TABLE 1.

Standard errors of prediction for area-normalized spectra

	RMS	Cross-valid lin1	Cross-valid lin2	Cross-valid quad3	Direct valid	Frequencies in model	Model →% of the secondary structure
α-Helix	21.8	5.9	5.5	5.9	5.4	1545–1655–1613	−179.45 + 2.030 × A1545 + 0.431 × A1655 + 0.828 × A1613
β-Sheet	17.6	7.3	7.0	6.5	6.6	1656–1634–1691	26.31 − 0.451 × A1656 + 0.335 × A1634 + 0.586 × A1691
β-Turns	4.2	3.6	3.7	3.0	3.2	1677–1528–1577	−30.49 + 0.428 × A1676 + 0.322 × A1528 + 0.141 × A1577
Random	11.4	8.8	9.3	8.9	7.9	1544–1627–1691	178.56 − 1.578 × A1544 − 0.332 × A1627 − 0.735 × A1691
310-Helix	3.2	3.1	3.1	3.0	2.9	1631–1694–1625	5.077 − 0.198 × A1631 + 0.440 × A1694 + 0.160 × A1625
Random + turns + 310-helices	12.8	8.2	8.6	8.1	7.3	1549–1629–1513	224.544 − 2.010 × A1549 − 0.345 × A1629 − 0.775 × A1513

RMS, standard deviation of the structure distribution in the whole database. Cross-valid lin¹, standard deviation of the predicted-actual structure content; 50 linear models were built with the cross-validation procedure. Cross-valid lin², the same as the previous column, but only the closest 25 spectra (Euclidian distance) were included in the every model. Cross-valid quad³, standard deviation of the predicted-actual structure content (50 quadratic models were built with the cross-validation procedure). Direct valid, standard deviation of the prediction compared with the actual structure content; one linear model was built including all 50 proteins (only one model including the 50 proteins was built). Frequencies in model, the best three frequencies in the linear model. Model, the equation yielding the secondary structure content in percentage corresponding to the Direct valid column. The equations report the constants found in direct mode since, in cross-validation mode, there are as many equations as there are spectra in the database. Note that the best value for the linear model between Tables 1 and 2 is in boldface type.