Table 3.

Predictive performance on the TESTlarge dataset

	Predictors	Material production (MF)			Purification (PF)			Crystallization (CF)			Diffraction-quality crystallization (CR)
		Average	±std	p-value	Average	±std	p-value	Average	±std	p-value	Average	±std	p-value
AUC	fDETECT	0.63	±0.05		0.65	±0.05		0.54	±0.05		0.62	±0.04
	Crysalis	0.62	±0.05	0.269	0.58	±0.05	<0.001	0.55	±0.05	0.234	0.60	±0.04	<0.001
MCC	fDETECT	0.11	±0.07		0.19	±0.07		0.12	±0.09		0.19	±0.07
	Crysalis	0.10	±0.08	0.011	0.10	±0.07	<0.001	0.03	±0.08	<0.001	0.16	±0.06	<0.001
Accuracy	fDETECT	75.3	±2.0		74.2	±2.3		66.9	±3.4		59.7	±3.3
	Crysalis	74.8	±2.1	0.012	70.9	±2.3	<0.001	63.6	±2.9	<0.001	58.1	±3.2	<0.001

We report average AUC, MCC and accuracy and their corresponding standard deviations over 100 bootstrap tests (each test is based on 25% of randomly chosen proteins). Statistical significance of differences between fDETECT and Crysalis was measured with paired t-test; the measured values are normal, which we verified based on the Anderson-Darling test at 0.05 significance. The best results that are not significantly different with each other (p-value >0.05) for each outcome are given in bold font