Table 1.

Results of information theoretic mathematical model selection on integrated HIV DNA per million CD4+ T cells

Rank	Model	ΔLL	N rates	ΔAIC	ΔBIC
1	Differentiation with skips: subsets can proliferate and die and are connected from least to most differentiated but additional connections are possible (e.g., TN > TCM).)	0	11	0	0
2	Constrained differentiation with skips: same as 1 but with limits on maximal differentiation rates (no greater than cell turnover) based on biological plausibility.	10.5	11	10.5	10.4
3	Linear differentiation: subsets can proliferate and die and are connected from least to most differentiated.	25.4	9	17.4	10.9
4	Carrying capacity: integrated HIV DNA in each subset is assumed to have an equilibrium value such that levels away from this value return through logistic growth/shrinking.	28.7	10	24.7	21.4
5	Linear differentiation linked to proliferation: a mathematical formulation in which some proportion of proliferation leads to differentiation.	44.9	10	40.9	37.6
6	No differentiation: subsets can only proliferate and die.	84	5	60	40.6
7	Constrained linear differentiation: same as #3 but with limits on maximal differentiation rates based on biological plausibility.	75	9	67	60.5
8	Carrying capacity 2: same as #4 with a different mathematical form for equilibration.	73.2	10	69.2	65.9
9	Only differentiation: subsets have no proliferation/death or net repopulation rates.	113.1	4	85.1	62.5
10	Forced clearance: repopulation rates must be negative, and no differentiation is included.	136.4	5	112.4	93

Constrained differentiation with skips was chosen as the optimal model (see bolded rank 2) as best BIC given biologically realistic parameters. Δ denotes differences from the absolute best model (rank 1). N rates is included to indicate model complexity (more estimated rates is more complex).

LL log likelihood, AIC Akaike information criterion, BIC Bayesian information criterion.