Expected prediction accuracy of PHD

Figure captions:

• Dataset: Cross-validation experiment on 705 sequence unique protein chains (largest sequence unique subset of PDB in May, 1996, i.e., no two proteins in that set have more than 25% pairwise identical residues).
  
• Fig. 1: Distribution of prediction accuracy for per-residue accuracy (percentage of residues predicted correctly in either of the three states helix, strand, rest).
  
  • Fig. 1a: Subset of proteins with less than 100 residues.
  • Fig. 1b: Subset of proteins with less than 200 residues.
  • Fig. 1c: Subset of proteins with more than 200 residues.
• Reliability of correctly predicting residues.
  Given is the expected per-residue accuracy for residues with a reliability index (RI) above a given cutoff.
  • Fig. 2a: Overall per-residue accuracy (three states: H, E, L). For example, a level of accuracy comparable to homology modelling is reached for 48% of all residues by PHDsec (RI >6).
  • Fig. 2b: Per-residue accuracy for correctly predicting helix, strand, and other
    Percentage of observed: number of correctly predicted residues in either state (H,E,L) / number of residues observed in that state (H,E,L)
  • Fig. 2c: Per-residue accuracy for correctly predicting helix, strand, and other
    Percentage of predicted: number of correctly predicted residues in either state (H,E,L) / number of residues predicted in that state (H,E,L)
• Reliability of correctly predicting accessibility.
  Fig. 2d: Overall per-residue accuracy (two states: buried/exposed). For example, a level of accuracy comparable to homology modelling is reached for 50% of all residues by PHDsec (RI >2).
• Fig. 3: Distribution of prediction accuracy for the per-segment accuracy (percentage of overlapping segments, defined in Rost et al., 1994).
  
• Fig. 4: Distribution of prediction accuracy for 'bad' predictions, i.e. residues observed in helix and predicted in strand or vice versa.