数量结构-活动关系
主成分分析
偏最小二乘回归
分子描述符
结合亲和力
人工神经网络
化学
人工智能
计算生物学
计算机科学
机器学习
立体化学
生物
生物化学
受体
作者
Suzanne Sirois,Denis Giguère,René Roy
出处
期刊:Medicinal Chemistry
日期:2006-09-01
卷期号:2 (5): 481-489
被引量:20
标识
DOI:10.2174/157340606778250252
摘要
This study presents the first QSAR model for Galectin-3 glycomimetic inhibitors based on docked structures to the carbohydrate recognition domain (CRD). Quantitative numerical methods such as PLS (Partial Least Squares) and ANN (Artificial Neural Networks) have been used and compared on QSAR models to establish correlations between molecular properties and binding affinity values (Kd). Training and validation of QSAR predictive models was performed on a master dataset consisting of 136 compounds. The molecular structures and binding affinities (Kd) (136 compounds) were obtained from the literature. To address the issue of dimensionality reduction, molecular descriptors were selected with PLS contingency approach, ANN, PCA (Principal Component Analysis) and GA (Genetic Algorithms) for the best predictive Galectin-3 binding affinity (Kd). Final sets comprising 56, 31 and 35 descriptors were obtained with PLS, PCA and ANN, respectively. The objective of this prototype QSAR model is to serve as a first guideline for the design of novel and potent Gal-3 selective inhibitors with emphasis on modification at both C-3 and at O-3 positions [1]. Keywords: Galectin-3, 3D-QSAR, glycomimetics, Neural-Network
科研通智能强力驱动
Strongly Powered by AbleSci AI