We develop a novel and general approach to estimating the accuracy of protein multiple sequence alignments without knowledge of a reference alignment, and use our approach to address a new problem that we call
. For protein alignments, we consider twelve independent features that contribute to a quality alignment. An accuracy estimator is learned that is a polynomial function of these features; its coefficients are determined by minimizing its error with respect to true accuracy using mathematical optimization. We evaluate this approach by applying it to the task of parameter advising: the problem of choosing alignment scoring parameters from a collection of parameter values to maximize the accuracy of a computed alignment. Our estimator, which we call
imator”), yields a parameter advisor that on the hardest benchmarks provides more than a 20% improvement in accuracy over the best default parameter choice, and outperforms the best prior approaches to selecting good alignments for parameter advising.