Biclustering binary data and its application in traditional Chinese Medicine

The biomedical dataset is usually expressed as matrix containing only one and zero. This paper addresses searching for blocks that (mostly) contain ones in such a dataset where there may exist meaningless ones considered as noise and overlapping situation between blocks. To this end, we propose a new biclustering algorithm through defining a suitable clustering similarity measurement and integrating the hierarchical clustering algorithm to discover high-quality biclusters. A powerful feature of our approach is that biclusters with different numbers of rows and columns can be detected varying from many rows to a few columns and in turn, since the user can guide the search towards biclusters by defining at first different cluster thresholds for specific dimensions. Several experiments and comparisons based on both synthetic datasets and real datasets are adopted to demonstrate its efficiency in finding the biclusters and achieving significant improvements when compared to other basic greedy strategies. In addition, the approach is applied to the traditional Chinese medical researches in order to indicate its practical value.