Factorgrams: A tool for visualizing multi-way associations in biological data

Effective visualization of biological data is often critical for subsequent analy- sis. The popular clustergram/dendrogram visualization rearranges rows and columns of a data matrix so as to highlight clusters of similar responses, but assumes each row or column belongs to only one cluster and cannot associate each row or column with multiple clusters. Such multi-way associations oc- cur frequently, e.g., when a gene plays multiple biological roles. We describe the 'factorgram' visualization, which rearranges the data into an expanded view, associating each row (or column) with multiple clusters of rows (or columns) and elucidating potentially new biological relationships. Factor- grams for mouse gene expression and yeast synthetic-lethal gene-interaction datasets detect a larger number of statistically-significant clusters than clus- tergrams, plus a larger number of clusters enriched for gene ontology annota- tions. Experimentally-verified associations previously identified by manual rearrangement of rows and columns not grouped together by clustergrams, are readily identified by the factorgram.