雄激素受体干扰物的活性预测模型的研究

In silico prediction of binding affinity for the androgen receptor ligands

目的在于定量预测雄激素受体干扰物活性,并确定最佳建模方法。选择150个分子作为数据集,随机选38个分子作为检验集,其它分子为训练集。每个化合物分子计算了193个分子参数。通过采用多元线性回归和主成分回归等方法,建立数学模型,并用验证集检验了所建模型的预测能力。结果发现逐步筛选法和主成分分析方法所建模型都表现出较强的预测能力(应用于检验集的相关系数分别为R=0.61,R=0.52)。以上研究将有助于新药雄激素受体抑制剂的筛选和开发。

To quantitatively predict the activity of androgen receptor ligands, a dataset composed of 150 diversified compounds was built based on multivariate linear regression analysis and principal component analysis （PCA）. 38 compounds served as test set and the rest as training set. For each molecule, 193 molecular indices were calculated, From the analysis of the models, the predictability of stepwise method（R = 0.61 ） for molecular descriptors and PCA （R = 0. 52） are satisfactory. All these models are helpful for aiding screening and development of androgenic compounds.