Research On Virtual Screening Methods For Drugs Targeting Cancer-associated G Protein Coupled Receptors

Cancer is one of the most difficult diseases in medicine,which affects human health and seriously threatens human life.Nowadays,drug discovery for treating cancers become more and more important.As far as we all known,there are more than 300 G protein-coupled receptors(GPCRs)have been proven to closely related with cancer.Drug design targeting these GPCRs is essential for treating cancer,and accurate prediction of biological activities of ligands is critical for the screening and optimization of hit compounds for cancer treatment.In this thesis,a novel virtual screening method called ELVS for drug leads targeting cancer-related G protein-coupled receptors is proposed based on ensemble learning with the base classifier of inductive matrix completion.Each base classifier was implemented on the basis of traditional matrix completion where the eigenvector information of GPCRs and ligands was integrated into to the matrix in order to obtain better prediction performance.The pipeline of our method is as follows:(1)To generate multiple data sets and the relationship matrix between GPCR and ligands by replacement sampling;(2)To obtain the feature matrix of GPCRs by the word2vec program and obtain the ECFPs features with 512 dimensions for compounds.(3)Load the GPCR-Ligands relationship matrix,GPCRs feature matrix and ligand matrix information into the base predictors for training the optimal models;(4)To obtain the final prediction models by ensemble learning with these base predictors.The experimental results on 15 representative cancer-related GPCRs datasets.We choose two common evaluation indicators the correlation coefficient(R~2)and the root mean square error(RMSE)to evaluate the results of regression prediction.The experimental results show that our method has obtained the best performance when comparing with machine learning method and other matrix completion methods.In addition,our method is generally superior to the deep learning prediction method,we obtained the average increase of 15.4%on R~2 and the average decrease of 11.3% on RMSE.