Skip to main content
. Author manuscript; available in PMC: 2012 Dec 17.
Published in final edited form as: IEEE/ACM Trans Comput Biol Bioinform. 2008 Apr-Jun;5(2):291–300. doi: 10.1109/TCBB.2007.1074

Fig. 2.

Fig. 2

Results of leave-one-out cross validation on R5X4 versus R5 or X4 corecptor usage prediction. (a) Ability of the best neural networks to discriminate between D-tropic samples (0) and R5 or X4 samples (1). (b) ROC curve showing predictive performance on the held-out samples. The best probability of detection (P(D)) of 0.80 is achieved with probability of false alarm (P(FA)) of slightly > 0.2 (area under the ROC curve A(z) = 0.760).