Rather than devise another algorithm. we examin??ed several assumptionsthat under??lie many pre??dictions: First. most algorithms use sequence align??ments of evolutionarily related pro??teins as part of their input. Posi??tions that show few changes (conserved) are presumed to be criti??cal for function; the ration??ale is that mutations at conserved positions must have been so dama??ging that they were selected against. Second. a set of sub??sti??tution rules (Box 1). derived from decades of laboratory experiments is either explicitly or implicitly included in most analyses. These rules are rea????sonably suc??cessful for predicting mutational outcomes at conserved positions. perhaps because lab experi??ments have been highly biased to conserved positions.2