摘要
为构建基于深度学习的微管蛋白秋水仙碱位点抑制剂(CBSIs)预测模型,进行CBSIs的活性预测和药物虚拟筛选,我们收集了1482个结构多样性的靶向微管蛋白秋水仙碱位点的抑制剂和非抑制剂,以分子指纹和分子图为特征表述,采用图卷积神经网络深度学习方法,建立分类预测模型。对所建立模型的预测结果进行比较,发现了一个最优预测模型(Model-Chemprop),它在测试集上的敏感度(SE)值为0.9109、特异性(SP)值为0.8125、总体准确度(Q)值为87.92%、AUC值为0.891。因此,基于深度学习建立的最优模型可以作为虚拟筛选工具,用于新型CBSIs的活性预测和发现,以及靶向富集库的构建。
In silico models for predicting the activity of tubulin colchicine binding site inhibitors(CBSIs)and non-CBIS based on deep learning were constructed.We firstly collected a total of 1482 CBSIs and non-CBSIs with multi-scaffolds from ChEMBL datebase.In silico classification models were established based upon molecular fingerprint and graph features using deep learning methods of graph convolution neural network.Compared the prediction results of our established models,we found the best prediction model(Model-Chemprop)from the test set achieved a sensitivity(SE)of 0.9109,a specificity(SP)of 0.8125,an overall prediction accuracy(Q)of 87.92%and AUC values of 0.891.Therefore,the established optimal deep learning-based models could be used as virtual screening tools for discovery and predicting of new CBSIs and the construction of CBSIs targeted enrichment libraries.
作者
邓燕红
蔡涵萱
张建华
黄汉辉
王领
DENG Yan-hong;CAI Han-xuan;ZHANG Jian-hua;HUANG Han-hui;WANG Ling(Department of Pharmacy,The Third Affiliated Hospital of Guangzhou Medical university,Guangzhou 510150,China;Joint International Research Laboratory of Synthetic Biology and Medicine,School of Biology and Biological Engineering,South China University of Technology,Guangzhou 510006,China)
出处
《化学研究与应用》
CAS
CSCD
北大核心
2020年第12期2192-2198,共7页
Chemical Research and Application
基金
广东省医学科学技术研究基金项目(B2018126)资助
广州市卫生和计划生育科技项目(20181A010058)资助
广州医科大学附属第三医院科研基金项目(2016Y07)资助。
