Table 2.

The best models for PSSM-based features based on a subset with 1000 training data points

PSSM features	Sensitivity	Specificity	Precision	F1
PSSM[i − 2, i + 2]	0.29	0.89	0.37	0.32
(Features 281–380)
PSSM[i − 3, i + 3]	0.28	0.93	0.48	0.35
(Features 261–400)
PSSM[i − 4, i + 4]	0.30	0.94	0.54	0.38
(Features 241–420)
PSSM[i − 5, i + 5]	0.32	0.93	0.52	0.40
(Features 221–440)

The feature numbers, as listed in Supplementary Table S3, are given in parentheses. These groups are nested so that the second group contains the first, and so on, up to the last group, which consists of all PSSM-based features. The predictive performance was comparable among the different groups, and although the inclusion of larger scoring windows improved performance somewhat, the improvement was statistically insignificant.