Classical and dominance-based rough sets in the search for genes under balancing selection

Since the time of Kimura's theory of neutral evolution at molecular level the search for genes under natural selection is one of the crucial problems in population genetics. There exists quite a number of statistical tests designed for it, however, the interpretation of the results is often hard due to the existence of extra-selective factors, such as population growth, migration and recombination. The author, in his earlier work, has proposed the idea of multi-null hypotheses methodology applied for testing the selection in ATM, RECQL, WRN and BLM genes - the foursome implicated in human familial cancer. However, because of high computational effort required for estimating the critical values under nonclassical null hypotheses, mentioned strategy is not an appropriate tool for selection screening. The current article presents novel, rough set based methodology, helpful in the interpretation of the tests outcomes applied only versus classical nulls. The author considers for this purpose both classical and dominance based rough set frameworks. None of rough set based methods requires long-lasting simulations and, as it is shown in a paper, both give reliable results. The advantage of dominance based approach over classical one is more natural treatment of statistical test outcomes, resulting in better generalization without necessity of manual incorporating the domain-dependent reasoning to the process of knowledge processing. However, in testing this gain in generalization proved to be at the price of a slight loss of accuracy.