Protein engineering is typically performed via either rational design or combinatorial library screening. This project sought to develop strategies which effectively harness the strengths of the two methodologies, and apply them to engineering of model proteins. A set of computational algorithms, SPARTIC and LibDesign, were developed which used structural information to guide design of targeted libraries. SPARTIC uses a Monte Carlo algorithm coupled with a structural model and an energetic scoring function to find ensembles of amino acid combinations that are compatible with a target backbone. The LibDesign algorithm then scores degenerate codon libraries based on these ensembles in terms of total library size and completeness, from which a library with an optimal size/score combination can be selected for construction. The libraries are then constructed using gene assembly mutagenesis and screened for the property of interest.

We used computational library design strategies, including SPARTIC and LibDesign, to engineer variants of blue fluorescent protein (BFP), a short-wavelength mutant of Aequorea victoria green fluorescent protein (GFP) that is too dim and photobleaches too easily for routine use. Mutants isolated from structurally targeted libraries had considerably higher whole-cell fluorescence, improved quantum yield, and longer photobleaching half-lives than the parental sequences. These mutants are useful for multi-color labeling experiments and as Förster resonance energy transfer (FRET) partners to longer-wavelength fluorescent proteins such as GFP or red fluorescent variants.