时间：2019年11月1日（周五）   13：45- 15：15
　　Time：November 1, 2019,   1:45-3:15 PM
　　地点：西湖大学云栖校区3号楼312会议室
　　Venue:Meeting Room 312, 3th Floor, Build 3, Yunqi Campus
　　主持人：生命科学学院 郑厚峰博士
　　Host：Dr. Hou-Feng Zheng,  PI of School of Life Sciences, Westlake University
　　主讲嘉宾/Speaker:
　　

　　Prof. Yi-Hsiang Hsu
　　Co-Director of the GeneticEpi Program, Marcus Institute for Aging Research,Beth Israel Deaconess Medical Center, Boston, MA.
　　

主讲人简介/Biography:
　　Prof. Yi-Hsiang Hsu is the Co-Director of the GeneticEpi Program, Marcus Institute for Aging Research,Beth Israel Deaconess Medical Center, Boston, MA.He is an Associate Professor at Harvard MedicalSchool and at Program for Quantitative Genomics,Program of Molecular and Integrative PhysiologicalSciences, Harvard School of Public Health, Boston, MA. He is also an Associate Faculty at the BROADInstitute of MIT and Harvard, Cambridge, MA.Prof. Hsu is a Statistical Geneticist/ComputationalBiologist and was trained at Harvard. He received hisdoctoral degree in statistical genetics/populationgenetics at Harvard University in 2005. He got his MDtraining in oncology.Prof. Hsu has led multiple large-scale NGS and GWAS projects on metabolic relevant disorders,cardiovascular disorders, musculoskeletal disorders and neuropsychiatric disorders. He haspublished 120 papers in subject of human genetics and is regarded as one of the leading expertsin field of using sophisticated statistical genetics and deep-learning machine-learning to conductadvanced statistical analysis.
　　

摘要/Abstract
　　With the recent advance in technologies, omics data (such as genome, transcriptome, epigenomics, proteome and metabolome) are accessible and can provide biological insight of disease pathophysiology and mechanisms of action (MOA) on drug targets. In this talk, I will demonstrate an approach that we recently developed to incorporate 3D genomic structure and chromatin states of gene regulatory landscapes in a deep learning framework to predict functions of disease-associated variants and their underlying targeted genes.  This significantly increased our understanding and evaluation of biological importance for those otherwise unknown genetics variants.  It allows us to prioritize high-impact variants and their targeted genes for developing new drug intervention.
　　In addition to common diseases, we also applied deep learning approaches to cancer immunotherapies. Neoantigens, generated by cancer-specific DNA alterations that are only expressed in cancer cells, represent an optimal target for the immune system and make possible a new class of personalized cancer vaccines. To identify neoantigens, we applied the CNN model to predict peptide-MHC binding affinity based on the interaction properties of pairwise amino acids. Our model has been shown to have better predictive accuracy than the current state-of-the-art methods when tested with benchmark data, and its performance maintains at a reasonable level even for alleles with little training data.
　　Research highlights:
　　Genetic determinants of common aging relevant disorders:Using population-based wholegenome sequencing, exome-sequencing and GWAS approaches to identify sequence variantsthat are associated with complex traits/disorders.
　　Precision medicine on immuno-oncology:To predict neoantigens from somatic mutations forpotential use in cancer vaccine therapies, the CNN deep learning machine learning approach isapplied to estimate the binding affinity between patients’ HLA type and mutant peptides. Themutated peptides are estimated from somatic mutations. The same approaches are also appliedto predict the immunogenicity of the MHC-peptide complex to T-cells.
　　Human primary cell-specific gene regulatory circuits:Using chromatin confirmation capture HIC seq (3D genomics structure), ATAC-seq, Chip-seq and RNA-seq as well as functional genomicsapproaches to build human primary cell-specific genomic regulatory circuits and to identify noncoding causal variants from NGS and GWAS loci.
　　Statistical methodology development:Develop multi-phenotype analytical approaches(PheWAS approaches) using aggregative results from GWAS or from other large-scale highdimensional association analyses.
　　AI machine learning approaches:To understand MOA, using AI machine learning deep learningapproaches to integrate genomics, transcriptomics, proteomics and metabolomics resources forbetter understanding relationships between the biological components that work together todrive a complex pathophysiological process of common diseases across signal pathways,organisms and even species on a global scale.
　　

联系人/Contact：
　　生命科学学院
　　于文越 yuwenyue@westlake.edu.cn