深度学习在药物活性预测研究中的应用

A Review of Deep Learning Application on Drug Activity Prediction

药物从研发到临床应用需要耗费较长的时间,研发期间的投入成本可高达十几亿元。而随着医药研发与人工智能的结合以及生物信息学的飞速发展,药物活性相关数据急剧增加,传统的实验手段进行药物活性预测已经难以满足药物研发的需求。借助算法来辅助药物研发,解决药物研发中的各种问题能够大大推动药物研发进程。传统机器学习方法尤其是随机森林、支持向量机和人工神经网络在药物活性方面能够达到较高的预测精度。深度学习由于具有多层神经网络,模型可以接收高维的输入变量且不需要人工限定数据输入特征,可以拟合较为复杂的函数模型,应用于药物研发可以进一步提高各个环节的效率。在药物活性预测中应用较为广泛的深度学习模型主要是深度神经网络(deep neural networks,DNN)、循环神经网络(recurrent neural networks,RNN)和自编码器(auto encoder,AE),而生成对抗网络(generative adversarial networks,GAN)由于其生成数据的能力常常被用来和其他模型结合进行数据增强。近年来深度学习在药物分子活性预测方面的研究和应用综述表明,深度学习模型的准确度和效率均高于传统实验方法和传统机器学习方法。因此,深度学习模型有望成为药物研发领域未来十年最重要的辅助计算模型。

It takes a long time for a drug to go from research and development to clinical application,and the investment cost during the period can reach one billion yuan.The combination of medicine and artificial and the development of big data of biochemistry contribute to sharply increasing drug activity data,and traditional experimental methods for drug activity prediction and discovery are hard to meet the needs of drug research and development.Algorithms are used to assist drug development and solve various problems during the process to significantly accelerate drug development.Traditional machine learning methods,especially random forests,support vector machines,and artificial neural networks,can improve drug activity prediction accuracy.Due to the multi-layer neural networks of deep learning,the model can process high-dimensional input variables and there is no need to limit the amount of input data characteristics manually.Deep learning can build a more complex function,and its application in drug research and development can further improve the efficiency of each step of drug research.Widely used deep learning models in drug activity are mainly DNN(deep neural networks),RNN(recurrent neural networks),and AE(auto encoder).GAN(generative adversarial networks)is often used in combination with other models for data enhancement due to its ability to generate data.Researches and applications of deep learning in drug molecule activity prediction in recent years showed that the accuracy and efficiency of deep learning models were higher than traditional experimental methods and traditional machine learning methods.Therefore,deep learning is expected to become the most critical auxiliary calculation model in drug research and development in the next decade.