Method and system for predicting membrane protease-mediated co-metabolic transformation of small molecules, and electronic device

Abstract

The application relates to the field of machine learning, and provides a membrane protein enzyme-mediated small molecule co-metabolic conversion prediction method and system and an electronic device. The method comprises the following steps: constructing a simulation system of a target membrane protein, obtaining the structure of a substrate and the structure of a to-be-tested small molecule, performing molecular docking to obtain a candidate initial binding conformation; performing molecular dynamics simulation, and extracting dynamic time sequence features based on a trajectory; after screening the dynamic time sequence features and extracted static features, splicing is performed to obtain a data set; a pre-constructed machine learning model is trained based on a training set; a test set is input into the trained machine learning model, and a membrane protein enzyme-mediated small molecule co-metabolic conversion prediction result is obtained. The application can solve the defects of inaccurate prediction, redundant features and unknown mechanism in the related art, and can realize efficient and accurate prediction of the membrane protein enzyme-mediated small molecule co-metabolic conversion process.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and in particular to a method, system, and electronic device for predicting small molecule co-metabolism transformation mediated by membrane proteases. Background Technology

[0002] Membrane proteins are widely involved in key physiological processes such as transmembrane transport of substances, signal transduction, substrate recognition, and catalytic transformation in organisms, and are important research subjects in environmental remediation, biotransformation, drug discovery, and enzyme engineering optimization.

[0003] Current research on small molecule transformation mediated by membrane proteins mainly relies on the following methods: ① Experimental screening: Observing whether the target small molecule is transformed through in vitro or whole-cell experiments. This method is reliable, but suffers from long cycles, high costs, and low throughput. ② Molecular docking: Using protein and small molecule structures to score docking and estimate binding capacity. However, this method is mostly based on a single static conformation, making it difficult to accurately reflect the dynamic behavior of membrane proteins in their native membrane environment. ③ Molecular dynamics simulation: Molecular dynamics simulation can provide information on the dynamic behavior of proteins and small molecules over time. This method saves experimental costs, but molecular dynamics trajectories are long-term, multivariate, and highly redundant data; direct modeling can lead to significant dimensionality pressure and overfitting risks. ④ Machine learning: Training models using static indicators such as the physicochemical properties and structural features of small molecules to predict whether they can be transformed by a certain type of protein. However, this type of method generally ignores the dynamic conformational features that change over time during protein-small molecule interactions, or cannot efficiently handle high-dimensional time series, resulting in uninterpretable dynamic features and limited prediction accuracy and mechanism explanation capabilities.

[0004] In summary, existing technologies have the following main shortcomings: ① Lack of membrane environment: Many studies investigate membrane proteins in aqueous environments, failing to consider the influence of the membrane bilayer on protein conformational stability and substrate entry pathways, leading to simulation results deviating from the true native state. ② Insufficient utilization of dynamic information: Traditional predictions rely heavily on static descriptors, failing to fully utilize the dynamic behavioral differences in protein-small molecule interactions. ③ Difficulty in efficiently modeling trajectory data: Molecular dynamics trajectories are long-term, multivariate, and highly redundant data; direct modeling would bring enormous dimensionality pressure and the risk of overfitting. ④ Inability to explain the underlying mechanisms: Existing methods often only provide prediction results, failing to explain "why the small molecule can or cannot be converted," and cannot provide clear guidance for subsequent mechanism research or process design. Summary of the Invention

[0005] This invention provides a method, system, and electronic device for predicting small molecule cometabolism mediated by membrane proteases, which addresses the shortcomings of inaccurate prediction, redundant features, and unclear mechanisms in related technologies. The solution of this application can achieve efficient and accurate prediction of small molecule cometabolism mediated by membrane proteases.

[0006] This invention provides a method for predicting small molecule cometabolism transformation mediated by membrane proteases, comprising: A simulation system for the target membrane protein was constructed, and the structures of the substrate and the small molecule to be tested were obtained. Molecular docking was then performed to obtain candidate initial binding conformations. Molecular dynamics simulations were performed on the candidate initial binding conformations, and dynamic time series features were obtained based on trajectory extraction. After filtering the dynamic time series features and the extracted static features, the data set is obtained by splicing them together. The dataset is divided into a training set and a test set, and a pre-built machine learning model is trained based on the training set. The test set is input into the trained machine learning model to obtain the prediction results of small molecule cometabolism transformation mediated by membrane protease.

[0007] According to the prediction method for membrane protease-mediated small molecule co-metabolism transformation provided by the present invention, the step of screening the dynamic time series features and the extracted static features, and then concatenating them to obtain a dataset, includes: Perform correlation analysis on the static features and remove features whose correlation meets the set criteria; Low-variance feature removal is performed on the dynamic time series features; After concatenating the static features and the dynamic time series features, a correlation analysis is performed again to remove features whose correlation meets the set criteria.

[0008] According to the prediction method for membrane protease-mediated small molecule co-metabolism transformation provided by the present invention, the steps of constructing a simulation system of the target membrane protein, obtaining the structure of the substrate and the structure of the small molecule to be tested, and performing molecular docking to obtain candidate initial binding conformations include: Obtain the three-dimensional structure and membrane localization signals of the target protein, and construct a membrane protein simulation system; To obtain the structure of the substrate and the structure of the small molecule to be tested; A stable binding conformation was obtained by molecular docking the substrate structure with the membrane protein simulation system. Molecular docking is performed between the structure of the small molecule to be tested and the stable binding conformation to determine the candidate initial binding conformation.

[0009] According to the prediction method for membrane protease-mediated small molecule cometabolism provided by the present invention, the step of performing molecular dynamics simulation on the candidate initial binding conformation and obtaining dynamic time series features based on trajectory extraction includes: The simulated system of the target membrane protein, as well as the structure of the substrate and the structure of the small molecule to be tested, are transformed into a molecular topological system; The energy of the molecular topological system is minimized, and isothermal and isobaric equilibrium is achieved. Conduct molecular dynamics simulations for a preset duration, record the coordinates and velocities of all atoms, and obtain trajectory data; Dynamic time series features are extracted from the trajectory data.

[0010] According to the prediction method for membrane protease-mediated small molecule co-metabolism transformation provided by the present invention, the method further includes obtaining dynamic time series features based on trajectory extraction, followed by: Frequency domain dimensionality reduction is performed on the dynamic time series features.

[0011] According to the prediction method for membrane protease-mediated small molecule co-metabolism transformation provided by the present invention, the frequency domain dimensionality reduction of the dynamic time series features includes: The dynamic time series features are converted into frequency domain data features; The frequency domain data is divided into several frequency bands according to the frequency range of the aforementioned frequency domain data characteristics; Based on the proportion of each frequency band, the dynamic time series features are compressed into the corresponding frequency band energy proportion.

[0012] According to the prediction method for membrane protease-mediated small molecule co-metabolic transformation provided by the present invention, the method further includes, after obtaining the prediction result of membrane protease-mediated small molecule co-metabolic transformation: Calculate the SHAP value of each feature for each sample in the dataset to obtain the SHAP matrix; Visual analysis results are generated based on the SHAP matrix, and these results are used to perform interpretability analysis on the prediction results.

[0013] According to the prediction method for membrane protease-mediated small molecule co-metabolism transformation provided by the present invention, the generation of visual analysis results based on the SHAP matrix includes: The absolute values ​​of the SHAP matrix are averaged, and a bar chart is drawn according to the descending order of features to obtain the SHAP feature importance ranking chart; Using the SHAP value of each feature as the ordinate and the original value of the feature as the abscissa, the data points of all samples are superimposed to create a SHAP scatter plot. By selecting the two features with the highest importance ranking, a SHAP dependency interaction graph of second-order interaction is drawn.

[0014] This invention also provides a prediction system for membrane protease-mediated small molecule cometabolism transformation, and a prediction method for membrane protease-mediated small molecule cometabolism transformation, comprising: The system construction module is used to construct a simulated system of the target membrane protein, obtain the structure of the substrate and the structure of the small molecule to be tested, and perform molecular docking to obtain candidate initial binding conformations; The feature extraction module is used to perform molecular dynamics simulations on the candidate initial binding conformations and obtain dynamic time series features based on trajectory extraction. The collection construction module is used to filter the dynamic time series features and the extracted static features and then concatenate them to obtain a data set; The model training module is used to divide the dataset into a training set and a test set, and to train a pre-built machine learning model based on the training set. The transformation prediction module is used to input the test set into the trained machine learning model to obtain the prediction results of small molecule co-metabolism transformation mediated by membrane protease.

[0015] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a prediction method for small molecule co-metabolism transformation mediated by any of the above-described membrane proteases.

[0016] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a prediction method for small molecule co-metabolism transformation mediated by any of the membrane proteases described above.

[0017] The present invention also provides a computer program product, comprising a computer program that, when executed by a processor, implements a prediction method for small molecule cometabolism mediated by any of the membrane proteases described above.

[0018] The method for predicting small molecule co-metabolism mediated by membrane proteases provided in this invention can efficiently generate the initial enzyme-substrate binding conformation through molecular docking. Then, using molecular dynamics simulations, the conformational change trajectory of membrane proteases in a dynamic environment is systematically captured at the atomic level, and dynamic time-series features containing catalytic mechanism information are extracted. Simultaneously, static structural features are integrated, and after feature screening and splicing, a high-quality dataset with both physical interpretability and information richness is constructed. Based on this, through the training and prediction of a machine learning model, rapid virtual screening of massive sequence spaces and substrate combinations is achieved, significantly reducing the workload and time cost of traditional experimental screening. This method effectively overcomes the bottlenecks of low throughput and long cycles of purely experimental methods, compensates for the shortcomings of traditional molecular dynamics simulations in efficiently traversing sequence spaces, and avoids the problems of lack of dynamic information and insufficient interpretability in purely data-driven models. Ultimately, it achieves efficient, accurate, and mechanistically supported prediction of small molecule co-metabolism mediated by membrane proteases. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the prediction method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the frequency domain dimensionality reduction process provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the prediction system provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the physical structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0022] Figure 1 This is a flowchart illustrating the prediction method provided in an embodiment of the present invention.

[0023] like Figure 1As shown, this embodiment provides a method for predicting membrane protease-mediated small molecule cometabolism transformation, including: Step 101: Construct a simulation system of the target membrane protein, obtain the structure of the substrate and the structure of the small molecule to be tested, and perform molecular docking to obtain candidate initial binding conformations; Specifically, it can obtain the three-dimensional structure and membrane localization signals of the target protein, and construct a membrane protein simulation system; To obtain the structure of the substrate and the structure of the small molecule to be tested; A stable binding conformation was obtained by molecular docking the substrate structure with the membrane protein simulation system. Molecular docking is performed between the structure of the small molecule to be tested and the stable binding conformation to determine the candidate initial binding conformation.

[0024] Step 102: Perform molecular dynamics simulation on the candidate initial binding conformation and obtain dynamic time series features based on trajectory extraction; The dynamic time series features obtained in this step may include, but are not limited to: Protein overall conformation shift sequence; Conformation shift sequence of small molecules relative to proteins; Small molecule self-conformation change sequence; Sequence of key residue distance changes; Sequence of distance changes between small molecules and key atoms or metal centers of active sites; A sequence of changes in the number of hydrogen bonds; Local structural domain dynamic fluctuation sequence.

[0025] During molecular dynamics simulations, the simulation system of the target membrane protein, as well as the structure of the substrate and the structure of the small molecule to be tested, can be transformed into a molecular topological system. The energy of the molecular topological system is minimized, and isothermal and isobaric equilibrium is achieved. Conduct molecular dynamics simulations for a preset duration, record the coordinates and velocities of all atoms, and obtain trajectory data; Dynamic time series features are extracted from the trajectory data.

[0026] Specifically, this may include the following steps: Topology file generation: Based on preset force field parameters (such as protein force fields like CHARMM36 and AMBER99SB, lipid force fields like Slipids or POPC, and water models like TIP3P), topology files for proteins, membranes, substrates, and small molecules to be tested are generated to ensure compatibility with molecular dynamics simulation programs. For example, protein PDB structure files obtained from the PDB protein database are converted into GRO format files so that they can run in the GROMACS program. The purpose of this step is to convert experimental structural data into molecular topological descriptions that can be recognized by the simulation program, define the mass, charge, bonding parameters, and non-bonded interaction parameters of each atom, and provide a force field basis for subsequent energy calculations and dynamic integration.

[0027] Energy minimization: The energy of the constructed system is minimized using the steepest descent method or the conjugate gradient method. The purpose of this step is to eliminate unreasonable atomic contacts, bond length / bond angle distortions, and steric hindrance conflicts that may be introduced during the system construction process, so that the system energy is reduced to a local minimum, a stable initial conformation is obtained, and the system collapses due to excessive virtual forces in subsequent dynamic simulations caused by unreasonable initial structure.

[0028] Isothermal equilibration (NVT ensemble): Pre-equilibration simulations are performed under constant particle number, constant volume, and constant temperature conditions. The system temperature is gradually increased to the target temperature (e.g., 310 K) using velocity rescaling or a Nosé-Hoover thermostat, and positional constraints are imposed on protein backbone atoms to protect the overall protein structure. The purpose of this step is to allow solvent and lipid molecules to rearrange and fully relax at the target temperature, filling gaps in the system while maintaining the integrity of the protein structure, thus providing a temperature-stable initial state for subsequent pressure equilibration.

[0029] Isobaric equilibration (NPT ensemble): Pre-equilibration simulations are performed under constant particle number, constant pressure, and constant temperature conditions. A Parrinello-Rahman or Berendsen isobaric device is used to maintain the system pressure at the target value (e.g., 1 bar), and the positional constraints on the proteins are gradually released. The purpose of this step is to adaptively adjust the simulation chamber size to a reasonable density, ensuring that parameters such as the lipid bilayer area / lipid ratio and water molecule density reach equilibrium values ​​consistent with experiments, thereby allowing the membrane protein-membrane system to reach thermodynamic steady state in its natural membrane environment.

[0030] Production Operation and Trajectory Acquisition: Perform molecular dynamics simulations of the finished product under the NPT ensemble for a preset duration (e.g., 100 ns, 500 ns, or longer), removing all positional constraints, and record the atomic coordinates and velocities of the system at preset time intervals (e.g., every 10 ps or every 100 ps). The purpose of this step is to fully sample the conformational fluctuations and dynamic behavior of the membrane protein-substrate-small molecule complex on nanosecond to microsecond timescales in a simulated environment close to physiological conditions. This includes processes such as overall protein motion, local domain fluctuations, ligand binding posture changes, and the formation and breakage of key interactions, providing rich time-resolved dynamic information for subsequent feature extraction.

[0031] Trajectory data processing and storage: The collected trajectory files are preprocessed to remove potential periodic boundary effects and preserve key parameters of the system at multiple time steps, including protein conformation, ligand positions, atomic coordinates, interaction energies (van der Waals energy, electrostatic interaction energy, etc.), hydrogen bond information, and distance information. The purpose of this step is to transform the raw binary trajectories generated by the simulation into a structured dataset that can be directly used in subsequent feature extraction steps, ensuring the accuracy and repeatability of feature extraction.

[0032] In practice, after obtaining the dynamic time series data, frequency domain dimensionality reduction can be performed on the dynamic time series data. Specifically, the dynamic time series features can be converted into frequency domain data features. The frequency domain data is divided into several frequency bands according to the frequency range of the aforementioned frequency domain data characteristics; Based on the proportion of each frequency band, the dynamic time series features are compressed into the corresponding frequency band energy proportion.

[0033] For example, frequency ranges can be divided into low-frequency, mid-frequency, and high-frequency bands. The low-frequency band represents large-scale conformational changes in proteins or ligands, the mid-frequency band represents local conformational fine-tuning, and the high-frequency band represents atomic-level local perturbations. Ultimately, each dynamic time series feature is compressed into its corresponding low-frequency, mid-frequency, and high-frequency energy percentages.

[0034] Figure 2 This is a schematic diagram of the frequency domain dimensionality reduction process provided in an embodiment of the present invention.

[0035] like Figure 2 As shown, the low-frequency band can be represented by 10-dimensional low-frequency features, the mid-frequency band by 90-dimensional mid-frequency features, and the high-frequency band by 4900-dimensional high-frequency features.

[0036] Step 103: After filtering the dynamic time series features and the extracted static features, the data set is obtained by splicing them together. Static features can be extracted from the interactions between substrates, small molecules, and membrane proteins, including but not limited to: Basic physicochemical properties of small molecules: molecular weight, number of rotatable bonds, polar surface area, hydrophobicity coefficient, etc. Small molecule electronic activity characteristics: highest occupied orbital energy, lowest unoccupied orbital energy, orbital gap, electronic energy, dipole moment, polarizability, etc. Static characteristics of protein-small molecule interactions: total binding energy, van der Waals energy, electrostatic interaction energy, polar solvation energy, and nonpolar solvation energy; Statistical characteristics of local fluctuations in proteins: mean fluctuations of amino acid residues, mean fluctuations of key atoms, and mean fluctuations of key structural domains.

[0037] In practical applications, correlation analysis can be performed on the static features to remove features whose correlation meets the set criteria. Low-variance feature removal is performed on the dynamic time series features; After concatenating the static features and the dynamic time series features, a correlation analysis is performed again to remove features whose correlation meets the set criteria.

[0038] Specifically, when performing correlation analysis and redundant feature removal, the Pearson correlation coefficient and Spearman correlation coefficient between any two features in the candidate feature set can be calculated. When the absolute value of the correlation coefficient between two features is greater than a preset threshold (e.g., 0.85), they are considered to be highly linearly correlated and belong to a highly redundant feature pair. For highly redundant feature pairs, one of them is deleted, and the other is retained as a representative feature. The threshold of 0.85 is based on the fact that in the field of feature engineering, an absolute value of the correlation coefficient exceeding 0.85 is generally considered to indicate a very strong linear relationship between the two variables. Retaining both will not provide additional independent information to the model, but will instead increase noise and computational cost. In practice, this threshold can also be flexibly adjusted within the range of 0.80 to 0.90 according to the size of the dataset and the needs of the model.

[0039] When removing low-variance features from static features, the variance of each feature in the dataset can be calculated. If the variance of a feature is lower than a preset threshold (e.g., 0.05), it indicates that the value of that feature is almost constant in the vast majority of samples, its ability to distinguish between different categories of samples is extremely low, and its contribution to the prediction task is negligible; therefore, it should be deleted. The 0.05 threshold is determined by considering the magnitude and distribution of each feature in the dataset (in this invention, each feature has been standardized or normalized to [0,1] or a range with a mean of 0 and a standard deviation of 1 before being input into the model). When the feature variance is lower than 0.05, it indicates that the value fluctuation range of that feature is less than 5% of the overall scale range, and it has almost no ability to distinguish between positive and negative samples. This threshold references common feature selection practices: features with near-zero variance are typically removed. For standardized data, the variance threshold is generally between 0.01 and 0.10. Based on the actual situation of the constructed dataset and the desired feature dimension (usually 20-100 dimensions), this invention determined 0.05 as the optimization threshold after multiple cross-validation evaluations. It is worth noting that this threshold can also be appropriately adjusted according to the feature distribution characteristics of different datasets to ensure that sufficient discriminative information is retained while removing noisy features.

[0040] Step 104: Divide the dataset into a training set and a test set, and train the pre-built machine learning model based on the training set; The machine learning models used in this embodiment can be logistic regression models, decision tree models, random forest models, support vector machine models, etc.

[0041] Step 105: Input the test set into the trained machine learning model to obtain the prediction results of membrane protease-mediated small molecule co-metabolism transformation.

[0042] In an exemplary embodiment, the process of obtaining the predicted results of membrane protease-mediated small molecule co-metabolism transformation further includes: Calculate the SHAP value of each feature for each sample in the dataset to obtain the SHAP matrix; Visual analysis results are generated based on the SHAP matrix, and these results are used to perform interpretability analysis on the prediction results.

[0043] The results of the visibility analysis may include: The absolute values ​​of the SHAP matrix are averaged, and bar charts are drawn according to descending feature order to obtain a SHAP feature importance ranking chart, showing which features contribute the most to the prediction results at the global level. The purpose of this chart is to clearly present the key factors affecting the small molecule transformation ability and their relative importance. A scatter plot of the SHAP value of each feature is created by overlaying all data points from all samples on the ordinate and the original feature value on the abscissa. The color of each point represents the value of another related feature. The purpose of this plot is to reveal the nonlinear relationship between feature values ​​and predicted contributions—for example, it can observe the trend that "the greater the proportion of low-frequency energy, the more positive the SHAP value," thereby interpreting biologically significant patterns such as "the greater the conformational flexibility of proteins at large scale, the more favorable it is for small molecule transformation." Two features with the highest importance ranking were selected, and a SHAP dependency interaction diagram of second-order interaction was drawn to observe how the influence of one feature on the prediction changes with the value of the other feature. The purpose of this diagram is to identify synergistic or antagonistic effects between features and to reveal the molecular mechanisms of multi-factor coupling regulation, such as "only when the proportion of low-frequency energy is relatively high, the increase in the number of hydrogen bonds has a significant promoting effect on conversion".

[0044] For a single small molecule sample, a waterfall plot visually illustrates the direction and magnitude by which each feature raises or lowers the predicted value of the sample from the baseline. Red arrows indicate positive contributions (promoting the determination of convertibility), while blue arrows indicate negative contributions (promoting the determination of non-convertibility). The purpose of this plot is to provide case-by-case "decision attribution" for any small molecule of interest, precisely answering "what are the reasons for the molecule being determined to be convertible (or non-convertible), and which specific factors played a decisive role."

[0045] Steps 101 to 103 in the above embodiments completed the construction of a high-quality modeling dataset from raw data, and steps 104 to 105 completed the training and prediction output of the machine learning model. However, simply obtaining the classification result of "a certain small molecule can be transformed or cannot be transformed" is not enough to guide subsequent enzyme engineering, substrate design, or mechanism research—it is also necessary to answer "what factors led to this prediction result." The purpose of this embodiment is to perform post-trained interpretability analysis on the machine learning model, trace the direction and magnitude of the contribution of each feature to the prediction result, reveal the key molecular-level factors affecting membrane protein-mediated small molecule transformation, and transform the prediction result from a "black box output" into a transparent conclusion that is "traceable, interpretable, and can guide experiments."

[0046] In addition to generating visual analysis results based on the SHAP matrix as described above, the built-in feature importance evaluation function of the trained machine learning model can also be invoked to output the global contribution ranking of each feature to the model's predictions. For tree ensemble models (such as Random Forest, XGBoost, LightGBM, etc.), feature importance can be calculated by weighting the reduction in Gini impurity or the number of times a feature is selected as the basis for splitting in a splitting node; for linear models (such as logistic regression), feature importance can be reflected by the absolute value of the standardized regression coefficients. The purpose of this step is to initially identify the set of features that have the greatest impact on the model's decision from a global perspective and obtain a macro-ranking of the importance of each feature.

[0047] In summary, the prediction method for membrane protease-mediated small molecule co-metabolism provided in this application has at least the following beneficial effects: Prediction accuracy is significantly improved: After adding dynamic frequency features, the model's AUC, F1 score, and accuracy are significantly higher than those of traditional methods that only use static features; Strong physical interpretability: FFT dimensionality reduction preserves the frequency-conformation correspondence, and the features have clear biological and physical significance, which can directly support mechanism analysis; High computational efficiency: Effectively solves the problem of dimensionality explosion in molecular dynamics time series, reducing model training costs; Close to natural state: Based on transmembrane environment modeling, the true physiological conformation of membrane proteins is restored, improving the reliability of simulation and prediction; High versatility: Applicable to prediction of various membrane protein-small molecule interactions and transformations, and can be extended to multiple scenarios such as environmental remediation and drug development.

[0048] The prediction system for small molecule cometabolism mediated by membrane proteases provided by this invention is described below. The prediction system for small molecule cometabolism mediated by membrane proteases described below can be referred to in correspondence with the prediction method for small molecule cometabolism mediated by membrane proteases described above.

[0049] Figure 3 This is a schematic diagram of the prediction system provided in an embodiment of the present invention.

[0050] like Figure 3 As shown, the prediction system for membrane protease-mediated small molecule cometabolism transformation provided in this embodiment includes: System construction module 301 is used to construct a simulated system of the target membrane protein, obtain the structure of the substrate and the structure of the small molecule to be tested, and perform molecular docking to obtain candidate initial binding conformations; The feature extraction module 302 is used to perform molecular dynamics simulation on the candidate initial binding conformation and obtain dynamic time series features based on trajectory extraction; The collection construction module 303 is used to filter the dynamic time series features and the extracted static features and then concatenate them to obtain a data set; The model training module 304 is used to divide the dataset into a training set and a test set, and to train a pre-built machine learning model based on the training set. The transformation prediction module 305 is used to input the test set into the trained machine learning model to obtain the prediction results of membrane protease-mediated small molecule co-metabolism transformation.

[0051] The specific implementation method of the prediction system for membrane protease-mediated small molecule cometabolism provided in this embodiment can be carried out with reference to the above embodiment, and will not be repeated here.

[0052] Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4 As shown, the electronic device may include: a processor 410, a communication interface 420, a memory 430, and a communication bus 440, wherein the processor 410, the communication interface 420, and the memory 430 communicate with each other via the communication bus 440. The processor 410 can call logical instructions in the memory 430 to execute a prediction method for membrane protease-mediated small molecule co-metabolic transformation, the method including: A simulation system for the target membrane protein was constructed, and the structures of the substrate and the small molecule to be tested were obtained. Molecular docking was then performed to obtain candidate initial binding conformations. Molecular dynamics simulations were performed on the candidate initial binding conformations, and dynamic time series features were obtained based on trajectory extraction. After filtering the dynamic time series features and the extracted static features, the data set is obtained by splicing them together. The dataset is divided into a training set and a test set, and a pre-built machine learning model is trained based on the training set. The test set is input into the trained machine learning model to obtain the prediction results of small molecule cometabolism transformation mediated by membrane protease.

[0053] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a 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.) to execute all or part of the steps of the methods of the various embodiments of the present invention. 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.

[0054] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the prediction method for membrane protease-mediated small molecule co-metabolism provided by the above methods, the method comprising: A simulation system for the target membrane protein was constructed, and the structures of the substrate and the small molecule to be tested were obtained. Molecular docking was then performed to obtain candidate initial binding conformations. Molecular dynamics simulations were performed on the candidate initial binding conformations, and dynamic time series features were obtained based on trajectory extraction. After filtering the dynamic time series features and the extracted static features, the data set is obtained by splicing them together. The dataset is divided into a training set and a test set, and a pre-built machine learning model is trained based on the training set. The test set is input into the trained machine learning model to obtain the prediction results of small molecule cometabolism transformation mediated by membrane protease.

[0055] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for predicting membrane protease-mediated small molecule co-metabolism transformation provided by the methods described above, the method comprising: A simulation system for the target membrane protein was constructed, and the structures of the substrate and the small molecule to be tested were obtained. Molecular docking was then performed to obtain candidate initial binding conformations. Molecular dynamics simulations were performed on the candidate initial binding conformations, and dynamic time series features were obtained based on trajectory extraction. After filtering the dynamic time series features and the extracted static features, the data set is obtained by splicing them together. The dataset is divided into a training set and a test set, and a pre-built machine learning model is trained based on the training set. The test set is input into the trained machine learning model to obtain the prediction results of small molecule cometabolism transformation mediated by membrane protease.

[0056] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0057] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0058] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting small molecule co-metabolism transformation mediated by membrane proteases, characterized in that, include: A simulation system for the target membrane protein was constructed, and the structures of the substrate and the small molecule to be tested were obtained. Molecular docking was then performed to obtain candidate initial binding conformations. Molecular dynamics simulations were performed on the candidate initial binding conformations, and dynamic time series features were obtained based on trajectory extraction. After filtering the dynamic time series features and the extracted static features, the data set is obtained by splicing them together. The dataset is divided into a training set and a test set, and a pre-built machine learning model is trained based on the training set. The test set is input into the trained machine learning model to obtain the prediction results of small molecule cometabolism transformation mediated by membrane protease.

2. The method for predicting membrane protease-mediated small molecule co-metabolism transformation according to claim 1, characterized in that, The process of filtering the dynamic time series features and the extracted static features, and then concatenating them to obtain a data set, includes: Perform correlation analysis on the static features and remove features whose correlation meets the set criteria; Low-variance feature removal is performed on the dynamic time series features; After concatenating the static features and the dynamic time series features, a correlation analysis is performed again to remove features whose correlation meets the set criteria.

3. The method for predicting membrane protease-mediated small molecule co-metabolism transformation according to claim 1, characterized in that, The construction of the simulated system for the target membrane protein, the acquisition of the structure of the substrate and the structure of the small molecule to be tested, and the molecular docking to obtain candidate initial binding conformations include: Obtain the three-dimensional structure and membrane localization signals of the target protein, and construct a membrane protein simulation system; To obtain the structure of the substrate and the structure of the small molecule to be tested; A stable binding conformation was obtained by molecular docking the substrate structure with the membrane protein simulation system. Molecular docking is performed between the structure of the small molecule to be tested and the stable binding conformation to determine the candidate initial binding conformation.

4. The method for predicting small molecule co-metabolism transformation mediated by membrane protease according to claim 1, characterized in that, The step of performing molecular dynamics simulations on the candidate initial binding conformations and obtaining dynamic time series features based on trajectory extraction includes: The simulated system of the target membrane protein, as well as the structure of the substrate and the structure of the small molecule to be tested, are transformed into a molecular topological system; The energy of the molecular topological system is minimized, and isothermal and isobaric equilibrium is achieved. Conduct molecular dynamics simulations for a preset duration, record the coordinates and velocities of all atoms, and obtain trajectory data; Dynamic time series features are extracted from the trajectory data.

5. The method for predicting membrane protease-mediated small molecule co-metabolism transformation according to claim 1, characterized in that, The process of obtaining dynamic time series features based on trajectory extraction then includes: Frequency domain dimensionality reduction is performed on the dynamic time series features.

6. The method for predicting small molecule co-metabolism transformation mediated by membrane protease according to claim 1, characterized in that, The frequency domain dimensionality reduction of the dynamic time series features includes: The dynamic time series features are converted into frequency domain data features; The frequency domain data is divided into several frequency bands according to the frequency range of the aforementioned frequency domain data characteristics; Based on the proportion of each frequency band, the dynamic time series features are compressed into the corresponding frequency band energy proportion.

7. The method for predicting membrane protease-mediated small molecule co-metabolism transformation according to claim 1, characterized in that, The method for obtaining the predicted results of membrane protease-mediated small molecule co-metabolism transformation also includes: Calculate the SHAP value of each feature for each sample in the dataset to obtain the SHAP matrix; Visual analysis results are generated based on the SHAP matrix, and these results are used to perform interpretability analysis on the prediction results.

8. The method for predicting membrane protease-mediated small molecule co-metabolism transformation according to claim 7, characterized in that, The generation of visual analysis results based on the SHAP matrix includes: The absolute values ​​of the SHAP matrix are averaged, and a bar chart is drawn according to the descending order of features to obtain the SHAP feature importance ranking chart; Using the SHAP value of each feature as the ordinate and the original value of the feature as the abscissa, the data points of all samples are superimposed to create a SHAP scatter plot. By selecting the two features with the highest importance ranking, a SHAP dependency interaction graph of second-order interaction is drawn.

9. A predictive system for membrane protease-mediated small molecule cometabolism, employing the predictive method for membrane protease-mediated small molecule cometabolism according to any one of claims 1-8, characterized in that, include: The system construction module is used to construct a simulated system of the target membrane protein, obtain the structure of the substrate and the structure of the small molecule to be tested, and perform molecular docking to obtain candidate initial binding conformations; The feature extraction module is used to perform molecular dynamics simulations on the candidate initial binding conformations and obtain dynamic time series features based on trajectory extraction. The collection construction module is used to filter the dynamic time series features and the extracted static features and then concatenate them to obtain a data set; The model training module is used to divide the dataset into a training set and a test set, and to train a pre-built machine learning model based on the training set. The transformation prediction module is used to input the test set into the trained machine learning model to obtain the prediction results of small molecule co-metabolism transformation mediated by membrane protease.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the prediction method for membrane protease-mediated small molecule cometabolism transformation as described in any one of claims 1-8.

Images