Performance Comparison of Clustering Methods for Gene Family Data

Abstract

Clustering gene sequences into families is important for understanding and predicting gene function. Many clustering algorithms and alignment-free similarity measures have been used to analyze gene family. The clustering results can be influenced by the similarity measure and clustering algorithm used. We compare the results from running four commonly used clustering methods, including K-means, single-linkage clustering, complete-linkage clustering and average-linkage clustering, on three alignment-free similarity measures. We try to find out which method should provide the best clustering result based on real-world gene family datasets. Experiment results show that average-linkage clustering with our similarity measure, DMk, performed best.