PNNL

Abstract

Statistical methods to assign the significance of an alignment under a null distribution have been well studied. However assessing uncertainty of a proposed alignment, defining the confidence in this alignment, and assessing its overall quality have received considerably less attention. We introduce a novel global credibility metric to assess how well any single alignment describes the overall alignment space. We use this metric to identify “centroid” alignments that seek to optimize an alignment’s credibility by minimizing sums of distance from posterior weighted space alignments. Furthermore, we show that there is no principled basis for expecting popular optimal alignment procedures to yield alignments with high credibility, and show that they often don’t. Lastly, we describe sequence alignment visualization procedure to communicate the local credibility of an alignment at the base-pair level.