CN107516012A - A Structure Descriptor Based on Calculation of 3D Molecular Structure of Organic Compounds - Google Patents

Abstract

The invention discloses a kind of structured descriptor calculated based on organic compound molecule 3-D solid structure, belong to compound Quantitative Study of Structure Property relation research method technical field.Purpose is by being training set from part known compound, structural characterization is carried out to training set sample using the descriptor, then such compound structure property relation (QSPR/QSAR) model is built with appropriate mathematical method (multiple linear regression (MLR), PLS (PLS)), for a certain property of the similar unknown compound of simulation and forecast.Method comprises the following steps：Skeleton non-hydrogen atom classification in step 1 organic compound molecule；Step 2 carries out parametrization dyeing to different non-hydrogen atoms；Step 3 builds the relation between different types of non-hydrogen atom by reciprocal function；Organic compound molecule structure is optimized minimum energy state by step 4, obtains the space coordinates of non-hydrogen atom, and structured descriptor is calculated in application program.By establishing the relational model between compound structure descriptor and certain property, can the accurately similar organic compound of simulation and forecast property, the QSPR/QSAR researchs for organic compound have very high reference value.

Description

A Structure Descriptor Based on Calculation of 3D Molecular Structure of Organic Compounds

technical field

The invention specifically relates to a structural descriptor based on the calculation of the three-dimensional molecular structure of an organic compound, and belongs to the technical field of compound quantitative structure-property relationship (QSPR/QSAR) research methods.

Background technique

The structure of the compound determines the properties, and the properties are the reflection of the structure of the compound. The construction of quantitative relationships between molecular structures and properties requires the introduction of corresponding structural descriptors. For a long time, researchers have done a lot of meaningful work in this area. By using the geometric structure, topological properties, connection characteristics and various physical and chemical parameters of the molecule for structural description, and then establishing a QSPR/QSAR model to predict various properties of the compound, see the paper: Generalized Correlation Index for Quantification of Persistent Environmental Pollutants Study on structure-chromatographic retention relationship [J]. Analytical Chemistry, 2006, 34(8): 1096-1100. However, the above methods are all two-dimensional (2D) structure descriptors, which are difficult to reproduce the three-dimensional structure of molecules in real space, and it is difficult to distinguish structures such as cis-trans isomerism. Three-dimensional (3D) descriptors have developed rapidly and become the mainstream of QSPR/QSAR molecular structure characterization, mainly including WHIM index and CoMFA, see the paper: MS-WHIM, new 3D theoretical descriptors derived from molecular surface properties: a comparative 3D QSAR study in a series of steroids.J Comput-AidedMol Des, 1997, 11:79-92; Investigation of structural requirements for inhibitory activity at the rat and housefly picrotoxinin binding sites in ionotropic GABA receptors using DISCOtech and CoMFA, Chemosphere, 2004-689: 8. The WHIM index is obtained through the weighted transformation of atomic space coordinates by different physical quantities to generate invariant variables for rotation and translation. The calculation process is quite complicated and difficult to be widely used. The disadvantage of CoMFA is that when conducting a group of molecular research, it is first necessary to superimpose the spatial structure of the sample molecules and overlap the conformation. In addition, the process of spatial grid division, variable number control, and potential field probe selection is complicated and difficult, and the workload is heavy. , and there are many uncertain factors, which cannot be ignored. Therefore, it is of great significance to construct a simple and easy-to-understand three-dimensional (3D) descriptor based on the three-dimensional structure of compound molecules, but there is no very effective and simple method yet.

Contents of the invention

Therefore, aiming at the above-mentioned deficiencies in the prior art, the purpose of the present invention is to provide a simple, understandable and effective parametric characterization method (structural descriptor) for the molecular structure of compounds for QSPR/QSAR research. In specific applications, select some known compounds as the training set, use the descriptor to characterize the structure of the training set samples, and then use appropriate mathematical methods (multiple linear regression (MLR), partial least squares regression (PLS)) to construct This type of compound quantitative structure-property relationship (QSPR/QSAR) model is used to predict a certain property (such as chromatographic retention, toxicity, migration characteristics, degradability, drug efficacy, biological activity, etc.) Research provides references.

Method of the present invention comprises the following steps:

Step 1 Classification of skeleton non-hydrogen atoms in organic compound molecules

Non-hydrogen atoms in organic compounds form molecules according to different connection methods (chemical bonds), ignoring the influence of non-skeleton hydrogen atoms, non-hydrogen atoms in molecules can be divided into A1, A2, A3, The four types of A4 are connected to 1, 2, 3, and 4 non-hydrogen atoms, respectively.

Step 2 Parametric coloring of different non-hydrogen atoms

The characteristics of non-hydrogen atoms in molecules are mainly determined by factors such as the number of valence electrons and the number of electron layers. Therefore, the parametric dyeing value of non-hydrogen atoms is obtained by using the following formula to perform parametric dyeing on different non-hydrogen atoms.

Z _i ＝[m _i (n _i -1)(X _C /X _i ) ^1/2 -h _i ] ^1/2 Formula 1

In the formula, n _i is the number of electron shells of non-hydrogen atom i, m _i is the number of electrons in the outermost shell, Xi is the Pauling electronegativity of carbon atoms, _{h i is} the number of hydrogen atoms directly connected to it; X _C is carbon The Pauling electronegativity of the atom.

Step 3 Construct the relationship between different kinds of non-hydrogen atoms through the reciprocal function

The relationship between the non-hydrogen atoms in the molecule is not a specific interaction between the atoms, but to reflect two aspects: the degree of closeness is consistent with the change trend of the Z _i value of the non-hydrogen atoms, and it is opposite to the change trend of the distance between the two. Usually the reciprocal shape function can meet this requirement, and the following formula is used to express the relationship between different non-hydrogen atoms.

r _ij is the relative distance between the non-hydrogen atoms i and j in the molecule (that is, the ratio of the space distance between them to the length of the carbon-carbon single bond); n and l are the types of the non-hydrogen atoms. The 4 types of non-hydrogen atoms in the compound molecule can produce 10 different related items: m ₁₁ , m ₁₂ , m ₁₃ , m ₁₄ , m ₂₂ , m ₂₃ , m ₂₄ , m ₃₃ , m ₃₄ , m ₄₄ , where m ₁₂ Indicates the relationship between the first type of non-hydrogen atoms and the second type of non-hydrogen atoms in the molecule, similarly m ₂₃ indicates the relationship between the second type of non-hydrogen atoms and the third type of non-hydrogen atoms in the molecule, and so on. The 10 different relationship items are recorded as x ₁ , x ₂ , x ₃ , x ₄ , x ₅ , x ₆ , x ₇ , x ₈ , x ₉ and x ₁₀ , so that a maximum of 10 molecular structures will be generated for the research sample related variables.

Step 4: optimize the molecular structure of the organic compound to the lowest energy state, obtain the spatial coordinates of non-hydrogen atoms, and use the program to calculate the structure descriptor.

Use ChemOffice 8.0 to construct the initial three-dimensional structure of organic compound molecules, use the MOPAC semi-empirical quantum chemistry software that comes with Chem 3D to finally optimize the molecular structure at the AM1 level (cutoff value 0.001kJ·mol ^-1 ), and obtain the molecular structure of each atom. Spatial position coordinates. Input the spatial position coordinates and parametric dyeing value of each atom in the molecule into the self-compiled C language application program for processing, and obtain the molecular structure descriptor.

When specifically using the structural descriptor of the present invention, it is also necessary to use the variables obtained in the above steps to firstly use stepwise regression to screen the variables according to the significance of the variables, and then use the selected variable combination as the dependent variable X, and the properties of the compound as Dependent variable Y, use appropriate mathematical methods (multiple linear regression (MLR), partial least squares regression (PLS)) to construct a quantitative structure-property relationship (QSPR/QSAR) model for this type of compound, and then simulate the properties of similar unknown compounds predict.

The beneficial effect of the present invention is that: the present invention provides a structural descriptor based on the calculation of the three-dimensional molecular structure of the organic compound, which is simple and easy to understand, and can parametrically characterize the molecular structure of the organic compound. Use appropriate mathematical methods (multiple linear regression (MLR), partial least squares regression (PLS)) to construct a quantitative structure-property relationship (QSPR/QSAR) model of the compound, and the correlation coefficient (R) of the model and the correlation coefficient (R) of the interactive test _CV ) are ideal, revealing the structural factors affecting the properties of the compounds to a certain extent. The model can more accurately predict the related properties of similar unknown compounds, and has a high reference value for the QSPR/QSAR research of organic compounds.

Description of drawings

Fig. 1 is o-xylene molecular structure in the embodiment

Fig. 2 is the variation situation of the MLR model correlation coefficient (R/R _CV ) in the variable screening process in the embodiment

Fig. 3 is the change situation of variable selection process MLR model standard deviation (SD/SD _CV ) in the embodiment

Fig. 4 is 37 samples in the embodiment in the first two principal component score space scatter distribution diagrams of PLS;

Fig. 5 is the correlation diagram of predicted value and experimental value in the embodiment;

detailed description

The organic compound structure descriptor of the present invention can be used for the simulation prediction of compound multiple properties (such as chromatographic retention value, toxicity, migration characteristics, degradability, drug efficacy, biological activity, etc.), the present invention is applied to compound below in conjunction with accompanying drawing The simulation prediction of chromatographic retention time will be explained in detail, and similar methods can be used when other properties of compounds need to be simulated and predicted.

Experimental Materials

In this example, the aroma components of 37 wild jasmine flowers are selected as research samples, and the gas chromatography retention time of the compound is represented by t _R. The experimental value is taken from the paper: Analysis of the aroma components of wild jasmine flowers by solid-phase microextraction GC-GC [J]. Fine Chemical Industry , 2007, 24(2): 159-161. The compounds and their GC retention times (t _R ) are listed in Table 1.

Table 1

experimental method

1) Compound molecular structure characterization

The chromatographic retention time (t _R ) of organic compounds is not only related to the measurement factors, but also related to the molecular structure. The type and number of atoms constituting the compound, the connection mode between atoms, etc. will all have an impact on t _R . Non-hydrogen atoms form molecules according to different connection methods (chemical bonds). Neglecting the influence of non-skeletal hydrogen atoms, the non-hydrogen atoms in the molecule can be divided into four types according to the number of non-hydrogen atoms connected to them, A1, A2, A3, and A4, which respectively represent 1, 2, 3, and 4 non-hydrogen atoms. Atoms connected, such as secondary carbon atoms connected to two non-hydrogen atoms belong to the A2 atom type. The characteristics of non-hydrogen atoms in molecules are mainly determined by factors such as the number of valence electrons and the number of electron layers. Therefore, the following formula (1) is used to perform parametric dyeing on different non-hydrogen atoms to obtain the parametric dyeing value of non-hydrogen atoms.

Z _i ＝[m _i (n _i -1)(X _C /X _i ) ^1/2 -h _i ] ^1/2 (1)

In the formula, n _i is the number of electron shells of non-hydrogen atom i, m _i is the number of electrons in the outermost shell, Xi is the Pauling electronegativity of carbon atoms, _{h i is} the number of hydrogen atoms directly connected to it; X _C is carbon The Pauling electronegativity of the atom.

The relationship between the non-hydrogen atoms in the molecule is not a specific interaction between the atoms, but to reflect two aspects: the degree of closeness is consistent with the change trend of the Z _i value of the non-hydrogen atoms, and it is opposite to the change trend of the distance between the two. Usually the reciprocal shape function can meet this requirement, and the following formula (2) is used to express the relationship between different non-hydrogen atoms.

r _ij is the relative distance between the non-hydrogen atoms i and j of the molecule (that is, the ratio of the space distance between them to the length of the carbon-carbon single bond); n and l are the types of the non-hydrogen atoms. The 4 types of non-hydrogen atoms in the compound molecule can produce 10 different related items: m ₁₁ , m ₁₂ , m ₁₃ , m ₁₄ , m ₂₂ , m ₂₃ , m ₂₄ , m ₃₃ , m ₃₄ , m ₄₄ , where m ₁₂ Indicates the relationship between the first type of non-hydrogen atoms and the second type of non-hydrogen atoms in the molecule, similarly m ₂₃ indicates the relationship between the second type of non-hydrogen atoms and the third type of non-hydrogen atoms in the molecule, and so on. The 10 different relationship items are recorded as x ₁ , x ₂ , x ₃ , x ₄ , x ₅ , x ₆ , x ₇ , x ₈ , x ₉ and x ₁₀ , so that a maximum of 10 molecular structures will be generated for the research sample related variables.

Use ChemOffice 8.0 to construct the initial three-dimensional structure of organic compounds, use the MOPAC semi-empirical quantum chemistry software that comes with Chem 3D to finally optimize the molecular structure at the AM1 level (cutoff value 0.001kJ·mol ^-1 ), and get the space of each atom Position coordinates. The spatial position coordinates and parametric dyeing values of each atom were input into the self-compiled C language application program for processing, and the molecular structure descriptor was obtained.

Here, o-xylene is taken as an example to illustrate the calculation of structure descriptors. First number each non-hydrogen atom in the compound molecule, and then determine the atom type of each non-hydrogen atom. For o-xylene, there are 8 non-hydrogen atoms in the molecule (see Figure 1), and their atom types are A3, A3, A2, A2, A2, A2, A1 and A1. Calculate x ₁ , x ₁ is the relationship between the first type of non-hydrogen atoms and the first type of non-hydrogen atoms in the molecule, that is, the relationship between the two atoms No. 7 and No. 8. In the optimal conformation of o-xylene (see Figure 1), the spatial coordinates of No. 7 and No. 8 atoms are (-0.3524, 2.0282, -0.0202) and (2.1627, 0.5775, 0.0627) respectively, and their spatial distance is 0.2905nm , then r _7,8 =0.2905/0.1540=1.8864. In addition, the dyeing value is calculated according to formula (1), Z7=Z8=1.0000, so the value of x ₁ is calculated as follows: x ₁ =m ₁₁ =1.0000×1.0000/1.8864 ² =0.2810. Likewise, several other structure descriptor values can be calculated.

The above method is used to parametrically characterize the structure of the research sample, and the compound structure descriptor is obtained. Since the relationship between the fourth type of non-hydrogen atoms does not exist in the sample, 9 non-all "0" structure descriptors are finally obtained, listed in Table 2.

Table 2

QSPR modeling and testing

Multiple linear regression (MLR) and partial least squares regression (PLS) methods were used to build the model respectively, and the model was tested interactively by "leave one out". It is generally believed that the modeling correlation coefficient (R) is between 0.8-1.0, indicating that the model is highly correlated; the cross-test correlation coefficient R _CV ≥ 0.7, indicating that the model has good robustness and predictive ability. The multicollinearity of each variable in the model is evaluated by variance inflation factors (VIF), the definition of VIF is: VIF=(1-r ² ) ^-1 , where r is the correlation between an independent variable and other variables degree (judgment criteria: VIF value greater than 5.0, serious collinearity among variables, and unreliable correlation equation), if the diagnosis finds that the VIF value of a certain variable in the model is too large, continue to reduce variable modeling.

Firstly, stepwise regression (SMR) is used to extract variables according to the significance of the variables, and then the selected variable combination is used as the independent variable X, and the gas chromatography retention time of the compound (t _R ) is used as the dependent variable Y, and then multiple linear regression (MLR) is used. ) to establish a model, variable screening and corresponding results are shown in Table 3 and Figure 2 and Figure 3.

table 3

A good forecasting model not only has good fitting ability for internal samples, but also has strong predictive ability for external samples. Therefore, when selecting a model, try to choose a model with a large cross-test correlation coefficient ( _RCV ) while ensuring a good fitting effect on internal samples, so as to ensure that the model has strong predictive ability and stability. It can be seen from Table 3, Figure 2, and Figure 3 that the model obtained by modeling the variable combination (x ₃ , x ₅ , x ₇ ) selected in the third step of the stepwise regression (SMR) should be selected. At this time, the compound gas chromatography retention time The regression equation between (t _R ) and the structure descriptor is:

t _R ＝0.2095+0.4030 x ₃ +0.2251 x ₅ +0.2249 x ₇ (3)

N = 37, R = 0.9403, SD = 1.5498, F = 84.0232; R _CV = 0.9186, SD _CV = 1.7994 F _CV = 59.4872.

The modeling correlation coefficient (R) reached 0.9403 (between 0.8-1.0), and the cross-test correlation coefficient (R _CV ) reached the maximum value of 0.9186 (greater than 0.7), indicating that the model is highly correlated, robust and predictive. The standard deviation (SD) was 1.5498, and the standard deviation of the interactive test (SD _CV ) was 1.7994, both of which were small, indicating that the prediction accuracy of the model was high. The variance inflation factor (VIF) of x ₅ in the variable is the largest at 1.3379, which is much smaller than the standard of 5.0, indicating that there is almost no collinearity among the model variables. The standardized regression coefficients of x ₃ , x ₅ , and x ₇ are 0.6190, 1.0776, and 0.1581, respectively, indicating that x ₅ has the greatest impact on the chromatographic retention time of the compound, followed by x ₃ , and x ₇ has relatively little impact on the chromatographic retention time of the compound. x ₅ corresponds to the relationship between the second type of non-hydrogen atoms, indicating that the more the second type of non-hydrogen atoms in the compound, the greater the chromatographic retention time.

The molecular structure descriptors screened in the third step of stepwise regression were used as the X variable, and the compound gas chromatography retention time (t _R ) was used as the dependent variable Y, and Simca-P11.5 was used to establish a partial least squares regression (PLS ) model, and the “leave one out method” was used to conduct an interactive test on the obtained PLS model. The final PLS model contains two principal components (A), and the regression equation between the gas chromatographic retention time (t _R ) of the compound and the original structure descriptor is formula (4).

t _R ＝0.3275+0.3790 x ₃ +0.2243 x ₅ +0.3157 x ₇ (4)

The correlation coefficient R of the model is 0.9381 (between 0.8-1.0), and R _CV is 0.8899 (greater than 0.7). Each standard deviation SD and SD _CV were 1.5362 and 1.7436, respectively. The above shows that the model has good fitting effect, good robustness and strong predictive ability. Figure 4 is a scatter diagram of the scores of the first two principal components of 37 samples in the PLS. The Hotelling T ² ellipse is a 95% confidence confidence circle. It can be seen that most of the sample points fall within the confidence circle. The results show that the structural description Symbols can properly represent the molecular structure characteristics of organic compounds and make correct responses in statistical models.

The MLR model and the PLS model simulated and predicted the gas chromatographic retention time (t _R ) of the sample, and the predicted values are listed in Cal1 and Cal2 in Table 1, respectively. Figure 5 is the correlation diagram between the model predicted value and the experimental value. It can be seen from Figure 5 that all sample points are distributed on the 45° diagonal or close to the diagonal, indicating that the predicted value is very close to the experimental value. The overall prediction effect is good, and the prediction results are highly reliable.

Compared with the existing technology, the constructed structural descriptor is a three-dimensional (3D) molecular structural descriptor calculated based on the three-dimensional structure of the molecule. The hydrogen atom coloring value takes into account rich structural information such as principal quantum number, electronegativity, number of outermost electrons, number of connected hydrogen atoms, etc.

The non-hydrogen atoms in the molecule are classified and parametrically colored, and the spatial distance relationship between different non-hydrogen atoms is used as a molecular structure descriptor to parametrically characterize the structure of some organic compounds. Stepwise regression (SMR), multiple linear regression (MLR), and partial least squares regression (PLS) were used to construct the relationship model between compound structure and gas chromatographic retention time (t _R ), and the correlation coefficient (R) of the model and the interaction test were used. The correlation coefficients (R) are ideal, revealing to some extent the structural factors affecting the retention time (t _R ) of compounds in gas chromatography. The model can accurately predict the gas chromatography retention time (t _R ) of volatile organic compounds in plant essential oils, and has a high reference value for the QSPR/QSAR research of organic compounds.

The above is a preferred embodiment of the present invention. It should be pointed out that the molecular structure descriptor described in the present invention can be applied to the toxicity, migration characteristics, degradation of compounds in addition to the simulation prediction of gas chromatography retention time (t _R ). For the simulation prediction of various properties such as sex, drug efficacy, biological activity, etc., for those of ordinary skill in the art, without departing from the principle of the present invention, some improvements and modifications can also be made. These improvements and modifications It should also be regarded as the protection scope of the present invention.

Claims (1)

1. a kind of structured descriptor calculated based on organic compound molecule three-dimensional structure, it is characterised in that methods described includes Following steps：

Skeleton non-hydrogen atom classification in step 1 organic compound molecule

Non-hydrogen atom forms molecule by different connected modes (chemical bond) in organic compound, ignores the shadow of non-skeleton hydrogen atom Ringing, the non-hydrogen atom of intramolecular can be divided into the class of A1, A2, A3, A4 tetra- according to its non-hydrogen atom number connected, represent respectively with 1, 2nd, 3,4 non-hydrogen atoms are connected.

Step 2 carries out parametrization dyeing to different non-hydrogen atoms

The feature of non-hydrogen atom in the molecule, mainly determined by factors such as its valence electron number, the electronics numbers of plies, thus using following formula pair Different non-hydrogen atoms carries out parametrization dyeing and obtains the parametrization dye number of non-hydrogen atom.

Z_i=[m_i(n_i-1)(X_C/X_i)^1/2-h_i]^1/2Formula one

N in formula_iFor the non-hydrogen atom i electronics number of plies, m_iFor outermost electron number, X_iFor the Pauling electronegativity of carbon atom, h_iFor with Its number of hydrogen atoms being directly connected to；X_CFor the Pauling electronegativity of carbon atom.

Step 3 builds the relation between different types of non-hydrogen atom by reciprocal function

Relation in molecule between non-hydrogen atom is not certain specific interaction between atom, but to reflect its level of intimate With non-hydrogen atom Z_iThe change trend of value is consistent and the two aspect situations opposite with the change trend of both distances.Usual shape reciprocal Function can meet this requirement, carry out expressing the relation between different non-hydrogen atoms using following formula.

r_ijIt is relative distance (i.e. the ratio between space length and carbon-carbon single bond bond distance's value table between the two of non-hydrogen atom i, j in molecule Show)；N and l is the affiliated type of non-hydrogen atom.4 class non-hydrogen atoms can produce 10 kinds of different relational terms in compound molecule： m₁₁、m₁₂、m₁₃、m₁₄、m₂₂、m₂₃、m₂₄、m₃₃、m₃₄、m₄₄, wherein m₁₂Represent first kind non-hydrogen atom and the second class non-hydrogen in molecule Relation between atom, similarly m₂₃The relation between the second class non-hydrogen atom and the 3rd class non-hydrogen atom in molecule is represented, with this Analogize.10 kinds of different relational terms are designated as x respectively₁、x₂、x₃、x₄、x₅、x₆、x₇、x₈、x₉And x₁₀, so at most will for research sample Produce 10 structured descriptors related to molecular structure.

Organic compound molecule structure is optimized minimum energy state by step 4, obtains the space coordinates of non-hydrogen atom, Structured descriptor is calculated in application program.

The initial volumetric structure of organic compound molecule is built using ChemOffice 8.0, the MOPAC half carried with Chem 3D Experience quantum chemistry software final optimization pass in AM1 levels obtains molecular structure (cutoff value 0.001kJmol-1), and obtains The locus coordinate of each atom.The locus coordinate of each atom in molecule and parametrization dye number input is self-editing C language application program is acted upon, and obtains Molecular structure descriptor.