Transcript CS689-domains - TAMU Computer Science Faculty Pages

Domains
• typical size ~100-200 amino acids
– mean=160 residues
– balance surface area to volume (hydrophobics in core)
• modularity, though insertions are possible
• a, b, a+b, a/b (“wound” bab, parallel strands)
• classic folds: globins, immunoglobulins, TIMbarrels, NBDs
• beta-sandwiches/clamshells, helix-bundles
• helical coiled-coils (collagen, lambda repressor)
• beta-barrels, beta-propellers
• domain insertions
– (the strange case of malate synthase...)
see 4 domains on CATH page:
http://www.cathdb.info/chain/1n8wA
C-terminal cap
active site in
mouth of
beta-barrel
• How small can a protein be and still have
structure?
–
–
–
–
–
–
–
no hydrophobic core
glucagon (30 res, a-helix); dis-ordered in solution
unraveling, conformational sampling
NMR studies of peptides?
10-aa SCF recognition peptide
disorder of p53 fragment in soln by NMR
on the contrary, 17-residue fragment from N-terminal
domain of ubiquitin folds into beta-hairpin on its own
• Zarella et al, Protein Science, 1999
Structure Superposition Algorithms
• least-squares
– Aij =∑PkiQkj – product over 2 sets of coords, P and Q
– R = (AtA)1 / 2A−1 – rotation that minimizes RMSD
– assumes translated to centers-of-mass
• Kabsch rotation algorithm (1976, Acta) – (equiv. to SVD)
Lij are Lagrange multipliers.
determine by solving:
let mij be eigenvalues
and aij be eigenvectors of RTR:
• MacKay (1984, Acta) – quaternions (solve linear system)
• SSAP (Orengo and Taylor)
– dynamic programming to minimize inter-molecular distance vectors
between Cb atoms
– pairs must be known a priori
• DALI (Holm and Sander)
– aligns scalar distance plots
– significance: z-scores:
z=(s-m)/s>7.0
– compare to scores from random
alignments
– beware of effect of length of
aligned/rejected; shorter->better
score
• VAST (Gibrat and Bryant)
– aligns secondary structure elements
– graph theory algotrithm – finds maximal clique in graph of
consistent alignable pairs of vectors
• LOCK (Singh and Brutlag)
– hierarchical, distances + SS elements
• rigid bodies can’t always be aligned well
• CE (combinatorial extension; Shindalyov&Bourne)
– identifies similar local fragments (3-5aa), extends them
– more tolerant of flexible regions
• SSM (Krissinel and Henrick)
– subgraph isomorphism
• must preserve topology?
LOCK (Singh and Brutlag)
Fold Families
•
•
•
•
•
•
clustering
PDBSelect and COG are based on homology only
FSSP - based on DALI score
SCOP – manually curated (by Alexy Muzrin)
CATH (Orengo and Thornton)
Pfam – based on HMMs (more details later)
SCOP (Sep 2007)
Number of
folds
Number of
superfamilies
Number of
families
All alpha proteins
259
459
772
All beta proteins
165
331
679
Alpha and beta proteins (a/b)
141
232
736
Alpha and beta proteins (a+b)
334
488
897
Multi-domain proteins
53
53
74
Membrane and cell surface
proteins
50
92
104
Small proteins
85
122
202
1086
1777
3464
Total
(beware of large-family bias when averaging over protein database)
Fold Recognition
• sequence alignment (homology)
– position-dependent profiles from multiple alignment
(Gribskov, McLachlan, Eisenberg, 1987), scores based
on sum of Dayhoff similarity over observed residues at
each pos.
• 3D profiles
• threading
• HMMs
• Convergence vs. Divergence
Sander and Schneider (1991) Database of Homology-Derived Protein Structures and
the Structural Meaning of Sequence Alignment.
Chothia, C. (1993). One thousand families for the molecular biologist.
3D Profiles (Eisenberg et al.)
• Given that you have a sequence threaded onto a
known structure, how well does it fit the fold?
– originally: residues scored by 18 environment classes
(Bowie, Luthy, Eisenberg, 1991)
– similarity of amino acids in model to structure
(homology, position-dependent distribution)
– tolerance of buried vs. surface exposure
– suitability of residues in secondary structures
– residue pair potentials (likelihood of contacts at 4-10A
radius shells) (Wilmanns and Eisenberg, 1993)
18 environment classes =
{E,P1,P2,B1,B2,B2}x{helix,sheet,coil}
Threading (for Fold Recognition)
• find optimal mapping of residues in sequence to model
• higher computational complexity that sequence alignment,
or can also be done by dynamic programming?
• Lathrop (Prot Eng, 1994; JMB, 1996) - showed that
threading is NP-complete when non-local effects are taken
into account (reduction to 3SAT)
• fold evaluation:
–
–
–
–
3D profiles
packing (steric conflicts, voids)
energy (molecular mechanics force field)
statistical (side-chain contacts, Sippl)
• PHYRE (Sternberg) – 3D-PSSM search
• THREADER (David Jones) – dynamic programming
• RAPTOR (Jinbo Xu) – integer programming (constraints)
Pfam, Hidden Markov Models (HMMs)
(Sonnhammer, Eddy, and Durbin, 1997)
Viterbi algorithm (forward/backward)
training: maximum likelihood, EM
HMM for
628 globins
(lines
indicate
most
frequentlyused
transitions)
1YBA – PDGH tetramer
1BEF - protease
Linkers
• definition:
4FAB - immunoglobulin
– do not pack against well-defined domains (lack
contact; not necessarily exposed, though)
– can’t count on sequence between known domains
– flexible, lack regular secondary structure (not always
coil; helical linkers exist)
– rich in Pro, Ala, charged residues; lack of Gly
• George and Heringa (2002)
• Bae, Mallick, Elsik (2005) – HMM (accuracy ~ 67%)
• Tanaka, Yokayama, Kuroda (2006) – length dependence
– significant frequency deviations were observed for glycine, proline,
and aspartic acid in short linker and nonlinker loops, whereas
deviations were observed for aspartic acid, proline, asparagine,
and lysine in long linker and nonlinker loops.
all fragments
length <= 9 aa
length > 9 aa
• DomCut (Suyama & Ohara, 2003)
– uses differences in amino acid composition
between the intra- and interdomain regions
to predict domain boundaries
• Armadillo (Dumontier et al., 2005)
– local smoothing of aa propensity index by
FFT; calculates Z-score