Hyun Kil Shin

Korea Institute of Toxicology, South Korea

Title: Beyond structural diversity of organic molecules: electron configuration fingerprint for inorganic bulk materials and engineered nanomaterials

---

Biography

Dr. Hyun Kil Shin is an expert in cheminformatics particularly in development of machine learning (ML) or deep learning (DL) model based on molecular structure datasets. His strong support on safe-by-design concept led him to participate in diverse research projects such as drug-induced liver toxicity prediction model development, biocidal active substance neurotoxicity prediction model development, and development of AI model designing safe compounds. He is currently a researcher in Korea Institute of Toxicology (KIT). As image data is one of most abundant data set, he also works with image data in research projects such as smartphone deployable animal skin disease diagnosis model development and atopy dermatitis region detection model development.

Abstract

Artificial intelligence (AI) models have been broadly applied in drug discovery; however, applicability domain (AD) of AI models is mainly focused on organic molecules so far since 1) majority of available database is composed of organic molecules, and 2) cheminformatics tools mainly handle organic molecules alone. Particularly, lack of appropriate cheminformatics tools for inorganic molecules becomes a significant technical obstacle that should be overcome in order to expand AD of AI models over wider range of chemical space beyond structural diversity of organic molecules. In order to provide more cheminformatics tools for inorganic compounds, electron configuration fingerprint (EC FP) was developed as a first fingerprint designed for inorganic compounds. Furthermore, size-dependent EC FP (SDEC FP) is designed by considering particle size in EC FP calculation to develop fingerprint for engineered nanomaterials (ENMs) whose structural diversity is much complicated than bulk inorganic materials due to compositional complexity in core, doping, and coating part of ENMs in different sizes. By applying EC FP, artificial neural network (ANN) models for prediction of physicochemical properties of inorganic compounds were developed based on composition of inorganic compounds alone. The models were developed with dataset containing almost all atoms in the periodic table to make reliable prediction on inorganic compounds with diverse atomic compositions. ANN models with EC FP outperformed other possible descriptors calculated from composition of inorganic compounds. SDEC FP was applied to develop prediction models for cytotoxicity and zeta potential of ENMs in diverse composition and shapes. Given that previous studies developed models applicable for one specific type of ENMs such as metal oxide, metal, coated, or carbon-based ENMs, the models developed with SDEC FP achieved breakthrough by developing general models applicable to any composition and shape of ENMs.