Abstract
To engineer the enantioselectivity of lipase from Burkhorderia cepacia KWI-56, a combinatorial library involving mutations at four amino acid residues in its hydrophobic substrate binding pocket was constructed by SIMPLEX (single-molecule-PCR-linked in vitro expression) and screened for (R)- and (S)-configurations of p-nitrophenyl 3-phenylbutyrate. Some combinations of amino acid substitutions in the four positions of the lipase were found as effective for changing the enantio-preference from the (S)-form substrate to the (R)-form (J. Mol. Biol. 331,585,2003). Here, the relative enantioselectivity and the amino acid residues' data of the obtained 17 variants in the above experiments have been analyzed by knowledge informational analysis using a fuzzy neural network (FNN), in which a rule on the interconnection between inputs and outputs was automatically acquired by learning without any structural information of the protein and the substrate. Hydrophobicity, van der Waals volume and electrical effect of amino acid residue in each position was provided as inputs for the FNN analysis and a set of rules was extracted indicating that van der Waals volume of position 167 has the most significance for the enantioselectivity of the lipase, as well as predicting likely variants that were not found in the actual screening. Then the real construction of such variants has shown that the FNN analysis can figure out the real world of the protein sequence combination space. The collaboration between high-throughput screening and bioinformatic technology has proved to be quite powerful to explore a novel protein with targeted functions from the vast space of protein sequences.
