Abraham Michael H, Gil-Lostes Javier, Fatemi Mohammad
Department of Chemistry, University College London, 20 Gordon Street, London WC1H OAJ, UK.
Eur J Med Chem. 2009 Jun;44(6):2452-8. doi: 10.1016/j.ejmech.2009.01.009. Epub 2009 Jan 20.
A large database of milk/plasma ratios, M/P, for 179 drugs and hydrophobic environmental pollutants has been constructed from literature data. Application of linear analyses shows that drugs preferentially partition into the aqueous and the protein phases of milk, but that the pollutants partition into the fat phase. No useful linear equation could be obtained for the entire 179 compound data set, but an artificial neural network with only five Abraham descriptors as input resulted in errors in log(1+M/P) of only 0.0574, 0.116 and 0.093 log units for a training set of 135 compounds, an internal test set of 22 compounds and an external test set of 22 compounds respectively. These errors correspond to 0.203, 0.193 and 0.334 log units respectively when transformed into errors in log(M/P).
我们根据文献数据构建了一个包含179种药物及疏水性环境污染物的奶/血浆比率（M/P）的大型数据库。线性分析的应用表明，药物优先分配到奶的水相和蛋白相中，而污染物则分配到脂肪相中。对于整个179种化合物的数据集，无法得到有用的线性方程，但一个仅以五个亚伯拉罕描述符作为输入的人工神经网络，对于由135种化合物组成的训练集、22种化合物组成的内部测试集和22种化合物组成的外部测试集，其log(1+M/P)的误差分别仅为0.0574、0.116和0.093对数单位。当转换为log(M/P)的误差时，这些误差分别对应于0.203、0.193和0.334对数单位。
