Prediction Of Toxicity Of Compound HERG Based On Deep Learning And Molecular Fingerprinting

With the development of computer technology,computer aided drug design has been widely used in people’s daily life.The application of computer technology can improve the success rate of experiments,which helps us save a lot of time,money and manpower,and will provide a broader space for new drug research and development.At present,the use of computer aided drug design has become the focus of human health research.Therefore,based on the structure of compounds,a new compound cardiotoxicity prediction model is established by using the deep learning algorithm.Human ether-a-go-go-related gene(hERG)channel blockade by small molecules is a big concern during drug development in the pharmaceutical industry.Blockade of hERG channels may cause prolonged QT intervals,which can lead to severe cardiotoxicity and is a major cause of many drug development failures.Thus,evaluating the hERG-blocking activity of small molecule compounds is important for successful drug development.Currently,many computational methods have been used to screen small molecule compounds with potential hERG-related toxicity.However,most methods usually consider one single feature type during model training,which may limit the predictive ability.Based on this basis,this thesis proposes a deep learning model named as DMFGAM implemented with molecular fingerprint and graph attention mechanism to predict the hERG channel blocking activity of small molecules.Firstly,DMFGAM integrates a variety of molecular fingerprint features and the graph features extracted from the multi-head GAT model,then the fused features are sent into the fully connected neural network to obtain the final classification results.The result in 5-fold cross-validation illustrates that DMFGAM achieves AUC of 0.894 and ACC of 0.817 on the dataset independently constructed by our own,which is better than those of existing four state-of-the-art computational methods.Meanwhile,DMFGAM also gain comparable results on three external validation sets,which demonstrate the reliability of our method.We believe DMFGAM can serve as an effective tool to predict hERG channel blockers in the early stages of drug discovery and development.