Properties of DMEs identified by various models using mutational effect summary data.

Preliminary models fitted to summary data of individual strain s value fits. Models that attempt to fit unidentified mutations either as additional mutations drawn from the same reflected gamma distribution as SNMs, or as an independent reflected gamma distribution to individual unidentified mutations, improve fit only marginally. The “two-gamma” model produces uninformative parameter estimates, with poor computational likelihood estimation across the likelihood profile (due to confounded parameters). The resulting non-monotonic likelihoods lead to uninterpretable confidence interval bound estimation for many parameters (question marks). The “Gaussian” model, which accounts for unidentified mutations as a Gaussian distribution representing the effects of these mutations across strains, performs better than a model that only accounts for SNMs. Parameter values for each model shown with 95% confidence intervals; ΔAIC is calculated relative to the “SNMs only” model.
(XLSX)