Small molecule drug metabolism and toxicity prediction method based on quantum chemistry

Abstract

The invention discloses a small molecule drug metabolism and toxicity prediction method based on quantum chemistry, and belongs to the field of drug research and development. The method comprises the following steps: constructing a data set containing molecular structure, metabolism and toxicity data of the small-molecular drug; optimizing a molecular structure based on quantum chemistry, and calculating parameters such as front track energy and charge distribution; constructing a metabolic pathway prediction model through Monte Carlo simulation in combination with the parameters and a biotransformation rule, and outputting potential metabolites and pathway probabilities; integrating quantum chemical parameters, molecular fingerprints and physicochemical properties, training a toxicity prediction sub-model by using a gradient boosting tree algorithm, and outputting a toxicity probability value and type; and integrating results to generate a correlation prediction report, and verifying the model through an external test set. The method improves the prediction precision, reduces the experiment cost, shortens the research and development period, and provides a tool for the safety evaluation of candidate drugs.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of drug research and development, in particular to a small molecule drug metabolism and toxicity prediction method based on quantum chemistry. BACKGROUND

[0002] In the process of drug research and development, the metabolism behavior and toxicity of small molecule drugs are key factors determining whether the drug can be successfully listed. Traditional drug metabolism and toxicity evaluation mainly relies on in vitro experiments and animal models, which has problems such as long cycle, high cost, low prediction accuracy caused by species difference. With the development of computer technology, prediction models based on computing methods have gradually become an important auxiliary tool for drug research and development, but the existing methods still have the following shortcomings: 1. Metabolic pathway prediction mainly relies on empirical rules or simple molecular fingerprints, ignoring the influence of molecular electronic structure on reaction activity, leading to missed or misjudged potential metabolites; 2. Toxicity prediction models are mostly based on molecular physicochemical properties or database matching, lacking dynamic correlation analysis of drug metabolite toxicity, and it is difficult to reflect the change rule of toxicity in the metabolism process; 3. The application of quantum chemistry parameters is mostly limited to single property prediction, and there is no systematic integration of biological transformation rules and machine learning algorithms, and the prediction accuracy and reliability need to be improved.

[0003] Therefore, it is urgent to develop a method integrating quantum chemistry theory, biological transformation rules and machine learning algorithms to realize efficient and accurate prediction of small molecule drug metabolism pathway and toxicity, shorten the drug research and development cycle and reduce the research and development risk. SUMMARY

[0004] The purpose of the present application is to overcome the shortcomings of the prior art and provide a small molecule drug metabolism and toxicity prediction method based on quantum chemistry, which realizes accurate prediction of small molecule drug metabolism behavior and toxicity by systematically integrating quantum chemistry parameter calculation, metabolic pathway prediction model, toxicity prediction sub-model and result integration analysis.

[0005] To achieve the above purpose, the technical scheme adopted by the present application is as follows: A small molecule drug metabolism and toxicity prediction method based on quantum chemistry, comprising the following steps: S1. Constructing a small molecule drug data set: The data set contains molecular structure information (such as SMILES format or molecular structure file) of small molecule drugs, known metabolic pathway data (including metabolic enzymes, reaction type, metabolite structure) and toxicity experiment data (such as cytotoxicity, acute toxicity, genotoxicity, etc. qualitative / quantitative data).

[0006] The data preprocessing step includes: Standardizing the molecular structure information, unifying the molecular structure representation format, removing redundant hydrogen atoms and unreasonable chemical bonds; Using a rule-based screening method to remove duplicate data and samples with an experimental data missing rate of more than 30%; For partially missing metabolic intermediate state data, supplement through molecular dynamics simulation (such as GROMACS software), with a simulation condition of 300K constant temperature and 1atm constant pressure, and a simulation duration of no less than 100ns, and extract the intermediate conformation as supplementary data.

[0007] S2. Quantum chemistry parameter calculation: Based on the principle of quantum chemistry, the structure of the small molecule drug is optimized, specifically using the density functional theory method, and the geometric structure is optimized by Gaussian software at the B3LYP functional and 6-31G(d,p) basis set level, and the convergence threshold is set to 10⁻ 5 Hartree / Å in the optimization process, and the optimization result is verified by frequency analysis to be a potential energy surface minimum point (no imaginary frequency or only a negligible small imaginary frequency).

[0008] The calculated quantum chemistry parameters include: Frontier orbital energy (highest occupied molecular orbital energy HOMO, lowest unoccupied molecular orbital energy LUMO); Charge distribution (atomic net charge, Mulliken charge, natural bond orbital charge); Bond energy (bond dissociation energy, bond order); Solvation energy (solvation free energy in water and lipid environments); Molecular vibration frequency, dipole moment, and polarizability; All parameters are exported as JSON or CSV structured data formats for subsequent model construction.

[0009] S3. Metabolic pathway prediction model construction: Combine the quantum chemistry parameters and biotransformation rules to construct a metabolic pathway prediction model, specifically including: 3.1 Establishment of biotransformation rule library: The biotransformation rules include phase I metabolic reaction rules and phase II metabolic reaction rules: The phase I metabolic reaction rules cover the atom transfer and bond breaking rules of oxidation (such as hydroxylation, epoxidation), reduction (such as nitro reduction, carbonyl reduction), and hydrolysis (such as ester hydrolysis, amide hydrolysis) reactions, and clearly define the active site characteristics (such as electron-rich centers, easy-to-break bond energy thresholds) of the reaction; Phase II metabolism reaction rules cover the site selection rules of conjugation reactions such as glucuronidation, sulfation, and glutathione conjugation (such as the nucleophilic threshold of hydroxyl and amino groups).

[0010] 3.2 Potential reaction site screening: Match the reaction site charge density (such as oxidation sites with atomic net charge > 0.2 e) and bond energy (easy-to-break bonds with bond dissociation energy < 350 kJ / mol) in quantum chemical parameters with the reaction activity threshold in biotransformation rules to screen potential reaction sites.

[0011] 3.3 Metabolic pathway prediction: Predict the probability of each path by Monte Carlo simulation, with no less than 1000 simulations. Each simulation is based on the random selection of reaction types according to the activity probability of the reaction site. The output includes the potential metabolite structures of small molecule drugs (represented in SMILES format) and the probability of each metabolic pathway (a value between 0 and 1).

[0012] S4. Toxicity predictor model construction: Based on quantum chemical parameters, construct a toxicity predictor model for the potential metabolites and the original drug molecule, specifically including: 4.1 Input feature selection: Input features include quantum chemical parameters calculated in step S2, molecular fingerprints (such as ECFP4 fingerprints), and physicochemical property parameters (molecular weight, octanol-water partition coefficient logP, number of hydrogen bond donors / acceptors, topological polar surface area, etc.).

[0013] 4.2 Model training: Use gradient boosting tree algorithm (such as XGBoost) to train the toxicity predictor model, with known toxicity experimental data (such as median lethal dose LD50, toxicity grade label) as labels. Through 5-fold cross-validation, optimize model hyperparameters (learning rate 0.01-0.1, tree depth 3-10, iteration number 100-1000) to make the prediction accuracy of the model reach more than 85%.

[0014] 4.3 Model output: The toxicity predictor model outputs toxicity probability values (between 0 and 1, with higher values indicating higher toxicity risk) and toxicity types (such as hepatotoxicity, nephrotoxicity, and cardiotoxicity).

[0015] S5. Result integration and report generation: Integrate the output results of the metabolism pathway prediction model and the toxicity predictor model to generate a metabolism-toxicity correlation prediction report for small molecule drugs, including: Original drug molecule metabolism pathway diagram (showing the conversion relationship of original drug → intermediate metabolites → final products and the probability of each path in the form of a flowchart). A comparison table of quantum chemical parameters of each metabolite (comparison of key parameters such as HOMO / LUMO energy, bond energy, etc. with toxicity correlation); Toxicity probability ranking (ranking all metabolites and parent drugs according to toxicity probability value from high to low); Structural warning label of high-risk metabolites (marking high toxicity correlation sites such as specific functional groups or atoms in the molecular structure).

[0016] S6. Model verification: An external independent test set is used to verify the metabolic pathway prediction model and the toxicity prediction sub-model: The accuracy rate of metabolic pathway prediction (the matching rate of predicted metabolites and experimental results) is not less than 80%; The F1 value of toxicity type prediction is not less than 0.8; The Pearson correlation coefficient between toxicity probability value and experimental data is not less than 0.75.

[0017] Compared with the prior art, the present application has the following beneficial effects: 1. The present application introduces quantum chemical parameters (such as frontier orbital energy, charge distribution) to accurately describe the influence of molecular electronic structure on metabolic reaction activity, improves the accuracy of metabolic pathway prediction, and reduces the missed judgment of potential metabolites; 2. The toxicity prediction sub-model constructed integrates quantum chemical parameters, molecular fingerprints and physicochemical properties, and realizes the quantitative prediction of metabolite toxicity by combining machine learning algorithm, overcoming the limitation of traditional method which only relies on parent drug toxicity data; 3. Through metabolic-toxicity correlation analysis, a systematic prediction report is generated, which provides whole-chain information from metabolic pathway to toxicity risk for drug researchers, helps to early screen high-safety candidate drugs, and reduces research and development cost. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 is the overall flowchart of the method of the present application; Figure 2 is the construction logic diagram of the metabolic pathway prediction model; Figure 3 is the training and verification flowchart of the toxicity prediction sub-model. DETAILED DESCRIPTION

[0019] The present application will be further described in detail below in combination with specific embodiments.

[0020] Aspirin (Aspirin) is used as the research object, and the metabolic pathway and toxicity of aspirin are predicted by the method of the present application.

[0021] Step S1: Constructing a data set We collected the molecular structure of aspirin (SMILES: CC(=O)OC1=CC=CC=C1C(=O)O), its known metabolic pathway (mainly hydrolysis to salicylic acid, followed by glucuronidation), and toxicity data (such as gastrointestinal toxicity and hepatotoxicity experimental data). After removing duplicate samples, we retained 500 valid data.

[0022] Step S2: Calculation of quantum chemical parameters The structure of the aspirin molecule was optimized using Gaussian 16 software and the B3LYP / 6-31G(d,p) method. The optimized HOMO energy was -6.8 eV, the LUMO energy was -1.2 eV, the ester bond energy was 320 kJ / mol, the net charge of the hydroxyl oxygen atom was -0.5 e, and the solvation energy (aqueous phase) was -15 kcal / mol.

[0023] Step S3: Metabolic Pathway Prediction Based on the biotransformation rule base, ester bonds (bond energy 320 kJ / mol < threshold 350 kJ / mol) were screened as potential hydrolysis sites, and hydroxyl groups (net charge of oxygen atom -0.5 e) were screened as potential glucuronidation sites. Monte Carlo simulation (1000 times) predicted a hydrolysis pathway probability of 92%, generating the metabolite salicylic acid (SMILES: OC1=CC=CC=C1C(=O)O), with a further glucuronidation pathway probability of 88%.

[0024] Step S4: Toxicity Prediction Using the quantum chemical parameters (such as HOMO / LUMO, charge distribution), ECFP4 fingerprint, and physicochemical properties (logP=1.2 for aspirin and logP=2.0 for salicylic acid) as inputs, a trained XGBoost model predicted that the probability of gastrointestinal toxicity of aspirin was 0.72 and the probability of hepatotoxicity of salicylic acid was 0.65.

[0025] Step S5: Report Generation The generated metabolism-toxicity correlation report shows that the main metabolic pathway of aspirin is hydrolysis → glucuronidation, and salicylic acid is a high-risk metabolite. Its hepatotoxicity is related to the high charge density (-0.5 e) of the hydroxyl group, and this site needs to be optimized in drug design.

[0026] Step S6: Model Validation External test set validation showed that the accuracy of metabolic pathway prediction was 85%, and the F1 value of toxicity prediction was 0.82, which met the preset standards.

[0027] The above embodiments are merely preferred embodiments of the present invention and are not intended to limit the present invention. The scope of protection of the present invention is determined by the appended claims.

Claims

1. A method for predicting the metabolism and toxicity of small molecule drugs based on quantum chemistry, characterized in that, Includes the following steps: S1. Construct a small molecule drug dataset, which includes molecular structure information, known metabolic pathway data, and toxicity experimental data of the small molecule drugs; S2. The structure of the small molecule drug is optimized based on the principles of quantum chemistry, and the quantum chemical parameters of the molecule are calculated. The quantum chemical parameters include frontier orbital energy, charge distribution, bond energy and solvation energy. S3. Combining the quantum chemical parameters and biotransformation rules, construct a metabolic pathway prediction model, and output the potential metabolite structure of small molecule drugs through the metabolic pathway prediction model; S4. For the potential metabolites and original drug molecules, a toxicity prediction sub-model is constructed based on quantum chemical parameters. The toxicity prediction sub-model is trained by a machine learning algorithm and is used to output the toxicity probability value and toxicity type. S5. Integrate the outputs of the metabolic pathway prediction model and the toxicity prediction sub-model to generate a metabolic-toxicity correlation prediction report for small molecule drugs.

2. The method according to claim 1, characterized in that, In step S2, the structural optimization adopts the density functional theory method, specifically, the geometric structure is optimized at the level of B3LYP functional and 6-31G(d,p) basis set, and the optimization result is verified to be the potential energy surface minimum point through frequency analysis.

3. The method according to claim 1, characterized in that, In step S3, the biotransformation rules include phase I metabolic reaction rules and phase II metabolic reaction rules. The phase I metabolic reaction rules cover the atomic transfer and bond breaking rules of oxidation, reduction and hydrolysis reactions, while the phase II metabolic reaction rules cover the binding site selection rules of glucuronidation and sulfation.

4. The method according to claim 1, characterized in that, In step S4, the machine learning algorithm is the gradient boosting tree algorithm, and the input features of the toxicity prediction sub-model also include molecular fingerprint and physicochemical property parameters, including molecular weight, lipid-water partition coefficient and number of hydrogen bond donors / acceptors.

5. The method according to claim 1, characterized in that, In step S5, the metabolism-toxicity correlation prediction report includes: a metabolic pathway diagram of the original drug molecule, a comparison table of quantum chemical parameters of each metabolite, a toxicity probability ranking, and structural warning labels for high-risk metabolites.

6. The method according to claim 1, characterized in that, In step S1, the construction of the small molecule drug dataset also includes a data preprocessing step: standardizing the molecular structure information, removing duplicate data and samples with a missing experimental data rate of more than 30%, and supplementing some missing metabolic intermediate data through molecular dynamics simulation.

7. The method according to claim 3, characterized in that, In step S3, the process of constructing the metabolic pathway prediction model includes: matching the charge density of the reaction site in the quantum chemical parameters with the reactivity threshold in the biotransformation rules, screening potential reaction sites, and then predicting the probability of occurrence of each pathway through Monte Carlo simulation.

8. The method according to claim 4, characterized in that, In step S4, the training process of the toxicity prediction sub-model includes: using quantum chemical parameters, molecular fingerprints and physicochemical properties as input features, using known toxicity experimental data as labels, and optimizing the model hyperparameters through 5-fold cross-validation to achieve a prediction accuracy of over 85%.

9. The method according to claim 1, characterized in that, The method also includes a model validation step: using an external independent test set to validate the metabolic pathway prediction model and the toxicity prediction sub-model, wherein the accuracy of metabolic pathway prediction is not less than 80% and the F1 value of toxicity type prediction is not less than 0.

8.

10. The method according to claim 1, characterized in that, In step S2, the quantum chemical parameters also include the vibrational frequency, dipole moment and polarizability of the molecule. These parameters are calculated using Gaussian software and exported as structured data.

Images