Research On Protein Classification Based On Improved Hybrid Multi-label Classifier

Protein is not only the material basis of life,but also the main undertaker of life activities.In recent years,with the increasing number of protein sequences in the database,the conventional biophysical technology is extremely cumbersome,expensive and error-prone.It is urgent that the technology of protein function or structure classification is studied based on bioinformatics.A new predictor,called MF-EFP,has been developed that can be used to deal with the systems containing both mono-functional and multi-functional enzymes by introducing “improved Hybrid Multi-label Classifier” and “neighbor score”.In order to verify the performance of the designed predictor,the five cross-validation was performed with MF-EFP on a benchmark dataset of enzymes classified into the following 7 functional classes:(1)EC 1 Oxidoreductase,(2)EC 2 Transferase,(3)EC 3 Hydrolase,(4)EC 4 Lyase,(5)EC 5 Isomerase,(6)EC 6 Ligase,(7)EC 7 Translocases,where the enzymes contained are less than 90% redundant.The performance of MF-EFP is better than the existing predictor,which is proved by experiment.As a user-friendly web-server,MF-EFP is freely accessible to the public at the web-site http://www.jcibioinfo.cn/MF-EFP.A multi-label predictor for membrane protein function is designed by introducing a multi-label neural network algorithm based on Re LU activation function.The loss function commonly used in single-label learning(such as cross-entropy)is replaced with a multi-label cross-entropy loss function to meet the needs of multi-label dataset.In order to verify the performance of the designed predictor,the five cross-validation was performed with multi-label algorithm on a benchmark dataset of membrane proteins classified into the following 8 functional classes:(1)single-pass type I membrane,(2)single-pass type II membrane,(3)single-pass type III membrane,(4)single-pass type IV membrane,(5)multi-pass membrane,(6)lipid-anchor membrane,(7)GPI-anchor,(8)peripheral membrane,where the membrane proteins contained are less than 75% redundant.There are methods that have better prediction effects than existing predictors after experimental verification.