Alternative talk for structural genomics

1
Alternative talk for structural genomics
Note these are conversions of the slides that I
actually presented
2
Computational biology bioinformatics _at_NESG
3
We can infer 3D similarity from sequence. We
cannot systematically predict novel folds.
4
Inferring 3D similarity from sequence

• we can infer 3D similarity from sequence!
• below line ??
• 30 no good for long proteins!

lt2.5Å rmsd over 80 of domain
gt6Å rmsd
C Sander R Schneider 1991 Proteins, 9, 56-68 B
Rost 1999 Prot Engng, 12, 85-94
5
Build sequence-structure families
6
Defining sequence-structure families
7
EVA comparative modelling
Marc Marti-Renom Andrej Sali (UCSF) http//eva.c
ompbio.ucsf.edu/eva/cm/ http//cubic.bioc.columbi
a.edu/eva
Accuracy
Coverage
V Eyrich, MA Marti-Renom, D Przybylski, A Fiser,
F Pazos, A Valencia, A Sali B Rost (2001)
Bioinformatics 17, 1242-3 MA Marti-Renom, MS
Madhusudhan, A Fiser, B Rost, A Sali (2002)
Structure 10, 435-440
8
How much do we cover today
9
Structural residue coverage in reality
J Liu B Rost 2002 Bioinformatics, 18, 922-933
10
Pan-genomic target selection Need to chop into
domains!
11
Scooping families from proteomes, in practice

• Goal for sequence-structure clusters
• All in cluster share fold
• All with similar sequence and similar fold in
  same cluster
• Problems
• Domains
• Overlap

12
Choose targets single-linkage clustering
100,000 eukaryotic proteins (yeast, fly, worm,
weed, human) 22 112 clusters 46 318 in largest
cluster NONSENSE!

• no cluster on full-length proteins we
  MUST chop into structural domain-like fragments

Liu, Hegyi, Acton, Montelione Rost 2003
Proteins, in press Liu Rost 2004 Proteins, 55,
678-686
13
CHOP proteins into structural domains
Liu Rost 2004 Proteins, 55, 678-686
14
Structural domain-like fragments for entire
proteomes
Single-domain proteins 61 in PDB 28 in 62
proteomes
Liu, Hegyi, Acton, Montelione Rost 2003
Proteins, 56, 188-200 Liu Rost 2004 Proteins,
55, 678-686
15
How many clusters? (3,000-16,000) Prokaryotes
enough? (yes)
16
To take or not to take
Take if gt 50 globular residues and no known 3D
17
Renormalize today0, end100
Today0 line
PSI3100 line
J Liu B Rost 2002 Bioinformatics, 18, 922-933
18
Structural genomics will make a difference
0 today 100 what remains to be done
residues
fragments
proteins
19
Pfam 5000 vs. use all
residues
fragments
proteins
20
Eukaryotes needed to complete coverage?
How many proka only? How many euka only? How many
mixed?
21
We cannot only do prokaryotes!
22
Adolescent already successful!
23
Consortia in adolescence successful!
24
Target selection really successful!
StrX
PDB all
25
Significant contribution to unique!
Every third unique protein from Structural
Genomics
26
Does multiplexing help?
Date 2003-07-28
4
Multiplex DOUBLES success rate!
27
Structural Genomics changes ... e.g. our map of
sequence space
28
Dynamic reorganization through 3D!
29
Surprising conclusions

• Most proteins multi-domain
• Most domains about 100 residues long
• 10,000 sequence-unique structures double the
  structural coverage
• BUTStructural genomics can do that before 2010
• Structural Genomics has already changed the way
  we organize structure space, more to come

30
(No Transcript)
31
Xhttp//www.rostlab.org

• PredictProtein PP X/predictprotein/
• META-PP X/meta/submit_meta.html
• EVA X/eva/
• services X/services/
• LOCnet X/services/locnet/
• PredictNLS X/predictNLS/
• ISIS X/services/isis/
• databases X/db/
• PEP X/db/PEP/
• CellCycleDB X/db/cellcycledb/
• NMPdb X/db/nmpdb/