| For many protein design problems, limited understanding of the relationship between sequence and function necessitates searching through a library of proteins to find the properties of interest. To accelerate this process, molecular models and optimization algorithms can be combined to design diverse libraries enriched in folded proteins. I apply this strategy to site-directed recombination, in which an alignment of p homologs is partitioned into f blocks, and the resulting gene fragments are combinatorially assembled to create a library with pf chimeric sequences. To design the fragments, I present a dynamic programming algorithm that minimizes the average energy of the library, subject to constraints on fragment length. This algorithm works for any pairwise residue potential, several of which are compared for their ability to predict which chimeras retain the parental function and/or fold. The alignments of folded and unfolded chimeras are used to generate sequence-function relationships via logistic regression, a technique for fitting models to binary data. Compared to methods developed for alignments of naturally occurring proteins, logistic regression more readily distinguishes true interactions from correlations between strongly stabilizing but non-interacting residues. |
