Predicting regions prone to protein aggregation based on SVM algorithm

摘要

The phenomenon of protein aggregation has been associated with several neurodegenerative diseases, such as Parkinson's and Alzheimer's. Computational tools have been used to predict regions prone to aggregate in proteins with relative success. We have developed a tool called MAGRE for such predictions, based on the machine learning and sliding window techniques. We have applied the Support Vector Machine algorithm to generate classification models. In order to accomplish classification training, we adopted information of primary structure - protein sequence - from the Amyloid Data Bank. We have implemented two predictor categories according to protein structural information: General and Folding Class. We have selected the best performances of the sliding windows method and considered the folding class in order to develop the predictor. We conducted testing with randomly selected protein sequences from the PDB data bank - MAGRE's performance was compared with two predictors from literature: Aggrescan and Zyggregator, being considered satisfactory.