Research On Conformation Consistency Of Molecular Docking Based On Machine Learning And Prediction Of Interaction Mode Of MHTT/Molecular Glue/LC3 Ternary Complex

Virtual screening plays an important role in drug development.It uses computational methods to test the top-ranked compounds without the need to test all the compounds in the compound library,which significantly accelerates the efficiency of lead compound discovery.However,with the gradual expansion of the explorable chemical space(currently about 1063),even if the top 0.1%to 2.5%of small molecules can be obtained through the virtual screening method for further experimental verification,the size of the retained compound database is still huge.Therefore,optimizing the virtual screening evaluation method has important scientific significance and application value.The consensus docking conformation of a compound refers to virtual screening through the use of different molecular docking software,and the RMSD calculation of the binding conformations of the top-ranked compounds.If the RMSD between the compound’s binding conformations from the output of different molecular docking programs is less than a certain threshold(such as 2 A),it is believed that the compound has a consensus docking conformation.Whether it has the consensus conformation can be used as a new evaluation method for virtual screening,which helps to reduce the size of the data set and at the same time increase the enrichment rate of the active compound.However,this process requires a lot of computer resources.Therefore,this paper proposes to apply the machine learning method to the research of compound consensus docking conforamation,in order to quickly and effectively perform virtual screening.The first part of this article uses machine learning methods to predict whether a compound has a consistent docking conformation.First,evaluate the ability of multiple programs to restore the experimental structure and the ability of the conformational restoration experiment with molecular docking consistency.Then,through a variety of choices,select the conformation of the results of the Vina and rDock docking procedures to obtain the molecular docking conformation conformance data set.Then construct a prediction model to predict whether the compound has a consensus docking conformation,and evaluate the accuracy,sensitivity and AUC of the prediction model on the external test set.Finally,the results showed that the FNN model has the best predictive ability,with the highest AUC value of 0.806,and the accuracy rate of 0.816.The model is selective to the target system,and performs relatively well in the kinase system,and the generalization ability of the model can be further explored.In the second part of this article,the interaction mode of the mHTT/small molecule/LC3 ternary complex is predicted by using the methods of molecular docking,homology modeling and molecular dynamics simulation.The accumulation of mutant huntingtin protein(mHTT)can cause Huntington’s disease,so the elimination of mHTT is critical for the treatment of Huntington’s disease.LC3 is an autophagy-related protein,which can degrade mHTT during autophagy.The latest research shows that small molecule "glue"can "bond" LC3 and mHTT,accelerating the removal of mHTT.In this paper,computer simulations found that the small molecule glue binds to the Phell and LC3 HP2 pockets of the mHTT N17 domain to "bond" the two proteins,which provides a reasonable three-dimensional method for the screening and optimization of novel and potent small molecule adhesives.