Models of deletion for visualizing bacterial variation: an application to tuberculosis spoligotypes

We found that inferred mutations of spoligotypes frequently involve the loss of a single or very few adjacent spacers. Using a second-order variant of Akaike's Information Criterion, we selected the Zipf model as the basis for resolving ambiguities in the ancestry of spoligotypes. We developed a method to construct graphs of spoligotypes (which we call spoligoforests). To demonstrate this method, we applied it to a tuberculosis data set from Cuba and compared the method to some existing methods.We propose a new approach in analyzing relationships of M. tuberculosis isolates using spoligotypes. The spoligoforest recovers a plausible history of transmission and mutation events based on the selected deletion model. The method may be suitable to study markers based on loci of similar structure from other bacteria. The groupings and relationships in the spoligoforest can be analyzed along with the clinical features of strains to provide an understanding of the evolution of spoligotypes.The visualization of relationships among genotypes of bacterial isolates is a useful approach to addressing both evolutionary and epidemiological questions. Inferences from graph representations of data assist in understanding the patterns of transmission of bacterial pathogens. Presently, there are two approaches to visualization. The first class of methods is sequence-based, and these methods often produce phylogenetic trees, or dendrograms. These trees are used extensively to represent relatedness of isolates that have been identified by almost any typing procedure. However, the models of sequence evolution upon which phylogenetic methods depend are not appropriate for many markers that are not sequence-based. Also, because many of these markers evolve rapidly enough to generate intra-specific variation, it is preferable to show direct relationships between genotypes. Bacterial isolates often cluster into the same genotype, and dendrograms are not suited to showing these clusters.The se