Thursday, August 2, 2007 - 2:50 PM
S186
Protein engineering without high-throughput screening: Practical approaches using machine learning algorithms
Sridhar Govindarajan, Alan Villalobos, Claes Gustafsson, Jon Ness, and Jeremy Minshull. DNA2.0 Inc,, 1430 SuiteE, O'Brien Drive, Menlo Park, CA 94025
Protein engineering using directed evolution has been a very useful technology in optimizing proteins for activities of interests. Directed evolution usually involves the creation of large libraries of variants that are screened in high-throughput. Often, the high-throughput screens serve as a surrogate assay for the real property/activity of interest. The best members of the library are then selected for repeated cycles of change, recombination and selection. The end result is a protein that is highly active as measured by the surrogate assay and often show no improvement in properties that are measured under commercial conditions. As an alternative, researchers began developing tiered and multi assay systems to counter the problem. We present protein design algorithms that require testing of a small number of variants in the most commercially relevant assay.