Homology ModelingSlide 2Sequence AlignmentThreadingMODELLER (Sali)Side-chain re-packingSCWRL 3.0Loop ModelingSlide 9ScoringDFIRE (Yaoqi Zhou)MinimizationHomology Modeling•comparative modeling vs. ab initio folding•alignment (check gaps)•threading•loop building•re-packing side-chains in core, DEE, SCWRL•fold evaluation/scoring •statistical potentials (pot. of mean force) - DFIRE •minimization•servers: Swiss-model•ab initio folding –Rosetta (Baker)–MONSSTER (Skolnick)–I-TASSER (Zhang)Sequence Alignment•critical step•gaps should be in loops (check in final model)•dynamic programming (Smith-Waterman)–LALIGN: http://www.ch.embnet.org/software/LALIGN_form.html–adjust gap parameters: gap-open penalty>x?gap-extension penalty<x? x=average match score–could also adjust substitution matrix (PAM250, BLOSUM62)•use PSI-Blast to include info from homologs–iterative: retrieves homologs, refines search...•use HMM to align to familyThreading•use info about 3D structure to improve alignment•local secondary structure, solvent-accessibility•3D profiles (Eisenberg)•3D-PSSM/Phyre (Sternberg, Lawrence Kelley)•THREADER•RAPTORMODELLER (Sali)•references–A. Šali and T. L. Blundell. Comparative protein modelling by satisfaction of spatial restraints. J. Mol. Biol. 234, 779-815, 1993.–A. Fiser, R. K. G. Do and A. Š ali. Modeling of loops in protein structures. Protein Science 9, 1753-1773, 2000. –Fiser A, Sali A. (2003). Modeller: generation and refinement of homology-based protein structure models. Methods Enz. 374:461-91.•loop-modeling via dynamics •evaluation: –>30% identity?–stereochemistry: Procheck –contacts/exposure: ProSA (Sippl, 1993) – distance-based pair potentialsSide-chain re-packing•mutations cause steric conflicts (and voids)–changing rotamers can relieve conflicts –adjacent side-chains are coupled–multiple changes might be required–combinatorial search: exhaustive versus Monte Carlo (Holm & Sander, 1992)•DEE (Dead-End Elimination) –pruning method –pre-processing, singles, pairs–Desmet, Mayo–reduction in branching factor?• interesting application: use DEE todetermine rotamer populations for tryptophans; use to predict fluorescence quenching times (Hellings 2003, BiophysJ)•rigid backbone assumption–how important is backbone flexibility?–also sample alternative backbone conformations at each site –(Georgiev and Donald, 2007)SCWRL 3.0•Canutescu et al. (2003)–Dunbrack BBdep rotamer library–de-couple interaction graph into bi-connected componentsrepresenting local dependencies•TreePack (Xu and Berger, JACM 2006)–geometric neighborhood graph decomposition; up to 90x fasterside-chain interactions:energy of configuration:Loop Modeling•two approaches:1. MD/conformational sampling2. templates from loop library•accuracy depends on length: 2-4 (turns), 4-8, >8 (ab initio) •importance in immunoglobulins (hyper-variable loops in antigen-binding region)•modeling loops via molecular dynamics –Monte Carlo conformational search using a FF/energy function, high temp MD: 800K (Bruccoleri & Karplus, 1990)–“Does Conformational Free Energy Distinguish Loop Conformations in Proteins?”•templates from loop library (examples from existing structures)–amino acid similarities–fit to stems: •C distance, vectors (i-1:i,j,j+1), carbonyls?, anglesScoring•statistical potentials –knowledge-based poten (Sippl, 1990)–potential of mean force–residue-based potentials (e.g. C-C contact distance, or centers-of-mass)•atomic pairwise potentials –(Lu & Skolnick, 2001)–capture side-chain interactions better–discriminate correct folds better–z-score of true fold vs. decoys (gapless threading)DFIRE (Yaoqi Zhou)•Distance-scaled Finite Ideal-gas REference state –Zhou and Zhou (Prot. Sci, 2002)–all-atom potential–Nexp(i,j,r) will not increase in r2 as in an infinite system– =1.57 gives best correlation with density in radial shells–improves ability to recognize correct fold versus decoys–see also: •RAPDF (Samudrala and Moult, 1998)•DOPE (Shen and Sali, 2006)fair???instead, assume:Minimization•a logical step, however...•one of the conclusions from CASP4 (Baker):–minimization generally made models worse (took predicted structures farther from native)–threshold: •minimization works if rmsd<2Å, •but ab initio models are often 4-6Å rmsd–backbone adjustments
View Full Document