Ambivalent covariance models

Background Evolutionary variations let us define a set of similar nucleic acid sequences as a family if these different molecules execute a common function. Capturing their sequence variation by using e. g. position specific scoring matrices significantly improves sensitivity of detection tools. Members of a functional (non‐coding) RNA family are affected by these variations not only on the sequence, but also on the structural level. For example, some transfer‐RNAs exhibit a fifth helix in addition to the typical cloverleaf structure. Current covariance models – the unrivaled homology search approach for structured RNA – do not benefit from structural variation within a family, but rather penalize it. This leads to artificial subdivision of families and loss of information in the Rfam database.