高等学校化学学报 ›› 1999, Vol. 20 ›› Issue (6): 861.

• 论文 • 上一篇    下一篇

循环子空间回归为喹喏酮N1位构效关系建模

孙卫国, 陈德钊, 陈亚秋, 叶宁, 胡上序   

  1. 浙江大学化学工程系, 杭州 310027
  • 收稿日期:1998-09-29 出版日期:1999-06-24 发布日期:1999-06-24
  • 通讯作者: 陈德钊
  • 作者简介:孙卫国,男,27岁,博士研究生.
  • 基金资助:

    浙江省自然科学基金(批准号:295027)资助课题

Modeling with Cyclic Subspace Regression for Quantitative Structureactivity Relationships of 1-Substituted 7-[3-[(ethylamino)methy]-1-pyrrolidinyl]

SUN Wei-Guo, CHEN De-Zhao, CHEN Ya-Qiu, YE Ning, HU Shang-Xu   

  1. Department of Chemistry Engineering, Zhejiang University, Hangzhou, 310027
  • Received:1998-09-29 Online:1999-06-24 Published:1999-06-24

摘要: 拟从线性泛函的角度分析循环子空间回归(CSR).CSR方法将从自变量参数矩阵和因变量向量中提取成分,循环地构造并扩张Krylov子空间,且以此作为源空间,运用最小二乘准则解得映射到因变量实空间的线性泛函.整个求解过程包容了最小二乘回归(LSR)、主成分回归(PCR)、偏最小二乘回归(PLS)以及其它中间的回归方法.然后由预报能力的强弱,从中确定最佳回归模型.本文应用CSR方法为喹喏酮N1位抗菌构效关系建模,效果良好.

关键词: 循环子空间回归, 构效关系, N1喹喏酮, 建模

Abstract: The cyclic subspace regression(CSR) was analyzed from the viewpoint of linear functiona LIN the paper. By extracting components from independent variable and dependent variable parameter matrixes, CSR circularly configured and expanded Krylov subspace, and ultimately a real function was obtained which was mapped into independent variable real space according to the least square criterion. The whole solving process subsumes LSR, PCR, PLS and other medial regression algorithms. The optimal modeLIs selected from which has the minimum predictive error and the maximum stability. By using modeling with CSR for quantitative structure-activity relationships of N1 quinolone antibacterial, preferable results were obtained.

Key words: Cyclic subspace regression, Quantitative structure-activity relationship, N1Quino-lone antibacterial, Modeling

中图分类号: 

TrendMD: