Fig. 8.
MICS prediction accuracy for five types of β-turns. Prediction performance scores for type I (a and a'), II (b, b'), I′ (c,c'), II′ (d, d') and VIII (e, e') β-turns for all available tetrapeptides obtained by the trained ANN in the validation datasets (see Methods). (a–e) Matthews correlation coefficient (MCC, red), Qpred (green) and Qobs (blue), as a function of the P4k threshold value used for a positive prediction of a type k β-turn. These scores are used to select the threshold (see SI Table S5) for a positive type k β-turn prediction. (a'–e') The numbers of the positively predicted β-turns are plotted (right y-axis, thin line with open squares) as a function of the binned P4k value; the ratio of true type k β-turns relative to the number of predictions (Qpred) (left y-axis, bold line with filled circles) shows the probability that any given tetrapeptide is correctly identified as a type k β-turn, from which fitted polynomials (Eqs. S3–S7) are derived (red solid lines) and used by MICS to convert the P4 scores into actual probabilities.