.

[0055] In some specific embodiments, determining the enthalpy change and entropy change of the binding process based on the N target conformation snapshots includes: (i) Construct a Hamiltonian for each of the target conformation snapshots, wherein the Hamiltonian includes: a quantum mechanical Hamiltonian and a molecular mechanical Hamiltonian.

[0056] (ii) Determine the total Hamiltonian based on the Hamiltonian of the N target conformation snapshots.

[0057] For example, the total Hamiltonian can be calculated using the formula Calculated.

[0058] in, The total Hamiltonian of the system; The quantum mechanical Hamiltonian for high-precision description of the QM region; The molecular mechanical Hamiltonian for the classical description of the MM region; The Hamiltonian is used to describe the interactions (including electrostatics, van der Waals forces, etc.) between the QM and MM regions.

[0059] (iii) Based on the variable quantum algorithm, determine the ground state electron energy of each target conformation snapshot according to the total Hamiltonian.

[0060] Here, the ground-state electronic energy of each target conformation snapshot can be calculated by executing variable quantum algorithms such as the Adaptive Derivative-Assembled Pseudo-Trotter anxious Variational Quantum Eigensolver (ADAPT-VQE). .

[0061] Here, the enthalpy change ΔH of the bonding process can be calculated using Boltzmann weighted average, as follows: (iv) Based on the Boltzmann weighted ensemble average algorithm, determine the ensemble average enthalpy of the bound state and the ensemble average enthalpy of the unbound state.

[0062] For example, the ensemble mean enthalpy can be calculated using the following formula. :

[0063] in, This represents the Boltzmann constant, and T represents the system temperature. Indicates the first The ground-state electron energy of a target conformation snapshot.

[0064] (v) Determine the enthalpy change of the bonding process based on the ensemble average enthalpy of the bonded state and the ensemble average enthalpy of the unbonded state; For example, the enthalpy change ΔH (i.e., the quantum-classical mixed enthalpy change) can be calculated according to the formula... Confirmed. Among them, Ensemble mean enthalpy representing the state of combination , Ensemble mean enthalpy representing the unbound state Here, the bound state and the unbound state refer to the state of the target molecule and the ligand.

[0065] (vi) Determine the entropy change of the combination process based on the interaction entropy method.

[0066] For example, the entropy change during the combination process can be efficiently estimated using existing MD trajectory data based on hybrid methods such as Interaction Entropy. .

[0067] In other words, once the iterative calculation meets the convergence condition, the final results can be integrated to calculate affinity, and a report can be generated.

[0068] For example, it can be based on the core thermodynamic formula. Calculate affinity; among which, This indicates the Gibbs free energy change (drug affinity is determined by the final...) (to characterize).

[0069] In some embodiments, the method further includes: if the convergence condition is not met, performing a target operation and re-executing the target step until the convergence condition is met; wherein the target step is: scheduling computing resources according to the at least two computing domains to obtain a predicted value of drug affinity between the target molecule and the ligand.

[0070] The target operation includes at least one of the following: (1) Expand the boundary range of the quantum computing region; (2) Increase the sampling time.

[0071] It should be noted that if convergence is not achieved, the target operation is performed, which means taking optimization measures. Specifically, this may include: adjusting the boundaries of the QM / MM region, increasing the classical sampling time, or performing additional quantum computations on specific difficult conformations.

[0072] In this embodiment, an iterative feedback loop can be used to perform self-diagnosis based on the intermediate results of the calculation and trigger a new round of optimization calculations until the results converge. This feedback loop can ensure the reliability of the final result.

[0073] like Figure 3 The diagram shown is a structural diagram of a drug affinity prediction system according to one embodiment of this application. The system executes the drug affinity prediction method provided in the embodiments of this application. The system mainly includes: an input interface module, a preprocessing module, an adaptive hybrid computing scheduling engine, a classical computing module, a quantum interface module, a result analysis and iterative optimization module, and an output module.

[0074] The input interface module is primarily responsible for receiving computational tasks submitted by users. For example, users can upload initial information (such as three-dimensional structure files of proteins and ligands) through a graphical interface or API.

[0075] Preprocessing module: mainly responsible for receiving structure files and automatically performing preparatory work before calculation.

[0076] For example, the preprocessing module may specifically include a binding pocket identification unit and a quantum effect priority prediction unit. The binding pocket identification unit is used to automatically identify and precisely define the binding pocket regions of a protein using a composite algorithm based on distance and solvent accessibility. The quantum effect priority prediction unit incorporates a quantum effect priority prediction network, QEP-Net, which is used for a novel, microscopic physical prediction task—predicting the quantum effect priority of each atomic region in a molecule—and using the prediction results to guide a multi-step physical simulation process.

[0077] The adaptive hybrid computing scheduling engine is the "brain" and decision-making center of the entire automated system. Unlike traditional linear workflow engines, this scheduling engine can dynamically and intelligently allocate computing resources (such as adjusting computing parameters). For example, based on the priority distribution map output by QEP-Net (including the priority of quantum effects of each atom), the scheduling engine can adaptively determine: (1) the size and boundary of the QM region to achieve on-demand allocation of computing resources; (2) the duration of classical molecular dynamics simulation and the intensity of enhanced sampling; and (3) the quantum computing algorithm most suitable for the current QM system scale (for example, ADAPT-VQE is used for small systems, and other approximation algorithms are switched for large systems). Among them, the scheduling engine can dynamically generate MD simulation parameters and submit them to the classical computing module according to the scale and chemical complexity of the QM region.

[0078] Classical computing module: Connected to one or more classical computer clusters (such as GPU clusters), this module is responsible for executing the configuration sampling task of the scheduling engine instructions.

[0079] Quantum Interface Module: A bridge connecting classical workflows and quantum computers. This module's functions include Hamiltonian construction, task formatting and submission, and result reception and parsing. This module can automatically construct a QM / MM Hamiltonian for each conformation and submit it to the quantum computer.

[0080] The Results Analysis and Iterative Optimization module is responsible for performing convergence analysis and uncertainty assessment on the preliminary calculation results. Specifically, if this module detects that the calculation variance is too large or has not reached the preset convergence threshold, it can send a feedback signal to the scheduling engine, suggesting iterative optimization.

[0081] Output module: After the calculation converges, it is responsible for integrating the final results, calculating the affinity prediction value, and generating a visual report that includes the results, confidence assessment, and iterative optimization process.

[0082] In this embodiment, an adaptive, iterative, automated closed-loop system capable of accurately and quantitatively calculating drug affinity was constructed. This system comprises an adaptive hybrid computing scheduling engine and a result analysis and iterative optimization module, forming a complete intelligent computing closed loop. This system combines high-precision quantum computing with efficient classical sampling. Through artificial intelligence methods (such as QEP-Net), it intelligently identifies the key molecular regions (i.e., QM regions) that most require quantum mechanical description. Based on the predictions of QEP-Net, it dynamically makes decisions to optimize the allocation of computing resources and establishes an adaptive, feedback-driven workflow. It self-optimizes based on intermediate results to ensure the convergence and reliability of the calculation results, thereby addressing the shortcomings of existing technologies in quantitative accuracy, intelligent resource allocation, and robustness of the computational process.

[0083] The following example uses a classic drug target—Abl kinase and its inhibitor imatinib—to illustrate how the drug affinity prediction method provided in this application can automatically calculate drug affinity.

[0084] First, the user uploads a three-dimensional structure file of the Abl kinase-imatinib complex (e.g., PDB ID: 1IEP) through the input interface module, and the drug affinity prediction system starts the calculation task.

[0085] Then, QEP-Net prediction and adaptive QM region partitioning are performed, mainly including: (1) Pocket identification: The preprocessing module automatically identifies the ATP binding pocket where imatinib is located.

[0086] (2) QEP-Net Prediction: The Quantum Effect Priority Prediction Unit (QEP-Net) analyzes the binding pocket. Because the imatinib molecule contains multiple aromatic rings and forms key hydrogen bond interactions with amino acids in the kinase hinge region, QEP-Net assigns very high priority scores (i.e., quantum effect priority) to atoms in these regions (e.g., the MET-318 residue in the hinge region, ASP-381 in the DFG motif, and the pyrimidine and benzene rings of imatinib itself).

[0087] like Figure 4As shown, proteins (i.e., Abl kinases) are displayed as gray cartoon bands, ligands (i.e., imatinib) are displayed as blue ball-and-stick models, and QEP-Net high-priority regions are highlighted as red stick models, representing key interacting residues (e.g., MET-318, ASP-381) that are most sensitive to quantum effects and identified by QEP-Net. These regions will be preferentially classified as QM regions for high-precision calculations.

[0088] (3) Adaptive partitioning: The adaptive scheduling engine receives these high scores and, in conjunction with the preset partitioning strategy, automatically partitions all the high-scoring atoms and the imatinib molecule itself into the QM region, while other aliphatic residues in the pocket are retained as the MM region.

[0089] Then, the process of scheduling engine-driven classical conformation sampling is performed, which mainly includes: (1) Parameter settings: The scheduling engine identified that the QM region of the system is large and contains a flexible P-loop. Therefore, it adaptively set a long (e.g., 500 nanoseconds) Gaussian accelerated molecular dynamics GaMD simulation duration and increased the bias potential energy of the P-loop region to ensure that the conformation of the flexible region is fully sampled.

[0090] (2) Conformation screening: After the GaMD simulation is completed, the system selects 200 representative conformation snapshots for subsequent calculations.

[0091] Next, we will perform enthalpy change and entropy change calculations, including the following two aspects: (1) Quantum computing: The scheduling engine submits the QM / MM Hamiltonians of the 200 selected conformations to the quantum computer in batches through the quantum interface module. The quantum computer can use the ADAPT-VQE algorithm to calculate the exact energy (i.e., the ground state electron energy) of each conformation and finally calculate the ensemble-averaged binding enthalpy change. .

[0092] (2) Entropy change calculation: The scheduling engine calls the Interaction Entropy method to calculate the entropy change using the existing MD trajectory data. .

[0093] Next, results analysis and iterative decision-making are performed. For example, the specific process is as follows: (1) Initial analysis: After the first round of calculations, the results analysis and iterative optimization module found that although the overall energy converged, the standard deviation of the enthalpy change from the P-loop region conformation (i.e. the standard deviation of the binding enthalpy change) was extremely large, resulting in the standard deviation of the binding enthalpy change (e.g. 1.2 kcal / mol) being higher than the preset convergence threshold (e.g. 0.8 kcal / mol).

[0094] (2) Iterative feedback: The result analysis and iterative optimization module sends feedback instructions to the scheduling engine. For example, the feedback instruction is: "For the conformation clusters in the 'open' state of P-loop, add 100 target conformation snapshots corresponding to the QM / MM Hamiltonian construction and quantum energy calculation tasks; among them, the 100 target conformation snapshots are obtained by supplementing and screening from existing molecular dynamics trajectories, or by screening from newly added molecular dynamics trajectories after extending the sampling time." Another example is "expand the QM region".

[0095] Specifically, if the existing MD / GaMD trajectory already contains enough P-loop open-state conformations, then 100 representative conformation snapshots are selected from the existing trajectory, and QM / MM Hamiltonian construction and ADAPT-VQE energy calculation are performed on these 100 snapshots; if the existing trajectory does not have enough samples of the conformation cluster, then the GaMD sampling is extended or the sampling intensity of the conformation cluster is increased, then 100 representative conformation snapshots are selected from the newly added trajectory, and then the corresponding quantum computing task is performed.

[0096] (3) Second round of calculation: The scheduling engine executes the instruction to perform supplementary calculations. After the calculation is completed, the result analysis and iterative optimization module analyzes the results again and finds that the standard deviation of the combined enthalpy change has dropped to 0.7 kcal / mol, which is lower than the preset convergence threshold of 0.8 kcal / mol, indicating that the convergence condition is met.

[0097] Finally, the results or reports will be automatically output, for example: After the iterative calculations converge, the output module integrates the final results. and The value was used to calculate the Gibbs free energy change. The system generates a final report, which, for example, includes the following: (1) Predicted drug affinity: =-12.5±0.7 kcal / mol.

[0098] (2) Visualization of key quantum effect regions identified by QEP-Net.

[0099] (3) Iterative optimization process record: used to show the uncertainty of the first round of calculation and how the second round of supplementary calculation improved the reliability of the results.

[0100] In the above embodiments of this application, the problem of accurate calculation is solved by introducing real quantum computing. For systems containing complex quantum effects (such as metal ions and halogen bonds), the prediction accuracy of this application can reduce the mean absolute error (MAE) by 40-50% compared with the purely classical method.

[0101] The drug affinity prediction method in this application does not simply use GNN to directly predict the final result, but rather realizes a new paradigm of AI-enabled physical simulation. Through the intelligent guidance of QEP-Net, the number of atoms in the required QM region can be reduced by 20-30% while achieving the same computational accuracy, thereby reducing the cost of expensive quantum computing by more than an order of magnitude.

[0102] It should be noted that in the field of new drug development, accurately predicting the affinity between drug molecules and targets is a core bottleneck. The drug affinity prediction method of this application directly addresses this core bottleneck, providing pharmaceutical companies and research institutions with a revolutionary computational tool that can significantly shorten the development cycle and reduce experimental trial-and-error costs. Especially in the early stages of drug discovery, it can complete virtual screening of large-scale compound libraries with extremely high efficiency and accuracy. This automated system can be deployed as a cloud service (SaaS) platform, providing new drug development companies with high-throughput, high-precision computational screening services (SaaS), significantly shortening the development cycle and reducing experimental trial-and-error costs. It can drive the entire pharmaceutical industry towards a more efficient and lower-cost intelligent R&D model, possessing significant commercial value and market demand.

[0103] In this embodiment, the drug affinity prediction system, which includes a QEP-Net prediction module, an adaptive hybrid computing scheduling engine, and a result analysis and iterative optimization module, not only achieves end-to-end automation, but also realizes true intelligence, adaptive decision-making, and closed-loop optimization through QEP-Net and the iterative optimization module. It can solve complex problems in actual scientific research calculations and ensure the accuracy and reliability of the final results.

[0104] The drug affinity prediction method described in this application first uses a QEP-Net to intelligently predict the priority of quantum effects in each region of the molecule to determine which regions most require high-precision quantum mechanical descriptions. Then, based on the prediction results of the QEP-Net, an adaptive hybrid computing scheduling engine is guided to allocate computing resources dynamically and optimally. Finally, through an iterative feedback loop, the convergence and reliability of affinity calculation results are ensured. This reduces the probability of requiring expert intervention and rework by about 80%, greatly improving the robustness and automation level of the calculation.

[0105] like Figure 5 As shown, one embodiment of this application also provides a drug affinity prediction device, the device comprising: The information acquisition module 510 is used to acquire first information, which includes: the three-dimensional structure information of the target molecule and the three-dimensional structure information of the ligand; The first processing module 520 is configured to determine the quantum effect priority of each atom in the binding pocket of the target molecule based on the first information, wherein the quantum effect priority is used to characterize the degree of contribution of the atom to the quantum effect in the interaction between the target molecule and the ligand. The second processing module 530 is used to divide each of the atoms into at least two computational domains according to the quantum effect priority of each atom, wherein the computational domains include: a quantum computing region and a classical computing region; The third processing module 540 is used to schedule computing resources according to the at least two computing domains to obtain the predicted value of drug affinity between the target molecule and the ligand.

[0106] In this embodiment, by calculating the priority of quantum effects, key atoms that contribute significantly to quantum effects can be accurately identified, overcoming the problems of wasted computing power or omission of key regions in traditional schemes. Based on the priority of quantum effects, atoms are dynamically divided into quantum computing regions and classical computing regions, which can realize the on-demand allocation and scheduling of computing resources. This reduces the computational dimension and time cost while ensuring prediction accuracy, achieving the optimal balance between accuracy and computing power consumption in drug affinity prediction.

[0107] Optionally, the first processing module 520 includes: The first processing unit is configured to obtain the binding pocket of the target molecule based on the first information; The second processing unit is used to construct a molecular graph corresponding to the binding pocket by taking each atom in the binding pocket as a node and the chemical bonds between each atom as edges, and to assign an initial feature vector to each node in the molecular graph. The third processing unit is used to input the molecular diagram into the quantum effect priority prediction network to obtain the quantum effect priority of each atom; In the quantum effect priority prediction network, the first... The node at the th The feature vector of layer +1 is obtained by aggregating a first feature vector and at least one second feature vector, wherein the first feature vector is the... The node at the th The feature vector of the layer, the second feature vector is the first layer. The node at the th Feature vectors of neighboring nodes of the layer and All are integers greater than or equal to 1.

[0108] Optionally, the second processing module 530 includes: The first partitioning unit is used to partition atoms whose quantum effect priority is greater than or equal to a first threshold into the quantum computing region; The second partitioning unit is used to partition atoms whose quantum effect priority is less than the first threshold into the classical computing region; The first threshold is set according to the distribution of quantum effect priority of each atom.

[0109] Optionally, the third processing module 540 includes: The parameter generation unit is used to generate molecular dynamics simulation parameters based on the characteristic information of the quantum computing region, wherein the characteristic information includes: scale information and / or chemical complexity information; A conformation sampling unit is used to perform conformation sampling based on the molecular dynamics simulation parameters to obtain at least two conformation snapshots, wherein the conformation sampling includes enhanced sampling of the quantum computing region; The conformational filtering unit is used to filter out N target conformational snapshots from the at least two conformational snapshots based on the second information of the at least two conformational snapshots, where N is a positive integer, and the second information includes energy information and structural information. The fourth processing unit is used to determine the enthalpy change and entropy change of the binding process based on the N target conformation snapshots, wherein the binding process is the process by which the target molecule and the ligand bind from the free state to form a complex; The fifth processing unit is used to perform convergence determination based on the enthalpy change; The sixth processing unit is used to determine the predicted value of drug affinity based on the enthalpy change and the entropy change if the convergence condition is met. The convergence condition includes: the standard deviation of the enthalpy change is less than a second threshold.

[0110] Optionally, the fourth processing unit includes: The first processing subunit is used to construct a Hamiltonian for each of the target conformation snapshots, the Hamiltonian including: a quantum mechanical Hamiltonian and a molecular mechanical Hamiltonian; The second processing subunit is used to determine the total Hamiltonian based on the Hamiltonians of the N target conformation snapshots. The third processing subunit is used to determine the ground state electron energy of each target conformation snapshot based on the total Hamiltonian using a variable quantum algorithm. The fourth processing subunit is used to determine the ensemble average enthalpy of the bound state and the unbound state based on the Boltzmann weighted ensemble average algorithm. The fifth processing subunit is used to determine the enthalpy change of the bonding process based on the ensemble average enthalpy of the bonding state and the ensemble average enthalpy of the non-bonding state. The sixth processing subunit is used to determine the entropy change of the combination process based on the interaction entropy method.

[0111] Optionally, the device further includes: The ninth processing subunit is used to execute the target operation and re-execute the target steps if the convergence condition is not met, until the convergence condition is met. The target step is: to schedule computing resources according to the at least two computing domains to obtain the predicted value of drug affinity between the target molecule and the ligand; The target operation includes at least one of the following: Expanding the boundaries of the quantum computing domain; Increase the sampling duration.

[0112] The apparatus provided in this application embodiment can execute the above method embodiment, and its implementation principle and technical effect are similar, so it will not be described again here.

[0113] It should be noted that the division of units in the embodiments of this application is illustrative and only represents one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units.

[0114] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a processor-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0115] One embodiment of this application also provides a processing device, which includes a processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the drug affinity prediction method as described in any of the preceding claims.

[0116] The specific implementation of the drug affinity prediction method, which is executed by the program running on the processor of the processing device, can be found in the detailed description of the drug affinity prediction method, and will not be repeated here.

[0117] In addition, specific embodiments of this application also provide a readable storage medium storing a program. When executed by a processor, this program implements the various processes of the above-described drug affinity prediction method embodiments and achieves the same technical effect. To avoid repetition, it will not be described again here. The readable storage medium can be any available medium or data storage device that the processor can access, including but not limited to magnetic memory (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO), etc.), optical memory (e.g., CD, DVD, BD, HVD, etc.), and semiconductor memory (e.g., ROM, EPROM, EEPROM, non-volatile memory (NAND FLASH), solid-state drive (SSD), etc.).

[0118] This application also provides a computer program product, including computer instructions. When executed by a processor, these computer instructions implement the various processes of the above-described drug affinity prediction method embodiments and achieve the same technical effects. To avoid repetition, they will not be described again here.

[0119] Optionally, embodiments of this application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0120] The computer program product described in this application includes computer instructions that, when executed by a processor, implement the various processes of the method embodiments shown above and achieve the same technical effects. To avoid repetition, these will not be repeated here.

[0121] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0122] It should be noted that many of the functional components described in this specification are referred to as modules / submodules in order to more specifically emphasize the independence of their implementation.

[0123] In this application embodiment, the module / submodule can be implemented in software so that it can be executed by various types of processors. For example, an identified executable code module may include one or more physical or logical blocks of computer instructions, which may be constructed as objects, procedures, or functions. Nevertheless, the executable code of the identified module does not need to be physically located together, but may include different instructions stored in different bits, which, when logically combined, constitute the module and achieve the module's intended purpose.

[0124] In practice, an executable code module can be a single instruction or many instructions, and can even be distributed across multiple different code segments, different programs, and across multiple memory devices. Similarly, operational data can be identified within the module and can be implemented in any suitable form and organized within any suitable data structure. This operational data can be collected as a single dataset or distributed across different locations (including different storage devices), and can exist, at least in part, solely as electronic signals within the system or network.

[0125] When a module can be implemented using software, considering the current level of hardware technology, modules that can be implemented in software can be implemented using hardware circuits by those skilled in the art to achieve the corresponding functions, without considering cost. These hardware circuits include conventional very-large-scale integrated circuits (VLSI) or gate arrays, as well as existing semiconductors such as logic chips and transistors, or other discrete components. Modules can also be implemented using programmable hardware devices, such as field-programmable gate arrays, programmable array logic, and programmable logic devices.

[0126] The above describes the preferred embodiments of this application. It should be noted that those skilled in the art can make several improvements and modifications without departing from the principles described in this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method of drug affinity prediction, characterized by, The method includes: Obtain first information, which includes: the three-dimensional structural information of the target molecule and the three-dimensional structural information of the ligand; Based on the first information, the quantum effect priority of each atom in the binding pocket of the target molecule is determined, and the quantum effect priority is used to characterize the degree to which the atom contributes to the quantum effect in the interaction between the target molecule and the ligand. Based on the quantum effect priority of each atom, each atom is divided into at least two computational domains, the computational domains including: a quantum computational domain and a classical computational domain; The predicted drug affinity between the target molecule and the ligand is obtained by scheduling computational resources according to the at least two computational domains.

2. The drug affinity prediction method according to claim 1, characterized in that, The step of determining the quantum effect priority of each atom within the binding pocket of the target molecule based on the first information includes: Based on the first information, obtain the binding pocket of the target molecule; Using each atom within the binding pocket as a node and the chemical bonds between the atoms as edges, a molecular graph corresponding to the binding pocket is constructed, and an initial feature vector is assigned to each node in the molecular graph. The molecular diagram is input into a quantum effect priority prediction network to obtain the quantum effect priority of each atom; In the quantum effect priority prediction network, the first... The node at the th The feature vector of layer +1 is obtained by aggregating a first feature vector and at least one second feature vector, wherein the first feature vector is the... The node at the th The feature vector of the layer, the second feature vector is the first layer. The node at the th The feature vectors of the neighboring nodes of the layer, and All are integers greater than or equal to 1.

3. The drug affinity prediction method according to claim 1, characterized in that, The division of each atom into at least two computational domains based on the quantum effect priority of each atom includes at least one of the following: Atoms with a quantum effect priority greater than or equal to a first threshold are assigned to the quantum computing region; Atoms with a quantum effect priority less than a first threshold are assigned to the classical computing region; The first threshold is set according to the distribution of quantum effect priority of each atom.

4. The drug affinity prediction method according to claim 1, characterized in that, The step of scheduling computing resources according to the at least two computing domains to obtain the predicted drug affinity between the target molecule and the ligand includes: Based on the characteristic information of the quantum computing region, molecular dynamics simulation parameters are generated, wherein the characteristic information includes: scale information and / or chemical complexity information; Based on the molecular dynamics simulation parameters, conformational sampling is performed to obtain at least two conformational snapshots, wherein the conformational sampling includes enhanced sampling of the quantum computing region; Based on the second information of the at least two conformational snapshots, N target conformational snapshots are selected from the at least two conformational snapshots, where N is a positive integer, and the second information includes: energy information and structural information; Based on the N target conformation snapshots, determine the enthalpy change and entropy change of the binding process, wherein the binding process is the process by which the target molecule and the ligand bind from the free state to form a complex; Based on the enthalpy change, a convergence determination is made; If the convergence condition is met, the predicted value of the drug affinity is determined based on the enthalpy change and the entropy change. The convergence condition includes: the standard deviation of the enthalpy change is less than a second threshold.

5. The drug affinity prediction method according to claim 4, characterized in that, The step of determining the enthalpy change and entropy change of the binding process based on the N target conformation snapshots includes: A Hamiltonian is constructed for each of the target conformation snapshots, the Hamiltonian including: a quantum mechanical Hamiltonian and a molecular mechanical Hamiltonian; The total Hamiltonian is determined based on the Hamiltonian of the N target conformation snapshots. Based on the variable quantum algorithm, the ground state electron energy of each target conformation snapshot is determined according to the total Hamiltonian; Based on the Boltzmann weighted ensemble average algorithm, the ensemble average enthalpy of the bound state and the ensemble average enthalpy of the unbound state are determined. The enthalpy change of the bonding process is determined based on the ensemble average enthalpy of the bonded state and the ensemble average enthalpy of the unbonded state. The entropy change of the combination process is determined based on the interaction entropy method.

6. The drug affinity prediction method according to claim 4, characterized in that, The method further includes: If the convergence condition is not met, execute the target operation and re-execute the target steps until the convergence condition is met. The target step is: to schedule computing resources according to the at least two computing domains to obtain the predicted value of drug affinity between the target molecule and the ligand; The target operation includes at least one of the following: Expanding the boundaries of the quantum computing domain; Increase the sampling duration.

7. A drug affinity prediction device, characterized in that, The device includes: The information acquisition module is used to acquire first information, which includes: the three-dimensional structural information of the target molecule and the three-dimensional structural information of the ligand; A first processing module is configured to determine the quantum effect priority of each atom within the binding pocket of the target molecule based on the first information, wherein the quantum effect priority is used to characterize the degree to which the atom contributes to the quantum effect in the interaction between the target molecule and the ligand. The second processing module is used to divide each of the atoms into at least two computational domains according to the quantum effect priority of each atom, wherein the computational domains include: a quantum computing region and a classical computing region; The third processing module is used to schedule computing resources according to the at least two computing domains to obtain the predicted value of drug affinity between the target molecule and the ligand.

8. A processing apparatus, characterized in that, It includes a processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the drug affinity prediction method as described in any one of claims 1 to 6.

9. A computer program product, characterized in that, It includes computer instructions that, when executed by a processor, implement the steps of the drug affinity prediction method as described in any one of claims 1 to 6.

10. A readable storage medium, characterized in that, The readable storage medium stores a program that, when executed by a processor, implements the steps of the drug affinity prediction method as described in any one of claims 1 to 6.

Images