Fig. 2. Higher-order covariance information improves the inference of epistasis.

Distribution of inferred epistasis using the full model inferred via HCMF (a), which includes higher-order covariance information, and using the independent model (b), in which off-diagonal terms of the covariance matrix are set to zero. c, Receiver operating characteristic (ROC) curve for identifying positive epistasis. The area under the curve (AUC) value is 0.90 for the full model, while it drops to 0.81 for the independent model. d, Analogous ROC and AUC values for identifying negative epistasis. The AUC values are 0.90 and 0.69 for full and independent models, respectively.