Protein Engineering Protocols - Mycobacteriology research center

6 Kono et al.patterning hydrophobic and polar sites, a 51-residue homeodomain motif (22).The group has also redesigned residue subsets within portions of a variety ofproteins (23–25). Functional properties, such as metal binding or catalysis, mayalso be included as elements of the design process (26–28). The elements andalgorithms of directed protein design has been the subject of several recentreviews (4,29,30).Despite some striking successes, computational methods for the directeddesign of sequences have limitations with respect to characterizing the featuresof protein sequences folding to a particular structure. Stochastic methods can beapplied to large proteins and permit many sites to be varied simultaneously, butthe computational times and resources for such calculations are extensive, evenfor small proteins. Directed methods will necessarily be sensitive to the energyor scoring function used, because they identify the optimum of a particularenergy function. All such energy functions, however, are necessarily approximate,and uncertainties in the energy function may not merit the search forglobal optima. Many naturally occurring proteins are not optimized. In fact,most proteins are only marginally stable, e.g., ∆G o < 10 kcal/mol for folding(31). In addition, sequences that function, e.g., those that bind another molecule,need not be the global optimum with respect to structural stability. It isimportant to develop methods complementary to those used for directed proteindesign, methods that reveal the features of sequences likely to fold to a particularstructure but which may not be structurally optimal. Such techniques maybe used in designing protein sequences. In addition, such computational methodsmay be applied to a new class of protein design studies, combinatorialexperiments, in which large numbers of proteins may be simultaneously synthesizedand screened.1.3. Probabilistic Approaches to Protein DesignIn the context of protein design, we refer to the use of site-specific aminoacid probabilities rather than specific sequences as “probabilistic proteindesign.” Probabilistic approaches, as opposed to directed or deterministicapproaches, are often used in the quantitative sciences for cases in which thereis only partial information regarding a problem. For protein design, such a probabilisticapproach is motivated by the complexities and uncertainties associatedwith the folding process. Protein folding is a complicated kinetic process, withmyriad interactions specifying the folded state. Each of the stabilizing noncovalentinteractions is of comparable magnitude and there seems to be no oneoverriding determinant of folding. The means of quantifying these interactionsare necessarily approximate (see Note 1). Probabilistic design methods alsodirectly provide very useful sequence information, particularly regarding structurallyimportant amino acids. The amino acid probabilities can guide the