Method for estimating human protoporphyrinogen oxidase (human PPO) inhibitory activity
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- PAN ADVANCED BUSINESS RESEARCH LLC
- Filing Date
- 2025-10-06
- Publication Date
- 2026-06-04
Smart Images
Figure 0007869935000001_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a technique for accurately estimating the presence or absence or degree of inhibitory activity of human protoporphyrinogen oxidase (human PPO) using a computer.
Background Art
[0002] Drug development has a serious problem in that, despite requiring enormous research and development costs and a long time, the probability of final approval is very low.
[0003] In recent years, there has been an increasing social demand worldwide to reduce and replace animal experiments from an ethical perspective in the development process of pharmaceuticals and chemical substances.
[0004] Human protoporphyrinogen oxidase (human PPO) is an essential enzyme in the heme biosynthesis pathway. Therefore, human PPO inhibitors are used in applications in photodynamic therapy (PDT) (Non-Patent Document 1), targets for anti-tumor drug development (Non-Patent Document 2), and comparison materials for the toxicity of chemical substances such as herbicides containing the PPO inhibitor diphenyl ether herbicide and non-PPO inhibitors Chlortoluron and Atrazine to human PPO (Non-Patent Document 3), etc., and the importance of its research is increasing.
[0005] In Non-Patent Document 3, Acifluorfen, a human PPO inhibitor (5-[2-chloro-4-(trifluoromethyl)phenoxy]-2-nitrobenzoic acid), Oxyfluorfen (2-chloro-1-(3-ethoxy-4-nitrophenoxy)-4-(trifluoromethyl)benzene), Lactofen ((1-ethoxy-1-oxopropan-2-yl) 5-[2-chloro-4-(trifluoromethyl)phenoxy]-2-nitrobenzoate), Fomesafen (5-[2-chloro-4-(trifluoromethyl)phenoxy]-N-methylsulfonyl-2-nitrobenzamide), Chlornitrofen(1,3,5-trichloro-2-(4-nitrophenoxy)benzene), Oxadiazon(5-tert-butyl-3-(2,4-dichloro-5-propan-2-yloxyphenyl)-1,3,4-oxadiazol-2-one), Oxadiargyl (5-tert-butyl-3-(2,4-dichloro-5-prop-2-ynoxyphenyl)-1,3,4-oxadiazol-2-one), Butafenacil(2-methyl-1-oxo-1-prop-2-enoxypropan-2-yl) 2-chloro-5-[3-methyl-2,6-dioxo-4-(trifluoromethyl)pyrimidin-1-yl]benzoate), Saflufenacil The human PPO inhibitory activity of 20 compounds, shown in the following structural formulas [Chemical Formula 1] to [Chemical Formula 20], including (2-chloro-4-fluoro-5-[3-methyl-2,6-dioxo-4-(trifluoromethyl)pyrimidin-1-yl]-N-[methyl(propan-2-yl)sulfamoyl]benzamide) and the non-PPO inhibitors Chlortoluron (3-(3-chloro-4-methylphenyl)-1,1-dimethylurea) and Atrazine (6-chloro-4-N-ethyl-2-N-propan-2-yl-1,3,5-triazine-2,4-diamine), has been measured as IC50 (molar concentration of a compound that inhibits PPO enzyme activity by 50%).
[0006] [ka]
[0007] [ka]
[0008]
change
[0009]
change
[0010]
change
[0011]
change
[0012]
change
[0013]
change
[0014]
change
[0015]
change
[0016]
change
[0017]
change
[0018]
change
[0019]
change
[0020]
change
[0021]
change
[0022]
change
[0023]
change
[0024]
change
[0025]
change
[0026] On the other hand, Bacillus subtilis is known as a PPO that exhibits resistance to Acifluorfen, and it has been reported that it is not inhibited even at a concentration of 100 μM, exhibiting very high resistance (Non-Patent Literature 4). Although the analog Acifluorfen-methyl (Chemical Formula 21) showed weak inhibition at the same concentration, it was still suggested to exhibit strong resistance (Non-Patent Literature 4). Furthermore, regarding genetically modified tobacco plants expressing Bacillus subtilis genes, non-genetically modified tobacco plants suffered significant damage at around 10 μM with Oxyfluorfen, but genetically modified tobacco plants did not reach that level even at 100 μM, and a concentration even higher than 100 μM was required to inflict 50% damage, so it has been reported that Bacillus subtilis PPO also exhibits resistance to Oxyfluorfen (Non-Patent Literature 5). Thus, although the sensitivity to Bacillus subtilis differs depending on the compound, it is known that there are clear interspecies differences in PPO inhibitory activity.
[0027] [ka]
[0028] PPO crystal structure data derived from human (Homo sapiens) containing acifluorfen can be obtained from Non-Patent Document 6, and PPO crystal structure data derived from Bacillus subtilis can be obtained from Non-Patent Document 7, and from the PDB, respectively, as description numbers 3NKS and 3I6D.
[0029] Furthermore, crystal structure data for PPO derived from the tobacco plant (Nicotiana tabacum) has been reported, including the three-dimensional structure of compound 10(4-Bromo-3-(5'-carboxy-4'-chloro-2'-fluorophenyl)-1-methyl-5-trifluoromethyl-pyrazol), a phenylpyrazole-based PPO inhibitor, as shown in the following structural formula (Chemical Formula 22) (Non-Patent Literature 8). The crystal structure data can be obtained from the Protein Data Bank (PDB) under description number 1SEZ.
[0030] [ka]
[0031] As control compounds, single-molecule crystal structure data for the non-PPO inhibitors Chlortoluron and Atrazine can be obtained from Non-Patent Document 9 and Non-Patent Document 10, respectively.
[0032] High-precision in silico estimation techniques are expected to be a promising alternative to animal experiments. However, conventional techniques are limited to derivatives or analogues of target compounds, resulting in limitations in estimation accuracy, which has hindered their widespread adoption. The molecular descriptors used in these conventional techniques generally refer to physicochemical properties such as hydrophobic parameters, electronic parameters, and steric parameters.
[0033] Regarding methods for predicting PPO inhibitors in particular, Patent Document 1 is a technical document disclosed in a patent application filed by the present applicant. This is an in silico PPO inhibitor prediction technology that applies frontier orbital theory (Non-Patent Document 12), and predicts PPO inhibitory activity based on the correlation between the LUMO distribution and PPO inhibitory activity for each compound, and predicts the degree of PPO inhibitory activity based on the correlation between the difference in variation of PPO inhibitory activity between different plant species and the degree of PPO inhibitory activity.
[0034] The PPO inhibition model in plants described in Patent Document 1 focuses on bimolecule-to-bolecule interactions between PPO inhibitors and oxidized FADs. This demonstrates that, in plant species such as corn and barley, the difference in PPO inhibitory activity due to differences in amino acid residues is not significant. By adding a dummy variable (I[barley]) that gives 1 to the PPO inhibitory activity value (pI50) of barley, the degree of PPO inhibitory activity between different plant species, corn and barley, can be predicted with high accuracy using the same regression equation.
[0035] However, since the method described in Patent Document 1 targeted bimolecular complexes, it was possible to predict with high accuracy the degree of PPO inhibitory activity due to species differences between plants such as corn and barley. However, it was difficult to apply this method to organisms with different amino acid residue compositions of enzymes and to estimate the selectivity between species due to differences in nearby amino acid residues, including those on the isoalloxazine ring of oxidized FAD. In contrast, the present invention estimates the selectivity between species for the first time by using the LUMO distribution shape to analyze the intermolecular interactions between oxidized FAD, one or more nearby amino acid residues, and three or more human PPO active site domains of a human PPO inhibitor.
[0036] Furthermore, Non-Patent Literature 11 showed that human PPO cannot function as a correctly folded, active enzyme without oxidized FAD, and revealed that in the absence of FAD, the protein precipitates as an insoluble aggregate and does not exhibit enzymatic activity. On the other hand, it was reported that when FAD binds, the three-dimensional structure of the enzyme is stabilized, and the correct structure necessary for catalytic reactions is formed, indicating that oxidized FAD is a core element constituting the active site of human PPO, interacting with nearby amino acid residues, establishing electron transfer pathways, and being essential for enabling oxidation reactions. Therefore, Non-Patent Literature 11 reported that oxidized FAD plays a central role in all aspects of human PPO structure formation, stabilization, and functional expression, and that when its binding is lost, both the enzyme's activity and stability are significantly reduced.
[0037] [ka]
[0038] In this invention, a new human PPO active site domain model was constructed, consisting of a candidate human PPO inhibitor, oxidized FAD, and one or more amino acid residues. Using this model, the correlation between electronic interactions within the active site domain and the human PPO inhibitory activity of known compounds was analyzed using a machine learning model to clarify how oxidized FAD controls redox reactions as an electron acceptor.
[0039] Specifically, in the PPO inhibition response of the human PPO active site domain model, we hypothesized that the presence of one or more amino acid residues would alter the LUMO distribution shape of the human PPO active site domain model due to the electron acceptor function of oxidized FAD, and performed machine learning analysis. From the results of this analysis, we observed the differences in the LUMO distribution shape of the human PPO active site domain model consisting of a human PPO inhibitor, oxidized FAD, and one or more amino acid residues. As a result, we found that there are two patterns: a first pattern in which the LUMO distribution shape remains unchanged and is maintained, and a second pattern in which the LUMO distribution shape changes or completely disappears.
[0040] We hypothesized that the LUMO distribution shape pattern of this human PPO active site domain model could serve as a novel molecular descriptor that directly reflects the presence or absence of inhibitory activity. By associating the second pattern described above with "present" inhibitory activity and the first pattern with "absent" inhibitory activity, we confirmed through machine learning that a correlation exists.
[0041] Furthermore, the present invention can be applied not only to the qualitative estimation of the "presence or absence" of inhibitory activity, but also to the quantitative estimation of its "degree." For example, a regression model was constructed using machine learning, with an index that quantifies the pattern of the LUMO shape and geometric features calculated from the three-dimensional structural data of the active site domain as explanatory variables, and the intensity of inhibitory activity of known compounds (e.g., pI50 value) as the dependent variable. As a result, it was confirmed that the degree of inhibitory activity of unknown candidate compounds can be estimated with high accuracy as a continuous value.
[0042] In other words, the key technical aspect of the present invention lies in using the LUMO distribution shape pattern of the entire PPO active site domain, including the candidate human PPO inhibitor, as a molecular descriptor for determining the presence or degree of human PPO inhibitory activity, rather than using existing physicochemical characteristic values.
[0043] Therefore, the present invention is based on a novel approach of using LUMO distribution shape patterns as a novel molecular descriptor in an in silico method. That is, unlike conventional approaches that use existing physicochemical properties as descriptors, the entire complex of three or more molecules, consisting of a candidate human PPO inhibitor, oxidized FAD, and a human PPO active site domain consisting of one or more amino acid residues, is the target of analysis. By using the LUMO distribution shape pattern, which reflects the changes in the electronic state of the entire domain, as an index for determining the presence or degree of human PPO inhibitory activity, highly accurate estimation is made possible while minimizing experimental resources. The inventors have diligently pursued research efforts based on this unique idea and have completed the present invention. [Prior art documents] [Patent Documents]
[0044] [Patent Document 1] Patent No. 7738230
[0045] [Non-Patent Document 1] VHFingar et al., Cancer Res., (1997), 57, 4551-4556 [Non-Patent Document 2] S.Bazzocco et al.,Clin.Cancer Res.,(2015),21,3695-3704 [Non-Patent Document 3] M. Jakubek et al.,Processes,(2021),9,383 [Non-Patent Document 4] AVCorrigall et al., Arch Biochem Biophys., (1998), 354(2), 261-7 [Non-Patent Document 5] KWChoi et al.,Bioscience,Biotechnology,and Biochemistry,(1998),62(3),558-560 [Non-Patent Document 6] X.Qin et al.,FASEB J.,(2011),25,653―664 [Non-Patent Document 7] X.Qin et al.J Struct Biol.,(2010),170,76-82 [Non-Patent Document 8] M.Koch et al.,The EMBO Journal,(2004),23,1720-1728 [Non-Patent Document 9] G. Pepe et al., Zeitschrift fur Kristallographie-New Crystal Structures, (2000) 215, 63-64 [Non-Patent Document 10] T.Le et al.,CrystEngComm,(2016),18,962-970 [Non-Patent Document 11] Z. Novakova et al. (2021),16(2),PLoS ONE,e0246249 [Non-Patent Document 12] "Chemical Reactions and Electron Orbitals" by Kenichi Fukui, Maruzen, 1976. [Overview of the Initiative] [Problems that the invention aims to solve]
[0046] Drug development faces a serious challenge: despite requiring enormous research and development costs and many years, the probability of final approval is extremely low. In particular, the "valley of death" problem, in which many candidate compounds are dropped during the transition from basic research to clinical trials, is partly due to the low accuracy of screening in the early stages of development. On the other hand, in silico methods are expected to be a promising alternative to animal experiments for physiologically active substances, including pesticides, but the low accuracy of estimation remains a challenge. Under these circumstances, the present invention aims to provide a new method for estimating inhibitory activity with high accuracy and stability by constructing a human PPO active site domain consisting of a PPO inhibitor, oxidized FAD, and one or more amino acid residues, taking into account the amino acid residue composition in the human PPO, and using the change pattern of the LUMO distribution shape of the human P active site domain as an indicator, thereby eliminating uncertainties such as conventional numerical thresholds and expert subjectivity. [Means for solving the problem]
[0047] The inventors focused on frontier orbital theory (Non-Patent Literature 12) and applied the principle that redox reactions of human PPO inhibitors, oxidized FAD, and the human PPO active site domain consisting of one or more amino acid residues are governed by the interaction between the highest occupied orbital (HOMO) of an electron donor molecule and the lowest unoccupied orbital (LUMO) of an electron acceptor molecule. In other words, they hypothesized that in the absence of a human PPO inhibitor, the redox reaction proceeds through the interaction of the substrate's HOMO with the LUMO of oxidized FAD in the active site domain. They considered the key to inhibiting this reaction to be the function of oxidized FAD as an electron acceptor, i.e., the state of its LUMO. When candidate human PPO inhibitor compounds were introduced into the human PPO active site domain, the inventors observed how the spatial distribution of LUMOs in the vicinity, including on the isoalloxazine ring of oxidized FAD, changed. As a result, they found that there are two patterns: a first pattern in which the LUMO distribution is substantially maintained, and a second pattern in which the LUMO distribution changes or completely disappears.
[0048] Furthermore, considering that the shape pattern of the LUMO distribution could serve as a novel molecular descriptor that directly reflects the presence or degree of inhibitory activity, we designated the second pattern as "present" and the first pattern as "absent," and analyzed the correlation using machine learning, confirming a high correlation.
[0049] In other words, the core of the technical concept of this invention lies in using the LUMO distribution shape pattern itself, which is expressed as the entire human PPO active site domain including the candidate human PPO inhibitor, rather than individual physicochemical properties, as a molecular descriptor for determining the presence or degree of activity. It should be noted that finding the optimal combination of amino acid residues constituting the active site domain in this invention is by no means self-evident; for example, the total number of combinations of selecting one or more amino acid residues from the 16 candidate residues is 2 to the power of 16 This amounts to 65,536 - 1 = 65,535 possibilities. Therefore, the solution presented by this invention was not discovered by chance or through trial and error using a large number of people. Rather, it was the first discovery of the LUMO distribution shape pattern of the electronic state in the human PPO active site domain, leading to the completion of the invention. In other words, this invention relates to the following technology. (1) A method for estimating the presence or degree of inhibitory activity of a candidate compound on human protoporphyrinogen oxidase (human PPO), characterized by comprising the following steps (i) to (iv) performed by a computer: (i) A step of constructing three-dimensional structural data of the active site domain, comprising the candidate compound, oxidized FAD which is a coenzyme of human PPO, and one or more amino acid residues that constitute the active site of human PPO; (ii) A step of calculating the spatial distribution of the lowest unoccupied orbitals (LUMOs) of the active site domain; (iii) The spatial distribution of the calculated LUMO is The process involves classifying a compound into one of several patterns, including a first pattern in which the three-dimensional shape of the LUMO distribution in the vicinity of the isoaloxazine ring of the oxidized FAD is maintained at the same spatial distribution as the standard, based on the state in which the candidate compound has been removed from the active site domain, and comparing it to the standard, and including a first pattern in which the shape of the LUMO distribution is changed or completely disappears. This classification is performed automatically using a machine learning model that has been trained with quantitative numerical information obtained by calculating the similarity of images or extracting three-dimensional shape features from the three-dimensional data of the spatial distribution of the LUMO as an index that quantifies the pattern of the LUMO shape, with the index as the explanatory variable and the classification of the pattern as the dependent variable. process; (iv) Based on the classification result of the pattern obtained in step (iii) or the index, perform at least one of the following processes (a) and (b) to estimate the presence or degree of inhibitory activity of the candidate compound: (a) A process to qualitatively estimate the presence or absence of inhibitory activity of the candidate compound by taking the classification result as input, determining that human PPO inhibitory activity is "present" if the classification result is the second pattern, and determining that human PPO inhibitory activity is "absent" if the classification result is the first pattern. (b) A process to quantitatively estimate the degree of inhibitory activity of the candidate compound using a regression model that has been trained with the above-mentioned index as an explanatory variable and the intensity of inhibitory activity of known compounds as the dependent variable. (2) The one or more amino acid residues are Arg97, Arg168, Gly169, Val170, Phe171, Ala172, Phe331, Gly332, His333, Leu334, Leu344, Gly345, Ile346, Val347, Met368 and Ile419 The method according to (1), characterized in that it is selected from the group consisting of 16 residues. (3) The method further includes a comparative step in which the spatial distribution of LUMO of candidate compounds is calculated based on the three-dimensional structural data of PPO of Bacillus subtilis, and the interspecies selectivity of the candidate compounds is verified using the difference from the spatial distribution of LUMO in human PPO as an indicator. (1) Methods used. (4) The calculation of the spatial distribution of the LUMO of candidate compounds based on the PPO three-dimensional structural data of Bacillus subtilis is performed using an active site domain in which amino acid residues are selected from the group consisting of seven residues: Leu68, Lys71, Gly175, Ile176, Tyr177, Met413, and Val448. (3) Methods used. (5) A compound database consisting of numerous candidate compounds is used to screen for compounds that possess human PPO inhibitory activity. A computer-based screening process for this purpose. It is a method, (1) Using the estimation method described above Estimate A method characterized by including the step of selecting candidate compounds that are presumed to have human PPO inhibitory activity as subjects for experimental verification. (6)(1) A method characterized by using the estimation results obtained by the estimation method described above as an indicator, comparing them with the inhibitory activity of existing human PPO inhibitors or predetermined criteria, and selecting a human PPO inhibitor from a group of candidate compounds. (7)The existing human PPO inhibitors used for comparison include at least one of the following: Acifluorfen, Oxyfluorfen, Lactofen, Fomesafen, Chlornitrofen, Oxadiazon, Oxadiargyl, Butafenacil, and Saflufenacil. (6) Methods used. (8) By comparing the inhibitory activity (IC50, etc.) of the candidate compound measured by experimental methods with the estimated results, Human PPO inhibitors Make a choice (6) Methods used. (9)(1) An image processing system that performs the method described in [the previous method], characterized by including a LUMO image reading and alignment unit, a preprocessing unit for the FAD neighbor region, an inference unit equipped with the machine learning model, and a postprocessing and display unit that outputs the activity estimation result. (10)(1) The method described above can be used by a computer. to A program to be executed. (11)(10) A non-temporary recording medium on which the program described above is stored. [Effects of the Invention]
[0050] The present invention provides a technology for objectively and accurately estimating the presence or degree of human PPO inhibitors, and achieves the following remarkable effects. (1) By constructing a human PPO active site domain that includes one or more amino acid residues constituting the human PPO active site in the model, we considered the differences in combinations of one or more amino acid residues in the vicinity, including on the isoalloxazine ring of oxidized FAD. As a result, we clarified that differences caused by human PPO inhibitors can be classified by the distribution shape pattern of the LUMO according to frontier orbital theory, and it is possible to estimate with high accuracy the presence or absence or degree of activity of human PPO inhibitors, which was difficult to distinguish, as well as differences in resistance between species. This realizes highly accurate in silico estimation for screening of physiologically active substances, including pharmaceuticals and pesticides. This will enable the efficient narrowing down of promising compounds, overcome the "valley of death" problem in drug development, and greatly contribute to shortening the research and development period and reducing costs as a powerful alternative to animal testing conducted for safety confirmation in humans in the research and development of physiologically active substances. (2) Because shape pattern recognition that does not depend on numerical thresholds is used, the judgment process can be automated by machine learning, and high reproducibility can be obtained. (3) Because the basis for the determination is based on the visual shape changes of the LUMO, the reason for the determination is easy to understand and explain intuitively. As a result, the determination does not depend on the subjective opinion of experts, and anyone can perform objective classification work, contributing to an improvement in the reliability of the entire model. (4) Further detailed analysis of the LUMO disappearance pattern in this method (e.g., "complete disappearance" or "change") suggests that it can be extended not only to qualitative estimation to determine the presence or absence of inhibitory activity, but also to a quantitative model that continuously estimates its strength. [Brief explanation of the drawing] [Figure 1] This figure shows the LUMO distribution of the oxidized FAD in human PPO and the 16-amino acid residue (without the inhibitor Acifluorfen) of the human PPO active site domain. [Figure 2] This figure shows the LUMO distribution of the inhibitor Acifluorfen, oxidized FAD, and the 16-amino acid residue human PPO active site domain in human PPO. [Figure 3]This figure shows the LUMO distribution of the oxidized FAD in human PPO and the 9-amino acid residue (without the inhibitor Acifluorfen) of the human PPO active site domain. [Figure 4] This figure shows the LUMO distribution of the inhibitor Acifluorfen, oxidized FAD, and the 9-amino acid residue human PPO active site domain in human PPO. [Figure 5] This figure shows the LUMO distribution of the oxidized FAD in human PPO and the 10-amino acid residue (without the inhibitor Acifluorfen) of the human PPO active site domain. [Figure 6] This figure shows the LUMO distribution of the inhibitor Acifluorfen, oxidized FAD, and the 10-amino acid residue human PPO active site domain in human PPO. [Figure 7] This figure shows the LUMO distribution of the inhibitor oxyfluorfen, oxidized FAD, and the 10-amino acid residue human PPO active site domain in human PPO. [Figure 8] This figure shows the LUMO distribution of the oxidized FAD and the 7-amino acid residue (without the inhibitor Acifluorfen) of the active site domain of Bacillus subtilis PPO. [Figure 9] This figure shows the LUMO distribution of the inhibitor Acifluorfen, oxidized FAD, and the 7-amino acid residue active site domain of Bacillus subtilis PPO. [Figure 10] This figure shows the LUMO distribution of the inhibitor Acifluorfen, oxidized FAD, and the 4-amino acid residue active site domain of Bacillus subtilis PPO. [Figure 11] This figure shows the LUMO distribution of the inhibitor Oxyfluorfen, oxidized FAD, and the 4-amino acid residue active site domain of Bacillus subtilis PPO. [Figure 12] This figure shows the LUMO distribution of the inhibitor Acifluorfen-methyl, oxidized FAD, and the 4-amino acid residue active site domain of Bacillus subtilis PPO. [Figure 13]This figure shows the change in the pattern value of the LUMO distribution shape for human PPO inhibitory activity pI50. [Best Mode for Carrying Out the Invention]
[0051] In this specification, "human PPO active site domain" refers to the functional and structural unit that constitutes the active site of human PPO. Specifically, it refers to candidate human PPO inhibitor compounds, oxidized FAD which is a coenzyme of human PPO, and three-dimensional structures containing one or more amino acid residues.
[0052] (1) Method for estimating human PPO inhibitors according to the present invention The estimation method of the present invention is characterized by performing the following steps using a computer. First, (i) construct three-dimensional structural data of a candidate human PPO inhibitor compound, an oxidized FAD coenzyme, and an active site domain containing one or more amino acid residues. Next, (ii) calculate the spatial distribution of the LUMO of this active site domain by quantum chemical calculation. Then, (iii) classify the LUMO distribution shape into one of several predetermined patterns (for example, a pattern in which the LUMO shape of the vicinity including the isoalloxazine ring of the oxidized FAD is "retained" and a pattern in which it "changes or disappears"). Finally, (iv) use this classified pattern as a novel molecular descriptor and estimate the presence or degree of human PPO inhibitory activity of the candidate compound using a machine learning model. The machine learning model used for this estimation is prepared by pre-training the correlation between the inhibitory activity of known compounds and the LUMO shape pattern.
[0053] The general inhibitory activity of human PPO inhibitors is expressed by IC50 (the molar concentration of a compound that inhibits human PPO enzyme activity by 50%), and a smaller value indicates stronger inhibitory activity. The IC50 values of the 20 compounds from Formula 1 to Formula 20 were obtained from Non-Patent Literature 3 and used for correlation analysis using machine learning. In addition, the pI50 value (the logarithm of the reciprocal of the molar concentration of a compound that inhibits human PPO enzyme activity by 50%) was calculated based on the IC50, and a regression model was constructed using machine learning to represent the degree of inhibitory activity as a continuous value, and used for correlation analysis.
[0054] Quantum chemical calculations should ideally be performed using the ab inertial molecular orbital method / density functional theory calculation program GAMESS to calculate the LUMO distribution shape of the conformations of human PPO inhibitors, oxidized FADs, and a human PPO active site domain model consisting of one or more amino acid residues. These calculations should be performed using a basis set larger than 6–31G(d) with the B3LYP functional of the DFT (Density Functional Theory) method.
[0055] The method for representing the LUMO distribution shape as a molecular descriptor can be performed according to the following standard procedure, classifying it into one of several predetermined qualitative patterns (for example, a pattern in which the LUMO distribution shape in the vicinity, including the isoalloxazine ring of oxidized FAD, is "retained" without change, and a pattern in which it "changes or completely disappears"). The classification of the calculated spatial distribution shape pattern of LUMO can be performed not only by visual inspection, but also automatically by quantitative methods, such as known image recognition techniques for calculating image similarity, or techniques for extracting three-dimensional shape features. Examples of such quantitative measurements include the volume, surface area, sphericity, and change in the centroid position of the visualized LUMO distribution. By applying these values as explanatory variables to a machine learning model, the classification can be automated.
[0056] To clarify the correlation of the present invention, a decision tree classifier was constructed using the IC50 values of 20 compounds (Chemical Formulas 1 to 20) described in Non-Patent Literature 3) to perform a correlation analysis between human PPO inhibitory activity and LUMO shape distribution. The performance was evaluated using test data. As a result, there were 10 true positives (TP) and 10 true negatives (TN), with 0 false positives (FP) and false negatives (FN). This achieved a classification performance of 100% in terms of sensitivity, specificity, and accuracy. This is an extremely high performance, corresponding to an area under the ROC curve (AUC) of 1.0, and similarly high robustness can be expected even when using cross-validation. This result shows that the method of the present invention, which involves pattern classification of LUMO distribution shape, can objectively and accurately estimate the presence or absence of human PPO inhibitory activity.
[0057] Furthermore, the 20 compounds listed in Chemical Formulas 1 to 20 described in Non-Patent Document 3 were targeted, and their LUMO distribution patterns were visually classified and quantified as follows: "complete disappearance" was assigned a value of 2, "change observed" a value of 1, and "no change" a value of 0. Next, this quantified pattern was used as the explanatory variable, and the pI50 value (logarithm of the reciprocal of the molar concentration of a compound that inhibits human PPO enzyme activity by 50%), which represents the intensity of inhibitory activity of each compound, was used as the dependent variable. Linear regression analysis was then performed using machine learning. As a result, a statistically very significant correlation was observed between the two, and the following highly accurate regression equation was obtained. pI50 = 1.21 * (LUMO distribution shape pattern value) + 4.19 (Significance F (p value) = 6.43 × 10 -7 , multiple coefficient of determination (R 2 )=0.756) These results demonstrate that by constructing a regression model using LUMO shape patterns as explanatory variables, it is possible to accurately estimate the degree of inhibitory activity of unknown compounds as a continuous value.
[0058] The correlation analysis was performed using a machine learning model. In the case of a classification model used in this invention to determine the presence or absence of inhibitory activity, the machine learning model is not limited to decision trees, but any known classification algorithm such as support vector machines (SVMs), random forests, and neural networks can be applied. Similarly, in the case of a machine learning regression model to estimate the degree of inhibitory activity, any known regression algorithm such as linear regression, decision tree regression, and support vector regression (SVR) can be applied.
[0059] In generating the aforementioned machine learning model and estimation model using the machine learning regression model, in addition to the LUMO distribution shape, parameters such as hydrophobic parameters, electronic parameters, and steric parameters related to the PPO inhibitor can be appropriately added as molecular descriptors. Examples of such molecular descriptors include, but are not limited to, the LogP value indicating the hydrophobic parameter of the compound, σ (Hammett constant) indicating the electronic parameter, and MR, Es, and STERIMOL values indicating the steric parameter. Therefore, by combining these molecular descriptors, it is possible to further improve the accuracy of the estimation model.
[0060] (2) Selection method for human PPO inhibitors according to the present invention The method for selecting human PPO inhibitors according to the present invention is characterized by selecting human PPO inhibitors using the presence or degree of human PPO inhibitory activity obtained by the estimation method described in (1) above as an indicator. As a specific example, first, candidate compounds are selected from a compound database containing a large number of compounds, such as PubChem, ChEMBL, and ZINC, based on assay data, etc. Next, the presence or degree of human PPO inhibitory activity is estimated for these candidate compounds using the estimation method of the present invention. Then, based on the estimation results, promising compounds are selected as subjects for experimental verification. This selection can be made, for example, by comparing the estimation results with the activity of existing human PPO inhibitors (such as acifluorfen) or predetermined reference values. In addition, the inhibitory activity (such as IC50) of the selected candidate compounds is measured experimentally, and the measured value is compared with the estimation result to ultimately select a superior human PPO inhibitor.
[0061] (3) A system for carrying out the present invention The method of the present invention can be suitably carried out by an image processing system including a computer on which a dedicated program is installed. The system may be configured to include, for example, a reading and alignment unit that reads three-dimensional structural data of LUMO images and performs alignment for comparison, a preprocessing unit that extracts FAD neighbor regions to be compared, an inference unit that implements the machine learning model described in claim 5, etc., and determines the presence and degree of inhibitory activity, and a postprocessing and display unit that displays the estimation results in an easy-to-understand manner for the user.
[0062] The present invention will be further described in detail below with reference to examples, but it goes without saying that the present invention is not limited to these examples. [Examples] [Example 1]
[0063] <Construction of three-dimensional structural data of the human PPO active site domain> In this example, to specifically construct the human PPO active site domain and identify differences in LUMO distribution shape and spatial distribution patterns depending on the presence or absence of a human PPO inhibitor, 16 amino acid residues located within 4.00 Å of the heavy atom of acifluorfen, a human PPO inhibitor, were extracted from the crystal structure data of human PPO (PDB ID: 3NKS) and are shown in Table 1 in order of proximity. Acifluorfen and oxidized FAD were combined with one or more of these amino acid residues to construct a human PPO active site domain model, which was used for correlation analysis of the presence or absence of human PPO inhibitory activity. [Table 1] [Example 2]
[0064] <Calculation of LUMO distribution shape in a human PPO active site domain control model> From the crystal structure data of human PPO (PDB ID: 3NKS), a control model of the human PPO active site domain without the human PPO inhibitor Acifluorfen was constructed to define the LUMO distribution shape in the absence of an inhibitor, as a comparative reference for changes in the LUMO distribution shape. This control model consists of 16 amino acid residues selected in Example 1 and oxidized FAD. Specifically, after constructing the control model from the crystal structure data of human PPO (PDB ID: 3NKS), hydrogen was added using the open-source modeling software Molby, and structural optimization of all interatomic distances, bond angles, and dihedral angles was performed using the semi-empirical molecular orbital calculation method PM3. Using the optimized conformation, single-point calculations were performed using the DFT method's B3LYP functional and 6-31G(d) basis functions to determine the LUMO shape. The calculation results were visualized using the open-source modeling software Avogadro with the isosurface value set to 0.05 (Figure 1). This was used as the criterion for the "LUMO shape retention" pattern (no change) corresponding to "no inhibitory activity". [Example 3]
[0065] <Classification of LUMO distribution patterns based on differences in amino acid residue combinations in the human PPO active site domain> Next, in Example 2, to compare with the LUMO distribution shape of the human PPO active site domain containing all 16 residues, including the inhibitor Acifluorfen, which was used as a control model, the LUMO distribution shape was calculated in the same manner as in Example 1. The spatial distribution was clearly different from that without the inhibitor (Figure 1), and the LUMO distribution shape was classified as altered (Figure 2), suggesting that the LUMO distribution shape changed as a result of the inhibition of the redox reaction of human PPO by Acifluorfen. Furthermore, Table 2 shows the relationship between the LUMO distribution shape and combinations of amino acid residues from 1 to 16. As a result, the LUMO distribution shape showed "complete disappearance" or "alteration" for all combinations of amino acid residues from 1 to 16, which we considered to be the basis for Acifluorfen exhibiting human PPO inhibitory activity in any combination of amino acid residues from 1 to 16. Moreover, these examples are merely representative examples, and similar "complete disappearance" or "alteration" of the LUMO distribution shape was confirmed for other combinations of amino acid residues from 1 to 16. These results show that the LUMO distribution shape of the human PPO active site domain changes similarly for a large number of amino acid residue combinations from 1 to 16, and that the LUMO distribution shape pattern can be defined as either "completely absent" or "changed" as corresponding to the presence of human PPO inhibitory activity. Thus, by including a wide range of amino acid residues near the inhibitor, a model that sensitively reflects inhibitory activity can be constructed. Therefore, it can be concluded that the present invention's approach of extracting amino acid residues located within 4.00 Å of the heavy atom of Acifluorfen to construct the active site domain is one of the most effective methods in this invention. [Table 2] [Example 4]
[0066] <Creation of human PPO inhibitor candidate compounds and oxidized FAD complexes> In this embodiment, in order to construct a machine learning model and analyze the correlation between the pattern classification of LUMO distribution shape and the presence or absence of human PPO inhibitor activity using machine learning, oxidized FAD complexes were prepared with candidate human PPO inhibitor compounds for the 20 compound groups shown in [Chemical Formula 1] to [Chemical Formula 20], for which the human PPO inhibitor activity IC50 has been reported in Non-Patent Literature 3, using the following method.
[0067] The acifluorfen complex, a two-molecule complex consisting of acifluorfen and oxidized FAD obtained from the crystal structure data of human PPO (PDB ID: 3NKS), was created using the molecular modeling software Winmostar64Bit(FREE)V11.6.1.
[0068] By modifying the Acifluorfen complex by introducing substituents and other structural changes, we created Oxyfluorfen, Lactofen, Fomesafen, and Chlornitrofen complexes, each consisting of two molecules of a diphenyl ether herbicide and oxidized FAD.
[0069] Next, using the single-molecule crystal structure (Non-Patent Literature 9) of the non-human PPO inhibitor Chlortoluron, the Chlortoluron complex was created by superimposing three CNC atoms from the 1-position of the benzene ring of 3-chloro-4-methylphenyl in the amide bond portion of Chlortoluron onto three COC atoms from the ether bond portion of the Acifluorfen complex. This created a Chlortoluron complex, a two-molecule complex of Chlortoluron and oxidized FAD. Similarly, using the single-molecule crystal structure (Non-Patent Literature 10) of the non-human PPO inhibitor Atrazine, the Atrazine complex was created by superimposing three CNC atoms from the 4th, 5th, and 6th positions of the s-triazine ring onto three COC atoms from the ether bond portion of the Acifluorfen complex. This created an Atrazine complex, a two-compound complex of Atrazine and oxidized FAD.
[0070] Furthermore, the Oxadiazon and Oxadiargyl complexes were prepared as follows: First, the PPO crystal structure data from the tobacco plant (Nicotiana tabacum) with description number 1SEZ was obtained from the PDB. From its conformational data, Oxadiazon and a 1SEZ inhibitor with a hetero 5-membered ring structure similar to Oxadiazon and Oxadiargyl were constructed. Only the conformation of the bimolecular complex of (4-BROMO-3-(5'-CARBOXY-4'-CHLORO-2'-FLUOROPHENYL)-1-METHYL-5-TRIFLUOROMETHYL-PYRAZOL) and oxidized FAD was extracted and designated as the 1SEZ complex. This was done to obtain the initial configuration of an inhibitor with a different structure from Acifluorfen. Since the subsequent LUMO calculation would use amino acid residues derived from human PPO, the oxidized FAD of the 1SEZ complex and the Acifluorfen complex were superimposed with isoaloxazine rings, and the Acifluorfen of the Acifluorfen complex derived from the human PPO crystal structure was replaced with the 1SEZ inhibitor to create the 3NKS-1SEZ complex. Then, based on this 3NKS-1SEZ complex, the Oxadiazon complex and the Oxadiargyl complex were created by modifying the substituents and other structures using Winmostar.
[0071] For Butafenacil, Saflufenacil, and Compound 1, the heterosix-membered rings were superimposed with the benzene rings of Acifluorfen that have 2-Cl, 4-CF3, respectively, to create Butafenacil, Saflufenacil, and Compound 1 complexes. Similarly, for Compounds 2 through 9, the benzene rings of Acifluorfen that have 2-Cl, 4-Cl were superimposed with the benzene rings of Acifluorfen that have 2-Cl, 4-CF3, respectively, to create Compound 2 complexes to Compound 9 complexes. [Example 5]
[0072] <Analysis of human PPO inhibitor activity of candidate human PPO inhibitor compounds using a machine learning model - in the case of 9 amino acid residues> First, as shown in Table 2, the human PPO active site domain of Acifluorfen, consisting of 9 amino acid residues (Arg97, Ile419, Ile346, Val347, Leu334, His333, Gly332, Phe331, Arg168) that showed a change in LUMO distribution shape due to Acifluorfen, and complexes prepared for the 20 compounds from [Chemical Formula 1] to [Chemical Formula 20] were superimposed on the isoaloxazine rings of oxidized FAD using Winmostar. Acifluorfen in the human PPO active site domain of Acifluorfen was replaced with the compounds from [Chemical Formula 1] to [Chemical Formula 20], respectively, to create the human PPO active site domains from [Chemical Formula 1] to [Chemical Formula 20], and the LUMO distribution shape was calculated in the same manner as in Example 2. The LUMO shape of each compound was classified into "complete disappearance," "changed," or "unchanged" based on the patterns defined in Example 2 and Example 3, and compounds with an inhibitory activity IC50 of 3.86 μM or less were classified as "active" (Table 3). Figure 3 shows the LUMO distribution shape of the human PPO active site domain control model without acifluorfen, calculated in the same manner as in Example 2, and Figure 4 shows the results for the "changed" human PPO active site domain containing acifluorfen, calculated in the same manner as in Example 3. This classification (labeling) was performed by two evaluators who independently performed visual verification, and only those whose judgments matched were used. [Table 3]
[0073] A decision tree classifier was constructed using this dataset, and its performance was evaluated on the test data. As a result, there were 10 true positives (TP) and 3 true negatives (TN), with false positives (FP) and false negatives (FN) being as follows: 7 matter and 0 This resulted in sensitivity, specificity, and positive predictive value of 100%, 30%, and 58.8%, respectively. While this initial model had high sensitivity at 100%, its specificity was low at 30%, leading to the misidentification of inactive compounds. [Example 6]
[0074] <Analysis of human PPO inhibitor activity of candidate human PPO inhibitor compounds using a machine learning model - for compounds with 10 amino acid residues>
[0075] Therefore, by adding Gly169 to the above 9 amino acid residues, we further investigated the structure of the human PPO active site domain in order to solve the above problem, and found that making it 10 amino acid residues is extremely important for improving accuracy. Specifically, using the human PPO active site domain consisting of the above 10 amino acid residues (Arg97, Arg168, Gly169, Phe331, Gly332, His333, Leu334, Ile346, Val347, Ile419), we calculated the LUMO distribution shape of the 20 compounds from [Chemical Formula 1] to [Chemical Formula 20] in the same manner as in Example 2. The LUMO shape of each compound was classified as either "complete disappearance," "changed," or "unchanged" based on the patterns defined in Example 2 and Example 3, and compounds with an inhibitory activity IC50 of 3.86 μM or less were classified as "active" (Table 4). Figure 5 shows the LUMO distribution shape of the human PPO active site domain control model without acifluorfen, calculated in the same manner as in Example 2, and Figure 6 shows the result of "complete disappearance" of the human PPO active site domain with acifluorfen, calculated in the same manner as in Example 3. This classification (labeling) was performed by two evaluators who independently performed visual comparisons, and only those whose judgments matched were used. [Table 4]
[0076] As shown in Table 4, in the case of Oxyfluorfen, the LUMO did not completely disappear, and the "changed" pattern is shown in Figure 7. However, the model of the present invention correctly classified this as having inhibitory activity. For other compounds with human PPO inhibitory activity, all compounds showed either "complete disappearance" or "changed." This indicates that the method of the present invention accurately estimated more analog and diverse LUMO distribution shapes as "changed," in addition to the digital "complete disappearance" and "no change" shape patterns. Furthermore, all compounds in the group with no change in LUMO distribution shape were estimated to have no human PPO inhibitory activity. Thus, the LUMO distribution shape is effective as a molecular descriptor for estimating human PPO inhibitory activity.
[0077] To further analyze these results using machine learning, a decision tree classifier was constructed using this dataset, and its performance was evaluated on test data. As a result, there were 10 true positives (TP), 10 true negatives (TN), and 0 false positives (FP) and false negatives (FN). This achieved a classification performance of 100% in sensitivity, specificity, and positive predictive value. This is a high performance corresponding to an area under the ROC curve (AUC) of 1.0, and similarly high robustness can be expected even when using cross-validation. From these results, it can be concluded that a method of classifying LUMO distribution shapes can objectively and accurately estimate the presence or absence of human PPO inhibitory activity.
[0078] Comparing the results of Example 5 (9 amino acid residues) (specificity 30%) with those of Example 10 (specificity 100%), it is clear that adding just one amino acid residue, Gly169, significantly improved classification performance. Gly169 is only one amino acid residue located near the active site, and the presence or absence of this single amino acid residue strongly influences the LUMO distribution shape pattern of the entire human PPO active site domain, fundamentally improving classification accuracy. This result demonstrates a synergistic effect that is difficult to predict simply by adding one amino acid residue. [Example 7]
[0079] <Construction of three-dimensional structural data of the active center domain of Bacillus subtilis> This example directly compared models of the PPO active site domain of Bacillus subtilis, which is resistant to acifluorfen, with that of the human PPO active site domain, demonstrating interspecies selectivity. The PPO crystal structure data from Bacillus subtilis containing acifluorfen used for comparison can be obtained from the PDB as description number 3I6D. Similar to Example 1, seven amino acid residues located within 4.00 Å of the heavy atom of the Bacillus subtilis PPO inhibitor acifluorfen were extracted and are shown in Table 5 in order of proximity. From these seven amino acid residues located within 4.00 Å, a Bacillus subtilis active site domain model derived from 3I6, which is sterically significantly different from the human PPO active site domain, was constructed. [Table 5] [Example 8]
[0080] <Calculation of LUMO distribution shape in a Bacillus subtilis PPO active center domain control model> Based on the above crystal structure data of Bacillus subtilis PPO (PDB ID: 3I6D), to define the LUMO distribution shape in the absence of a PPO inhibitor, as a control for the change in the LUMO distribution shape of the Bacillus subtilis PPO active site domain, a Bacillus subtilis PPO active site domain without acifluorfen, consisting of the coenzyme oxidized FAD and the above 7 amino acid residues, was constructed in the same manner as in Example 2, and the LUMO distribution shape was calculated and shown in Figure 8. This was used as the standard for the "no change" pattern of "LUMO shape retention" corresponding to resistance to Bacillus subtilis PPO inhibitory activity. [Example 9]
[0081] <Pattern classification of LUMO distribution shape in the PPO active site domain of Bacillus subtilis> Next, to compare with the control model defined in Example 8, which does not contain the inhibitor Acifluorfen (Figure 8), we introduced the inhibitor Acifluorfen into the Bacillus subtilis PPO active site domain containing all seven amino acid residues and calculated the LUMO distribution shape. As shown in Figure 9, the LUMO distribution shape was classified as "changed." Furthermore, as a result of examining other amino acid residue combinations, as shown in Table 6, we obtained an extremely interesting result: in the case of a specific combination of four amino acid residues (Leu68, Lys71, Ile176, Met413), the LUMO distribution shape did not change despite the presence of the inhibitor Acifluorfen (Figure 10). Moreover, an important point here is that the "complete disappearance" pattern observed in the human PPO active site domain model (Example 3) was not observed in any combination in the Bacillus subtilis PPO active site domain model. From this, we considered that the fundamental mechanism by which Bacillus subtilis exhibits resistance to Acifluorfen is that the LUMO distribution shape remains "unchanged" or "changed," and does not reach "complete disappearance." We believe that this difference in the behavior of the LUMO pattern demonstrates the effectiveness of the present invention in enabling the estimation of interspecies selectivity. [Table 6] [Example 10]
[0082] <Calculation of LUMO distribution shape for diphenyl ether-based PPO inhibitors against Bacillus subtilis> First, for Acifluorfen, Oxyfluorfen, and Acifluorfen-methyl, which have been reported to have resistance to Bacillus subtilis, the change in LUMO distribution shape was calculated using the Bacillus subtilis PPO active site domain consisting of four amino acid residues (Leu68, Lys71, Ile176, Met413) that showed "no change" in LUMO distribution shape with Acifluorfen, as shown in Table 7, in the same manner as in Example 3. In addition to Acifluorfen shown in Figure 10, the LUMO distribution shape of the Bacillus subtilis PPO active site domain containing Oxyfluorfen is shown in Figure 11. When this result was compared with the control LUMO distribution shape obtained in Example 8 (Figure 8), it showed "no change" in LUMO distribution shape, similar to Figure 8. When this was compared with the results for Acifluorfen and Oxyfluorfen obtained from the human PPO active site domain in Example 6, it clearly demonstrated consistency between the experimentally observed resistance between Bacillus subtilis and humans, where it was effective in humans but not in Bacillus subtilis. This suggests that the difference in LUMO distribution shape resulting from differences in the amino acid residue composition of the active site domains of resistant Bacillus subtilis and humans is the cause of the resistance between them, and that this allows for the estimation of interspecies resistance between Bacillus subtilis and humans. Similar results were obtained for Acifluorfen-methyl (Chemical Formula 21), a methyl ester of Acifluorfen, which also showed "no change" in LUMO distribution shape (Figure 12), and was correctly estimated. Acifluorfen-methyl is not included in the group of 20 compounds used in the correlation analysis of Examples 5 and 6. However, this result also suggests that this method can estimate the interspecies selectivity of any candidate compound, not just the above compounds. The above results are summarized in Table 7. From these results, it was found that the sensitivity of PPO inhibitory activity by Acifluorfen, Oxyfluorfen, and Acifluorfen-methyl in human PPO and Bacillus subtilis PPO is consistent with the experimental results. [Table 7]
[0083] The above results prove that the configuration of the present invention is extremely effective in estimating interspecies selectivity, that is, even for PPO active center domains of different species, the change in inhibitory activity can be pattern-recognized by the LUMO distribution shape. That is, as a result of comparing the active center domains of human PPO and Bacillus subtilis PPO, in Bacillus subtilis showing resistance, the LUMO distribution shape either did not change or remained with some change, and no pattern of complete disappearance was observed. This is in clear contrast to the result that complete disappearance was observed in human PPO showing sensitivity. It was concluded that this difference in the degree of disappearance as to whether or not the LUMO distribution shape reaches complete disappearance is one of the causes of resistance to the drug and is the root mechanism of interspecies selectivity.
Example 11
[0084] <Automatic Classification Based on Quantitative Feature Quantities of LUMO Distribution Shape> The pattern classification of the LUMO distribution shape in the estimation method of the present invention was verified by a method that is more objective and automatic without relying on human vision. Specifically, for the human PPO active center domain consisting of 10 amino acid residues used in Example 6, the LUMO distributions of the control model without an inhibitor and the model containing Acifluorfen as an inhibitor were obtained as three-dimensional images.
[0085] Next, using known three-dimensional image analysis software, a plurality of geometric feature quantities (volume, surface area, sphericity, center of gravity position, etc.) describing the three-dimensional shape were calculated from each LUMO distribution image. As a result, clear differences were observed in the values of the calculated feature quantities between the control model and the model containing Acifluorfen.
[0086] Then, using these geometric feature quantities as explanatory variables and the presence or absence of inhibitory activity (based on the IC50 value) as the objective variable, they were applied to the same decision tree classifier as constructed in Example 6. As a result, the number of false positives (FP) and false negatives (FN) was zero, and a classification performance of 100% was achieved in all of sensitivity, specificity, and accuracy.
[0087] From the above results, it was demonstrated that by using the geometric feature quantity of the LUMO distribution shape, it is possible to automatically estimate the presence or absence of inhibitory activity objectively and with high accuracy without the need for visual determination.
Example 12
[0088] <Quantitative estimation of the degree of inhibitory activity based on the LUMO shape pattern> In addition, the present invention is applicable not only to the estimation of the "presence or absence" but also to the estimation of the "degree" of inhibitory activity. For example, starting from using the pattern of the LUMO shape (e.g., "complete disappearance" as 2, "change" as 1, "no change" as 0) as an explanatory variable, based on one or more geometric feature quantities calculated from the three-dimensional structure data of the human PPO active center domain, with the intensity of the inhibitory activity (pI50 value, etc.) of known compounds as the objective variable, a regression model of machine learning such as decision tree regression or linear regression is constructed. By inputting the LUMO shape pattern and geometric feature quantities calculated for an unknown candidate compound into this regression model, the degree of inhibitory activity can be quantitatively estimated as a continuous value.
[0089] The pattern results of the LUMO distribution shapes of the compound groups from [Chemical Formula 1] to [Chemical Formula 20] in Table 4 of Example 6 were digitized with "complete disappearance" as 2, "change" as 1, and "no change" as 0.
[0090] Next, with this digitized pattern of the LUMO distribution shape as the explanatory variable and the pI50 value (the logarithm of the reciprocal of the molar concentration of the compound that inhibits 50% of the human PPO enzyme activity) obtained from the inhibitory activity IC50 of each compound described in Non-Patent Document 3 as the objective variable, a machine learning regression model was constructed and regression analysis was performed.
[0091] As a result, there was a statistically extremely significant correlation between the two, and a significant F (p-value) = 6.43×10 -7 (This is an extremely significant regression analysis result close to almost 0), and furthermore, the multiple determination coefficient (R 2 ) was 0.756, and the following highly accurate regression equation was obtained. pI50 = 1.21*(LUMO distribution shape pattern value) + 4.19
[0092] Figure 13 shows a linear regression equation for PPO inhibitors that have PPO inhibitory activity against humans, with the horizontal axis representing the pattern value of the LUMO shape distribution as a "molecular descriptor" and the vertical axis representing the human PPO inhibitory activity (pI50) of the compound. This result clearly demonstrates that by constructing a regression model with LUMO shape patterns as explanatory variables, it is possible to estimate the degree of inhibitory activity of unknown compounds with high accuracy as a continuous value. [Industrial applicability]
[0093] This invention is a technology for accurately estimating the presence or degree of inhibitory activity that a chemical substance exerts on human protoporphyrinogen oxidase (human PPO) on a computer. Its industrial applicability is broad and will bring about significant changes in the following fields. First, in the development of pharmaceuticals such as antitumor drugs, it will increase the probability of success in research and development and drastically reduce the enormous cost and time by efficiently narrowing down only the promising compounds from a vast number of candidate compounds in the early stages of development. Second, in the safety evaluation of physiologically active substances such as pesticides, it will be possible to evaluate the toxicity that physiologically active substances exert on the human body without relying on animal experiments, thus solving ethical issues related to animal protection and contributing to the alternative to animal experiments. Furthermore, as an application, it is extremely useful in the development of pesticides, including herbicides, and will support the design and development of safe and effective compounds that do not affect the human body and selectively inhibit only the PPO of target weeds or fungi.
Claims
1. A method for estimating the presence or degree of inhibitory activity of a candidate compound against human protoporphyrinogen oxidase (human PPO), characterized by comprising the following steps (i) to (iv) performed by a computer: (i) A step of constructing three-dimensional structural data of an active site domain, comprising the candidate compound, oxidized FAD which is a coenzyme of human PPO, and one or more amino acid residues that constitute the active site of human PPO; (ii) A step of calculating the spatial distribution of the lowest unoccupied orbitals (LUMOs) of the active site domain; (iii) A step of classifying the calculated spatial distribution of LUMO into one of a plurality of patterns, including a first pattern in which the three-dimensional shape of the LUMO distribution is maintained at the same level in spatial distribution as the standard, and a second pattern in which the shape of the LUMO distribution changes or completely disappears, based on the manner of three-dimensional shape change of the LUMO distribution in the vicinity including the isoalloxazine ring of the oxidized FAD when compared with the state in which the candidate compound is removed from the active site domain, the classification being performed automatically using a machine learning model that has been trained with quantitative numerical information obtained by calculating the similarity of images or extracting three-dimensional shape features from the three-dimensional data of the spatial distribution of LUMO as an index that quantifies the pattern of the LUMO shape, with the index as an explanatory variable and the classification of the pattern as the dependent variable; (iv) A step of estimating the presence or degree of inhibitory activity of the candidate compound by performing at least one of the following processes (a) and (b) based on the classification result of the pattern obtained in step (iii) above or the index: (a) A process to qualitatively estimate the presence or absence of inhibitory activity of the candidate compound by taking the classification result as input, determining that human PPO inhibitory activity is "present" if the classification result is the second pattern, and determining that human PPO inhibitory activity is "absent" if the classification result is the first pattern. (b) A process to quantitatively estimate the degree of inhibitory activity of the candidate compound using a regression model that has been trained with the above-mentioned index as an explanatory variable and the intensity of inhibitory activity of known compounds as the dependent variable.
2. The method according to claim 1, characterized in that the one or more amino acid residues are selected from the group consisting of 16 residues: Arg97, Arg168, Gly169, Val170, Phe171, Ala172, Phe331, Gly332, His333, Leu334, Leu344, Gly345, Ile346, Val347, Met368, and Ile419.
3. The method according to claim 1, further comprising a comparison step of calculating the spatial distribution of LUMO of a candidate compound based on three-dimensional structural data of PPO of Bacillus subtilis, and verifying the interspecies selectivity of the candidate compound using the difference from the spatial distribution of LUMO in human PPO as an indicator.
4. The method according to claim 3, characterized in that the calculation of the spatial distribution of LUMO of candidate compounds based on the PPO three-dimensional structural data of Bacillus subtilis is performed using an active site domain in which the amino acid residues are selected from the group consisting of seven residues: Leu68, Lys71, Gly175, Ile176, Tyr177, Met413, and Val448.
5. A computer-based screening method for screening compounds having human PPO inhibitory activity from a compound database consisting of numerous candidate compounds, the method comprising the step of selecting candidate compounds that are estimated to have human PPO inhibitory activity using the estimation method described in claim 1 as subjects for experimental verification.
6. A method characterized by selecting a human PPO inhibitor from a group of candidate compounds by comparing the estimation result obtained by the estimation method described in claim 1 with the inhibitory activity of existing human PPO inhibitors or predetermined criteria.
7. The method according to claim 6, wherein the existing human PPO inhibitor used for comparison includes at least one of Acifluorfen, Oxyfluorfen, Lactofen, Fomesafen, Chlornitrofen, Oxadiazon, Oxadiargyl, Butafenacil, and Saflufenacil.
8. The method according to claim 6, which involves selecting a human PPO inhibitor by comparing the inhibitory activity (IC50, etc.) of the candidate compound measured by experimental methods with the estimated result.
9. An image processing system that performs the method described in claim 1, characterized in that it includes a LUMO image reading and alignment unit, a preprocessing unit for the FAD neighborhood region, an inference unit equipped with the machine learning model, and a postprocessing and display unit that outputs the activity estimation result.
10. A program for causing a computer to perform the method described in claim 1.
11. A non-temporary recording medium on which the program described in claim 10 is recorded.