Abstract: In order to reduce the rate of the failure of the new drugs’ clinical experiment and enhance drugs’ ADME/T properties，it’s necessary to consider the combinative activity of drugs and HSA（Human Serum Albumin） and increase the prediction accuracy of drugs and HSA，so that proper components can be chosen to make new drugs.Genetic algorithm was used to choose molecular descriptors and Support Vector Machine classification method was employed to predict the combinative activity between drugs and HSA.On the basis of research in SVM，the method of searching optimized SVM parameter by PSO（Particle Swarm Optimization） algorithm was presented and the ends were compared with Bilinear Grid Search Method.It is shown that this model has a good prediction ability and the accuracy is 86%.

Key words: Support Vector Machine（SVM）, Human Serum Albumin（HSA）, Genetic Algorithm（GA）, Particle Swarm Optimization（PSO）, Bilinear Grid Search Method（BGSM）

摘要： 为降低新药在临床实验的失败率，提高新药的ADME特性，在制药初期需考虑预测药物与HSA（人血清白蛋白）的结合能力，以提高预测正确率为目标并因此选取合适的制药成分。首先采用遗传算法对分子描述符进行筛选，然后利用支持向量机（SVM）模型预测药物与HSA的结合能力。针对支持向量机的分类精度和泛化能力取决于核函数参数选取的特点，提出了基于粒子群优化算法（PSO）的SVM核参数优化选择法。通过PSO自动获取SVM最佳核参数，并将结果同双线性网格搜索法比较，结果表明，建立的模型对药物与HSA之间的结合能力有较高的预测性，正确率达到86%。

关键词: 支持向量机（SVM）, 人血清白蛋白（HSA）, 遗传算法（GA）, 粒子群优化算法（PSO）, 双线性网格搜索法（BGSM）