Pairwise Conditional Random Fields for Protein Function Prediction

Abstract

Protein Function Prediction (PFP) is considered one of the complex computational problems where any protein can simultaneously belong to more than one class. This issue is known as Multi-label classification problem in pattern recognition. Multi-label data sets are kind of data where each instance belongs to more than one class. This feature differentiates multi-label classification from the standard types of data classification. One of the challenges in multi-label classification data is correlation between the labels. This feature makes the issue cannot be classified into distinct sets of classification divided. Another major challenge is the high dimensional data in some applications. This paper presents a new method for the Protein Function Prediction and classification of multi-label data using conditional random fields. More specifically, the proposed approach is a method based on Pairwise Conditional Random Fields which considered the relationship of the labels. After introducing the Pairwise Conditional Random Fields optimization problem and solving it, the proposed method is evaluated under different criteria and the results confirm higher performance compared to available multi-label classifiers.