| 植田 秀樹 | M, 2回目発表 | 計算システムズ生物学 | 金谷 重彦, 峠 隆之(BS), 小野 直亮, 黄 銘 |
|
title: Precise prediction of cleavage site of Alzheimer's disease relevant γ-secretase from the amino acid sequence by machine learnings
abstract: Alzheimer's disease-associated amyloid β-peptide (Aβ) is generated by γ-secretase cleavage. γ-Secretase cleavage mechanistically occurs at multiple sites in a successive fashion. Although the mechanisms of γ-secretase to produce Aβ species is intensively studied, the specificity of γ-secretase cleavage is still elusive. In the present study, we have made a cleavage prediction model for γ-secretase to interpret the cleavage mechanism based on physicochemical and biochemical factors. We accumulated 24 substrates proteins by scientific references. Mathematical models for estimating amounts for 35 Aβ precursor protein fragments generated in the process of the successive cleavage by γ-secretase were created by regression models based on the information about peptide sequences in the vicinity of cleavage sites. In concrete, we exhaustively examined 5050 regression models using 101 machine learning methods, 10 types of peptide properties, and 5 window size reflecting the vicinity of cleavage sites. Here the ten types of peptide properties are constructed by the first 10 principal components (PC1-PC10) for 553 amino acid indices. Strikingly, we obtained five models with the highest recognition rate (87.5%) for prediction start and end cleavage sites of the 24 substrate proteins. Factor loading structure of PC9 supports that amino acid properties associated with secondary structure contribute to recognize cleavage site by γ-secretase. language of the presentation: Japanese | |||
| 川崎 聡大 | M, 2回目発表 | 計算システムズ生物学 | 金谷 重彦, 峠 隆之(BS), 黄 銘 |
|
title: Unsupervised Learning for Arrhythmia Detection Using Electrocardiogram
abstract: Holter electrocardiogram (Holter ECG), an instrument for long-term ECG measurement over 24 hours, is useful to discover transient arrhythmias which occurs infrequently and would been not seen in short-term measurement. Along with the development of machine learning, its application to automatic arrhythmia detection has been expected to reduce doctors’ workload on diagnosis for Holter ECG. Although some studies using supervised learning methods such as CNN and RNN have reported that these models could reach high accuracy performance, insufficient labeled data is the bottleneck on identification the signal of Holter ECG data. In this study, we conducted experiments to examine the performance of IMSAT which is an unsupervised learning model, in ECG classification. In the results, we have found that IMSAT model could classify ECG data into normal beat, ventricular and atrial arrhythmia. A more accurately model can be expected by task-specific adjustments for the important components of IMSAT model. language of the presentation: Japanese | |||
| 弘世 和久 | M, 2回目発表 | 計算システムズ生物学 | 金谷 重彦, 峠 隆之(BS), モハマドアルタフルアミン, 黄 銘 |
|
title: Identification of molecular pathway for Schizophrenia and Bipolar Disorder based on biomarkermetabolites
abstract: Schizophreniais disorder that prevents communication, while bipolar disorder is disorderthat frecently changes emotion, mania state and depression. These are severebrain diorders and have resistance to medicine and some similar symptoms. If disordersbecome chronic, it contributes to suicide and autonomic ataxia. Here, we took biomarkermetabolites, proteins, associated with health state deeply. In additon,weconstructed Protein-Protein Interactions network(PPI,PPIN) and intergrated eachPPI on disorders . In order to analyze topological properties and community ofPPI, we used cytoscape, visualize and analysis tool, and DPclusO, graphclustering. As result, we determined density 0.4 clustering from 0.1 to 0.9 andwe found that density 0.4 clustering had 17 significant clusters and 136proteins. So network analysis provides the infomation, showing associationbetween clusters and metabolites. This research will contribute to inventingnew drug, early detection. language of the presentation: Japanese | |||
| HU JIEYING | M, 2回目発表 | 計算システムズ生物学 | 金谷 重彦, 峠 隆之(BS), 小野 直亮, 黄 銘 |
|
title: *** An Integrated hERG Dataset Assemble Molecular Graph Convolution Neural Network Model for The Prediction of Cardiotoxicity ***
abstract: *** Toxicity is a central issue in the development of new drugs, there are numbers of drugs failing in clinical trials or even need to be taken off of the market because of the toxic effects. Cardiotoxicity has become a leading cause for drug failure. In recent years, many in silico hERG models have shown their positive efforts on predicting hERG channel toxicity at the early steps of the drug discovery process. CNN can be conducted on the chemical structural graphs directly and it has demonstrated its good performance on images recognition and classification in the last decade. In this research, we proposed a molecular graph convolutional neural network to develop QSPR model for predicting hERG channel adverse cardiac effects for drug virtual screening and drug design at the early steps of the drug discovery process. *** language of the presentation: *** English *** | |||