Prediction Method Research Of Protein Based On Protein Evolutionary Information

In the last five to ten years. Computer science has been playing a significant role inmedicine and bioscience, has made and making a great contribution. At post-genome era,there are still lots of unknown information to be decoded using computer technique. It is soattractive to predict and classify the specific protein with low-cost fast.In this paper, we use the primary structure of protein as the basic information. Then wemake highly sensitive alignment by running PSI-BLAST. We achieve the position specificscoring matrix which contains the evolutionary information of the protein. We achieve thefeature vector after applying auto covariance transform on the matrix. We predictbioluminescent proteins and extracellular matrix proteins using KNN and support vectormachine. We achieve a good result on predicting the bioluminescent protein. Tested by10-fold cross-validation and independent test, the accuracy of the proposed model reaches85.17%for the training dataset and90.71%for the testing dataset respectively. And as weknown, it is the best prediction method on the bioluminescent protein. In the prediction ofextracellular matrix proteins, we just achieve the result close by the best method. On thetesting dataset, we achieved an accuracy of74.43%. We built an online service to predictbioluminescent proteins from the non-bioluminescent proteins. And it is helpful toresearchers.