Online Regression Algorithm with Nonlinear Transformation and Its Application to Prediction of Protein-protein Interaction Strengths

• Item type: SIG Technical Reports(1)
• 公開日: 2017-07-10
• タイトル: Online Regression Algorithm with Nonlinear Transformation and Its Application to Prediction of Protein-protein Interaction Strengths
• タイトル / 言語: en
• タイトル: Online Regression Algorithm with Nonlinear Transformation and Its Application to Prediction of Protein-protein Interaction Strengths
• 言語: eng
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: National Institute of Technology, Matsue College
• 著者所属: Presently with Kyoto University
• 著者所属: Presently with Riken
• 著者所属(英): en
• 著者所属(英): National Institute of Technology, Matsue College
• 著者所属(英): en
• 著者所属(英): Presently with Kyoto University
• 著者所属(英): en
• 著者所属(英): Presently with Riken
• 著者名: Morihiro, Hayashida; Mayumi, Kamada; Hitoshi, Koyano
• 著者名(英): Morihiro, Hayashida; Mayumi, Kamada; Hitoshi, Koyano
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: In a living molecular cell, protein-protein interactions have various important roles. In particular, we focus on interaction strengths that provide useful knowledge to understand complicated cellular networks, and several prediction methods have been developed. In our previous work, new feature space mappings based on protein domains were proposed, and support vector regression and relevance vector machine were employed. The combination of the mapping and the supervised regression method outperformed the existing methods. In this work, online learning algorithms, the regression passive-aggressive (PA) and adaptive regularization of weights for regression with covariance reset (ARCOR) algorithms, are examined. Furthermore, nonlinear transformation to a linear regression formula is introduced, and ensemble learning is examined. For evaluation, we performed three-fold cross-validation computational experiments, and took the root mean square error (RMSE). The RMSE by our proposed method was smaller than those by the existing methods. The result implies that our method combining online regression algorithms with nonlinear transformation and sequences of domain regions is useful.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: In a living molecular cell, protein-protein interactions have various important roles. In particular, we focus on interaction strengths that provide useful knowledge to understand complicated cellular networks, and several prediction methods have been developed. In our previous work, new feature space mappings based on protein domains were proposed, and support vector regression and relevance vector machine were employed. The combination of the mapping and the supervised regression method outperformed the existing methods. In this work, online learning algorithms, the regression passive-aggressive (PA) and adaptive regularization of weights for regression with covariance reset (ARCOR) algorithms, are examined. Furthermore, nonlinear transformation to a linear regression formula is introduced, and ensemble learning is examined. For evaluation, we performed three-fold cross-validation computational experiments, and took the root mean square error (RMSE). The RMSE by our proposed method was smaller than those by the existing methods. The result implies that our method combining online regression algorithms with nonlinear transformation and sequences of domain regions is useful.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AN10505667
• 書誌情報: 研究報告数理モデル化と問題解決（MPS）; 巻 2017-MPS-114, 号 7, p. 1-4, 発行日 2017-07-10
• ISSN / 収録物識別子タイプ: ISSN
• ISSN / 収録物識別子: 2188-8833
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会