Gene clustering is very important for extracting underlying biological information of gene expression data. Currently, SOM (self-organizing maps) is known as one of the most popular neural networks applied for gene clustering. However, SOM is sensitive to the initialization of neurons' weights. In this case, biologists may need to spend a lot of time in repeating experiments until they obtain a satisfactory clustering result. In this paper, we apply QBSOM (query-based SUM) to tackle the drawbacks of SOM. We have tested the proposed method by several kinds of real gene expression data. Experimental results show that QBSOM is superior to SOM in not only the time consumed but also the result obtained. Considering the gene clustering result of YF (yeast full) dataset, QBSOM yields 17% less in MSE (mean-square-error) and 68% less in computation cost compared with SOM. Our experiments also indicate that QBSOM is particularly adaptive for clustering high dimensional data such as the gene expression data. It is better than SUM for system convergence. (C) 2010 Elsevier Ltd. All rights reserved.
