1. The Protein Data Bank (PDB)
PDB is the principal repository for protein structures
Established in 1971
Accessed at or simply
Currently contains over 32,000 structure entities
Updated 9/05
Page 287

2. PDB content growth (www.pdb.org)
structures
year
Fig. 9.6
Page 281

3. PDB holdings (September, 2005)
29,876 proteins, peptides
1,338 protein/nucl. complexes
1,500 nucleic acids
13 carbohydrates
32,727 total
Table 9-2
Page 281

4. Protein Data Bank gateways to access PDB files Swiss-Prot, NCBI, EMBL
CATH, Dali, SCOP, FSSP
databases that interpret PDB files
Fig. 9.10
Page 285

5. Access to PDB through NCBI
You can access PDB data at the NCBI several ways.
Go to the Structure site, from the NCBI homepage
Use Entrez
Perform a BLAST search, restricting the output
to the PDB database
Page 289

6. Access to PDB through NCBI
Molecular Modeling DataBase (MMDB)
Cn3D (“see in 3D” or three dimensions):
structure visualization software
Vector Alignment Search Tool (VAST):
view multiple structures
Page 291

7. Fig. 9.15
Page 290

8. Fig. 9.15
Page 290

9. Fig. 9.16
Page 291

10. Fig. 9.16
Page 291

11. Fig. 9.16
Page 291

12. Fig. 9.16
Page 291

13. Fig. 9.16
Page 291

14. Fig. 9.17
Page 292

15. Access to structure data at NCBI: VAST
Vector Alignment Search Tool (VAST) offers a variety
of data on protein structures, including
-- PDB identifiers
-- root-mean-square deviation (RMSD) values
to describe structural similarities
-- NRES: the number of equivalent pairs of
alpha carbon atoms superimposed
-- percent identity
Page 294

16. Many databases explore protein structures
SCOP
CATH
Dali Domain Dictionary
FSSP
Page 293

17. Structural Classification of Proteins (SCOP)
SCOP describes protein structures using a
hierarchical classification scheme:
Classes
Folds
Superfamilies (likely evolutionary relationship)
Families
Domains
Individual PDB entries
Page 293

18. Class, Architecture, Topology, and
Homologous Superfamily (CATH) database
CATH clusters proteins at four levels:
C Class (a, b, a&b folds)
A Architecture (shape of domain, e.g. jelly roll)
T Topology (fold families; not necessarily homologous)
H Homologous superfamily
Page 293

19. SCOP statistics (September, 2005)
Class # folds # superfamilies # families
All a
All b
a/b
a+b …
Total
Table 9-4
Page 298
a/b = parallel b sheets
a+b = antiparallel b sheets

20. Fig. 9.23
Page 298

21. Fig. 9.24
Page 299

22. Fig. 9.25
Page 300

23. Fig. 9.25
Page 300

24. Fig. 9.26
Page 301

25. Fig. 9.27
Page 302

26. Fig. 9.28
Page 303

27. Dali Domain Dictionary
Dali contains a numerical taxonomy of all known
structures in PDB. Dali integrates additional data
for entries within a domain class, such as secondary
structure predictions and solvent accessibility.
Page 302

28. Fig. 9.29
Page 303

29. Fig. 9.30
Page 304

30. Fig. 9.30
Page 304

31. Fig. 9.30
Page 304

32. Fold classification based on structure-structure
alignment of proteins (FSSP)
FSSP is based on a comprehensive comparison of
PDB proteins (greater than 30 amino acids in length).
Representative sets exclude sequence homologs
sharing > 25% amino acid identity.
The output includes a “fold tree.”
Page 293

33. Fig. 9.31
Page 305

34. FSSP: fold tree
Fig. 9.32
Page 306

35. Fig. 9.33
Page 307

36. Fig. 9.34
Page 307

37. Approaches to predicting protein structures
There are about >20,000 structures in PDB, and
about 1 million protein sequences in SwissProt/
TrEMBL. For most proteins, structural models
derive from computational biology approaches,
rather than experimental methods.
The most reliable method of modeling and evaluating
new structures is by comparison to previously
known structures. This is comparative modeling.
An alternative is ab initio modeling.
Page

38. Approaches to predicting protein structures
obtain sequence (target)
fold assignment
comparative
modeling
ab initio
modeling
Fig. 9.35
Page 308
build, assess model

39. Comparative modeling of protein structures
[1] Perform fold assignment (e.g. BLAST, CATH, SCOP);
identify structurally conserved regions
[2] Align the target (unknown protein) with the template.
This is performed for >30% amino acid identity
over a sufficient length
[3] Build a model
[4] Evaluate the model
Page 305

40. Errors in comparative modeling
Errors may occur for many reasons
[1] Errors in side-chain packing
[2] Distortions within correctly aligned regions
[3] Errors in regions of target that do not match template
[4] Errors in sequence alignment
[5] Use of incorrect templates
Page 306

41. Comparative modeling
In general, accuracy of structure prediction depends
on the percent amino acid identity shared between
target and template.
For >50% identity, RMSD is often only 1 Å.
Page 306

42. Fig. 9.36
Page 308
Baker and Sali (2000)

43. Comparative modeling
Many web servers offer comparative modeling services.
Examples are
SWISS-MODEL (ExPASy)
Predict Protein server (Columbia)
WHAT IF (CMBI, Netherlands)
Page 309