DNA barcoding has recently emerged for fast taxonomic classification of species using molecular biomarkers. Different from traditional classification scheme, DNA barcode often involves a small number of samples in each class, likely leading to a phenomenon known as overfit. To evaluate the efficacy of a biomarker based on a given meaningful multiple sequence alignment, we use a metric-based information measure that identifies converging interdependence on statistically significant sites. Experiments show that for the identified sites, when the convergent information between sites in the biomarker is small, its classification information is also small, whereas when it is high, then the information of the class is high. The correlation between these two types of pattern indicates the importance of selecting informative sites, in order for the biomarker to be effective as an identification barcode.