Peng-Jie Jing1, Hong-Bin Shen1. 1. Institute of Image Processing and Pattern Recognition, Shanghai Jiao Tong University and Key Laboratory of System Control and Information Processing, Ministry of Education of China, Shanghai 200240, China Institute of Image Processing and Pattern Recognition, Shanghai Jiao Tong University and Key Laboratory of System Control and Information Processing, Ministry of Education of China, Shanghai 200240, China.
Abstract
MOTIVATION: The existing methods for genetic-interaction detection in genome-wide association studies are designed from different paradigms, and their performances vary considerably for different disease models. One important reason for this variability is that their construction is based on a single-correlation model between SNPs and disease. Due to potential model preference and disease complexity, a single-objective method will therefore not work well in general, resulting in low power and a high false-positive rate. METHOD: In this work, we present a multi-objective heuristic optimization methodology named MACOED for detecting genetic interactions. In MACOED, we combine both logistical regression and Bayesian network methods, which are from opposing schools of statistics. The combination of these two evaluation objectives proved to be complementary, resulting in higher power with a lower false-positive rate than observed for optimizing either objective independently. To solve the space and time complexity for high-dimension problems, a memory-based multi-objective ant colony optimization algorithm is designed in MACOED that is able to retain non-dominated solutions found in past iterations. RESULTS: We compared MACOED with other recent algorithms using both simulated and real datasets. The experimental results demonstrate that our method outperforms others in both detection power and computational feasibility for large datasets. AVAILABILITY AND IMPLEMENTATION: Codes and datasets are available at: www.csbio.sjtu.edu.cn/bioinf/MACOED/.
MOTIVATION: The existing methods for genetic-interaction detection in genome-wide association studies are designed from different paradigms, and their performances vary considerably for different disease models. One important reason for this variability is that their construction is based on a single-correlation model between SNPs and disease. Due to potential model preference and disease complexity, a single-objective method will therefore not work well in general, resulting in low power and a high false-positive rate. METHOD: In this work, we present a multi-objective heuristic optimization methodology named MACOED for detecting genetic interactions. In MACOED, we combine both logistical regression and Bayesian network methods, which are from opposing schools of statistics. The combination of these two evaluation objectives proved to be complementary, resulting in higher power with a lower false-positive rate than observed for optimizing either objective independently. To solve the space and time complexity for high-dimension problems, a memory-based multi-objective ant colony optimization algorithm is designed in MACOED that is able to retain non-dominated solutions found in past iterations. RESULTS: We compared MACOED with other recent algorithms using both simulated and real datasets. The experimental results demonstrate that our method outperforms others in both detection power and computational feasibility for large datasets. AVAILABILITY AND IMPLEMENTATION: Codes and datasets are available at: www.csbio.sjtu.edu.cn/bioinf/MACOED/.