Understanding the genotype-phenotype association is a fundamental problem in genetics. A major open problem in mapping complex traits is identifying a set of interacting genetic variants (such as single nucleotide polymorphisms or SNPs) that influence disease susceptibility. Logic regression (LR) is a statistical approach that has been proposed to model interactions of SNPs. Several LR-based association detection approaches have been developed in the past. However, existing LR-based approaches are insufficient in handling noisy and increasingly larger data. In this paper, we first develop a relational clustering approach for handling noisy data, where we reduce noise by filtering out unrelated SNPs. We then propose a parallel fish-agent LR approach to speed up the computation. The basic idea of our approach is using multiple fish-agents that explore the model space independently. At each iteration, agents in the same or different clusters communicate with others to achieve faster convergence to the global optimal solutions. Simulation results show that our approach significantly speeds up the LR computation over existing approaches. Also, our results show that our approach achieves good performance in dealing with noise.