应用人工神经网络预测氢化可的松的溶解度

Prediction of Solubilities of Hydrocortisone in Various Solvents via Artificial Neural Network

以量子化学方法在密度泛函B3LYP/6-31G(d)水平上计算得到含有电负性原子的溶剂水、醇类、酮类、酯类、氯代烷烃共17种溶剂的结构参数:最高占用轨道能(EHOMO)、分子最低空轨道能(ELUMO)、分子偶极矩(μ)、分子总能量(Etotal)、最正原子净电荷(q+)、最负原子净电荷(q-)。采用误差反向传播(BP)算法的三层人工神经网络,确定隐含层节点数为7。建立了EHOMO、ELUMO、μ、Etotal、q+、q-、摩尔体积(VM)、介电常数(ε)、温度(T)共9个参数与氢化可的松在不同温度下不同溶剂中溶解度之间的关系模型。运用此神经网络模型可预测不同分离条件下氢化可的松的溶解度,平均预测相对误差为7.0%。

The structure parameters EHOMO,ELUMO,q^＋,q^-,μ,Etotal of seventeen solvents such as water,alcohols,aldehydes,esters,fluoroalkanes were calculated at B3LYP/6-31G（d） level.By means of error back-propagation（BP） algorithm artificial neural network（ANN） and 7 hidden layer units,the relationship between each of EHOMO,ELUMO,q^＋,q^-,μ,Etotal,VM,ε,T and solubility of hydrocortisone in various solvents at different temperatures were established.The solubilities of hydrocortisone under various conditions were predicted by virture of ANN with an average relative error of 7.0%.