An Evolutionary Approach to Hard Test Case Generation for Shortest Common Superstring Problem

The shortest common superstring problem has important applications in computational biology (e.g. genome assembly) and data compression. This problem is NP-hard, but several heuristic algorithms proved to be efficient for this problem. For example, for the algorithm known as GREEDY it was shown that, if the optimal superstring has the length of N, it produces an answer with length not exceeding 3.5N. However, in practice, no test cases were found where the length of the answer is greater than or equal to 2N. For hard test case generation for such algorithms the traditional approach assumes creating such tests by hand. In this paper, we propose an evolutionary algorithm based framework for hard test case generation. We examine two approaches: single-objective and multi-objective. We introduce new test case quality measures and show that, according to these measures, automatically generated tests are better than any known ones.