Prediction Of LRRK2 Inhibitor Activity Based On Regression Method

Parkinson’s disease is a chronic progressive neurodegenerative disease,which has been considered as the second most prevalent neurodegenerative disease.Researchers have found that genetic factors play a more and more important role in Parkinson’s disease.Now it has been found that there are 7 disease-causing genes specifically related to Parkinson’s disease,and LRRK2 gene mutation is most likely to play a role in typical late-stage Parkinson’s disease.LRRK2（Leucine rich repeat kinase 2）is a protein kinase encoded by PARK8 gene.The most common LRRK2 mutation is at G2019S（GS）site.Researchers have designed many LRRK2 inhibitors to treat Parkinson’s disease based on this site.In this paper,based on the activity of these inhibitors（pIC50）and their two-dimensional or three-dimensional structural characteristics,quantitative structure-activity relationship（QSAR）analysis was conducted by MOE software to establish training set,verification set and test set,and the predictive ability of these three subsets was tested by partial least square method.In this paper,QSAR analysis modeling is carried out from three aspects,hoping to obtain a set of specific descriptors to predict new small molecules.（1）Using 2D descriptor to conduct QSAR modeling for this small molecule.Finally,the following conclusions are drawn:first of all,weight analysis is adopted for the selection of descriptors,that is,the first two descriptors of weight are selected.By eliminating outliers and removing useless descriptors,the training set R2 is close to 0.9±0.02,but the predictive ability of the verification set and test set is not satisfactory,because the training set overfits and does not add 3D descriptors.Lead to the following.（2）After adding 3D descriptor,R2 of the training set is no different from the above.The predictive ability of the verification set and test set is still not good,and the absolute value of the difference between the experimental value pIC50 and the predicted value pIC50 is greater than 1.Then we introduced a new scheme:the method of selecting descriptors first and then subdividing,that is,selecting appropriate descriptors first,scoring the whole compound,and then dividing three subsets.（3）The final results are obtained through the above methods:the molecular descriptor USES 10 2D descriptor,which are a_don,a_nO,a_ICM,GCUT_SMR₃,KierAl,lip_don,Slogp_VSA5,SMR,SMR_VSA4,SMR_VSA5,and the training set is R2=0.9118.Both the verification set and the test set have reached the standard,and the absolute value of the difference between the experimental value and the predicted value in the test set is less than 1.Finally,the formula of prediction model of LRRK2 inhibitor is:pIC50=-2.06042+0.87368×a_don+0.42089×a_nO+2.35157×a_ICM-0.13266×GCUT_SMR₃-0.19108×KierAl-0.85941×lip_don-0.01550×Slogp_VSA5+0.81823×SMR+0.02817×SMR_VSA4-0.00389×SMR_VSA5...