コロキアムB発表

日時: 9月12日(月)4限(15:10-16:40)


会場: L2

司会: 嶋利 一真
中川 翔太 M, 2回目発表 計算システムズ生物学 金谷 重彦, 松本 健一, 小野 直亮, MD.ALTAF-UL-AMIN, 黄 銘
title: Creation of a Generalized Model of Health Information from Chest X-Ray Images by Embedding Using Deep Learning.
abstract: Some studies have shown that diagnostic medical devices have a high affinity with deep learning and can classify pathologies with higher accuracy than radiologists. However, these models are specialized for one particular pathology classification and cannot be used for other pathologies, so they are not generalized models of health information. The purpose of this study is to create a model that can evaluate the classification and severity of two diseases as a preliminary step toward creating a generalized model of health information. Assuming that the generalized information is omitted because the classification model learns explicit pathology classification using pathology labels, we attempted to obtain this information by extracting latent features of pathology images that do not depend on the labels. In this study, we chose Vector-Quantized Variational Auto-Encoder (VQ-VAE) as an unsupervised learning model, acquired discrete feature vectors, and learned an embedded space with pathology information by supervised learning using these latent variables as input We will verify whether it is possible to evaluate the classification and degree of pathology based on the positional relationship in the feature space by learning an embedding space with pathology information using these latent variables as input.
language of the presentation:Japanese
 
村田 友真 M, 2回目発表 計算システムズ生物学 金谷 重彦, 松本 健一, 小野 直亮, MD.ALTAF-UL-AMIN, 黄 銘

title: Building a Regression Model of Ligand-Protein Interaction Using Deep Learning 

abstract: The development of drugs that can inhibit the spread of SARS-CoV-2, the causative agent of COVID-19 (a new type of coronavirus infection), is urgently needed. The most promising approach to inhibit SARS-CoV-2 is to target a protein that is required in the process from infection to proliferation and inhibit its function by administering a compound that binds to the protein. There are several candidate target proteins. In this research, we consider screening drug candidate compounds from 50,000 natural compounds in the natural products database KNApSAcK by targeting main protease (Mpro), which is required for the synthesis of essential proteins of SARS-CoV-2. In this presentation, we report on a deep learning model of molecular graph convolutional neural network to predict the binding strength with Mpro, which was constructed from existing experimental data, and describe our future plan.

language of the presentation: Japanese

研究題目: 深層学習を用いたリガンド-プロテイン相互作用の回帰モデルの構築

研究概要: 現在、COVID-19(新型コロナウイルス感染症)の原因ウイルスであるSARS-CoV-2の感染拡大を抑制できる薬の開発が一刻も早く望まれている。SARS-CoV-2の抑制のためには、感染から増殖に至るプロセスの中で必要とされているタンパク質をターゲットとして、それに結合する化合物を投与することでその機能を阻害することが有力な手法とされている。ターゲットとなるタンパク質の候補はいくつかあるが、本研究ではSARS-CoV-2の必須タンパク質の合成に関与するメインプロテアーゼ(Mpro)をターゲットとして、天然物データベースKNApSAcKの5万件の天然化合物から薬剤候補化合物をスクリーニングすることを考える。本発表では、既存の実験データから構築したMproとの結合強度を予測する分子グラフ畳み込みニューラルネットワークの深層学習モデルについて報告し、今後の方針を述べる。

 
里川 航亮 M, 2回目発表 計算システムズ生物学 金谷 重彦, 松本 健一, 小野 直亮, MD.ALTAF-UL-AMIN, 黄 銘
title: Segmentation from 3D PET images using deep learning
abstract:In recent society, many people have diabetes. Also, many people have kidney failure caused by high blood pressure and high blood sugar levels from diabetes. These diseases are diagnosed by using blood and urine tests, but PET image may be used to examine blood flow in the kidney. In this time, physicians manually segment the kidney parts from many images, so it will be a time consuming and burdensome task. In order to solve this problem, I propose a model which can distinguish kidney from 3D PET images, and I try to contribute to society by automating diagnose. In my research, the convolutional model was used for segmentation of the kidneys and abdominal aorta. Also, we expect to be applied to segmentation of other organs and estimation of the amount of blood flowing through the kidney.
language of the presentation: Japanese
発表題目: 深層学習を用いた3次元PET画像のセグメンテーション
発表概要:現代の社会では、多くの人が糖尿病を患っている。また、糖尿病による高血圧や高血糖状態が原因で、腎不全を患っている患者も多い。これらの疾患の診断には血液検査や尿検査を用いることが多いが、PET 画像を用いて腎臓を流れる血液量を調べる場合がある。その際医師は、何枚もの画像から腎臓部分を手動でセグメンテーションをするため、医師たちにとって時間のかかる負担の大きい作業となっている。この問題を解決するために、3次元PET画像から腎臓部分を識別するモデルを提案し、診断の自動化を試みることで社会に貢献することを考える。本研究では、畳み込みモデルを用いて腎臓および腹部大動脈のセグメンテーションを行なった。このモデルが実現すると、他の臓器のセグメンテーションや腎臓を流れる血液量の推定などに応用されることが期待される。
 
AKTER MOST ATIKA M, 2回目発表 計算システムズ生物学 金谷 重彦, 松本 健一, 小野 直亮, MD.ALTAF-UL-AMIN, 黄 銘
title:Drug Repurposing for Inflammatory Bowel Disease(IBD)based on bipartite relations between drugs and IBD related genes
astract: Drug repurpusing, which treats new/other diseases using existing drugs, has become a much-admired tactic. It can also bereferred to as the re-investigation of the existing drugs that failed to indicate the usefulness for the new diseases. In this work, we mainly focus on finding inflammatory bowel disease (IBD) associated drugs by bi-clustering the drug-target interactions aided by known IBD risk genes. First a comprehensive bipartite network is constructed involving the drugs and their target proteins based collected data from Drug-Bank database. A bi-clustering algorithm BiClusO is then applied to the bipartite network for finding high density clusters. The presence of IBD risks genes in the clusters are examined and statistically significant clusters are determined which are then utilized for IBD drug repurposing. We have identified some potential IBD drugs based on network analysis. The IBD drugs we have identified are given below: Cisplatin, Etanercept, Oxaliplatin, VX-702, Carboplatin, Adalimumab, AV411, CRx-139, SCIO-469 and Chloroquine. Here we present an approach to identify IBD disease drugs. The approach can be generalized to find drugs for IBD diseases. Our proposed method will be helpful to understand the mechanisms of the way the drugs work.
language of the presentation: English