. Author manuscript; available in PMC: 2018 Sep 1.

Published in final edited form as: Hum Mutat. 2017 Jun 27;38(9):1109–1122. doi: 10.1002/humu.23267

Table 1.

Metrics of prediction performance for NAGLU and SUMO-ligase

Challenge^a		Prediction			Positive Control^b			Negative Control^c
Challenge^a		RMSD	Pearson's r	Spearman's rho	RMSD	Pearson's r	Spearman's rho	RMSD	Pearson’s r	Spearman’s rho
NAGLU		0.31	0.55	0.57	0.12	0.95	0.94	0.42	0.45	0.48
NAGLU w/o outliers		0.24	0.71	0.71	0.14	0.92	0.93	0.39	0.53	0.57
SUMO - Ligase Set 1	No Map	0.55	0.39	0.46	0.24	0.91	0.92	0.59	0.30	0.38
SUMO - Ligase Set 1	Mapped	0.55	0.39	0.46	0.24	0.91	0.92	0.59	0.30	0.38
SUMO -Ligase Set 2	No Map	0.63	0.35	0.46	0.25	0.90	0.89	0.57	0.31	0.39
SUMO -Ligase Set 2	Mapped	0.56	0.33	0.46	0.25	0.90	0.89	0.57	0.31	0.39
SUMO -Ligase Set 3	No Map	0.57	0.21	0.20	0.26	0.89	0.82	0.57	0.24	0.22
SUMO -Ligase Set 3	Mapped	0.59	0.18	0.20	0.26	0.89	0.82	0.57	0.24	0.22

In the SUMO-ligase challenge, there are two prediction sets, submission 1 with scaled prediction scores (No Map) and submission 2 (Mapped) with each predicted value mapped to the experimental value of closest rank.

In the positive control, the expected difference between experiment and prediction is estimated from the reported experimental errors. That is, a perfect prediction method could not be more accurate than this. See MATERIALS AND METHODS.

In the negative control, a prediction score was computed for each mutation based on amino acid frequency information only, using the equation described in MATERIALS AND METHODS. The resulting prediction scores were mapped to the experimental value of closest rank.