Learning Advisors for Multiple Sequence Alignment

While the multiple sequence alignment output by an aligner strongly depends on the parameter values used for its alignment scoring function (i.e. choice of gap penalties and substitution scores), most users rely on the single default parameter setting. A different parameter setting, however, might yield a much higher-quality alignment for a specific set of input sequences. The problem of picking a good choice of parameter values for a given set of input sequences is called parameter advising. A parameter advisor has two ingredients: (i) a set of parameter choices to select from, and (ii) an estimator that estimates the accuracy of a computed alignment; the parameter advisor then picks the parameter choice from the set whose resulting alignment has highest estimated accuracy. Our estimator Facet (Feature‑based Accuracy Estimator) is a linear combination of real-valued feature functions of an alignment. We assume the feature functions are given as well as the universe of parameter choices from which the advisor’s set is drawn. For this scenario we define the problem of learning an optimal advisor by finding the best possible set, or estimator, for a collection of training data of reference alignments. Learning optimal advisors is NP‑complete. For the advisor sets problem, we develop a fast approximation algorithm that finds near optimal sets. For the advisor estimator problem, we have an efficient method for finding the coefficients for the estimato