摘要
为提高抗HIV活性预测的精度,采用深度学习算法,提出一种基于深度信念网络的抗HIV活性预测方法。利用BP神经网络任意精度逼近非线性函数的优点,结合多个受限玻尔兹曼机(restricted Boltzmann machine,RBM)进行非监督逐层贪婪模式训练学习,建立深度信念网络算法模型(deep belief network,DBN)。将抗HIV活性的高温超导源数据(HTS raw data)、高温超导抑制剂(HTS%inhibition(20μM))、最大测定信号释放比率(mean max signal)等特征作为DBN模型的输入,抗HIV平均活性值作为该模型的输出,设计实验对模型进行训练及验证,实验结果表明,DBN模型对抗HIV活性的预测均方根误差小,预测精度高,平均预测精度为93.82%,适用于抗HIV活性评估。
To improve the accuracy of the anti-HIV activity, a prediction method using deep belief network was proposed. Using the advantage of BP neural network which could approach the nonlinear function with any accuracy, combining multiple restricted Boltzman machine for layer-by-layer training and learning based on unsupervised greed method, a DBN algorithm model for the accuracy of the anti-HIV activity prediction was established. The anti-HIV activity~ s HTS raw data, HTS~ inhibition （20 μM）, mean max signal, etc. were taken as the input of the DBN network, and output corresponded to the average value of the anti-HIV activity, and experiment was designed for training and verifying the model Experimental results show that the root mean square prediction error is small, the accuracy of the prediction is high, and the average prediction precision of the anti-HIV activity using the DBN model is 93.82%, which is applicable to evaluating the anti-HIV activity.
出处
《计算机工程与设计》
北大核心
2017年第1期226-230,237,共6页
Computer Engineering and Design
基金
国家自然科学基金项目(31160341)
新疆研究生科研创新基金项目(XJGRI2015034)
关键词
抗HIV活性
深度信念网络
非监督
精度
预测
anti-HIV activity
deep belief network
unsupervised
accuracy
prediction
