Transcript erice long talk - WHAT IF servers

Power and weakness
of data
Power: data + software +
bioinformatician = answer.
Weakness: Data errors. Data
poorly understood. Poor
software. Never enough
data. Few bioinformaticians
available.
Laerte about structures:
sequence , Gert
“Use the Force, Luke”
Signals in Sequences
The number of sequences
available for analysis rapidly
approaches infinite.
We need new ways to look
at all this information.
The First Law:
First law of sequence
analysis:
A conserved residue
is important.
With thousands of
aligned sequences:
Second law of sequence
analysis:
A very conserved residue
is very important.
Signals in sequences:
Conserved, CMA, variable
QWERTYASDFGRGH
QWERTYASDTHRPM
QWERTNMKDFGRKC
QWERTNMKDTHRVW
Black = conserved
White = variable
Green = correlated
mutations(CMA)
Sequence Signals
Three types of information
from multiple sequence
alignments:
1) Conservation
2) Correlation
3) Variability
Artefacts
Wrong sequence signals
can result from:
Not enough sequences
Too conserved sequences
Too variable sequences
Over-alignment
Over-interpretation
Recalcitrant residues
Sequence Entropy
20
Ei = S pi ln(pi)
i=1
Sequence Variability
Sequence variability is the
number of residue types
that is present in more than
0.5% of the sequences.
Entropy - Variability
Evolution = try everything
(and keep what works well)
Variability = Chaos
(try everything)
Entropy
= Information
(keep what works well)
Entropy - Variability
Variability is result of
DNA trying everything.
Entropy is the protein’s
break on evolutionary
speed.
Ras Entropy - Variability
11 Red
12 Orange
22 Yellow
23 Green
33 Blue
Ras Location
11
12
22
23
33
Red
Orange
Yellow
Green
Blue
Protease
Entropy - Variability
11 Red
12 Orange
22 Yellow
23 Green
33 Blue
Protease Location
11
12
22
23
33
Red
Orange
Yellow
Green
Blue
Globin
Entropy - Variability
11 Red
GPCR
12 Orange
22 Yellow
23 Green
33 Blue
Globin Location
11
12
22
23
33
Red
Orange
Yellow
Green
Blue
And now for drug design:
GPCR
11 Red
12 Orange
22 Yellow
23 Green
33 Blue
GPCRs: (Membrane facing
amino acids left out)
11
12
22
23
33
Red
Orange
Yellow
Green
Blue
Summary
Given many sequences:
Every residue’s role known.
Signaling paths detectable.
Two step evolutionary
model:
First main site, soon after
modulator site.
Beyond the summary
Sequence -> structure -> function
is wrong. It should be:
Structure -> sequence -> function.
And, because active sites are at
the surface, conserved residues
are at or near the surface.
Beyond the summary
Why do all TIM-barrel
enzymes have the
functional residues at
the C-terminal side of
the strands?
Beyond the summary
23 Green: Modulator
Up to 18 residue types
22 Yellow: Core
Up to 14 residue types
12 Orange: Around main site
12
11
23 33
Up to 8 residue types
22
11 Red: main site
Up to 4 residue types
The weakness of data
Data errors.
Poor software.
Data poorly understood.
Never enough data.
Few bioinformaticians
around.
The weakness of data
WHAT_CHECK
Rob Hooft
www.cmbi.kun.nl/gv/servers/
www.cmbi.kun.nl/gv/pdbreport/
Structure validation
Everything that can go
wrong, will go wrong,
especially with things as
complicated as protein
structures.
Why ?
Why does a sane (?) human
being spend fourteen years
to search for twelve million
errors in the PDB?
Because:
All we know about proteins
is derived from PDB files.
If a template is wrong the
model will be wrong.
Errors become smaller
when you know about them.
What do we check?
Administrative errors.
Crystal-specific errors.
NMR-specific errors.
Really wrong things.
Improbable things.
Things worth looking at.
Ad hoc things.
Error detection
Detecting errors is one
thing
fixing them another…
We try not to say about the
structure that it is wrong,
but we try to say what is
wrong about the structure.
How difficult can it be?
How difficult can it be?
Your best check:
Planarity
Little things hurt big
Improbable things
How wrong is wrong?
Our errors
Four sigma: 12.000 false
positives.
Administrative errors
misunderstood.
Improbable is not wrong.
Poor data makes errors
unavoidable.
Bugs.
Contact Probability
Contact Probability
DACA
DACA
DACA
DACA
DACA
Contact probability box
Using contact
probability
His, Asn, Gln ‘flips’
Where are the protons?
Hydrogen bond network
Hydrogen bond force field
Hydrogen bond force field
15% should be flipped
Summary
Everything that could go wrong
has gone wrong.
Errors are on a ‘sliding scale’.
Error detection can detect a lot,
but surely not everything (yet).
Beyond the summary,
For Drug Design:
Forget: High throughput.
Forget: Docking.
Forget: Structure in
absence of many, many
sequences.
First gather and digest all
Beyond the summary,
For Drug Design:
First know your enemy,
then defeat it.
Thanks to:
Laerte Oliveira
Florence Horn
Francisco
Rob Hooft
Wilma Kuipers
Bob Bywater
Nora vd Wenden
Mike Singer
Ad IJzerman
Margot Beukers
Amos Bairoch
Sao Paulo
San
Delft
Weesp
Copenhagen
The Hague
Boston
Leiden
Leiden
Geneva