Grid-based prediction of torsion angle probabilities of protein backbone and its application to discrimination of protein intrinsic disorder regions and selection of model structures

Table 3 Accuracy for four angles, 5° for each bin, using different combinations of features groups in M2 on training dataset TR4590 with 10-fold cross validation. The number in parentheses is the accuracy of matching the native angles to one of the top five predicted angle bins

Method	φ (Top 5)	ψ (Top 5)	θ (Top 5)	τ (Top 5)
Angles-based features(Angles)^a	0.200(0.629)	0.183(0.608)	0.374(0.823)	0.180(0.542)
Structure-based features(Struct)^b	0.193(0.602)	0.176(0.583)	0.363(0.804)	0.174(0.521)
PSSM-based features(PSSM)^c	0.188(0.588)	0.168(0.555)	0.353(0.784)	0.167(0.493)
Angles+PSSM	0.202(0.633)	0.186(0.613)	0.377(0.826)	0.184(0.545)
Angles+Struct	0.201(0.632)	0.185(0.611)	0.376(0.825)	0.182(0.544)
PSSM+Struct	0.198(0.622)	0.183(0.603)	0.373(0.819)	0.180(0.534)
All features of M2 model	0.203(0.636)	0.187(0.616)	0.379(0.828)	0.185(0.547)

^apredicted angle feature group (φ and ψ angles and Cα-atom-based angle θ and rotational angle τ). ^b Structure-based feature group: predicted secondary structure probability, relative solvent accessibility, half-sphere exposure, and contact numbers. ^c PSSM based feature group: the features from PSSM profile

ISSN: 1471-2105