Human Protein Function Prediction Enhancement Using Decision Tree Based Machine Learning Approach

Abstract

The interrelated complex protein sequence databases are brimming day by day with the rapid advancement in technology. The sophisticated computational techniques are required for the extraction of data from these huge loads, so that refined extracted information can be easily deployable for the progress of mankind. The human protein function prediction (HPFP) is the relevant research area whose identification or function prediction leads to the discovery of drugs, detection of disease, crop hybridization, etc. Numerous approaches are present these days for HPFP because of its wide and versatile nature of this domain. The Decision tree (DT) based white box Machine Learning (ML) approaches is enriched with computational techniques to grab the information from this important research area. This study uses the decision tree based machine learning approach together with a sequence derived features (SDF’s) extraction from the human protein sequence in order to predict the protein function. The experiment has been performed by manually extracting the human protein classes and sequences from HPRD (human protein reference database) [1]. Thereafter extract the SDF’s from the sequences with the help of proposed HP-SDFE server as well as with the help of web servers and the DT based different classifiers such as boosting, winnowing, pruning etc. has been used for HPF prediction. The efficacies of different DT classifiers are examined and compared with the existing benchmark. The importance of input configurations together with enhanced SDF’s has been thoroughly examined, which leads the individual molecular class prediction accuracy to 97%. The proposed methodology is also applicable in other similar research areas.