A quantum chemistry and machine learning combined high-throughput screening method for zircon collectors

Abstract

The present application discloses a high-throughput screening method for zircon collectors by combining quantum chemistry and machine learning, including: establishing a molecular database R of flotation reagents; ionizing the molecules in the molecular database R to obtain an ionized molecular database A; obtaining a molecular database RA; hydroxylating the metal sites of zircon minerals to form a zircon target molecular database T to obtain a molecular database RT; obtaining quantum chemical parameters and log files; extracting molecular structural property parameters; training a machine learning model to predict the reaction Gibbs free energy; screening flotation reagents; and obtaining a targeted zircon flotation collector. The present application can quickly predict the reaction Gibbs free energy between molecules in the molecular database RA and different ions through a mathematical model, thereby quickly and efficiently realizing reagent screening. Compared with traditional reagent screening schemes, the screening efficiency is high and the accuracy rate is high.

Description

Technical Field

[0001] The present invention relates to the technical field of flotation reagents, and in particular to a high-throughput screening method combining quantum chemistry and machine learning for a zircon collector. Background Art

[0002] Zircon is an indispensable raw material for traditional metallurgy, the emerging ceramic industry, and precision casting. It is also a carrier mineral for uranium, an important nuclear raw material. Traditional zircon extraction mainly uses physical sorting, and pays less attention to sorting and separation methods based on surface chemistry. Flotation collectors enhance the efficiency of flotation separation of target minerals by strengthening the hydrophobicity of the mineral surface. However, the surface of zircon minerals is highly hydrophilic, and specific flotation collectors are required to selectively increase the hydrophobicity of the zircon surface. The existing technology uses the traditional trial and error method of reagents to screen flotation collectors, which not only has a low success rate and low screening efficiency, but also has the risk of synthesizing unknown toxic substances, resulting in huge waste of resources and safety issues.

[0003] Therefore, it is necessary to develop a safe, rapid and high-throughput method for screening zircon collectors. Summary of the invention

[0004] The main technical problem solved by the present invention is to provide a high-throughput screening method combining quantum chemistry and machine learning for zircon collectors, so as to solve the problem in the prior art of using traditional trial-and-error method to screen flotation collectors, which not only has a low success rate and low screening efficiency, but also has the risk of synthesizing unknown toxic substances, resulting in huge waste of resources and safety issues.

[0005] In order to solve the above technical problems, the technical solution adopted in this application is:

[0006] A quantum chemistry and machine learning combined high-throughput screening method for zircon collectors, comprising:

[0007] S1: establishing a molecular database R of flotation reagents, wherein the molecules in the molecular database R are an alkyl-functional group molecule set composed of a typical reagent functional group and an alkyl group;

[0008] S2 ionizes the molecules in the molecular database R to obtain an ionized molecular database A;

[0009] S3 takes the molecular database R and the ionized molecular database A as the initial structure, and creates corresponding Gaussian or ORCA input files for the molecules in the initial structure to obtain the molecular database RA;

[0010] S4 hydroxylates the metal sites of zircon minerals to form a zircon target molecule database T;

[0011] S5 docking the molecules in the molecular database R with the molecules in the zircon target molecular database T to obtain alkyl-functional group-metal ion hydroxylates, thereby obtaining a molecular database RT;

[0012] S6 optimizes and analyzes the molecules in the molecular database RA, the molecular database T and the molecular database RT at a calculation level not lower than B3LYP / DEF2TZVP, taking into account empirical dispersion and SMD solvation models, to obtain quantum chemical parameters and log files;

[0013] S7 uses NBO software to read the energy information and wave function information in the log file to obtain an output file, and extracts molecular structure property parameters by analyzing the output file;

[0014] S8, based on the molecular structure property parameters, training a machine learning model to predict the Gibbs free energy of the reaction of the molecules in the molecular database RA binding to the molecules in the zircon target molecular database T, and the Gibbs free energy of the reaction of the molecules in the molecular database RA binding to other metal ions;

[0015] S9 screens flotation agents according to the reaction Gibbs free energy to obtain targeted agent functional groups;

[0016] S10 assembles the targeted agent functional group with the hydrocarbon chain to obtain a targeted zircon flotation collector.

[0017] Furthermore, the ionization in S2 specifically includes: gradually deprotonating the molecules in the molecular database R to obtain all ionized molecules thereof;

[0018] Furthermore, the alkyl group is ethyl.

[0019] Furthermore, the zircon target molecule database T includes Zr 4+ , ZrOH 3+ , Zr(OH)2 2+ and Zr(OH)3 + .

[0020] Furthermore, the docking in S5 is specifically to dock the functional groups of the molecules in the molecular database R with the molecules in the zircon target molecular database T, and the docking distance is 2 angstroms.

[0021] Furthermore, the optimization in S6 is specifically as follows: using a method not lower than B3LYP and a 6-311G* basis set, taking into account the third generation of empirical dispersion, and under the SMD implicit water solvation model, the molecules in the molecular database RA, the zircon target molecular database T, and the molecular database RT are structurally optimized;

[0022] The analysis in S6 is specifically as follows: using a method not lower than B3LYP and a DEF2TZVP basis set, considering the effect of the third generation empirical dispersion, and under the SMD implicit water solvation model, vibration analysis and natural charge population analysis are performed on the molecules in the molecular database RA, the zircon target molecule database T, and the molecular database RT.

[0023] Furthermore, the molecular structure property parameters include: atomic type and number, frontier molecular orbital energy, atomic Mulliken charge population, natural charge population, APT charge population molecules, corrected thermodynamic free energy, structural electronic energy, zero-point vibrational energy, and the shortest bond length and second shortest bond length of the molecules in the molecular database RT.

[0024] Further, the S8 includes:

[0025] S81 takes the molecular structure property parameters and the corresponding molecular numbers as data, constructs a table document, stores the table document in a .csv format to facilitate data reading, and obtains a sample set;

[0026] S82 uses a quantum chemical method to calculate the Gibbs free energy of the reaction between the molecules in the RA and different ions to obtain data, and brings the data into the sample set to obtain a training set;

[0027] S83: taking 80% of the data in the training set for training the model and 20% of the data for testing the model, thereby enhancing the accuracy of the machine learning model and obtaining a mathematical model;

[0028] S84 brings the molecules in the molecular database RA into the mathematical model to obtain the Gibbs free energy of the reaction of the molecules in the molecular database RA binding to the molecules in the zircon target molecular database T, and the Gibbs free energy of the reaction of the molecules in the molecular database RA binding to other metal ions;

[0029] Preferably, the other metal ions include Fe 3+ 、Fe(OH) 2+ , and Fe(OH)2 + .

[0030] Furthermore, S9 includes: taking the Gibbs free energy G1 of the reaction between the molecule in the molecular database RA and the molecule in the zircon target molecule database T and the Gibbs free energy G2 of the reaction between the molecule and other metal ions, calculating the difference between G1 and G2 as the selectivity score, and screening the flotation agent according to the selectivity score.

[0031] Furthermore, before S6, the molecules in the molecular database R, the zircon target molecular database T and the molecular database RT are also roughly optimized, and the rough optimization is specifically: using the molecular dynamics tool of GaussView, Materials Studios, Jmol, Avogadro or ChemDraw to adopt the molecular mechanics method to screen molecules with stable configurations.

[0032] Furthermore, before S9, the targeted drug functional groups are further fine-optimized, and the fine-optimization specifically includes: eliminating the targeted drug functional groups that are toxic, unstable in composition, and incompatible in form;

[0033] Furthermore, the hydrocarbon chain in S9 is a hydrocarbon chain with a carbon chain length of eight.

[0034] Different from the prior art, the beneficial effects of this application are:

[0035] The present application optimizes and analyzes the molecules in the molecular database RA, the molecular database T and the molecular database RT through quantum chemical methods, obtains quantum chemical parameters and log files, extracts molecular structure property parameters and introduces them into the model, and uses a machine learning model to obtain a mathematical model for predicting the Gibbs free energy of the reaction. The mathematical model can quickly predict the Gibbs free energy of the reaction between the molecules in the molecular database RA and different ions, thereby realizing drug screening quickly and efficiently. Compared with traditional drug screening schemes, the screening efficiency is high and the accuracy is high. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a flowchart of training a machine learning model in one embodiment of the present application.

[0037] Figure 2 It is a linear correlation analysis diagram of the reaction free energy predicted by one embodiment of the present application and the reaction free energy predicted by quantitative chemistry. DETAILED DESCRIPTION

[0038] The present invention is described in detail below in conjunction with the embodiments.

[0039] This embodiment provides a high-throughput screening method for zircon collectors by combining quantum chemistry and machine learning, including:

[0040] S1 establishes a molecular database R of flotation reagents, and the molecules in the molecular database R are an alkyl-functional group molecule set composed of a typical reagent functional group and an alkyl group.

[0041] In one embodiment, the functional group is selected from 52 typical pharmaceutical functional groups, the alkyl group is selected from ethyl, and the ethyl short chain is combined with the functional group, which has a high cost performance. In other embodiments, the alkyl group can also be selected from other groups such as propyl.

[0042] S2 ionizes the molecules in the molecular database R to obtain an ionized molecular database A.

[0043] Ionization is to gradually deprotonate the molecules in the molecular database R to obtain all of their ionic molecules. For example, the carboxyl group has only one ionic state, that is, only one proton is removed to obtain -COO - ; Phosphoric acid can remove one or two protons to obtain -PO3H - And-PO3 2- Two ionic states.

[0044] S3 uses the molecular database R and the ionized molecular database A as the initial structure, and creates corresponding Gaussian or ORCA input files for the molecules in the initial structure to obtain the molecular database RA.

[0045] The specific molecular database RA is shown in the following table:

[0046]

[0047] S4 hydroxylates the metal sites of zircon minerals to form the zircon target molecule database T.

[0048] In one embodiment, the zircon target molecule database T includes Zr 4+ , ZrOH 3+ , Zr(OH)2 2+ and Zr(OH)3 + .

[0049] S5 docks the molecules in the molecular database R with the molecules in the zircon target molecular database T to obtain alkyl-functional group-metal ion hydroxylates, and obtains the molecular database RT.

[0050] The docking is specifically to dock the functional groups of the molecules in the molecular database R with the molecules in the zircon target molecular database T, and the docking distance is 2 angstroms.

[0051] S6 optimizes and analyzes the molecules in the molecular database RA, molecular database T and molecular database RT at a calculation level not lower than B3LYP / DEF2TZVP, taking into account the empirical dispersion and SMD solvation model, and obtains quantum chemical parameters and log files.

[0052] The optimization is as follows: using the B3LYP method and 6-311G* basis set, taking into account the third generation of empirical dispersion, and under the SMD implicit water solvation model, the molecules in the molecular database RA, the zircon target molecular database T, and the molecular database RT are structurally optimized;

[0053] The specific analysis is as follows: using no less than the B3LYP method and the DEF2TZVP basis set, considering the effect of the third-generation empirical dispersion, and under the SMD implicit water solvation model, vibration analysis and natural charge population analysis are performed on the molecules in the molecular database RA, the zircon target molecule database T and the molecular database RT.

[0054] In one embodiment, before S6, the molecules in the molecular database R, the zircon target molecular database T and the molecular database RT are also subjected to rough optimization, and the rough optimization is specifically: using the molecular dynamics tool of GaussView, Materials Studios, Jmol, Avogadro or ChemDraw to adopt the molecular mechanics method to screen the molecules with stable configuration, and eliminating the molecules with unstable configuration through rough optimization to avoid unnecessary calculation.

[0055] S7 uses NBO software to read the energy information and wave function information in the log file to obtain the output file. By analyzing the output file, the molecular structure property parameters are extracted.

[0056] The molecular structure property parameters include: atomic type and number, frontier molecular orbital energy, atomic Mulliken charge population, natural charge population, APT charge population molecules, corrected thermodynamic free energy, structural electronic energy, zero-point vibrational energy, and the shortest bond length and second shortest bond length of molecules in the molecular database RT.

[0057] S8 trains a machine learning model to predict the Gibbs free energy of the reaction between molecules in the molecular database RA and molecules in the zircon target molecular database T, as well as the Gibbs free energy of the reaction between molecules in the molecular database RA and other metal ions based on the molecular structure property parameters.

[0058] In one embodiment, the training of the machine learning model may adopt a gradient boosting regression algorithm, and the specific implementation process of the above steps includes:

[0059] S81 takes the molecular structure property parameters and the corresponding molecular numbers as data, constructs a table document, stores the table document in .csv format to facilitate data reading, and obtains a sample set;

[0060] S82 uses quantum chemical methods to calculate the Gibbs free energy of the reaction between molecules in RA and different ions, obtains data, and brings the data into the sample set to obtain the training set;

[0061] S83 takes 80% of the data in the training set for training the model and 20% of the data for testing the model to enhance the accuracy of the machine learning model and obtain a mathematical model;

[0062] S84 brings the molecules in the molecular database RA into the mathematical model to obtain the Gibbs free energy of the reaction between the molecules in the molecular database RA and the molecules in the zircon target molecular database T, as well as the Gibbs free energy of the reaction between the molecules in the molecular database RA and other metal ions.

[0063] In one embodiment, the process of training a machine learning model is as follows: Figure 1 shown.

[0064] S9 screens flotation agents based on the reaction Gibbs free energy to obtain targeted agent functional groups.

[0065] Specifically, in one embodiment, the above steps are implemented as follows:

[0066] Take the Gibbs free energy G1 of the reaction between the molecule in the molecular database RA and the molecule in the zircon target molecule database T, and the Gibbs free energy G2 of the reaction between the molecule and other metal ions, calculate the difference between G1 and G2 as the selectivity score, and screen the flotation agent based on the selectivity score.

[0067] In one embodiment, the selectivity scores of the molecules in the molecular database RA are compared with the zircon target molecule Zr(OH) n 4-n The selectivity scores of the four molecules in database T (n = 0, 1, 2, 3) are expressed as ΔG n Indicates that n refers to different Zr(OH) n 4-n (n=0,1,2,3), different Zr(OH) n 4-n The reaction Gibbs free energy G1 table (partial diagram) is as follows:

[0068] Other metal ions include Fe 3+ 、Fe(OH) 2+ , and Fe(OH)2 + ;

[0069] The obtained selectivity score G1-G2 table (partially shown) is as follows:

[0070]

[0071] The larger the parameter in the table, the more selective the agent is for other metal ions than for zirconium sites, and vice versa, the smaller the parameter, the stronger the effect on zirconium sites. It is easy to screen out the corresponding agents such as molecules 46, 44, and 41.

[0072] S9 assembles the targeted agent functional group with the hydrocarbon chain to obtain a targeted zircon flotation collector.

[0073] In one embodiment, the hydrocarbon chain is a hydrocarbon chain with a carbon chain length of eight. A hydrocarbon chain with a carbon chain length of eight can achieve better hydrophobicity without weakening the capture property of the functional group.

[0074] In one embodiment, before S9, the targeted drug functional groups are further fine-optimized, and the fine-optimization specifically includes: eliminating the targeted drug functional groups that are toxic, unstable in composition, and incompatible in form. For example, arsenic acid is a toxic molecule, and organic-metaphosphoric acid is a commonly used inhibitor, which is also unstable and does not meet the requirements, so it is eliminated.

[0075] The quantum chemical method, the scheme of this embodiment and the traditional screening method were used to screen the drugs respectively, and the time required for screening and the number of screened drugs were calculated to obtain the following table:

[0076]

[0077] It can be seen from the above table that the agent screening method of this embodiment has the highest screening efficiency and realizes high-throughput flotation agent screening.

[0078] The linear correlation of the reaction free energy predicted by this embodiment and the reaction free energy predicted by quantum chemistry is compared and analyzed to obtain Figure 2 .

[0079] Depend on Figure 2 It can be seen that the reaction free energy predicted by this embodiment has a good linear correlation with the reaction free energy predicted by quantum chemistry, and the drug molecules can be accurately screened out.

[0080] The above are only implementation modes of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A high-throughput screening method combining quantum chemistry and machine learning for zircon collectors, characterized in that: include: S1: establishing a molecular database R of flotation reagents, wherein the molecules in the molecular database R are an alkyl-functional group molecule set composed of a typical reagent functional group and an alkyl group; S2 ionizes the molecules in the molecular database R to obtain an ionized molecular database A; S3 takes the molecular database R and the ionized molecular database A as the initial structure, and creates corresponding Gaussian or ORCA input files for the molecules in the initial structure to obtain the molecular database RA; S4 hydroxylates the metal sites of zircon minerals to form a zircon target molecule database T, wherein the zircon target molecule database T includes Zr 4+ , ZrOH 3+ , Zr(OH)2 2+ and Zr(OH)3 + ; S5 docking the molecules in the molecular database R with the molecules in the zircon target molecular database T to obtain alkyl-functional group-metal ion hydroxylates, thereby obtaining a molecular database RT; S6 optimizes and analyzes the molecules in the molecular database RA, the molecular database T and the molecular database RT at a calculation level not lower than B3LYP / DEF2TZVP, taking into account empirical dispersion and SMD solvation models, to obtain quantum chemical parameters and log files; S7 uses NBO software to read the energy information and wave function information in the log file to obtain an output file, and extracts molecular structure property parameters by analyzing the output file; S8 trains a machine learning model to predict the Gibbs free energy of the reaction of the molecules in the molecular database RA with the molecules in the zircon target molecular database T, and the Gibbs free energy of the reaction of the molecules in the molecular database RA with other metal ions, according to the molecular structure property parameters, wherein the other metal ions include Fe 3+ 、Fe(OH) 2+ , and Fe(OH)2 + ; S9 screens flotation agents according to the reaction Gibbs free energy to obtain targeted agent functional groups; S10 assembles the targeted agent functional group with a hydrocarbon chain, wherein the hydrocarbon chain has a carbon chain length of eight, to obtain a targeted zircon flotation collector.

2. The high-throughput screening method of quantum chemistry and machine learning combination according to claim 1, characterized in that: The ionization in S2 specifically includes: gradually deprotonating the molecules in the molecular database R to obtain all of their ionized molecules.

3. The high-throughput screening method of quantum chemistry and machine learning combination according to claim 2, characterized in that: The alkyl group is ethyl.

4. The high-throughput screening method of quantum chemistry and machine learning combination according to claim 1, characterized in that: The docking in S5 is specifically to dock the functional groups of the molecules in the molecular database R with the molecules in the zircon target molecular database T, and the docking distance is 2 angstroms.

5. The high-throughput screening method of quantum chemistry and machine learning combination according to claim 1, characterized in that: The optimization in S6 is specifically as follows: using a method not lower than B3LYP and a 6-311G* basis set, taking into account the third generation of empirical dispersion, and under the SMD implicit water solvation model, the molecules in the molecular database RA, the zircon target molecular database T, and the molecular database RT are structurally optimized; The analysis in S6 is specifically as follows: using a method not lower than B3LYP and a DEF2TZVP basis set, considering the effect of the third generation empirical dispersion, and under the SMD implicit water solvation model, vibration analysis and natural charge population analysis are performed on the molecules in the molecular database RA, the zircon target molecule database T, and the molecular database RT.

6. The high-throughput screening method combining quantum chemistry and machine learning according to claim 1, characterized in that: The molecular structure property parameters include: atomic type and number, atomic Mulliken charge population, natural charge population, APT charge population molecules, corrected thermodynamic free energy, structural electronic energy, zero-point vibrational energy, and the shortest bond length and second shortest bond length of molecules in the molecular database RT.

7. The high-throughput screening method combining quantum chemistry and machine learning according to claim 1, characterized in that: The S8 includes: S81 takes the molecular structure property parameters and the corresponding molecular numbers as data, constructs a table document, stores the table document in a .csv format to facilitate data reading, and obtains a sample set; S82 uses a quantum chemical method to calculate the Gibbs free energy of the reaction between the molecules in the RA and different ions to obtain data, and brings the data into the sample set to obtain a training set; S83: taking 80% of the data in the training set for training the model and 20% of the data for testing the model, thereby enhancing the accuracy of the machine learning model and obtaining a mathematical model; S84 brings the molecules in the molecular database RA into the mathematical model to obtain the Gibbs free energy of the reaction between the molecules in the molecular database RA and the molecules in the zircon target molecular database T, as well as the Gibbs free energy of the reaction between the molecules in the molecular database RA and other metal ions.

8. The high-throughput screening method of quantum chemistry and machine learning combination according to claim 1, characterized in that: The S9 includes: taking the Gibbs free energy G1 of the reaction between the molecule in the molecular database RA and the molecule in the zircon target molecule database T and the Gibbs free energy G2 of the reaction between the molecule and other metal ions, calculating the difference between G1 and G2 as the selectivity score, and screening the flotation agent according to the selectivity score.

9. The high-throughput screening method of quantum chemistry and machine learning combination according to claim 1, characterized in that: Before S6, the molecules in the molecular database R, the zircon target molecular database T and the molecular database RT are also roughly optimized, and the rough optimization is specifically: using the molecular dynamics tool of GaussView, Materials Studios, Jmol, Avogadro or ChemDraw to adopt the molecular mechanics method to screen molecules with stable configurations.

10. The high-throughput screening method combining quantum chemistry and machine learning according to claim 1, characterized in that: Before S9, the targeted drug functional groups are further fine-tuned, and the fine-tuning is specifically to eliminate the targeted drug functional groups that are toxic, unstable in composition, and incompatible in form.

Images