Ranking SS Prediction Using CA Overlap Chester Shiu
- Slides: 10
Ranking SS Prediction Using CA Overlap Chester Shiu CS 273 May 31, 2005
Servers ROBETTA ¢ META-Basic ¢ Shotgun-INBGU ¢
ROBETTA ¢ ¢ ¢ Server Implementation of ROSETTA Attempts Homology Modelling, then fills in gap with 3 mers + simulated annealing Does not handle extremes very well!
Why? ¢ ROBETTA performs poorly at extremes Small – domain classification errors l Large – low contact order clustering? l ¢ Errors from poor homology identification and dependence on SA
META-Basic Not a Meta Server! ¢ Meta-Profile ¢ l Sequence AND Structure 6 PSI-BLAST iterations + RPS BLAST ¢ High Specificity ¢
Shotgun-INBGU ¢ Uses Sequence l Multiple Alignment l Profiles of Fold Libraries l Consensus from linear weighing of parameters ¢ Can pick out weak signal ¢
Methodology Pair-wise compare top ranked model from each algorithm. ¢ Select pair with highest score ¢ Rationale: If ROBETTA suffers homology error then other two should outweigh ¢
Scoring c. RMSD ¢ Livebench 3 D Score: ¢ exp(-ln(2)*d*d/(3*3)) l ¢ But only got 2/10 correct Number aligned Cα < 3 A l 9/10 correct!
The Erroneous Ranking 1 rr 9 – ATP-Dependent Protease ¢ c. RMSD roughly equidistant (15. 9 vs 16. 2) ¢ Low Data: 4 versus 3 overlaps at <3 A ¢
Why is 3 D Score so off? ¢ ¢ Lower penalty for high distance than c. RMSD, but still major Sequence alignment issue?
