Table 4.

Summary of GA-based model selection

Model #	True model	Selected model	Generation #	Shrinkage-related penalty	Hybrid	Runtime (h)	Fitness	Δ-fitness
1	1cmt, 1_abs, transit_cmt, lin_elim, prop_err	1cmt, 1_abs, lin_elim, prop_err	20	0	No	18.2	2370.6	– 65
		1cmt, 1_abs,, lin_elim, prop_err	11	0	Yes	11.1	2370.6	− 65
		1cmt, 1_abs,, lin_elim, prop_err	20	100	No	17.3	2376.5	− 150.7
		1cmt, 1_abs,, lin_elim, comb1_err	11	100	Yes	17.7	2376.7	− 150.5
2	1cmt, 1_abs, 0_abs, lin_elim, comb1_err	1cmt, 1_abs, lin_elim, comb2_err	20	0	No	22.8	2504.4	− 887.7
		1cmt, 1_abs, lin_elim, comb1_err	11	0	Yes	11.2	2505.7	− 886.4
		1cmt, 1_abs, lag, lin_elim, comb1_err	20	100	No	19.1	2522.2	− 856
		1cmt, 1_abs, lin_elim, comb2_err	11	100	Yes	10.7	2592.4	− 785.8
3	2_cmt, bolus, lin_elim, MM_elim, comb1_err	1_cmt, bolus, lin_elim, MM_elim, comb_err	20	0	No	16.2	1632.1	− 346.8
		1_cmt, bolus, lin_elim, MM_elim, comb1_err	11	0	Yes	14.2	1633.4	− 345.5
		1_cmt, bolus, lin_elim, MM_elim, comb2_err	20	100	No	17.3	1634.1	− 364.4
		1_cmt, bolus, lin_elim, MM_elim, comb2_err	11	100	Yes	15.6	1638.2	− 360.3
4	2_cmt, 1_abs, lag, lin_elim, add_err	1_cmt, bolus, lin_elim, add_err	20	0	No	20.5	4918.1	− 94.1
		1_cmt, bolus, lin_elim, MM_elim, prop_err	11	0	Yes	14.6	4928.4	− 83.8
		1_cmt, bolus, lin_elim, add_err	20	100	No	27.9	4921.5	− 201.4
		1_cmt, bolus, lin_elim, add_err	11	100	Yes	8.7	4921.5	− 201.4
5	3_cmt, bolus, lin_elim, comb1_err	1_cmt, bolus, lin_elim, comb1_err	20	0	No	14.2	2522.9	− 101.2
		1_cmt, bolus, lin_elim, comb1_err	11	0	Yes	12.2	2522.9	− 101.2
		1_cmt, bolus, lin_elim, comb1_err	20	100	No	18.1	2526.3	− 109.3
		1_cmt, bolus, lin_elim, comb1_err	11	100	Yes	9.5	2526.3	− 109.3

Δ delta, GA genetic algorithm, h hours, 1_cmt one compartment, 2_cmt two compartment, 3_cmt three compartment, 1_abs 1st order absorption, 0_abs 0 order absorption, lag lag time, transit_cmt transit compartments, lin_elim linear elimination, MM_elim Michaelis–Menten elimination, add_err additive error model, prop_err proportional error model, comb1_err combined1 error model, comb2_err combined2 error model. GA selection was considered successful if the best model in the last generation (selected model) had a fitness value smaller than the true model (negative Δ-fitness)