Representation and Data Structure of Genetic Algorithms for Protein Threading

Despite many years of efforts, a direct prediction of protein structure from sequence is still not possible. As a result, in the last few years researchers have started to address the ``inverse folding problem". How to identify and align a sequence to the fold with which it is most compatible, a process called Threading. In the two meetings in which protein folding predictions were objectively evaluated [Lemer et. al., 1995], it became clear that threading promises a real breakthrough, but that much improvement is still needed. Major effort is required in the algorithmic component to ensure that each sequence will be aligned in the optimal threading rather than as an approximation. We used the technique of Genetic Algorithms in order to significantly improve the ability of threading algorithms to find the optimal alignment of a sequence to a structure, i.e. the alignment with the minimum free energy. We designed a representation of the threading alignment as a string on which genetic operators can be effectively implemented, and investigated the appropriate data structure and parameter selection.