METHOD FOR PREDICTING ACTIVATION ENERGY USING AN ATOMIC FINGERPRINT DESCRIPTOR OR AN ATOMIC DESCRIPTOR

Description

BACKGROUND OF THE INVENTION

1. Field of the Invention

The present invention relates to a method for predicting the activation energy of phase I metabolism, mediated by CYP450 enzymes, using an effective atomic fingerprint descriptor or atomic descriptor.

2. Description of the Prior Art

The prediction of absorption, distribution, metabolism and excretion (ADME) properties of drugs is a very important technique to shorten the drug development period and to enhance the probability of success of drug development. Among the drug's ADME properties, drug metabolism is a key determinant of metabolic stability, drug-drug interactions, and drug toxicity.

Metabolic reactions can be divided according to the reaction mechanism into two categories: aliphatic hydroxylation and aromatic hydroxylation. Also, they can be divided according to the type of reaction into the following categories: N-dealkylation, C-hydroxylation, N-oxidation, O-dealkylation and the like. In aliphatic hydroxylation, the iron (Fe) of compound I in the active site of CYP450 (cytochrome P450) is substituted with the hydrogen of the substrate, so that the substrate becomes a radical. Then, a hydroxyl group binds to the substrate to form a metabolite. In aromatic hydroxylation, the iron of compound I binds to the substrate to form a tetrahedral intermediate, and then becomes detached from the substrate while giving a hydroxyl group to the substrate, thereby forming a metabolite.

The metabolism of the compound may occur at most positions to which hydrogen is bound. The possibility of reaction at each position depends on how the compound binds well to CYP450 and how the reactivity at the bound position is high. To determine accessibility, a docking study on CYP450 can be carried out, followed by calculation of binding affinity.

Prediction of the metabolisms of external substances is important in the early stage of new drug development. Particularly, the reaction rate and regioselectivity of phase I metabolism are very important pharmacokinetic characteristics, through which the toxicity of metabolites can be predicted.

Such reaction rate and regioselectivity can be predicted from activation energy, but existing methods depend on time-consuming quantum mechanical calculations and difficult experiments. For example, K. R. Korzekwa et al. (J. Am. Chem. Soc. 1990, 112, 7042) reported a method of predicting the activation energy for hydrogen abstraction by quantum mechanical calculation, and T. S. Dowers et al. (Drug Metab. Dispos. 2004, 32, 328) reported a method of predicting the activation energy of aromatic hydroxylation by quantum mechanical calculation. However, such quantum mechanical methods perform calculations in various molecular states, and thus cannot determine accurate activation energy due to the complexity resulting from the conformational difference between these states.

Accordingly, the present inventors have developed a novel, fast and accurate model which can predict the activation energy of phase I metabolism on the basis of only the characteristics of an external substrate using an atomic fingerprint descriptor Or an atomic descriptor, thereby completing the present invention.

SUMMARY OF THE INVENTION

It is an object of the present invention to provide a method for constructing a database of atomic fingerprint descriptors.

Another object of the present invention is to provide a method for predicting activation energy using an atomic fingerprint descriptor and an atomic descriptor.

Still another object of the present invention is to provide a method for predicting activation energy using an atomic descriptor.

Still another object of the present invention is to provide a method of predicting i) a metabolite, ii) the relative rate of metabolism, iii) the regioselectivity of metabolism, iv) the inhibition of metabolism, v) a drug-drug interaction, and vi) the toxicity of a metabolite, through the activation energy predicted by said methods.

To achieve the above objects, the present invention provides a method for constructing a database of atomic fingerprint descriptors, the method comprising the steps of:

(i) calculating the atomic fingerprint descriptor of a substrate, which is represented by the following equation 1;

(ii) predicting activation energy for an atomic position using an atomic descriptor;

(iii) predicting cytochrome P450-mediated metabolism using the predicted activation energy; and

(iv) comparing the predicted metabolism with experimental metabolism and storing whether the metabolism occurs:

Xabc  [Equation 1]

wherein X is the chemical symbol of an atom; a is a bond indicator that indicates the number of atoms bonded; b is a ring indicator that indicates whether the atom is part of a ring; and c is an aromatic indicator that indicates whether the atom is an aromatic atom.

The metabolism in step (iii) is aliphatic hydroxylation or aromatic hydroxylation.

Also, the metabolism in step (iii) is N-dealkylation, C-hydroxylation, N-oxidation or O-dealkylation.

The present invention can be applied to all CYP 450 enzymes, and it is apparent that the present invention can be applied particularly to human CYP 450 enzymes. The cytochrome P450 enzymes according to the present invention include, but are not limited to, CYP2E1, CYP3A4, CYP2B6, CYP2C8, CYP2C9, CYP1A1, CYP1A2, CYP2C19, CYP2D6, CYP1B1, and CYP2A6.

In another aspect, the present invention provides a method for predicting activation energy using an atomic fingerprint descriptor and an atomic descriptor, the method comprising the steps of:

(i) calculating the atomic fingerprint descriptor of a substrate, which is represented by the following equation 1;

(ii) comparing the calculated atomic fingerprint descriptor with the data, constructed by said method, to select an atomic position where cytochrome P450-mediated metabolism can occur; and

(iii) predicting activation energy for the selected atomic position using an atomic descriptor:

Xabc  [Equation 1]

wherein X is the chemical symbol of an atom; a is a bond indicator that indicates the number of atoms bonded; b is a ring indicator that indicates whether the atom is part of a ring; and c is an aromatic indicator that indicates whether the atom is an aromatic atom.

The metabolism in step (ii) is aliphatic hydroxylation or aromatic hydroxylation.

Also, the metabolism in step (ii) is N-dealkylation, C-hydroxylation, N-oxidation or O-dealkylation.

Examples of the cytochrome P450 enzyme include, but are not limited to, CYP2E1, CYP3A4, CYP2B6, CYP2C8, CYP2C9, CYP1A1, CYP1A2, CYP2C19, CYP2D6, CYP1B1, and CYP2A6.

In step (iii), the activation energy for cytochrome P450-mediated hydrogen abstraction from a substrate of the following formula 1 can be predicted using the atomic descriptors [δ_het], [max (δ_heavy)], [μ_C—H] and

[ ∑ i R . C .  α i ]  :

wherein the circle together with Fe—O indicates an oxyferryl intermediate; [δ_het] indicates the net atomic charge of a heteroatom in the alpha-position relative to the reaction center; [max(δ_heavy)] indicates the highest atomic charge in X¹, X² and X³ which are neither hydrogen nor helium; [μ_C—H] indicates the bond dipole of the carbon-hydrogen bond; and

[ ∑ i R . C .  α i ]

indicates the sum of the atomic polarizabilities of H, C, X¹, X² and X³.

According to the present invention, the atomic descriptors [δ_het] and [max(δ_heavy)] can be calculated, and activation energy can be calculated according to the following equation 1-1:

E_a^Habs—(B)=25.94+1.88*[δ_het]d+1.03*[max(δ_heavy)]  [Equation 1-1]

wherein E_a^Habs—(B) indicates activation energy required for abstraction of hydrogen attached to a carbon atom having a heteroatom in the alpha-position relative to the reaction center.

Also, according to the present invention, the atomic descriptors [μ_C—H] and

[ ∑ i R . C .  α i ]

can be calculated, and activation energy can be calculated according to the following equation 1-2:

E a Habs_  ( A ) = 28.50 - 2.22 * [ μ C - H ] + 1.12 * [ ∑ i R . C .  α i ] [ Equation   1  -  2 ]

wherein E_a^Habs—(A) indicates activation energy required for abstraction of hydrogen attached to a carbon atom having no heteroatom in the alpha-position relative to the reaction center.

In step (iii), the activation energy for tetrahedral intermediate formation in cytochrome P950-mediated aromatic hydroxylation for a substrate of the following formula 2 can be predicted using the atomic descriptors [δ_H] and [mean(α_alpha)]:

wherein the circle together with Fe—O indicates an oxyferryl intermediate; [δ_H] indicates the net atomic charge of the hydrogen of the substrate; and [mean(α_alpha)] indicates the mean value of the polarizabilities of adjacent carbon atoms.

According to the present invention, the atomic descriptors [δ_H] and [mean(α_alpha)] can be calculated, and activation energy can be calculated according to the following equations:

E_a^aro—o,p=21.34−0.75*[δ_H]−1.24*[mean)α_alpha)]  [Equation 2-1]

E_a^aro—m=22.14−0.68*[δ_H]−0.83*[mean(α_alpha)]  [Equation 2-2]

E_a^aro—0,2,3=21.02−1.49*[δ_H]−0.92*[mean(α_alpha)]  [Equation 2-3]

wherein E_a^aro—o,p indicates the activation energy for tetrahedral intermediate formation in a benzene having one substituent in the ortho/para-position; E_a^aro—m indicates the activation energy for tetrahedral intermediate formation in a benzene having one substituent in the meta-position; and E_a^aro—0,2,3 indicates the activation energy for tetrahedral intermediate formation in a benzene having 0, 3 or 3 substituents.

In another aspect, the present invention provides a method for predicting a metabolite using the activated energy predicted by said method. Herein, an atomic position having the lowest activation energy can be predicted as a position where metabolism occurs.

In still another aspect, the present invention provides a method of predicting a drug-drug interaction through the activation energy predicted by said method.

As used herein, the term “drug-drug interaction” refers to the effects that occur when two or more drugs are used at the same time. Such effects include changes in the kinetics of drug absorption by the intestinal tract, changes in the rate of detoxification and elimination of the drug by the liver or other organs, new or enhanced side effects and changes in the drug's activity. CYP2C9 which is a CYP isoform is one of the major enzymes that are involved in the phase I metabolism of drugs. The inhibition of this enzyme can result in an undesirable drug-drug interaction or drug toxicity [see Lin, J. H.; Lu, A. Y. H., Inhibition and induction of cytochrome P450 and the clinical implications. Clin. Pharmacokinet. 1998, 35 (5), 361-390]. Namely, if the activation energy of a substrate is relatively high, metabolism can be inhibited to result in the inhibition of CYP450 enzymes, thus causing an undesirable drug-drug interaction.

Also, metabolites, obtained by oxidation or reduction of substrates by cytochromes, can cause toxicities such as chemical carcinogenesis or mutagenesis, and for this reason, it is very important to predict metabolites, including substrate specificity for cytochromes (Vermeulen NPE, Donne-Op den Kelder G, Commandeur JNM. Molecular mechanisms of toxicology and drug design, in Trend in Drug Research, Proc. 7^th Noordwijkerhout-Camerino Symp., Claassen, V., Ed., Elsevier Science Publishers, Amsterdam, 1990, 253).

The present invention provides a method of predicting a metabolite of a CYP450 enzyme by predicting binding possibility using an atomic-type fingerprint descriptor, which includes the type of atom and the surrounding bond order, and by predicting reactivity using an atomic descriptor. The method of the present invention solves a time-consuming problem in predicting accessibility using the three-dimensional structure of a CYP450 enzyme and does not require any quantum mechanical calculation or experiment.

The atomic fingerprint descriptor for predicting the possibility of binding of a cytochrome P450 enzyme to a substrate can be expressed as follows:

The atomic fingerprint descriptor consists of: the element symbol of an atom; a bond order indicating the number of atoms bonded; a ring indicator that indicates whether the atom is part of a ring; and an aromatic indicator that indicates whether the atom is one included in an aromatic group. This expression method intuitively and simply expresses the type of atom and the surrounding bonding environment. However, the atomic fingerprint descriptor has its own information, but does not have the surrounding bonded atoms, and for this reason, the surrounding environment is reflected by writing the surrounding bonded atomic fingerprint descriptors therewith. The larger the connectivity, the more the information of the surrounding environment is included. However, if atomic fingerprint descriptors become excessively large, over-fitting can occur. In the present invention, when the information of atoms connected directly to the atomic fingerprint descriptor was used, the most efficient calculation results were shown.

If atomic fingerprint descriptors for all the atomic positions of a substrate are the same as the atomic fingerprint descriptions of the metabolism of the substrate used in a training set, it is determined to be “on”, and if not so, it is determined to be “off”. Then, since the positions where metabolic reactions can occur were determined, the prediction of reactivity is performed by calculating activation energy, and the relative order of priority is determined.

Prediction of the reactivity of cytochrome P450 enzymes with the substrates was carried out using the calculation methods described in Korean Patent Application No. 10-2008-0112389 (entitled “Method for predicting activation energy using effective atomic descriptors).

Finally, the prediction of metabolic reactions of cytochrome P450 enzymes with the substrates is performed through the prediction of binding possibility and the prediction of reactivity, and the activation energies of individual positions are calculated using reactivity prediction models. The activation energies are arranged in the order of lower to higher energy, and three positions having lower activation energies are determined to be positions at which metabolic reactions can occur. The analysis of the results is carried out by determining whether the two positions selected as described include an experimentally known metabolic position.

To achieve another object, the present invention provides a method of predicting the activation energy for CYP450-mediated hydrogen abstraction according to an equation including an effective atomic descriptor. This method of the present invention is fast and accurate and does not require any quantum mechanical calculation or experiment.

Hydrogen abstraction by a cytochrome P450 enzyme may be shown in the following reaction scheme 1:

wherein the circle together with Fe—O indicates an oxyferryl intermediate.

The present invention provides a method of predicting the activation energy for cytochrome P450-mediated hydrogen abstraction from a substrate of the following formula 1 using the atomic descriptors [δ_het], [max (δ_heavy)], [_C—H] and

[ ∑ i R . C .  α i ]  :

wherein the circle together with Fe—O indicates an oxyferryl intermediate; [δ_het] indicates the net atomic charge of a heteroatom in the alpha-position relative to the reaction center; [max(δ_heavy)] indicates the highest atomic charge in X¹, X² and X³ which are neither hydrogen nor helium; [μ_C—H] indicates the bond dipole of the carbon-hydrogen bond; and

[ ∑ i R . C .  α i ]

indicates the sum of the atomic polarizabilities of the atoms H, C, X¹, X² and X³. The present invention can be applied to all CYP 450 enzymes, and it is apparent that the present invention can be applied particularly to human CYP 450 enzymes. The cytochrome P450 enzymes according to the present invention include, but are not limited to, CYP2E1, CYP3A4, CYP2B6, CYP2C8, CYP2C9, CYP1A1, CYP1A2, CYP2C19, CYP2D6, CYP1B1, and CYP2A6.

In the method of predicting the activation energy, any C—H bond to a target molecule can be recognized as a position where metabolism can occur in the target molecule. If the C atom of the C—H bond of the target molecule is aliphatic carbon, it can be determined to be a position where hydrogen abstraction can occur.

In hydrogen abstraction by the CYP450 enzyme, the type of atom can be determined depending on whether a heteroatom is present or not in the alpha-position with respect to the reaction center (C—H where actual metabolism occurs).

If there is a heteroatom in the alpha-position relative to the reaction center, the atomic descriptors [δ_het] and [max(δ_heavy)] can be calculated, and the activation energy for hydrogen abstraction can be predicted according to the following equation 1-1:

E_a^Habs—(B)=25.94+1.88*[δ_het]d+1.03*[max(δ_heavy)]  [Equation 1-1]

wherein E_a^Habs—(B) indicates activation energy required for abstraction of hydrogen attached to a carbon atom having a heteroatom in the alpha-position relative to the reaction center.

If there is no heteroatom in the alpha-position relative to the reaction center, the atomic descriptors can be calculated, and the activation energy for hydrogen abstraction can be predicted according to the following equation [1-2]:

E a Habs_  ( A ) = 28.50 - 2.22 * [ μ C - H ] + 1.12 * [ ∑ i R . C .  α i ] [ Equation   1  -  2 ]

wherein E_a^Habs—(A) indicates activation energy required for abstraction of hydrogen attached to a carbon atom having no heteroatom in the alpha-position relative to the reaction center.

To achieve another object, the present invention provides a method of predicting the activation energy for CYP450-mediated aromatic hydroxylation according to an equation including an effective atomic descriptor. The method of the present invention is fast and accurate and does not require any quantum mechanical calculation or experiment.

The tetrahedral intermediate formation reaction in cytochrome P450-mediated aromatic hydroxylation may be shown in the following reaction scheme 2:

wherein the circle together with Fe—O indicates an oxyferryl intermediate.

The present invention provides a method of predicting the activation energy for tetrahedral intermediate formation in cytochrome P450-mediated aromatic hydroxylation for a substrate of the following formula 2 using the atomic descriptors [δ_H] and [mean (α_alpha)]:

wherein the circle together with Fe—O indicates an oxyferryl intermediate; [δ_H] indicates the net atomic charge of the hydrogen; and [mean(α_alpha)] indicates the mean value of the polarizabilities of adjacent carbon atoms.

The present invention may be applied to all CYP 450 enzymes, and it is apparent that the present invention can be applied particularly to human CYP 450 enzymes. The cytochrome P450 enzymes according to the present invention include, but are not limited to, CYP2E1, CYP3A4, CYP2B6, CYP2C8, CYP2C9, CYP1A1, CYP1A2, CYP2C19, CYP2D6, CYP1B1, and CYP2A6.

In the method of predicting the activation energy for tetrahedral intermediate formation, any C—H bond to a target molecule can be determined to be a position where metabolism can occur in the target molecule. Also, if the C atom of the C—H bond of the target molecule is aromatic carbon, it can be determined to be a metabolic position where aromatic hydroxylation can occur.

According to the present invention, the atomic descriptors [δ_H] and [mean(α_alpha)] can be calculated, and the activation energy can be predicted according to the following equations:

E_a^aro—o,p=21.34−0.75*[δ_H]−1.24*[mean)α_alpha)]  [Equation 2-1]

E_a^aro—m=22.14−0.68*[δ_H]−0.83*[mean(α_alpha)]  [Equation 2-2]

E_a^aro—0,2,3=21.02−1.49*[δ_H]−0.92*[mean(α_alpha)]  [Equation 2-3]

wherein E_a^aro—o,p indicates the activation energy for tetrahedral intermediate formation in a benzene having one substituent in the ortho/para-position; E_a^aro—m indicates the activation energy for tetrahedral intermediate formation in a benzene having one substituent in the meta-position; and E_a^aro—m indicates the activation energy for tetrahedral intermediate formation in a benzene having 0, 2 or 3 substituents.

In another aspect, the present invention provides a method of predicting the relative rate of metabolism (k) according to the following Arrhenius equation 2 using the activation energy predicted by said method:

k=Ae^−Ea/RT  [Equation 2]

wherein k is a reaction rate constant, A is a frequency factor, E_a is activation energy, R is a gas constant, and T is absolute temperature.

The reason why the above equation 2 was designed is because of the atomic fraction f=e^−Ea/RT exceeding activation energy. Namely, because only a molecule exceeding activation energy can cause a reaction, the reaction rate constant is determined by the ratio exceeding activation energy.

In another aspect, the present invention provides a method of predicting metabolic regioselectivity using the activation energy predicted by said method.

More specifically, the present invention provides a method of predicting the relative rate of metabolism according to the Arrhenius equation using the activation energy predicted by said method and predicting metabolic regioselectivity according to the following reaction scheme 3 and equation 3 using the predicted relative rate of metabolism:

P 1 P 2 = [ ES 1 ] [ ES 2 ]  k 5 k 6 [ Equation   3 ]

wherein P indicates the relative probability of formation of any metabolite of all possible metabolites of a substrate, E is an enzyme, S is a substrate, ES is an enzyme-substrate complex, [ES] is the concentration of the enzyme-substrate complex, and k is a reaction rate constant.

Namely, once the reaction rate of each atom in one molecule is determined according to the Arrhenius equation, the regioselectivity in the molecule can be determined, because metabolism occurs as the reaction rate decreases. [see Higgins, L.; Korzekwa, K. R.; Rao, S.; Shou, M.; Jones, J. P., An assessment of the reaction energetics for cytochrome P450-mediated reactions. Arch. Biochem. Biophys. 2001, 385, 220-230].

In still another aspect, the present invention provides a method of predicting the inhibition of metabolism using the activation energy predicted by said method. For example, it can be predicted that, if a substrate has relatively high activation energy, the substrate will not be metabolized, and thus will remain in the active site of CYP450 enzymes.

In still another aspect, the present invention provides a method of predicting a drug-drug interaction using the activation energy predicted by said method.

As used herein, the term “atomic fingerprint descriptor” refers to a value defined to express the type of atom and the surrounding bonding environment. It consists of the element symbol of an atom, a bond order indicating the number of atoms bonded, a ring indicator that indicates whether the atom is part of a ring, and an aromatic indicator that indicates whether the atom is one included in an aromatic group.

As used herein, the term “atomic descriptor” refers to a value defined to reflect the properties of an atom itself and the bonding environment of the atom. Examples of atomic descriptors that are used in the present invention include, but are not limited to, [δ_het], [max(δ_heavy)], [μ_C—H].

[ ∑ i R . C .  α i ] ,

[δ_H], [mean(α_alpha)], etc.

BRIEF DESCRIPTION OF THE DRAWINGS

The above and other objects, features and advantages of the present invention will be more clearly understood from the following detailed description taken in conjunction with the accompanying drawing, in which:

FIG. 1 is a flowchart showing a method of constructing a database of atomic fingerprint descriptors according to the present invention;

FIG. 2 is a flowchart showing a method of predicting activation energy using an atomic fingerprint descriptor and an atomic descriptor according to the present invention and predicting i) a metabolite, ii) the relative rate of metabolism, iii) the regioselectivity of metabolism, iv) the inhibition of metabolism, v) a drug-drug interaction, and vi) the toxicity of a metabolite;

FIG. 3 is a flowchart showing a method of predicting activation energy using atomic descriptors according to the present invention;

FIG. 4 shows the correlation between the quantum-mechanically calculated activation energy (QM E_a) for CYP450-mediated hydrogen abstraction and the activation energy (Predicted E_a) predicted according to the present invention; and

FIG. 5 shows the correlation between the quantum-mechanically calculated activation energy (QM E_a) for CYP450-mediated aromatic hydroxylation and the activation energy (Predicted E_a) predicted according to the present invention.

DETAILED DESCRIPTION OF THE INVENTION

Hereinafter, the elements and technical features of the present invention will be described in further detail with reference to examples. It is to be understood, however, that these examples are for illustrative purposes only and are not to be construed to limit the scope of the present invention. All literature cited herein is incorporated by reference.

EXAMPLES

Example 1

Construction of Database of Atomic Fingerprint Descriptors

As shown in FIG. 1, the present inventors constructed a database of atomic fingerprint descriptors through a training method comprising the following steps (see FIG. 1):

(i) calculating the atomic fingerprint descriptor of a substrate, which is represented by the following equation 1;

(ii) predicting activation energy for an atomic position using an atomic descriptor;

(iii) predicting cytochrome P450-mediated metabolism using the predicted activation energy; and

(iv) comparing the predicted metabolism with experimental metabolism and storing whether the predicted metabolism occurs:

Xabc  [Equation 1]

wherein X is the chemical symbol of an atom; a is a bond order that indicates the number of atoms bonded; b is a ring indicator that indicates whether the atom is part of a ring; and c is an aromatic indicator that indicates whether the atom is an aromatic atom.

Using the above-constructed database of atomic fingerprint descriptors, the possibility of reaction of the atomic fingerprint descriptor of a given substrate with each CYP450 isoform was analyzed.

TABLE 1
Results of analysis for possibility of reaction of a given substrate
using constructed atomic fingerprint descriptor database

	Atomic fingerprint
NO.	descriptors	CYP1A2	CYP2C9	CYP2D6	CYP3A4

1	C400C400H100H100H100	1	1	−1	1
2	C400C361C400H100H100	1	1	−1	1
3	C361C361C361H100	1	1	1	1
4	C400C361H100H100H100	1	1	1	1
5	C460C460H100H100N360	−1	1	1	1
6	C361C361H100N261	−1	−1	−1	1
7	C460C460C460C460H100	−1	−1	−1	−1
8	C460C361C460C460H100	−1	−1	−1	−1
9	C400C460H100H100H100	−1	−1	−1	−1
10	C400C400C400H100H100	1	1	1	1
11	C460C360C460H100H100	1	1	1	1
12	C360C360C460H100	1	0	−1	1
13	C300C360H100O100	0	0	0	1
14	C400C400C400H100N300	1	1	1	1
15	C400H100H100H100O200	1	1	1	1
16	C400C400H100H100N300	1	1	1	1
17	C400C351C400H100H100	−1	1	−1	−1
18	C361C351C361H100	1	−1	1	1
19	C351C351H100N351	−1	−1	−1	−1
20	C400H100H100H100N300	1	1	1	1
21	C460C400C460C460H100	0	−1	−1	−1
22	C460C460C460H100H100	−1	−1	1	1
23	C460C360C460C460H100	0	−1	0	−1

In Table 1 above, “1” indicates that, in a training set, there is a case in which a reaction occurred in a site having the relevant atomic fingerprint descriptor. “−1” indicates that, in a training set, there is no case in which a reaction occurred in a site having the relevant atomic fingerprint descriptor. “0” indicates that an atom having the relevant atomic fingerprint descriptor does not exist in a training set.

Example 2

Prediction of Metabolite of 2-Methoxyamphetamine Using the Prediction Method of the Present Invention

As shown in FIG. 2, the present inventors predicted activation energy using a method comprising the following steps (see FIG. 2):

(i) calculating the atomic fingerprint descriptor of a substrate, which is represented by the following formula 1;

(ii) comparing the calculated atomic fingerprint descriptor with the data, constructed by the method of Example 1, to select an atomic position where cytochrome P450-mediated metabolism can occur; and

(iii) predicting activation energy for the selected atomic position using an atomic descriptor:

Xabc  [Equation 1]

wherein X is the chemical symbol of an atom; a is a bond order that indicates the number of atoms bonded; b is a ring indicator that indicates whether the atom is part of a ring; and c is an aromatic indicator that indicates whether the atom is an aromatic atom.

After predicting the activation energy of 2-methoxyamphetamine using the above method, the metabolite of 2-methoxyamphetamine was predicted. 2-methoxyamphetamine has a chemical structure of the following formula 3:

First, the positions of carbon atoms having hydrogen at positions 1, 2, 3, 6, 7, 8, 9 and 10 were examined.

Then, the atomic fingerprint descriptors of positions 1, 2, 3, 6, 7, 8, 9 and 10 were calculated and compared with the atomic fingerprint descriptor database constructed in Example 10, thereby selecting an atomic position where metabolism may occur (see Table 1).

TABLE 2
Selection of atomic positions having the possibility of metabolism
through the comparison of atomic fingerprint descriptors

Atomic	Atomic fingerprint	Results of	Possibility
position	descriptor	comparison	of metabolism

Atom 1	C400C400H100H100H100	−1	Impossible
Atom 2	C400C400C400H100N300	1	Possible
Atom 3	C400C361C400H100H100	−1	Impossible
Atom 6	C361C361C361H100	1	Possible
Atom 7	C361C361C361H100	1	Possible
Atom 8	C361C361C361H100	1	Possible
Atom 9	C361C361C361H100	1	Possible
Atom 10	C400H100H100H100O200	1	Possible

Then, the activation energies of the atomic positions having the possibility of metabolism were calculated.

TABLE 3
Calculation of activation energies of atomic positions
having the possibility of metabolism (see Example 6)

	Atomic position	Activation energy
	Atom 2	22.93
	Atom 6	25.60
	Atom 7	27.42
	Atom 8	27.25
	Atom 9	27.30
	Atom 10	22.22

Then, atomic position 10 having the lowest activation energy was predicted as a position where a reaction occurs. Also, the following metabolite (formula 4) where O-dealkylation occurred at position 10 was predicted in the following manner.

Example 3

Prediction of Metabolite Using Only Reactivity Prediction Model

A metabolite was predicted only with a reactivity prediction model without considering the binding possibility of a substrate. When analysis was carried out using a method of selecting two positions having the highest possibility, a predictability of about 62-70% was generally shown.

TABLE 4
Results of metabolite prediction carried out
using only reactivity prediction model

	Na	Ncb	Nc/N (%)

	CYP1A2	144	101	70.1
	CYP2C9	119	83	69.7
	CYP2D6	146	91	62.3
	CYP3A4	196	128	65.3
	aNumber of substrates used in training;
	bNumber of substrates that accurately reproduced an experimentally known metabolism.

Example 4

Prediction of Metabolite Using Accessibility Prediction Model and Reactivity Prediction Model

In order to add the possibility of binding of various CYP450 enzymes to substrates, atomic fingerprint descriptors were used. A total of 185 atomic fingerprint descriptors were used, and the possibility of metabolism by each CYP450 isoform was analyzed. Using a combination of an accessibility prediction model and a reactivity prediction model, two positions having the highest possibility and experimentally known metabolic positions were comparatively analyzed, and the results of the analysis are shown in Table 5 below.

	TABLE 5
	Na	Ncb	Nc/N (%)

	CYP1A2	144	112	77.8
	CYP2C9	119	92	77.3
	CYP2D6	146	102	69.9
	CYP3A4	196	145	74.0
	aNumber of substrates used in training;
	bNumber of substrates that accurately reproduced an experimentally known metabolism.

Generally, a predictability of 70-78% was shown, and the predictability was more than 5% higher than that of Example 3 in which only the reactivity prediction model was used.

For reference, the substrates used in the metabolite prediction training according to each cytochrome P450 isoform in Tables 4 and 5 above are shown in the following Tables.

TABLE 6
Substrates used in training for prediction
of metabolites with CYP1A2 (144 cases)

	Substrate

1	1-ethylpyrene
2	1-methylpyrene
3	2,3,7-trichlorooxanthrene
4	(5S)-5-(3-hydroxyphenyl)-5-phenylimidazolidine-2,4-dione
5	(5S)-5-(4-hydroxyphenyl)-5-phenylimidazolidine-2,4-dione
6	7-ethoxy-4-(trifluoromethyl)-2H-chromen-2-one
7	7-ethoxycoumarin
8	7-ethoxyresorufin
9	7-methoxyresorufin
10	1-[(2S)-4-(5-benzylthiophen-2-yl)but-3-yn-2-yl]urea
11	aflatoxin-b1
12	all-trans-retinol
13	almotriptan
14	Ametryne
15	amitriptyline
16	amodiaquine
17	Antipyrine
18	Apigenin
19	atomoxetine
20	Atrazine
21	Azelastine
22	7-(benzyloxy)-4-(trifluoromethyl)-2H-chromen-2-one
23	Biochainin-a
24	bropirimine
25	Bufuralol
26	Bunitrolol
27	bupivacaine
28	Capsaicin
29	carbamazepine
30	Carbaryl
31	Carbofuran
32	Carvedilol
33	7-ethoxy-2-oxo-2H-chromene-3-carbonitrile
34	Celecoxib
35	chloroquine
36	chlorpromazine
37	chlorpropamide
38	Cilostazol
39	Cisapride
40	clomipramine
41	clozapine
42	2-chloro-3-(pyridin-3-yl)-5,6,7,8-tetrahydroindolizine-1-
	carboxamide
43	curcumin
44	cyclobenzaprine
45	dacarbazine
46	dimethyl 7,7′-dimethoxy-4,4′-bi-1,3-benzodioxole-5,5′-
	dicarboxylate
47	deprenyl
48	dextromethorphan
49	dibenzo-a-h-anthracene
50	diclofenac
51	dihydrodiol
52	dimethoxyisoflavone
53	dimethyloxoxanthene
54	dimmamc
55	domperidone
56	doxepin
57	eletriptan
58	ellipticine
59	estradiol-methyl-ether
60	estrone
61	etoricoxib
62	fenproporex
63	fluoxetine
64	flurbiprofen
65	formononetin
66	N-[2-(5-methoxy-1H-indol-3-yl)ethyl]-N-(propan-2-
	yl)propan-2-amine
67	galangin
68	genistein
69	2-[(R)-{[5-(cyclopropylmethoxy)pyridin-3-
	yl]methyl}sulfinyl]-5-fluoro-1H-benzimidazole
70	harmaline
71	harmine
72	hesperetin
73	imipramine
74	kaempferide
75	kaempferol
76	N-carbamimidoyl-4-cyano-1-benzothiophene-2-carboxamide
77	levobupivacaine
78	lidocaine
79	loratadine
80	4-(aminomethyl)-7-methoxy-2H-chromen-2-one
81	maprotiline
82	(2R)-1-(1,3-benzodioxol-5-yl)-N-ethylpropan-2-amine
83	(2R)-1-(1,3-benzodioxol-5-yl)-N-methylpropan-2-amine
84	3,8-dimethyl-3H-imidazo[4,5-f]quinoxalin-2-amine
85	melatonin
86	mephenytoin
87	methoxychlor
88	methoxychlor-mono-oh
89	methyleugenol
90	metoclopramide
91	mexiletine
92	mianserin
93	mirtazapine
94	(2S)-1-(4-methylphenyl)-2-(pyrrolidin-1-yl)propan-1-one
95	1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine
96	n-nitrosodiamylamine
97	naproxen
98	naringenin
99	nefiracetam
100	nn-dimethyl-m-toluamide
101	4-[methyl(nitroso)amino]-1-(pyridin-3-yl)butan-1-one
102	nordiazepam
103	nortriptyline
104	ochratoxin-a
105	olanzapine
106	olopatadine
107	(3S)-3-[3-(methylsulfonyl)phenyl]-1-propylpiperidine
108	oxycodone
109	perazine
110	perphenazine
111	phenytoin
112	1-methyl-6-phenyl-1H-imidazo[4,5-b]pyridin-2-amine
113	pimobendan
114	progesterone
115	propafenone
116	propanolol
117	prunetin
118	pyrazoloacridine
119	quinacrine
120	ropinirole
121	ropivacaine
122	rosiglitazone
123	safrole
124	sertraline
125	sildenafil
126	stilbene
127	(3Z)-3-[(3,5-dimethyl-1H-pyrrol-2-yl)methylidene]-1,3-
	dihydro-2H-indol-2-one
128	tacrine
129	tamarixetin
130	tangeretin
131	tauromustine
132	terbinafine
133	terbuthylazine
134	testosterone
135	theobromine
136	theophylline
137	tolperisone
138	N-(2,6-dichlorobenzoyl)-4-(2,6-dimethoxy-phenyl)-L-
	phenylalanine
139	trans-retinoic-acid
140	warfarin
141	zileuton
142	zolmitriptan
143	zolpidem
144	zotepine

TABLE 7
Substrates used in training for prediction of metabolites
5 with CYP2C9 (119 cases)

		Substrate

	1	2n-propylquinoline
	2	(5S)-5-(3-hydroxyphenyl)-
		5-phenylimidazolidine-
		2,4-dione
	3	(5S)-5-(4-hydroxyphenyl)-
		5-phenylimidazolidine-
		2,4-dione
	4	5-hydroxytryptamine
	5	2-(trans-4-tert-
		butylcyclohexyl)-3-
		hydroxynaphthalene-1,4-
		dione
	6	7-ethoxy-4-
		(trifluoromethyl)-2H-
		chromen-2-one
	7	7-ethoxycoumarin
	8	7-ethoxyresorufin
	9	9-cis-retinoic-acid
	10	1-[(2S)-4-(5-
		benzylthiophen-2-yl)but-3-
		yn-2-yl]urea
	11	aceclofenac
	12	Ametryne
	13	amitriptyline
	14	Antipyrine
	15	atomoxetine
	16	7-(benzyloxy)-4-
		(trifluoromethyl)-2H-
		chromen-2-one
	17	Biochainin-a
	18	Bufuralol
	19	Capsaicin
	20	carbamazepine
	21	Carvedilol
	22	7-ethoxy-2-oxo-2H-
		chromene-3-carbonitrile
	23	Celecoxib
	24	chlorpropamide
	25	Cisapride
	26	clomipramine
	27	Clozapine
	28	2-chloro-3-(pyridin-3-yl)-
		5,6,7,8-
		tetrahydroindolizine-1-
		carboxamide
	29	N,4-dimethyl-N-(1-phenyl-
		1H-pyrazol-5-
		yl)benzenesulfonamide
	30	2-[(3S,4R)-3-benzyl-4-
		hydroxy-3,4-dihydro-2H-
		chromen-7-yl]-4-
		(trifluoromethyl)benzoic
		acid
	31	cyclophosphamide
	32	dimethyl 7,7′-dimethoxy-
		4,4′-bi-1,3-benzodioxole-
		5,5′-dicarboxylate
	33	Deprenyl
	34	dexloxiglumide
	35	dextromethorphan
	36	Diazepam
	37	dibenzo-a-h-anthracene
	38	Diclofenac
	39	Diltiazem
	40	disopyramide
	41	doxepin
	42	eletriptan
	43	ellipticine
	44	estradiol
	45	estradiol-methyl-ether
	46	estrone
	47	etodolac
	48	etoperidone
	49	Fluoxetine.
	50	flurbiprofen
	51	fluvastatin
	52	N-[2-(5-methoxy-1H-indol-
		3-yl)ethyl-N-(propan-2-
		yl)propan-2-amine
	53	galangin
	54	2-[(R)-{(5-
		(cyclopropylmethoxy)pyridin-
		3-yl)methyl}sulfinyl]-5-
		fluoro-1H-benzimidazole
	55	harmaline
	56	harmine
	57	hydromorphone
	58	ibuprofen
	59	ifosfamide
	60	imipramine
	61	indomethacin
	62	kaempferide
	63	ketamine
	64	(1S,4S)-(6-dimethylamino-
		4,4-diphenyl-heptan-3-
		yl)acetate
	65	lansoprazole
	66	lidocaine
	67	loratadine
	68	lomoxicam
	69	losartan
	70	luciferin
	71	[(4E)-7-chloro-4-
		[(sulfooxy)imino]-3,4-
		dihydroquinolin-1(2H)-
		yl](2-
		methylphenyl)methanone
	72	mefenamic-acid
	73	melatonin
	74	meloxicam
	75	mephenytoin
	76	methadone
	77	methoxychlor-mono-oh
	78	methyleugenol
	79	mianserin
	80	midazolam
	81	mirtazapine
	82	4-({[(5S)-2,4-dioxo-1,3-thiazolidin-
		5-yl]methyl}-2-methoxy-N-[4-
		(trifluoromethyl)benzyl]benzamide
	83	(2S)-1-(4-methylphenyl)-2-
		(pyrrolidin-l-yl)propan-1-one
	84	n-nitrosodiamylamine
	85	naproxen
	86	nevirapine
	87	ochratoxin-a
	88	omeprazole
	89	oxybutynin
	90	oxycodone
	91	perazine
	92	perphenazine
	93	phenacetin
	94	phencyclidine
	95	phenprocoumon
	96	phenytoin
	97	piroxicam
	98	progesterone
	99	rosiglitazone
	100	(5Z)-7-[(1S,2R,3R,4R)-3-
		benzenesulfonamidobicyclo[2.2.1]
		heptan-2-yl]hept-5-enoic acid
	101	sertraline
	102	sildenafil
	103	7-chloro-N-({5-
		[(dimethylamino)methyl]cyclopent
		a-1,4-dien-l-yl}methyl)quinolin-4-
		amine
	104	tamarixetin
	105	tauromustine
	106	temazepam
	107	terbinafine
	108	testosterone
	109	theophylline
	110	tolbutamide
	111	torasemide
	112	N-(2,6-dichlorobenzoyl)-4-(2,6-
		dimethoxy-phenyl)-L-
		phenylalanine
	113	trans-retinoic-acid
	114	valdecoxib
	115	valsartan
	116	venlafaxine
	117	vivid-red
	118	warfarin
	119	zolpidem

TABLE 8
Substrates used in training for prediction
of metabolites with CYP2D6 (146 cases)

	Substrate

1	2-(piperazin-1-yl)pyrimidine
2	2-methoxyamphetamine
3	4-methoxyamphetamine
4	2-(5-methoxy-1H-indol-3-yl)-N,N-dimethylethanamine
5	5-methoxytryptamine
6	5-methoxytryptamine
7	7-ethoxycoumarin
8	all-trans-retinol
9	all-trans-retinol
10	amitriptyline
11	amodiaquine
12	aripiprazole
13	atomoxetine
14	atrazine
15	azelastine
16	biochainin-a
17	bisoprolol
18	N-({4-[(5-bromopyrimidin-2-yl)oxy]-3-
	methylphenyl}carbamoyl)-2-(dimethylamino)benzamide
19	brofaromine
20	bunitrolol
21	bupivacaine
22	capsaicin
23	carbamazepcapsaicinine
24	carbamazepcapsaicinine
25	carbofuran
26	C arvedilol
27	7-ethoxy-2-oxo-2H-chromene-3-carbonitrile
28	celecoxib
29	celecoxib
30	chlorpromazine
31	chlorpropamide
32	cibenzoline
33	cilostazol
34	cisapride
35	citalopram
36	clomipramine
37	clozapine
38	codeine
39	curcumin
40	cyclophosphamide
41	delavirdine
42	deprenyl
43	dextromethorphan
44	diclofenac
45	dihydrocodeine
46	diltiazem
47	dimmamc
48	domperidone
49	doxepin
50	2-(hydroxymethyl)-4-[5-(4-methoxyphenyl)-3-
	(trifluoromethyl)-1H-pyrazol-1-yl]benzenesulfonamide
51	eletriptan
52	ellipticine
53	estradiol
54	estrone
55	etoperidone
56	etoricoxib
57	eugenol
58	fenproporex
59	fluoxetine
60	fluvastatin
61	N-[2-(5-methoxy-1H-indol-3-yl)ethyl]-N-(propan-2-
	yl)propan-2-amine
62	galantamine
63	gefitinib
64	genistein
65	granisetron
66	harmaline
67	harmine
68	hydrocodone
69	hydromorphone
70	ibogaine
71	iloperidone
72	imipramine
73	cilostazol
74	(1S,4S)-(6-dimethylamino-4,4-diphenyl-heptan-3-yl)acetate
75	lidocaine
76	loratadine
77	4-(aminomethyl)-7-methoxy-2H-chromen-2-one
78	maprotiline
79	(2R)-1-(1,3-benzodioxol-5-yl)-N-ethylpropan-2-amine
80	(2R)-1-(1,3-benzodioxol-5-yl)-N-methylpropan-2-amine
81	melatonin
82	mequitazine
83	meta-chlorophenylpiperazine
84	methadone
85	methadone
86	methoxychlor-mono-oh
87	methoxyphenamine
88	methyleugenol
89	metoclopramide
90	metoprolol
91	mexiletine
92	mianserin
93	minaprine
94	mirtazapine
95	(2S)-1-(4-methoxyphenyl)-2-(pyrrolidin-1-yl)propan-1-one
96	(2S)-1-(4-methylphenyl)-2-(pyrrolidin-1-yl)propan-1-one
97	1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine
98	n-nitrosodiamylamine
99	nevirapine
100	4-[methyl(nitroso)amino]-1-(pyridin-3-yl)butan-1-one
101	nortriptyline
102	5-(diethylamino)-2-methylpent-3-yn-2-yl(2S)-2-cyclohexyl-
	2-hydroxy-2-phenylacetate
103	olanzapine
104	omeprazole
105	ondansetron
106	(3S)-3-[3-(methylsulfonyl)phenyl]-1-propylpiperidine
107	oxybutynin
108	oxycodone
109	perazine
110	perphenazine
111	phenacetin
112	phencyclidine
113	phenformin
114	phenytoin
115	pinoline
116	(2R)-1-(4-methoxyphenyl)-N-methylpropan-2-amine
117	procainamide
118	progesterone
119	promethazine
120	propafenone
121	propanolol
122	3-(2-chlorophenyl)-N-[(1S)-1-(3-
	methoxyphenyl)ethyl]propan-1-amine
123	reduced-dolasetron
124	ropivacaine
125	sertraline
126	sildenafil
127	sparteine
128	spirosulfonamide
129	7-chloro-N-({5-[(dimethylamino)methyl]cyclopenta-1,4-
	dien-1-yl}methyl)quinolin-4-amine
130	stilbene
131	stilbene
132	tangeretin
133	tauromustine
134	tegaserod
135	testosterone
136	theophylline
137	tolperisone
138	tramadol
139	traxoprodil
140	tropisetron
141	valdecoxib
142	venlafaxine
143	warfarin
144	yohimbine
145	zolpidem
146	zotepine

TABLE 9
Substrates used in training for prediction
of metabolites with CYP3A4 (196 cases)

	Substrate

1	1-ethylpyrene
2	1-methylpyrene
3	2n-propylquinoline
4	(5S)-5-(3-hydroxyphenyl)-5-phenylimidazolidine-2,4-dione
5	(5S)-5-(4-hydroxyphenyl)-5-phenylimidazolidine-2,4-dione
6	5-methylchrysene
7	1-ethoxycoumarin
8	7-methoxyresorufin
9	1-[(2S)-4-(5-benzylthiophen-2-yl)but-3-yn-2-yl]urea
10	1-[(2S)-4-(5-benzylthiophen-2-yl)but-3-yn-2-yl]urea
11	acetochlor
12	adinazolam
13	aflatoxin-b1
14	alachlor
15	alfentanil
16	all-trans-retinol
17	almotriptan
18	dextromethorphan
19	ambroxol
20	ametryne
21	amitriptyline
22	amodiaquine
23	androstenedione
24	apigenin
25	aripiprazole
26	atomoxetine
27	atrazine
28	azelastine
29	7-(benzyloxy)-4-(trifluoromethyl)-2H-chromen-2-one
30	bisoprolol
31	N-({4-[(5-bromopyrimidin-2-yl)oxy]-3-
	methylphenyl}carbamoyl)-2-(dimethylamino)benzamide
32	brotizolam
33	budesonide
34	bufuralol
35	bupivacaine
36	bupropion
37	capsaicin
38	carbamazepine
39	carbaryl
40	carbofuran
41	carvedilol
42	celecoxib
43	cerivastatin
44	chloroquine
45	chlorpropamide
46	cibenzoline
47	cisapride
48	citalopram
49	clobazam
50	clomipramine
51	clozapine
52	2-chloro-3-(pyridin-3-yl)-5,6,7,8-tetrahydroindolizine-
	1-carboxamide
53	cocaine
54	codeine
55	colchicine
56	(3S)-3-(6-methoxypyridin-3-yl)-3-{2-oxo-3-[3-(5,6,7,8-
	tetrahydro-1,8-naphthyridin-2-yl)propyl]imidazolidin-
	1-yl}propanoic acid
57	2-[(3S,4R)-3-benzyl-4-hydroxy-3,4-dihydro-2H-chromen-7-
	yl]-4-(trifluoromethyl)benzoic acid
58	diethyl({[(2R,4S,7S)-1]-ethyl-6-methyl-6,11-
	diazatetracyclo[7.6.1.0{circumflex over ( )}{2,7}.0{circumflex over ( )}{12,16}]hexadeca-
	1(15),9,12(16),13-tetraen-4-yl]sulfamoyl})amine
59	cyclobenzaprine
60	cyclophosphamide
61	dimethyl 7,7′-dimethoxy-4,4′-bi-1,3-benzodioxole-5,5′-
	dicarboxylate
62	delavirdine
63	deoxycholic-acid
64	deprenyl
65	deramciclane
66	dexamethasone
67	dexloxiglumide
68	dextromethorphan
69	dextropropoxyphene
70	(3S,8R,9S,10R,13S,14S)-3-hydroxy-10,13-dimethyl-
	1,2,3,4,7,8,9,11,12,14,15,16-
	dodecahydrocyclopenta[a]phenanthren-17-one
71	diazepam
72	dibenzo-a-h-anthracene
73	diclofenac
74	dihydrocodeine
75	dihydrodiol
76	diltiazem
77	disopyramide
78	domperidone
79	doxepin
80	ecabapide
81	eletriptan
82	ellipticine
83	eplerenone
84	estazolam
85	estradiol
86	estrone
87	etoperidone
88	etoricoxib
89	felodipine
90	fenproporex
91	fentanyl
92	finasteride
93	flucloxacillin
94	fluoxetine
95	fluvastatin
96	N-[2-(5-methoxy-1H-indol-3-yl)ethyl]-N-(propan-2-
	yl)propan-2-amine
97	gepirone
98	granisetron
99	2-[(R)-{[5-(cyclopropylmethoxy)pyridin-3-
	yl]methyl}sulfinyl]-5-fluoro-1H-benzimidazole
100	hydrocodone
101	hydromorphone
102	ibogaine
103	ifosfamide
104	iloperidone
105	imipramine
106	ketamine
107	ketobemidone
108	N-carbamimidoyl-4-cyano-1-benzothiophene-2-carboxamide
109	(1S,4S)-(6-dimethylamino-4,4-diphenyl-heptan-3-yl)acetate
110	laquinimod
111	levobupivacaine
112	lidocaine
113	lisofylline
114	ropinirole
115	loratadine
116	losartan
117	lovastatin
118	[(4E)-7-chloro-4-[(sulfooxy)imino]-3,4-dihydroquinolin-
	1(2H)-yl](2-methylphenyl)methanone
119	(2R)-1-(1,3-benzodioxol-5-yl)-N-ethylpropan-2-amine
120	(2R)-1-(1,3-benzodioxol-5-yl)-N-methylpropan-2-amine
121	melatonin
122	meloxicam
123	1-(4-methoxyphenyl)piperazine
124	methadone
125	methoxychlor
126	methoxychlor-mono-oh
127	metoclopramide
128	mianserin
129	midazolam
130	mirtazapine
131	4-{[(5S)-2,4-dioxo-1,3-thiazolidin-5-yl]methyl}-2-
	methoxy-N-[4-(trifluoromethyl)benzyl]benzamide
132	(2S)-1-(4-methylphenyl)-2-(pyrrolidin-1-yl)propan-1-one
133	1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine
134	mycophenolic-acid
135	n-nitrosodiamylamine
136	naringenin
137	nefiracetam
138	nn-dimethyl-m-toluamide
139	4-[methyl(nitroso)amino]-1-(pyridin-3-yl)butan-1-one
140	diethyl{4-[(4-bromo-2-
	cyanophenyl)carbamoyl]benzyl}phosphonate
141	nordiazepam
142	nortriptyline
143	5-(diethylamino)-2-methylpent-3-yn-2-yl(2S)-2-cyclohexyl-
	2-hydroxy-2-phenylacetate
144	ochratoxin-a
145	olanzapine
146	olopatadine
147	(1R,2R,10R,11S,14R,15R)-14-ethynyl-14-hydroxy-15-methyl-
	17-methylidenetetracyclo[8.7.0.0{circumflex over ( )}{2,7}.0{circumflex over ( )}{11,15}]hepta
	deca-6,12-dien-5-one
148	(3S)-3-[3-(methylsulfonyl)phenyl]-1-propylpiperidine
149	oxybutynin
150	oxycodone
151	perazine
152	perphenazine
153	phenacetin
154	phencyclidine
155	phenprocoumon
156	pimobendan
157	pradefovir
158	progesterone
159	propafenone
160	pyrazoloacridine
161	quinacrine
162	rebamipide
163	reboxetine
164	ropinirole
165	ropivacaine
166	roquinimex
167	safrole
168	safrole
169	salmeterol
170	senecionine
171	seratrodast
172	seratrodast
173	seratrodast
174	7-chloro-N-({5-[(dimethylamino)methyl]cyclopenta-1,4-
	dien-1-yl}methyl)quinolin-4-amine
175	tamarixetin
176	tamsulosin
177	tangeretin
178	tauromustine
179	temazepam
180	terbinafine
181	terbuthylazine
182	testosterone
183	theophylline
184	tramadol
185	trans-retinoic-acid
186	trazodone
187	triazolam
188	trofosfamide
189	tropisetron
190	valdecoxib
191
192
193	yohimbine
194	zaleplon
195	zolpidem
196	zotepine

Example 5

Comparison of Existing Metabolic Prediction Model with Prediction Model of the Present Invention

The present invention is an improved model compared to an existing metabolic prediction model.

The existing QSAR model (Sheridan R P, Korzekwa K R, Torres R A, Walker M J. J. Med. Chem. (2007) 50; 3173) and the present invention select two highly possible positions, and the MetaSite program (Cruciani G, Carosati E, Boeck B D, Ethirajulu K, Mackie C, Howe T, Vianello R. J. Med. Chem. (2005)48; 6970) selects three highly possible positions. Thus, these cannot be directly compared with each other, but as can be seen in Table 3 below, the present invention shows improved predictability.

TABLE 10
Comparison of existing metabolic prediction
model and inventive prediction model

	3A4	2D6	2C9	1A2

	QSAR modela	84%	70%	67%	—
	MetaSiteb	72%	86%	86%	75%
	Inventiona	74%	70%	77%	78%
	aselection of two highly possible positions
	bselection of three highly possible positions

Example 6

Prediction of Activation Energy Using Atomic Descriptors

6-1. Prediction of Activation Energy for Hydrogen Abstraction Using Atomic Descriptors

Hydrogen abstraction by a cytochrome P450 enzyme may be shown in the following reaction scheme 1:

wherein the cycle together with Fe—O indicates an oxyferryl intermediate.

In the present invention, the activation energy for cytochrome P450-mediated hydrogen abstraction from a substrate of the following formula 1 was predicted using the atomic descriptors [δ_het], [max(δ_heavy)]. [μ_C—H] and

[ ∑ i R . C .  α i ]  :

wherein the circle together with Fe—O indicates an oxyferryl intermediate;

E a Habs_  9  b   0 = 25.94 + 1.88 * [ δ het ] + 1.03 * [ max  ( δ heavy ) ] ; [ Equation   1  -  1 ]  E a Habs_  ( A ) = 28.50 - 2.22 * [ μ C - H ] + 1.12 * [ ∑ i R . C .  α i ] [ Equation   1  -  2 ]

wherein E_a^Habs—(B) indicates activation energy required for hydrogen attached to a carbon atom having a heteroatom (an atom other than carbon) in the alpha-position relative to the reaction center; E_a^Habs—(A) indicates activation energy required for hydrogen attached to a carbon atom having no heteroatom (an atom other than carbon) in the alpha-position relative to the reaction center; and [δ_het] indicates the net atomic charge of a heteroatom (an atom other than carbon) in the alpha-position relative to the reaction center; [max(δ_heavy)] indicates the highest atomic charge in X¹, X² and X³ which are neither hydrogen nor helium; [μ_C—H] indicates the bond dipole of the carbon-hydrogen bond; and

[ ∑ i R . C .  α i ]

indicates the sum of the atomic polarizabilities of the atoms H, C, X¹, X² and X³.

6-2. Prediction of Activation Energy for Tetrahedral Intermediate Formation in Aromatic Hydroxylation Using Atomic Descriptors

Tetrahedral intermediate formation reaction in cytochrome P450-mediated aromatic hydroxylation may be shown in the following reaction scheme 2:

wherein the circle together with O—FE indicates an oxyferryl intermediate.

In the present invention, the activation energy for tetrahedral intermediate formation in cytochrome P450-mediated aromatic hydroxylation of a substrate of the following formula was predicted using the atomic descriptors [δ_H] and [mean(α_alpha)]:

wherein the circle together with Fe—O indicates an oxyferryl intermediate;

E_a^aro—o,p=21.34−0.75*[δ_H]−1.24*[mean)α_alpha)]  [Equation 2-1]

E_a^aro—m=22.14−0.68*[δ_H]−0.83*[mean(α_alpha)]  [Equation 2-2]

E_a^aro—0,2,3=21.02−1.49*[δ_H]−0.92*[mean(α_alpha)]  [Equation 2-3]

wherein E_a^aro—o,p indicates the activation energy for tetrahedral intermediate formation in a benzene having one substituent in the ortho/para-position; E_a^aro—m the activation energy for tetrahedral intermediate formation in a benzene having one substituent in the meta-position; E_a^aro—0,2,3 indicates the activation energy for tetrahedral intermediate formation in a benzene having 0, 2 or 3 substituents; [δ_H] indicates the net atomic charge of the hydrogen; and [mean(α_alpha)] indicates the mean value of the polarizabilities of adjacent carbon atoms.

Example 7

Development of Model for Predicting Activation Energy for Hydrogen Abstraction

The activation energy for hydrogen abstraction is a good measure for predicting the regioselectivity of aliphatic hydroxylation and dehydroxylation in phase I metabolism.

wherein the circle together with Fe—O indicates an oxyferryl intermediate.

In order to model the above reaction, the activation energies of 431 cases of 119 molecules were calculated using the AM1 (Austin Model 1) molecular orbital method.

Herein, the term “cases” refers to the number of atoms. For example, if there are 3 molecules having 3, 4 and 7 atoms, respectively, there will be 14 cases of 3 molecules. The AM1 method is a semi-empirical method for quantum calculation of the electronic structures of molecules in computational chemistry and is a generalization of the modified neglect of differential diatomic overlap approximation (Dewar, M. J. S. et al., Journal of the American Chemical Society, 1985, 107, 3902).

The list of organic molecules calculated is shown in Table 11 below.

TABLE 11
Organic molecules used in training and verification
for hydrogen abstraction (119 organic molecules)
List of organic molecules

(3-amino-propyl)-dimethyl-amine	1-chloro-4-methyl-pentane
(3-bromo-propyl)-dimethyl-amine	1-chloro-butane
(3-chloro-propyl)-dimethyl-amine	1-chloro-heptane
(3-fluoro-propyl)-dimethyl-amine	1-chloro-hexane
(3-iodo-propyl)-dimethyl-amine	1-chloromethyl-3-methyl-benzene
1,2,3-trimethylbenzene	1-chloromethyl-4-methyl-benzene
1,2,4-trimethylbenzene	1-chloro-octane
1,2-difluoro-3-methyl-butane	1-chloro-pentane
1-bromo-2-methyl-benzene	1-chloro-propane
1-bromo-3-methyl-benzene	1-ethoxy-3-fluoro-benzene
1-bromo-4-methyl-benzene	1-ethyl-4-methylbenzene
1-bromo-4-methyl-pentane	1-fluoro-2,4-dimethyl-pentane
1-bromo-heptane	1-fluoro-2-methyl-benzene
1-bromo-hexane	1-fluoro-2-methyl-octane
1-bromo-octane	1-fluoro-3-methyl-benzene
1-bromo-pentane	1-fluoro-4-methyl-butane
1-bromo-propane	1-fluoro-4-methyl-benzene
1-chloro-2-methylbenzene	1-fluoro-4-methyl-heptane
1-chloro-3-methylbenzene	1-fluoro-4-methyl-pentane
1-chloro-4-methylbenzene	1-fluoro-butane
1,2,3-trimethylbenzene	Fluoro-benzene
1,2,4-trimethylbenzene	Iodo-benzene
1-ethyl-4-methylbenzene	mesitylene
1-methyl-2-propylbenzene	methoxybenzene
1-o-tolylpropan-1-one	m-xylene
2,4-difluoro-1-methylbenzene	n,4-dimethylbenzenamine
2-fluoro-phenylamine	o-xylene
2-methylanisol	phenol
3-fluoro-4-methylbenzeneamine	propylbenzene
3-fluoro-phenylamine	p-toluidine
4-ethoxy-aniline	p-xylene
4-ethoxy-phenol
4-fluoro-phenylamine
aniline
benzene
benzenethiol
chloro-benzene
cyanobenzene
ethoxybenzene
ethylbenzene

Such information was used to train and evaluate the empirical equations. These cases include methyl, primary, secondary and tertiary carbon atoms, etc., in various chemical environments.

The present inventors divided these cases into two types depending on whether electrically negative atoms (i.e. heteroatoms) exist around the breaking carbon-hydrogen bond.

Equations modeled with atomic descriptors through the correlation between effective atomic descriptors and quantum-mechanically calculated E_a for hydrogen abstraction are shown in Tables 12 and 13 below.

TABLE 12
Correlation between effective atomic descriptors and
quantum-mechanically calculated Ea for hydrogen abstraction
(the case of having no heteroatom in the alpha-position)

Atomic

Training set

	descriptor	Ra	RMSEb	Equation
	μC-H	0.88	0.63	EaHabs_(A) = 28.50 − 1.19*[μC-H]
	∑ i R . C .  α i	0.67	1.00	E a Habs  _  ( A ) = 28.50 - 0.90 *  [ ∑ i R . C .  α i ]
	Ra: correlation coefficient;
	RMSEb: root mean squared error.

TABLE 13
Correlation between effective atomic descriptors and
quantum-mechanically calculated Ea for hydrogen abstraction (the
case of having a heteroatom in the alpha-position)

Training set

Atomic descriptor	Ra	RMSEb	Equation

δhet	0.82	1.51	EaHabs_(B) = 25.94 +
			2.14*[δhet]
max(δheavy)	0.57	2.16	EaHabs_(B) = 25.94 +
			1.51*[max(δheavy)]
Ra: correlation coefficient;
RMSEb: root mean squared error.

The present inventors performed the training processes shown in Tables 12 and 13 above, thereby allowing linear equations to predict activation energy in various chemical environments using two normalized effective atomic descriptors suited to each case (equations 1-1 and 1-2 below).

Among these effective atomic descriptors, [δ_het], [max (δ_heavy)] and [μ_C—H] indicate the degree of weakness of the carbon-hydrogen bond, and

[ ∑ i R . C .  α i ]

indicates the stability of transition states. In the present invention, all transition states were verified through the analysis of frequencies.

FIG. 3 is a flowchart showing a method of predicting activation energy using the model of the present invention.

Specifically, the model for predicting the activation energy for CYP450-mediated hydrogen abstraction, developed in the present invention, comprises the following steps:

i) examining the metabolic position of a target molecule;

ii) determining the reaction type of the target molecule;

iii) determining the atomic type depending on whether there is a heteroatom in the alpha-position relative to the reaction center of hydrogen abstraction;

iv) if there is a heteroatom in the alpha-position, calculating the atomic descriptors [δ_het] and [max(δ_heavy)], and if there is no heteroatom in the alpha-position, calculating the atomic descriptors [μ_C—H] and

[ ∑ i R . C .  α i ] ;

v) normalizing the atomic descriptors; and

vi) predicting activation energy according to the following equations:

wherein the circle together with O—Fe indicates an oxyferryl intermediate;

E a Habs_  9  b   0 = 25.94 + 1.88 * [ δ het ] + 1.03 * [ max  ( δ heavy ) ] ;    R = 0.91 , R   M   S   E = 1.14 ,   n = 62 , P   value < 0.0001 ; [ Equation   1  -  1 ]  E a Habs_  ( A ) = 28.50 - 2.22 * [ μ C - H ] + 1.12 * [ ∑ i R . C .  α i ]    R = 0.95 , R   M   S   E = 0.43 ,   n = 224 , P   value < 0.0001 [ Equation   1  -  2 ]

wherein E_a^Habs—(B) indicates the activation energy required for abstraction of hydrogen attached to a carbon atom having a heteroatom (an atom other than carbon) in the alpha-position relative to the reaction center; E_a^Habs—(A) indicates activation energy required for abstraction of hydrogen attached to a carbon atom having no heteroatom (an atom other than carbon) in the alpha-position relative to the reaction center; [δ_het] indicates the net atomic charge of a heteroatom (an atom other than carbon) in the alpha-position relative to the reaction center; [max(δ_heavy)] indicates the highest atomic charge in X¹, X² and X³ which are neither hydrogen nor helium; [μ_C—H] indicates the bond dipole of the carbon-hydrogen bond; and

[ ∑ i R . C .  α i ]

indicates the sum of the atomic polarizabilities of the atoms H, C, X¹, X² and X³.

In equations 1-1 and 1-2 above, R: correlation coefficient; RMSE: root mean squared error; n: the number of atoms used in training; and P value: the significance of the correlation coefficient.

in step i), any C—H bond to the target molecule can be regarded as a position where metabolism can occur in the target molecule.

In step ii), if the carbon in any C—H bond to the target molecule is aliphatic carbon, it can be regarded as a position where H abstraction from the target molecule can occur.

In step iii), if there is a heteroatom in the alpha-position relative to the reaction center (C—H where actual metabolism occurs), equation 1-1 is used, and if there is no heteroatom in the alpha-position, equation 1-2 is used.

In step v), the term “normalization” refers to normalizing the mean of the values of atomic descriptors to zero (0) and the standard deviation to 1, from a statistical viewpoint. Namely, before prediction, normalization is carried out using the mean and standard deviation of the values of the atomic descriptors used in the training of the prediction model of the present invention.

As shown in FIG. 4, the activation energy predicted using the model of the present invention showed a high correlation with the quantum-mechanically calculated activation energy. 386 cases of 430 cases are within chemical accuracy (1 kcal per mol). Some inconsistent cases are attributable to interactions other than carbon-hydrogen-oxygen interactions during quantum mechanical calculation. Activation energies of various molecules in a gaseous state were calculated using Gaussian 03 [revision C.02. M. J. Frisch et al., Pittsburgh, Pa., USA, 2003].

Example 8

Verification of Activation Energy Predicted by Model for Predicting Activation Energy for Hydrogen Abstraction

Activation energies for hydrogen abstraction from the following four molecules, predicted using the prediction model of Example 7, were verified by comparison with experimental values:

TABLE 14
		Predicted	Metabolic rate	Experimental
		activation	induced from	metabolic
Molecule	#[a]	energy[b]	activation energy[c]	rate[d]

Hexane	1	26.89	4.1	4.5
	2	28.20	46.6	49
	3	28.16	49.3	46.5
Octane	1	29.69	8.2	2.5
	2	28.21	91.8	97.5
Ethylbenzene	1	30.32	0.1	0.2
	2	25.73	99.9	99.8
1-chloromethyl-4-	1	27.51	12.5	16.0
methyl-benzene	2	26.31	87.5	84.0
[a]# indicates the atomic number of each molecule in formula 2;
[b]activation energy predicted by the method of the present invention;
[c]metabolic rate induced by introducing the predicted activation energy [b] into the Arrhenius equation; and
[d]in vitro experimental metabolic rate.

The experimental metabolic rates of the molecules shown in Table 14 above are already known in the art. Specifically, the experimental metabolic rate of hexane can be found in the literature [Ken-ichirou MOROHASHI, Hiroyuki SADANO, Yoshiie OKADA, Tsuneo OMURA. Position Specificity in n-Hexane Hydroxylation by two forms of Cytochrome P450 in Rat liver Microsomes. J. Biochem. 1983, 93, 413-419]; the experimental metabolic rate of octane in the literature [Jeffrey P. Jones, Allan E. Rettie, William F. Trager. Intrinsic Isotope Effects Suggest That the Reaction Coordinate Symmetry for the Cytochrome P-450 Catalyzed Hydroxylation of Octane Is Isozyme Independent. J. Med. Chem. 1990, 33, 1242-1246]; the experimental metabolic rate of ethylbenzene can be found in the literature [Ronald E. White, John P. Miller, Leonard V. Favreau, Apares Bhattacharyya. Stereochemical Dynamics of Aliphatic Hydroxylation by Cytochrome P-450. J. AM. Chem. Soc. 1986, 108, 6024-6031]; and the experimental metabolic rate of 1-chloromethyl-4-methyl-benzene can be found in the literature [LeeAnn Higgins, Kenneth R. Korzekwa, Streedhara Rao, Magong Shou, and Jeffrey P. Jones. An Assessment of the Reaction Energetics for Cytochrome P450-Mediated Reactions. Arch. Biochem. Biophys. 2001, 385, 220-230].

As can be seen in Table 14 above, when the metabolic rates^[c] (induced by substituting into the Arrhenius equation the activation energies for hydrogen abstraction from the four molecules, hexane, octane, ethylbenzene and 1-chloromethyl-4-methyl-benzene, predicted according to the present invention) were compared with the experimental metabolic rates^[d], these metabolic rates showed similar tendencies. This suggests that the experimental metabolic rates can be predicted through the activation energies predicted according to the present invention.

Example 9

Development of Model for Predicting the Activation Energy for Tetrahedral Intermediate Formation in Aromatic Hydroxylation

The present inventors modeled tetrahedral intermediate formation serving as a good measure of the regioselectivity of aromatic hydroxylation in phase I metabolism.

wherein the circle together with Fe—O indicates an oxyferryl intermediate.

To model the above reaction, the activation energies of 85 cases of 31 benzene molecules in various chemical environments were calculated using the AM1 (Austin Model 1) molecular orbital method.

Herein, the term “cases” refers to the number of atoms. For example, if there are 3 molecules having 3, 4 and 7 atoms, respectively, there will be 14 cases of 3 molecules. The AM1 method is a semi-empirical method for quantum calculation of the electronic structures of molecules in computational chemistry and is a generalization of the modified neglect of differential diatomic overlap approximation (Dewar, M. J. S. et al., Journal of the American Chemical Society, 1985, 107, 3902).

The list of organic molecules calculated is shown in Table 15 below.

TABLE 15
Organic molecules used in training and verification for
tetrahedral intermediate formation (31 organic molecules)
List of organic molecules

	1,2,3-trimethylbenzene	Fluoro-benzene
	1,2,4-trimethylbenzene	Iodo-benzene
	1-ethyl-4-methylbenzene	mesitylene
	1-methyl-2-propylbenzene	methoxybenzene
	1-o-tolylpropan-1-one	m-xylene
	2,4-difluoro-1-methylbenzene	n,4-dimethylbenzenamine
	2-fluoro-phenylamine	o-xylene
	2-methylanisol	phenol
	3-fluoro-4-methylbenzenamine	propylbenzene
	3-fluoro-phenylamine	p-toluidine
	4-ethoxyaniline	p-xylene
	4-ethoxy-phenol
	4-fluoro-phenylamine
	aniline
	benzene
	benzenethiol
	chloro-benzene
	cyanobenzene
	ethoxybenzene
	ethylbenzene

Such information was used to train and evaluate the empirical equations. These cases were divided into three types: i) having one substituent in the ortho/para position; ii) having one substituent in the meta-position; and iii) having 0, 2 or 3 substituents.

Equations modeled with atomic descriptors through the correlation between effective atomic descriptors and quantum-mechanically calculated E_a for aromatic hydroxylation are shown in Tables 16, 17 and 18 below.

TABLE 16
Correlation between effective atomic descriptors and quantum-
mechanically calculated Ea for aromatic hydroxylation (the
case of having a substituent in the ortho-position)

Atomic

Training set

	descriptor	Ra	RMSEb	Equation
	δH	0.08	1.31	Eaaro—o.p = 14.67 +
				63.33*[δH]
	αalpha	0.57	1.07	Eaaro—o.p = 61.60 −
				26.53*[ αalpha]
	Ra: correlation coefficient;
	RMSEb: root mean squared error.

TABLE 17
Correlation between effective atomic descriptors and quantum-
mechanically calculated Ea for aromatic hydroxylation (the
case of having a substituent in the meta-position)

Atomic

Training set

	descriptor	Ra	RMSEb	Equation
	δH	0.03	0.56	Eaaro—m = −12.61 +
				333.87*[δH]
	αalpha	0.50	0.49	Eaaro—m = 132.75 −
				72.54*[ αalpha]
	Ra: correlation coefficient;
	RMSEb: root mean squared error.

TABLE 18
Correlation between effective atomic descriptors and
quantum-mechanically calculated Ea for aromatic hydroxylation
(the case of having 0, 2 or 3 substituents)

Atomic

Training set

	descriptor	Ra	RMSEb	Equation
	δH	0.69	0.95	Eaaro—0, 2, 3 = 70.00 −
				465.88*[δH]
	αalpha	0.05	1.31	Eaaro—0, 2, 3 = 17.65 +
				2.21*[ αalpha]
	Ra: correlation coefficient;
	RMSEb: root mean squared error.

The present inventors performed the training processes shown in Tables 16 to 18 above, thereby allowing linear equations to predict activation energy in various chemical environments using two normalized effective atomic descriptors suited to each case (equations 2-1, 2-2 and 2-3 below).

Among effective atomic descriptors which are used in the equations for predicting the activation energy for tetrahedral intermediate formation, [δ_H] determines the proximity between oxygenating species and substrate, and [(mean(α_alpha)] is related to the stability of transition states. In the present invention, para-nitrosophenoxy radical (PNR) was used as oxygenating species, and all transition states were verified through the analysis of frequencies.

FIG. 3 shows a flowchart showing a method of predicting activation energy using the model used in the present invention.

Specifically, the model for predicting the activation energy for tetrahedral intermediate formation in CYP450-mediated aromatic hydroxylation, developed in the present invention, comprises the following steps:

i) examining the metabolic position of a target molecule;

ii) determining the reaction type of the target molecule;

iii) if the reaction type in step ii) is determined to be aromatic hydroxylation, calculating the atomic descriptors [δ_H] and [mean(α_alpha)];

iv) normalizing the atomic descriptors; and

v) predicting activation energy according to the following equations:

wherein the circle together with Fe—O indicates an oxyferryl intermediate;

E_a^aro—o,p=21.34−0.75*[δ_H]−1.24*[mean)α_alpha)]

R=0.71, RMSE=0.95, n=16, P value=0.009;  [Equation 2-1]

E_a^aro—m=22.14−0.68*[δ_H]−0.83*[mean(α_alpha)]

R=0.88, RMSE=0.30, n=8, P value=0.026;  [Equation 2-2]

E_a^aro—0,2,3=21.02−1.49*[δ_H]−0.92*[mean(α_alpha)]

R=0.87, RMSE=0.65, n=33, P<0.0001  [Equation 2-3]

wherein E_a^aro—o,p indicates the activation energy for tetrahedral intermediate formation in a benzene having one substituent in the ortho/para position; E_a^aro—m indicates the activation energy for tetrahedral intermediate formation in a benzene having one substituent in the meta-position; E_a^aro—0,2,3 indicates the activation energy for tetrahedral intermediate formation in a benzene having 0, 2 or 3 substituents; [δ_H] indicates the net atomic charge of the hydrogen; and [mean(α_alpha)] indicates the mean of the polarizabilities of adjacent carbon atoms.

In equations 2-1, 2-2 and 2-3 above, R: correlation coefficient; RMSE: root mean squared error; n: the number of atoms used in training; and P value: the significance of the correlation coefficient.

As shown in FIG. 5, the activation energy predicted using the model of the present invention showed a high correlation with the quantum-mechanically calculated activation energy. 70 cases of 85 cases are within chemical accuracy (1 kcal per mol). Some inconsistent cases occurred because the model did not consider the ortho-, meta- and para-effects when modeling the benzene molecule having 0, 2 or 3 substituents. Activation energies of various molecules in a gaseous state were calculated using Gaussian 03 [revision C.02, M. J. Frisch et al., Pittsburgh, Pa., USA, 2003].

Example 10

Verification of Activation Energy Predicted by Model for Predicting Activation Energy for Tetrahedral Intermediate Formation in Aromatic Hydroxylation

The activation energies for tetrahedral intermediate formation for the following two molecules, predicted by the prediction model of Example 9, were verified by comparison with experimental values.

TABLE 19
		Predicted	Metabolic rate	Experimental
		activation	induced from	metabolic
Molecule	#[a]	energy[b]	activation energy[c]	rate[d]

Methoxybenzene	2	21.79	30.8	15-24
	3	22.41	11.1	1-3
	4	21.40	58.1	62-75
Chlorobenzene	2	22.81	8.0	17-19
	3	22.58	11.6	5-9
	4	21.39	80.4	71-79
[a]# indicates the atomic number of each molecule in formula 4;
[b]activation energy predicted by the method of the present invention;
[c]metabolic rate induced by introducing the predicted activation energy [b] into the Arrhenius equation; and
[d]in vitro experimental metabolic rate.

The experimental metabolic rates of the molecules shown in Table 19 above are already known in the art. Specifically, the experimental metabolic rate of methoxybenzene can be found in the literature [Robert P. Hanzlik, Kerstin Hogberg, Charles M. Judson. Microsomal hydroxylation of specifically deuterated monosubstituted benzenes. Evidence for direct aromatic hydroxylation. Biochemistry. 1984, 23, 3048-3055]; and the chlorobenzene can be found in the literature [H. G. Selander, D. M. Jerina, J. W. Daly. Metabolism of Chlorobenzene with Hepatic Microsomes and Solubilized Cytochrome P-450 Systems. Arch. Biochem. Biophys. 1975, 168, 309-321].

As can be seen in Table 19 above, when the metabolic rates^[c] (induced by substituting into the Arrhenius equation the activation energies for hydrogen abstraction from the two molecules, methoxybenzene and chorobenzene, predicted according to the present invention) were compared with the experimental metabolic rates^[d], these metabolic rates showed similar tendencies. This suggests that the experimental metabolic rates can be predicted through the activation energies predicted according to the present invention.

As described above, the method of the present invention can rapidly predict activation energy for phase I metabolites at a practical level without having to perform a docking experiment between any additional CYP450 and the substrate, or a quantum mechanical calculation, thereby making it easier to develop new drugs using a computer. Also, the present invention may propose a strategy for increasing the bioavailability of drugs through the avoidance of metabolites based on the possibility of drug metabolism. Furthermore, the method of the present invention proposes new empirical approaches which can also be easily applied to activation energies for various chemical reactions, and makes it possible to explain physical and chemical factors that determine activation energy. In addition, through the prediction of activation energy according to the present invention, it is possible to predict i) metabolic products, ii) the relative rate of metabolism, iii) metabolic regioselectivity, iv) metabolic inhibition, v) drug-drug interactions, and vi) the toxicity of a metabolite.

The present invention may be embodied in other specific forms without departing from its spirit or essential characteristics. The described embodiments are to be considered in all respects only as illustrative, and not restrictive. The scope of the invention is, therefore, indicated by the appended claims, rather than by the foregoing description. All changes which come within the meaning and range of equivalency of the claims are to be embraced within the scope of the present invention.