@techreport{oai:ipsj.ixsq.nii.ac.jp:00226916,
 author = {Yu, Morikawa and Kilho, Shin and Masataka, Kubouchi and Hiroaki, Ohshima and Yu, Morikawa and Kilho, Shin and Masataka, Kubouchi and Hiroaki, Ohshima},
 issue = {12},
 month = {Jul},
 note = {Metal-organic frameworks (MOFs) are coordination networks of metal ions or clusters bonded by organic ligands, and are expected to have a significantly wide application range due to their tunable micro-porosity and unprecedented surface area. While the diverse range of choices on combinations of metals and organic ligands is the origin of the tunable micro-porosity, it also makes deployment of optimal combinations difficult. The aim of this research is to propose a method to search candidates of optimal combinations leveraging machine learning techniques without actual synthesis of MOFs and as a result to make development of MOFs more effective and more efficient. In order to incorporate the structural features of organic ligands into investigation, this paper leverages graph kernels: the structural features should be represented faithfully by graphs of atoms, and graph kernels evaluate the similarity between graphs. Furthermore, we define a unified kernel to evaluate the similarity between MOFs, which is a weighted sum of an RBF kernel to evaluate the similarity of metals and the said graph kernel. With this unified kernel and the support vector machine classification algorithm, we develop a model that predicts whether a given combination of a metal and an organic ligand can result in an MOF with high specific surface area. Through the experiments with MOF data retrieved from the CoRE MOF dataset, we have verified that our proposed method can generate models with good predictive performance for six of the seven graph kernels well known in the literature., Metal-organic frameworks (MOFs) are coordination networks of metal ions or clusters bonded by organic ligands, and are expected to have a significantly wide application range due to their tunable micro-porosity and unprecedented surface area. While the diverse range of choices on combinations of metals and organic ligands is the origin of the tunable micro-porosity, it also makes deployment of optimal combinations difficult. The aim of this research is to propose a method to search candidates of optimal combinations leveraging machine learning techniques without actual synthesis of MOFs and as a result to make development of MOFs more effective and more efficient. In order to incorporate the structural features of organic ligands into investigation, this paper leverages graph kernels: the structural features should be represented faithfully by graphs of atoms, and graph kernels evaluate the similarity between graphs. Furthermore, we define a unified kernel to evaluate the similarity between MOFs, which is a weighted sum of an RBF kernel to evaluate the similarity of metals and the said graph kernel. With this unified kernel and the support vector machine classification algorithm, we develop a model that predicts whether a given combination of a metal and an organic ligand can result in an MOF with high specific surface area. Through the experiments with MOF data retrieved from the CoRE MOF dataset, we have verified that our proposed method can generate models with good predictive performance for six of the seven graph kernels well known in the literature.},
 title = {Prediction of Specific Surface Area of Metal-Organic Frameworks by Graph Kernels},
 year = {2023}
}