Prediction And Analysis For Anti-inflammatory Peptide Via Feature Selection Algorithm

In recent years,peptides have shown great potential as new diagnostic drugs and treatment of human diseases.Anti-inflammatory peptides(AIPs)provide a new therapeutic approach for autoimmune disorders and inflammatory diseases.Compared with small molecule nonspecific anti-inflammatory drugs,anti-inflammatory peptides have the characteristics of fewer side effects and more effectives.With the emergence of a large number of peptide data,how to accurately identify AIPs is of great significance to explore the internal mechanism of anti-inflammatory peptides and the treatment of inflammatory diseases.Despite the fact that an increasing number of AIP is validated in the traditional experiment methods,the experimental method is not only time-consuming but also very expensive.Therefore,in order to overcome the shortcomings of traditional experimental methods,the computational method based on machine learning was proposed to predict anti-inflammatory peptides.This method was more quickly and accurately for high-throughput data.In this paper,we established a corresponding machine learning prediction model to predict the AIPs.The specific work contents are as follows:1.Computational prediction and analysis for anti-inflammatory peptide via a hybrid feature selection technique.In order to comprehensively describe AIPs,8sequence features and 4 physicochemical features were fused to encode their information.However,this makes the training process of the model complicated and difficult.Feature selection algorithms can optimize the high-dimensional features and obtain the important features,thus speeding up the model training process,improving the model accuracy and enhancing the interpretability of the model.Here,for the AIPs prediction problem,a hybrid feature selection algorithm was proposed to screen and optimize the fusion high-dimensional features by integrating enhanced filter feature selection algorithm and wrapper feature selection algorithm.Finally,an AIPs prediction method PREDAIP(http://github.com/lindan1/PREDAIP),was developed by combining extremely randomized tree(ERT).The cross-validation results show that the hybrid feature selection algorithm greatly improves the prediction performance of the model compared with the fused high-dimensional features.PREDAIP is a reliable predictive tool for anti-inflammatory peptides compared to other available predictive tools in an independent set of tests.2.Prediction of AIPs based on efficient random feature selection algorithm.For the prediction of AIPs,new and effective peptide coding methods need to be further explored.Secondly,we proposed two novel data processing techniques,namely warm start and cool down,and embedded them into the traditional random feature selection(RFS)algorithm,to obtain an efficient random feature selection algorithm.We compared the performance of six traditional classifiers based on the benchmark dataset.Finally,using the optimized feature subset,and combined with random forests(RF)classifier,we developed a new prediction tools PREDAIP 2.0.The results of 10-fold cross validation and independent testing show that the predictive performance of PREDDAIP 2.0 is stable and reliable.The AIPs prediction tool can help researchers explore the intrinsic properties of AIPs and discover new anti-inflammatory peptides.