<p>High-order single nucleotide polymorphism (SNP) epistasis detection has become an important goal in genome-wide association study. However, detecting epistasis in high-dimensional SNP data leads to the combinatorial explosion problem, and the evaluation functions used to detect SNP combinations with disease associations often have biases towards specific disease models. To address these issues, this paper proposes a membrane computing-based multi-objective ant colony optimization algorithm (MC-MOACO) to efficiently detect high-order SNP interactions. The algorithm is divided into two stages: In the search stage, symmetric uncertainty (SU) is used to quantify the marginal effects of SNP loci, and loci with high marginal effects are selected. A dedicated population is created to conduct a local fine search on these high SU loci. Meanwhile, three other populations, each using complementary evaluation functions, are generated to explore the additive effects of all SNP loci. These populations are embedded within the membrane computing framework for parallel searching, and information exchange across membranes is facilitated through membrane communication rules to accelerate solution optimization. The <i>G</i>-test statistical method is used to further validate candidate solutions in the testing stage, reducing the error rate. Experiments are conducted on three high-order simulation disease models with different interaction defects, two large-scale simulated datasets, and two real-world datasets (age-related macular degeneration (AMD) and breast cancer (BC)) to assess the performance of the proposed algorithm. Experimental results show that MC-MOACO achieves higher discriminative power and faster search speed than five classical algorithms on simulated datasets. MC-MOACO successfully detects high-order SNP interactions throughout the entire genome on the actual AMD and BC datasets.</p>