서울대학교 생물정보학 협동과정 주최로 세미나를 아래와 같이 열고자 하오니 많은 참여 바랍니다.

일시:　2019년 11월 5일 화요일 오전 11시

연사: 남호정 교수님　(광주과학기술원)

장소:　서울대학교　300동(유회진 학술정보관) 611호

TITLE

AI-powered drug discovery and development

ABSTRACT

　The process of drug discovery is a challenging process considering its required time and efforts. Traditionally, developing one promising drug requires more than 10,000 initial compounds to be selected as candidates. Then these candidates undergo pre-clinical and clinical tests to be passed for several years. In recent years, the paradigm in drug development has been re-shaped in several ways. Many studies have shown the efficiency of artificial intelligence (AI) based drug development approaches that further increase the ability to predict and model the most relevant pharmacokinetic, metabolic, and toxicity endpoints, thereby accelerating the drug discovery process. This talk covers studies about the development of AI models for hit compound discovery and toxicity prediction in early drug discovery process. The first part of the talk is about the AI model that suggests hit compounds. We employ a convolutional neural network (CNN) on target protein raw sequences to capture local residue patterns participating in drug-target interactions. With the engineered features using CNN on protein sequences, our model shows improved prediction performance than previous protein descriptor-based models as well as than the previous deep learning model. By examining the pooled convolution results, we confirm that our model can detect binding regions of proteins for DTIs. Next, we developed an AI-based cardiotoxicity prediction model for screening human ether-a-go-go-related gene　(hERG) blockers. We developed a precise and interpretable hERG blocker prediction model by using deep learning with a self-attention approach. The validation result showed that the model was well-optimized and had high performance and the test set performance of the proposed model was higher than that of previous machine-learning-based models. Furthermore, we interpreted the calculated attention score vectors obtained from the proposed prediction model and demonstrated the critical structural patterns that are represented in hERG blockers.