Clustering Sequences in a Metric Space The MoBIoS Project

We are developing a [Molecular] Biological Information System (MoBIoS) based on metric space indices. Unfortunately, common similarity measures for sequence alignment do not form a metric-distance function. This is particularly vexing since the usual definition of edit distance does form a metric. Most clearly, the use of PAM log-odds matrices [2] yields higher similarity scores for more closely related sequences, an intuitively appealing result that reverses metric order. Further, log-odds scoring matrices contain negative values that can yield negative global alignment scores. This violates positivity. Use of PAM matrices also can violate symmetry and the triangle inequality.