tweaking structures working on the fiddly bits
play

Tweaking structures: working on the fiddly bits Kevin Karplus - PowerPoint PPT Presentation

Tweaking structures: working on the fiddly bits Kevin Karplus karplus@soe.ucsc.edu Biomolecular Engineering Department University of California, Santa Cruz Tweaking structures p.1/12 Outline of Talk Cost functions Clash detection


  1. Tweaking structures: working on the fiddly bits Kevin Karplus karplus@soe.ucsc.edu Biomolecular Engineering Department University of California, Santa Cruz Tweaking structures – p.1/12

  2. Outline of Talk Cost functions Clash detection Sidechain quality H bond quality Predicted backbone structure Conformation change operators JiggleSegment JiggleSubtree TweakHbond TweakPsiSegment TweakPsiSubtree TweakPhiSegment TweakPhiSubtree TweakPeptide TweakPsiPhiSegment TweakPsiPhiSubtree BackRub BigBackRub Tweaking structures – p.2/12

  3. Clash Detection Simple clash detector with minimum legal distance for each pair of atoms (no attractive term). Distances trained (badly) on large set of high-quality structures. Soft clash cost: ratio<1? 1-ratio**6 : 0 1 0.8 0.6 cost 0.4 0.2 0 0 0.2 0.4 0.6 0.8 1 dist/min_dist Tweaking structures – p.3/12

  4. Sidechain quality 9-dimensional space (where are CA-1, CA+1, and distal point on sidechain) Gaussian mixture for “rotamer library” Estimates joint probability of backbone and rotamer, not rotamer conditioned on backbone. Redoing our sidechains with SCWRL gives about equal number of better and worse models. Tweaking structures – p.4/12

  5. Our sidechains are not supremely good SAM-T08-server - SAM-T08-server-scwrl (all-atom rmsd) All whole chains 0.15 0.1 SAM-T08-server-scwrl better 0.05 0 -0.05 -0.1 SAM-T08-server better -0.15 1 10 100 SAM-T08-server-scwrl (all-atom rmsd) Tweaking structures – p.5/12

  6. Hbonds Different cost functions for different types of H-bonds. separation=3 separation=4 other separation 4 non-backbone classes Backbone H-bonds most carefully done: distance non-planarity asymmetry donor-acceptor-carbon angle Tweaking structures – p.6/12

  7. Predicted backbone structure Local structure prediction using neural nets Several alphabets: Bystroff’s φ - ψ classification de Brevern’s protein blocks CA-CA-CA-CA α torsion angle Cost function uses log P ( class | neural net ) P ( class | amino acid ) Tweaking structures – p.7/12

  8. Jiggle, TweakHbond JiggleSegment JiggleSubtree: make a small rigid transform. Move either the segment (between chain breaks) a subtree consisting of several segments. TweakHbondSegment or TweakHbondSubtree same idea, but donor or acceptor has very small change to position. other atoms of segment or subtree may move further Tweaking structures – p.8/12

  9. TweakPhi, TweakPsi, TweakPeptide TweakPhiSegment, TweakPhiSubtree, TweakPsiSegment, TweakPhiSegment: Make small change to angle. TweakPsi generally more productive than TweakPhi. TweakPeptide: rotates peptide plane about axis through two CA atoms. Tweaking structures – p.9/12

  10. TweakPsiPhi Segment/Subtree Make small change to psi angle and opposite change to next phi angle. Note: axes of rotations almost parallel on opposite ends of peptide plane. Moves rest of segment (or subtree) along circular arc without reorienting it much. Tweaking structures – p.10/12

  11. BackRub, BigBackRub Rotate chain segment about line between two CA atoms. For Backrub, CA are 2 apart on backbone. Reposition first and last peptide planes to try to get optimal superposition on new CA positions but old N and O positions (to preserve Hbonds). These tweaks tend to have high success rate and so can be done with larger angular changes than other tweaks. Tweaking structures – p.11/12

  12. Web sites CASP8 working files: http://www.soe.ucsc.edu/˜karplus/casp8/ List of my papers: http://www.soe.ucsc.edu/˜karplus/papers/paper-list.html These slides: http://www.soe.ucsc.edu/˜karplus/papers/ tweak-slides-2008-dec.pdf UCSC bioinformatics (research and degree programs) info: http://www.soe.ucsc.edu/research/compbio/ Tweaking structures – p.12/12

Recommend


More recommend