Date: 2018/07/17 (Tue.)
場所：奈良先端大 物質創成科学棟 小講義室F105
Location: NAIST, Materials Science Complex, Seminar Hall F105
13:00-13:10 Prof. Funatsu
13:10-14:00 Prof. Gasteiger
14:00-14:30 Dr. Ishikawa
14:35-15:05 Dr. Miyao
15:05-15:45 Dr. Ono and Dr. Eguchi
15:45-16:15 Dr. Hatanaka
Prof.Dr. Johann Gasteiger (University of Erlangen-Nürnberg; DSC, NAIST)
(University of Erlangen-Nürnberg・奈良先端大データ駆動型サイエンス創造センター)
Title: Prospects of Chemo- and Material Informatics in Industry
Computer-Chemie-Centrum University Erlangen-Nuremberg, 91052 Erlangen, Germany
Chemical Industry is in essence not selling chemicals but properties that are associated with these chemicals, be it a drug, a paint, or some plastic material, The first fundamental task of a chemist is therefore to correlate the desired property with a chemical structure or a material. Observe that any material basically is a chemical or a mixture of chemicals.
Many chemical or material properties are too complex to be directly calculable. In such a situation learning from data can be the only solution. The representation of chemical compounds or materials is essential for this learning process. Various approaches will be illustrated.
Once the chemical which is thought to have the desired property is selected, it has to be made. The planning of the production of a compound and the control of the processes chosen to make it are further areas for the application of chemoinformatics. 
 Applied Chemoinformatics – Achievements and Future Opportunities, T. Engel, J. Gasteiger (eds.), Wiley-VCH, Weinheim, ISBN 978-3-527-34201-3, May 2018.
Dr. Yasuaki Ishikawa (Assoc. Professor, Materials Informatics, Science and Technology, NAIST)
Title：Potential of informatics in development of electronic devices
Informatics is fascinating many researchers since it helps a lot to discover a new material, reaction mechanism, and even predict or design material properties. Device researchers use a lot of materials to complete their process, and develop devices which have a new function and/or superior characteristics, but a potential of the materials is not fully utilized in the devices, suggesting us that a designing of devices to maximize the performances is still a big challenge. In the device development, there is so many information such as device process parameters, materials, and characteristics. Informatics is possible to link every information, leading to helping a realization of a device having maximized performances.
Dr. Tomoyuki Miyao (Assoc. Professor, Materials Informatics, DSC, NAIST)
Informatics approach to determination of progression saturation for a synthesis campaign
Various types of questions that medicinal chemists face in their daily life can be reduced to a simple one: which compounds should be made next? Prediction methodologies in-silico including (quantitative) structure-activity relationship methods and molecular dockings are employed for trying to answer this question. Focusing only on one to-be-synthesized (virtual) compound, however, is not sufficient, when thinking about the whole synthesis campaign, where the number of synthesized molecules sometimes reaches a few thousands. In such cases, a method or metric for evaluating a synthesis campaign as a whole is clearly needed. In this talk, I will present an informatics approach to trying to solve this issue based on the comparison of distribution between already synthesized compounds and virtual compounds, which are potential candidates to be synthesized. We introduced two scores, which characterize a synthesis campaign from different points of view: one is from all synthesized compounds and the other from only active compounds. Based on the scores we developed, one might not only understand the progression of a synthesis campaign but also prioritize one campaign over other campaigns.
Dr. Naoaki Ono (Assoc. Professor, Data Science, DSC, NAIST) 小野 直亮（奈良先端大 データ駆動型サイエンス創造センター・先端科学技術研究科）
Dr. Ryohei Eguchi （Postdoctoral Fellow, Data Science, DSC, NAIST）
Classification of biosynthesis pathways of alkaloids using graph convolutional neural networks.
Various chemical descriptors such like molecular fingerprints have been long discussed to represent biochemical features, in order to embed molecular structures into a numerical space and quantify their activities. However, it is still difficult to predict bioactivities from molecular structures since it depends on the choices of those chemical descriptors.
Recently, machine learning methods based on Graph Convolutional Neural Networks (GCNN) have been proposed that can automatically optimize a model for molecular feature extraction from the given training sets. In this study, we introduce an application of GCNN to predict metabolic pathways of alkaloids, namely, one of the largest families of secondary metabolites in plants. We trained and tested GCNN model on 578 alkaloid compounds and the mean accuracy of 20 runs with random sampling is about 94% (Number of epoch: 200). The results showed that it is greatly expected that it will lead to an understanding of the evolution of metabolic system unique to organisms.
Dr. Miho Hatanaka (Assoc. Professor, Materials Informatics, DSC, NAIST）
Material design by efficient use of databases
In recent years, various databases including the first principle calculation results have been constructed. However, these databases are not fully utilized because the most of desired material properties (except for the band gaps, HOMO-LUMO gaps, and so on) are difficult to be computed directly. To overcome this problem, we need to establish the strategy to discover new materials by combining the databases and the information from mechanistic studies. In this talk, I will introduce two case studies about the luminescent thermometer and regioselective catalytic system. In the case of luminescent thermometer, whose emission color changes from green, yellow to red as the temperature increases, the key to control the color change was the energy level of the triplet excited state of the linker ligands, which connect metal complexes. In the case of regioselective catalytic system, the key to control the selectivity was the steric hindrance around the metal center in the catalyst. Based on these mechanistic information, the appropriate molecules were explored from the databases by focusing on the descriptors relating to the triplet energy and steric hindrance.