A Statistical Spacings Model to Infer Regions of Loss of Heterozygosity

We developed a statistical method that uses the concept of spacings, the distance between consecutive occurrences of an event, to take loss-of- heterozygosity (LOH) calls generated by SNP microarrays and infer regions of LOH. The proposed method is a useful data-reduction strategy, simplifies the multiple-testing problem, leads to theoretically conservative FDR estimates, and is robust against the inclusion of poor-quality data. In a recent study of LOH in therapy-related leukemia, the proposed method identified more markers in regions of LOH known by cytogenetics than did a widely used method based on hidden Markov modeling. Conceptually, the method may be generalized to infer the presence of other types of lesions as well.