Table 1.

Model selection to study (i) differences between the growth rate of different viruses (top) and (ii) epistatic interactions between viral gene segments (bottom). The starting maximal (a.k.a ‘beyond optimal’) models contained all interactions between fixed effects, as well as random effect structure estimated on both intercept and slope using lmer function in R (package lme4). They were simplified as recommended in (Crawley 2007; Zuur et al. 2009) to obtain a minimum adequate model usable for statistical analyses and interpretations.

Model objective	Maximal model	Minimum adequate model
Assess the effect of viruses	yield∼hour*virus + (hour|rep)	yield∼hour*virus-virus + (hour-1|rep)
	AIC: 340.70	AIC: 287.74
Assess the effect of viral gene segments	yield∼hour*seg2*seg4*seg7*seg8 + (hour|rep)	yield∼hour*seg2*seg4*seg7*seg8
	AIC: 340.70	-(seg2*seg4*seg7*seg8)
		-hour:seg2:seg4:seg7:seg8
		+ (hour-1|rep)
		AIC: 283.94

yield: log₁₀ of viral copies/ml supernatant; hour: hours post infection; virus: identity of viral construct; seg2: allele for segment 2 (‘Wt’ or ‘Ad’), likewise for seg4, seg7 and seg8; rep: identity of replicate infection, as a random effect.

AIC: Akaike’s Information Criterion, calculated as -2*log-likelihood + 2*n_parameters.