Table 1. Model comparison in real-world datasets.

We compare the predictive performance of flexTPC and Briere models. The best performing model has its values highlighted in bold. The model comparison criteria are indicated below the corresponding dataset. For datasets that were fit with a maximum likelihood approach (botrana, glacierbac, lhculex), we use mean leave one out cross-validated negative log-likelihood (LOOCV-nLL, lower is better) as the model comparison criterion to compare between the Briere1, Briere2, and flexTPC models. For the lhculex dataset, which was fit with a Bayesian approach, we use the Deviance Information Criterion (DIC, lower is better) as a model comparison criterion between a TPC functional form that was previously used in the literature to describe that trait (lit.function) and flexTPC.

Dataset
botrana	life stage	Briere1	Briere2	flexTPC
LOOCV-nLL	eggs	2.00	−1.79	−2.20
	instar 1	−1.46	−2.32	−2.64
	instar 2	−2.24	−2.88	−3.08
	instar 3	7.16	−1.55	−2.60
	instar 4	2.14	−2.61	−2.51
	instar 5	−0.48	−3.04	−3.12
	pupae	10.80	−2.95	−3.95
glacierbac	species (strain)	Briere1	Briere2	flexTPC
LOOCV-nLL	Arthrobacter sp (55)	0.27	−3.35	−3.61
	Arthrobacter sp (60)	2.53	−3.13	−3.79
	Pseudomonas (76)	inf	−2.23	−2.71
	Arthrobacter glacialis (137)	0.70	−3.49	−4.46
	Arthrobacter glacialis (158)	−2.64	−4.23	−5.68
abcoli	antibiotic	Briere1	Briere2	flexTPC
LOOCV-nLL	AMP	−0.97	−1.33	−2.08
	CLI	−1.43	−1.47	−1.67
	CPR	−0.78	−1.23	−1.70
	ERY	−0.34	−0.70	−1.21
	FOX	−1.87	−1.88	−2.36
	GEN	−0.80	−1.00	−1.68
	LVX	−1.50	−1.50	−1.74
	NTR	−1.72	−1.89	−2.10
	STR	−0.96	−1.24	−1.78
	TET	−1.65	−1.70	−1.92
	TMP	−1.90	−1.91	−2.59
	TOB	−1.08	−1.39	−2.10
	no drug	−0.96	−1.17	−2.67
lhculex		Culex pipiens
DIC	trait	lit. function	lit. model	flexTPC
	EV	quadratic	105.7	47.4
	pLA	quadratic	−17.3	−14.8
	MDR	Briere1	−313.9	−312.8
	lf	linear	280.7	280.0
		Culex quinquefasciatus
	trait	lit. function	lit. model	flexTPC
	EV	Briere1	979.0	370.2
	pLA	quadratic	−42.6	−39.2
	MDR	Briere1	−152.0	−149.5
	lf	linear	110.3	113.5