An incremental updating based fast phenotype structure learning algorithm

Abstract

Unsupervised phenotype structure learning is important in microarray data analysis. The goal is to (1) find groups of samples corresponding to different phenotypes (e.g. disease or normal), and (2) find a subset of genes that can distinguish different groups. Due to the large number of genes and a mass of noise in microarray data, the existing methods are often of some limitations in terms of efficicency and effectiveness. In this paper, we develop an incremental updating based phenotype structure learning algorithm, namely FPLA. With a randomly selected initial state, the algorithm iteratively tries three possible adjustments, i.e. gene addition, gene deletion and sample move, to improve the quality of the current result. Accordingly, four incremental updating based optimization strategies are devised to eliminate the redundancy computations in each iteration. Further, by utilizing a harmonic quality function, it improves the result accuracy by penalizing the "outlier" effect. The experiments conducted on several real microarray datasets show that FPLA outperforms the two representative competing algorithms on both effectiveness and efficiency. © 2014 Springer International Publishing Switzerland.