Skip to main content

Table 4 Results over the final EVA test

From: PCI-SS: MISO dynamic nonlinear protein secondary structure prediction

  Avg per residue Avg per chain
Method CC Q 3 BAD CC Q 3 BAD SOV
PHD 0.619 76.09 2.65 0.631 75.96 2.82 71.4
PHDpsi 0.619 76.09 2.65 0.631 75.96 2.82 71.4
PROF_king 0.577 72.65 3.70 0.594 72.86 3.38 66.8
PROFsec 0.651 77.72 2.36 0.659 77.70 2.49 75.3
PSIPRED-live (UniREF100 DB) 0.668 79.02 2.01 0.675 78.86 2.12 76.1
Sable 0.633 76.74 2.62 0.634 76.50 2.74 73.9
Sable2 0.651 77.95 1.86 0.644 77.45 2.05 73.0
SCRATCH (SSPro3) 0.616 76.07 3.24 0.622 76.15 3.20 70.6
SSPro4 0.643 77.69 2.43 0.642 77.58 2.46 72.0
Cascaded PCI 0.632 76.45 2.70 0.624 76.31 2.69 72.0
PSIPRED-local (frozen nr DB) 0.676 79.44 2.13 0.658 79.36 2.20 75.5
post-PCI(PCI &PSIPRED) 0.682 79.58 1.60 0.656 79.37 1.68 76.4
Bad-Score-Rule 0.679 79.59 1.98 0.659 79.45 2.10 75.7
  1. Results over the final EVA test set of 125 new protein chains dissimilar to all training data. CC denotes the average Matthews' correlation coefficient observed for the three classes. 'Avg per residue' results are calculated over the pool of all residues in the dataset whereas 'Avg per chain' results are compiled for each chain prior to computing the average. PSIPRED-local refers to the output of PSIPRED v2.45 run locally when provided with PSSM data generated from the filtered NCBI non-redundant nr database as frozen on 3 May 2004. PSIPRED-live indicates the performance of the actual PSIPRED server [31] as of the day that each protein chain was added to the EVA system. Bad-Score-Rule in the last row shows the results when a rule is applied to combine postPCI with PSIPRED to optimize Q3 score at the cost of BAD score (see text for details).