RAPTOR
Comparison of RAPTOR and
PSI-BLAST in recent CASPs
CASP (Critical Assessment of Techniques for Protein Structure Prediction) is a community wide experiment which is held every two years by NIH. All participatory 3D protein structure prediction servers will be assessed in the experiment. RAPTOR has been an active participant since CASP5 in 2002.
The targets used in CASP have been classified into two groups: Homology Modeling (for easy targets) and Fold Recognition (for hard targets). Servers are evaluated for each group.
The sum of the MaxSub scores for targets are listed below, measuring the accuracy of a server. The number of correctly recognized targets is also compared in the following tables.
| Homology Modeling Targets | ||
| Server | MaxSub Score | Correctly Predicted Targets |
| RAPTOR | 14.85 | 32 |
| PSI-BLAST | 18.56 | 32 |
| Fold Recognition Targets | ||
| Server | MaxSub Score | Correctly Predicted Targets |
| RAPTOR | 3.98 | 13 |
| PSI-BLAST | 0 | 0 |
As you can see, in CAFASP3, for homology modeling targets, RAPTOR and PSI-BLAST had the same number of correctly predicted targets. PSI-BLAST demonstrated better accuracy here than RAPTOR, as RAPTOR was designed for protein structure prediction, opposed to identifying known structures. The score here was also attributed to RAPTOR using a subset template library of PDB, ideal for unknown structure prediction, while PSI-BLAST searches the whole PDB, effective only for known structures.
For the fold recognition targets, RAPTOR correctly predicted 13 targets while PSI-BLAST predicted none. There was no significant sequence homology in this test, thus explaining why PSI-BLAST failed. RAPTOR still utilizes both sequence homology and structure homology when performing threading. RAPTOR uses structure homology to scan the template library to find many credible templates for the targets.
| Homology Modeling Targets | ||
| Server | MaxSub Score | Correctly Predicted Targets |
| RAPTOR | 19.92 | 34 |
| PSI-BLAST | 18.59 | 34 |
| Fold Recognition Targets | ||
| Server | MaxSub Score | Correctly Predicted Targets |
| RAPTOR | 6.34 | 19 |
| PSI-BLAST | 0.84 | 3 |
In CASP6, RAPTOR and PSI-BLAST had the same number of correct predictions for homology modeling. More importantly, RAPTOR turned out to have better accuracy than PSI-BLAST for these targets. This improved accuracy can be explained by the fact that the subset template library used by RAPTOR had increased significantly due to the increased size of the PDB.
Once again, in the fold recognition target competition, RAPTOR outperformed PSI-BLAST and had 19 correct predictions while PSI-BLAST had only 3.
The above tables show that for homology modeling targets RAPTOR and PSI-BLAST display fairly equivalent results, while for fold recognition targets, RAPTOR outperforms PSI-BLAST significantly.
Essentially, if you are performing tasks that require fold recognition, use RAPTOR. It provides superior protein structure prediction. Blast results generally fail, as they provide less than 25% accuracy, in situations where a structure is predicted.
