Fig. 3.

Histogram of the change in secondary structure prediction accuracy (Q₃) resulting from inclusion of low-resolution tertiary structure information from the native structure (black bars) or rosetta models (gray bars) in the neural network secondary structure prediction. The change in Q₃ is shown on the horizontal axis, with count (a) or sequence (b) shown on the vertical axis. The average improvement for the independent set of 137 structures (a)is7% and 5%, respectively, and for the combined set of LIVEBENCH6 and CAFASP3 targets (b) is 10% and 4%, respectively.