Automatic parameter selection for sequence similarity search

We show that simulated annealing search can be used to automatically select parameters and find highly similar data regions using a modified version of the DNA-DNA sequence similarity search program. We call this modified program AutoSimS. We use the average score of high-scoring chains to measure the goodness of the resulting sequence similarity search, and use adaptive simulated annealing to perform automatic search within a space of parameter values to maximize this goodness measure. We tested our program using pairs of DNA sequences, and the results show that although close-to-optimal parameter settings are very difficult to find manually, there are many different parameter settings that yield close-to-optimal search results. We suggest that our approach is able to successfully and automatically select parameters for programs used to finding close-to-optimal solutions, such as highly similar sequence regions.