Applied biological data mining based on improved K-means clustering algorithm and KNN classifier in protein sub-cellular localization

摘要

With the completion of the human genome sequencing, a large number of data especially amino acid sequences floods into biological database. How to analyze these data quickly and even predict the structure and function of protein correctly have become hot topics in recent years. In this paper, we mainly study K-means clustering algorithm and KNN classifier in amino acid sequences of complicated data, which are applied in the prediction of protein sub-cellular localization. In many cases, fuzzy boundary and unbalance are frequently appeared among biological data. The accuracy will be lower, if we make a prediction through traditional KNN and K-means clustering algorithm directly. Firstly, in order to make clear the unbalance, we propose the within-class thought to make sure that training samples in each class around the testing sample are selected and we introduce membership to tell which class the testing sample belongs to. Then, we bring in rough sets and membership to solve the fuzzy boundary. Particularly, we apply correlation coefficient in the rough sets to better reflect the relationship among data objects. The experimental results based on protein sub-cellular localization prediction show that the methods proposed newly better work than the traditional methods.