Artificial Immune Systems Perform Valuable Work When Detecting Epistasis in Human Genetic Datasets

摘要

We implement an Artificial Immune System (AIS) for epistasis detection in human genetic datasets. Our AIS outperforms previous attempts to solve the same problem by Penrod et al. by a factor of over 2.4 and performs at 81% of the power of the field standard exhaustive search, Multifactor Dimensionality Reduction (MDR). We show that the immune system performs best when 'paring down' large antibodies to more specific and accurate classifiers. This is promising as it shows that the AIS is doing valuable work, and needs not rely on a near-perfect antibody showing up by chance. We perform a receiver operator characteristic (ROG) analysis to further examine this property.