English
Top

Development of deep learning models for genomic data: novel applications for short-read structural variant detection and long-read basecalling

学友会セミナー

学友会セミナー:2020年01月08日

開催日時: 2020年01月08日 19:00 ~ 20:00
開催場所: 2号館 大講義室
講師: 張 耀中
所属: 東京大学医科学研究所
ヒトゲノム解析センターDNA情報解析分野・助教
演題: Development of deep learning models for genomic data: novel applications for short-read structural variant detection and long-read basecalling
概要:

Recently, with the rapid progress of artificial intelligence (AI)/deep learning (DL) technologies, the exploration of AI for high-performance medicine is on the way. Applications on specific personal medical data are emerging, which include medical images and electronic health records (EHRs). Focusing on integrating genomic data for better health and medical care, we develop DL models for genomic data to achieve more accurate and higher resolution performance. In this presentation, we introduce our two recent achievements of developing deep segmentation methods for genomic data, which covers challenging tasks in short-read and long-read sequencing. For the short-read, we established novel DL model in the framework of U-net to detect nucleotide-wise breakpoints of structural variants (SVs) inside of bins of pre-determined genomic regions, which cannot be resolved in the previous methods. For the long-read, we proposed new UR-net model to perform nanopore basecalling from a perspective of instance segmentation. Our experiments show that the proposed DL methods can be effective in handling noisy signals in whole genome sequencing data derived from cutting edge sequencing technologies, which helps to exceed previous limitations and achieve better performance.

世話人: 〇宮野 悟 (DNA情報解析分野)
 井元 清哉 (健康医療データサイエンス分野)