Apparatus and method for generating molecular structure
A method of generating a molecular structure includes generating, based on a predetermined number of a plurality of nodes, all possible two-dimensional (2D) graphs and a plurality of edges representing connections between the plurality of nodes and, for each 2D graph from among the generated 2D graphs, generating all possible molecular structures based on the 2D graph by substituting each of the plurality of nodes with a polygonal ring structure including carbon atoms and substituting the edges with bonds between polygonal ring structures. Also, a method includes substituting at least one of carbon atoms included in the polygonal ring structures with an atom other than a carbon atom.
1. A method of generating a molecular structure, the method comprising:
generating, based on a predetermined number of a plurality of nodes, all possible two-dimensional (2D) graphs including the predetermined number of nodes and a plurality of edges representing connections between the plurality of nodes;
for each 2D graph from among the generated 2D graphs, generating all possible molecular structures based on the 2D graph that meet a predetermined condition by:
substituting each of the plurality of nodes with a polygonal ring structure including carbon atoms,
substituting the plurality of edges with bonds between polygonal ring structures; and
substituting at least one of the carbon atoms included in the polygonal ring structures with an atom other than a carbon atom,
selecting some of the generated molecular structures;
calculating a specific property value with respect to each of the selected molecular structures; and
training a neural network by using a descriptor of the selected molecular structures as input data of the neural network and using the specific property value corresponding to each of the selected molecular structures as output data of the neural network,
wherein the predetermined condition comprises:
in the molecular structures corresponding to the 2D graphs, a number of peripheral ring structures surrounding a specific ring structure,
in the molecular structures corresponding to the 2D graphs, a number of ring structures bonded around a specific atom, and
in the 2D graphs, whether edges of a 2D graph are expressed within a plane without overlapping.
2. The method of claim 1 , wherein the substituting the at least one of the carbon atoms comprises:
substituting a first carbon atom from among the carbon atoms included in the polygonal ring structures with a metal atom; and
substituting at least one second carbon atom from among the carbon atoms included in the polygonal ring structures with any one of nitrogen (N), oxygen (O), phosphorus (P), sulfur (S), selenium (Se), and silicon (Si).
3. The method of claim 2 , wherein the first carbon atom is a carbon atom having a bond order of 2 from among the carbon atoms included in the polygonal ring structures.
4. The method of claim 3 , wherein a bond order of adjacent carbon atoms adjacent to the metal atom is 3.
5. The method of claim 1 , wherein each of the polygonal ring structures is any one from among a pentagonal ring structure and a hexagonal ring structure.
6. The method of claim 1 , wherein each of the generated 2D graphs comprises three to five nodes.
7. The method of claim 1 , wherein an adjacent polygonal ring structure adjacent to a polygonal ring structure including the metal atom is an aromatic ring.
8. The method of claim 1 , further comprising:
calculating the specific property value with respect to an arbitrary molecular structure from among the generated molecular structures by using at least some of the generated molecular structures and the neural network, the neural network being trained based on matching data obtained by matching each of the at least some of the generated molecular structures with a corresponding property value.
9. A non-transitory computer readable recording medium comprising a program, which when executed by a computer, performs the method of claim 1 .
10. An apparatus for generating a molecular structure, the apparatus comprising:
a memory storing at least one program; and
a processor configured to drive the apparatus for generating the molecular structure by executing the at least one program,
wherein the processor is further configured to:
generate, based on a predetermined number of a plurality of nodes, all possible two-dimensional (2D) graphs including the predetermined number of nodes and a plurality of edges representing connections between the plurality of nodes,
for each 2D graph from among the generated 2D graphs, generating all possible molecular structures based on the 2D graph that meet a predetermined condition by:
substituting each of the plurality of nodes with a polygonal ring structure comprising carbon atoms,
substituting the plurality of edges with bonds between polygonal ring structures, and
substituting at least one of the carbon atoms included in the polygonal ring structures with an atom other than a carbon atom,
select some of the generated molecular structures;
calculate a specific property value with respect to each of the selected molecular structures; and
train a neural network by using a descriptor of the selected molecular structures as input data of the neural network and using the specific property value corresponding to each of the selected molecular structures as output data of the neural network,
wherein the predetermined condition comprises:
in the molecular structures corresponding to the 2D graphs, a number of peripheral ring structures surrounding a specific ring structure,
in the molecular structures corresponding to the 2D graphs, a number of ring structures bonded around a specific atom, and
in the 2D graphs, whether edges of a 2D graph are expressed within a plane without overlapping.
11. The apparatus of claim 10 , wherein the processor is further configured to substitute a first carbon atom from among the carbon atoms included in the first molecular structures with a metal atom and substitute at least one second carbon atom from among the carbon atoms included in the polygonal ring structures with any one of nitrogen (N), oxygen (O), phosphorus (P), sulfur (S), selenium (Se), and silicon (Si).
12. The apparatus of claim 11 , wherein the first carbon atom is a carbon atom having a bond order of 2 from among the carbon atoms included in the polygonal ring structures.
13. The method of claim 12 , wherein a bond order of adjacent carbon atoms adjacent to the metal atom is 3.
14. The apparatus of claim 11 , wherein an adjacent polygonal ring structure adjacent to a polygonal ring structure comprising the metal atom is an aromatic ring.
15. The apparatus of claim 10 , wherein each of the polygonal ring structures is any one from among a pentagonal ring structure and a hexagonal ring structure.
16. The apparatus of claim 10 , wherein each of the generated 2D graphs comprises three to five nodes.
17. The apparatus of claim 9 , wherein the processor is further configured to calculate the specific property value with respect to an arbitrary molecular structure from among the generated molecular structures by using at least some of the generated molecular structures and a neural network, the neural network being trained based on matching data obtained by matching each of the at least some of the generated molecular structures with a corresponding property value.