Table 2.

Relative Accuracies of Backbone Conformational Distribution Predicted from Sequence^a

	All	PPII	β	αR	β‐I	αL	other
P(X,r)b	1.15	1.06	1.10	1.11	1.33	2.82	1.12
P(L‐X,r)c	1.19	1.08	1.12	1.17	1.43	3.25	1.13
P(X‐R,r)d	1.21	1.09	1.13	1.21	1.48	3.34	1.13
P(L‐X,r)×P(X‐R,r)/P(X,r)e	1.25	1.12	1.15	1.27	1.59	3.91	1.14
P(L‐X‐R,r)f	1.19	1.10	1.13	1.16	1.44	2.87	1.12
P({L}‐X‐{R},r)g	1.22	1.11	1.14	1.20	1.51	3.57	1.13
PANN(L‐X‐R,r)h	1.28	1.11	1.15	1.30	1.70	4.59	1.13

^aAccuracy of the prediction P for residue X to be located in region r of the Ramachandran map, with r = PPII, β, α_R, type I β‐turn (β‐I), α_R and other. The reported value is relative to the probability that any residue, regardless of type or neighbors, is found in that region.

^b P(X,r) is the predicted probability of X being located in r when taking the residue type of X into account, compared to the fractional population of r, regardless of residue type.

^c P(L‐X,r) prediction based on left and center residue.

^d P(X‐R,r) prediction based on center and right neighbor residue.

^ePrediction accuracy when effect of left, center and right residue are considered independently. Note that the denominator term, P(X,r), is needed for normalization.

^f P(L‐X‐R,r) prediction based on fractional occurrence of L‐X‐R tripeptides in region r of the training library.

^g P({L}‐X‐{R},r) prediction based on fractional occurrence of {L}‐X‐{R} tripeptides in region r of the training library, using grouping of left and right neighbor residues.

^h P^ANN(L‐X‐R,r) prediction based on trained artificial neural network.