Method for predicting enzyme-catalyzed reactions
View Patent ↗The reactivity of given metabolites is assessed using selected empirical atomic properties in the potential reaction center. Metabolic reactions are represented as biotransformation rules. These rules are generalized from the patterns in reactions. These patterns are not unique to reactants but are widely distributed among metabolites. Using a metabolite database, potential substructures are identified in the metabolites for a given biotransformation. These substructures are divided into reactants or non-reactants, depending on whether they participate in the biotransformation or not. Each potential substructure is then modeled using descriptors of the topological and electronic properties of atoms in the potential reaction center; molecular properties can also be used. A Support Vector Machine (SVM) or classifier is trained to classify a potential reactant as a true or false reactant using these properties.
1. A method of predicting the reactivity of a metabolite, comprising:
using a processor and software executable on the processor, performing steps comprising:
obtaining information about a class of metabolic chemical reactions;
dividing the class of reactions into a plurality of subclasses, each marked by a particular type of functional group transformation;
determining local structures of metabolite reaction centers for each transformation;
defining a set of reaction center patterns from the local structures; and
determining if the metabolite is a potential reactant in a subclass by determining if it has structural features compatible with a subclass-specific reaction center using a trained classifier.
2. The method of claim 1 , wherein determining if the metabolite is a potential reactant in a subclass by determining if it has structural features compatible with a subclass-specific reaction center using a trained classifier comprises:
training a classifier for each transformation;
calculating a set of descriptors selected from atomic properties for each atom in a reaction center and molecular properties;
inputting the set of descriptors to the classifier; and
predicting the reactivity of the metabolite to participate in a reaction.
3. The method of claim 1 , wherein the class is enzyme-catalyzed reactions.
4. The method of claim 3 , wherein one subclass is oxidoreductase-catalyzed reactions.
5. The method of claim 2 , wherein the class is enzyme-catalyzed reactions.
6. The method of claim 5 , wherein one subclass is oxidoreductase-catalyzed reactions.
7. The method of claim 1 , wherein the reaction center patterns are selected to match as many known reactants in the corresponding subclass as possible.
8. The method of claim 2 , wherein the reaction center patterns are selected to match as many known reactants in the corresponding subclass as possible.
9. The method of claim 2 , wherein the classifier is trained by providing positive and negative examples.
10. The method of claim 9 , wherein positive and negative examples are provided by identifying a set of structures among the metabolites in the class of metabolite reactions that contain each reaction center pattern and considering a structure to be a positive example if the metabolite containing it is known to participate in a reaction corresponding to the reaction center pattern the structure matches and considering it a negative example if it is not so known.
11. The method of claim 2 , wherein the classifier is a support vector machine.
12. The method of claim 11 , wherein the classifier is a linear-kernel support vector machine.
13. The method of claim 2 , wherein the descriptors are atomic properties.
14. The method of claim 13 , wherein the atomic properties include electric, energetic, steric and topological properties.
15. The method of claim 13 , wherein the atomic properties are graph potential, electrogeometrical state index, electrotopological state index, conjugated electrotopological state index, Gasteiger-Marsili partial charges, effective polarizability, and sigma electronegativity.
16. The method of claim 13 , further comprising molecular properties.
17. A system for predicting the reactivity of a metabolite, comprising:
a processor for (i) receiving information about a class of metabolic chemical reactions, (ii) dividing the class of reactions into a plurality of subclasses, each marked by a particular type of functional group transformation, (iii) determining local structures of metabolite reaction centers for each transformation, and (iv) defining a set of reaction center patterns from the local structures;
a classifier for predicting the reactivity of the metabolite to participate in a reaction, the classifier having been trained for each transformation; and
a descriptor calculator for calculating a set of descriptors selected from atomic properties for each atom in a reaction center and molecular properties, the set of descriptors being input to the trained classifier.
18. The system of claim 17 , wherein the classifier is a support vector machine.
19. The system of claim 17 , wherein the descriptors are atomic properties.
20. A system for predicting the reactivity of a metabolite, comprising:
a processor for (i) receiving information about a class of metabolic chemical reactions, (ii) dividing the class of reactions into a plurality of subclasses, each marked by a particular type of functional group transformation, (iii) determining local structures of metabolite reaction centers for each transformation, and (iv) defining a set of reaction center patterns from the local structures; and
means for determining if the metabolite is a potential reactant in a subclass by determining if it has structural features compatible with a subclass-specific reaction center.