Abstract
The recent development of high-throughput experimental methods has generated a large amount of protein interaction data, which is becoming the foundation for new biological discoveries. There are several methods developed for motif discovery, but these methods focus on detecting individual motifs rather than interacting motif pairs. The primary focus of this study is to predict reliable interacting motif pairs from combinations of protein features using a stochastic method. This paper describes an improved boosting algorithm for predicting interacting motif pairs of proteins and a method for generating negative interaction data for the algorithm.