powerpoint file available: ~jamesh/powerpoint/ oslo_2010.ppt
TRANSCRIPT
PowerPoint File available:
http://bl831.als.lbl.gov/
~jamesh/powerpoint/
Oslo_2010.ppt
http://ucxray.berkeley.edu/~jamesh/elves
Download Elves from:
Advanced Light Source
Beamline 8.3.1 staff
Acknowledgments
George MeigsGeorge Meigs
Jane Tanamachi
ALS 8.3.1 Team
Acknowledgements
8.3.1 PRT: Jamie CateCenter for Structure of Membrane Proteins
Membrane Protein Expression Center IICenter for HIV Accessory and Regulatory Complexes
W. M. Keck FoundationPlexxikon, Inc.
M D Anderson CRCUniversity of California Berkeley
University of California San FranciscoNational Science Foundation
University of California Campus-Laboratory Collaboration GrantHenry Wheeler
The Advanced Light Source is supported by the Director, Office of Science, Office of Basic Energy Sciences, Materials Sciences Division, of the US Department of Energy under contract No. DE-AC02-05CH11231 at Lawrence Berkeley National Laboratory.
Ken Frankel Chris Neilson Michael Blum Joe Ferrara
Elves examine images andset-up data processing
Elves run…
mosflmscalasolve
mlpharedm
arp/warp
Elven Automation
Elven Automation
Elves examine images andset-up data processing
Elves run…
mosflmscalasolve
mlpharedm
arp/warp
Elven Automation
Elves examine images andset-up data processing
Elves run…
mosflmscalasolve
mlpharedm
arp/warp
Conversational User Interface
user input ->
Conversational User Interface
user input -> process the data in /data/semet
Conversational User Interface
user input -> process the data in /data/semet
recognition blah blah blah blah /data/semet
Conversational User Interface
user input -> process the data in /data/semet
recognition blah blah blah blah /data/semet
simplify /data/semet
Conversational User Interface
user input -> process the data in /data/semet
recognition blah blah blah blah /data/semet
simplify /data/semet
preempt /data/semet contains image files
Conversational User Interface
user input -> process the data in /data/semet
report Elves will process from /data/semet/frame001.img to /data/semet/frame100.img Data were collected from 0º to 100º in 1º steps with an ADSC Q4 detector 100 mm from the crystal using 1.54 Å x-rays.
Conversational User Interface
user input -> process the data in /data/semet
report Elves will process from /data/semet/frame001.img to /data/semet/frame100.img Data were collected from 0º to 100º in 1º steps with an ADSC Q4 detector 100 mm from the crystal using 1.54 Å x-rays.
confirm Everything look okay? [Yes] ->
Conversational User Interface
user input -> process the data in /data/semet
report Elves will process from /data/semet/frame001.img to /data/semet/frame100.img Data were collected from 0º to 100º in 1º steps with an ADSC Q4 detector 100 mm from the crystal using 1.54 Å x-rays.
confirm Everything look okay? [Yes] ->user input distance was more like 110
Conversational User Interface
user input -> distance was more like 110
Conversational User Interface
user input -> distance was more like 110
recognition distance blah blah blah 110
Conversational User Interface
user input -> distance was more like 110
recognition distance blah blah blah 110
simplify DISTANCE 110
Conversational User Interface
user input -> distance was more like 110
report Elves will process from /data/semet/frame001.img to /data/semet/frame100.img Data were collected from 0º to 100º in 1º steps with an ADSC Q4 detector 110 mm from the crystal using 1.54 Å x-rays.
Conversational User Interface
user input -> distance was more like 110
report Elves will process from /data/semet/frame001.img to /data/semet/frame100.img Data were collected from 0º to 100º in 1º steps with an ADSC Q4 detector 110 mm from the crystal using 1.54 Å x-rays.
confirm Everything look okay? [Yes] ->
Conversational User Interface
user input -> distance was more like 110
report Elves will process from /data/semet/frame001.img to /data/semet/frame100.img Data were collected from 0º to 100º in 1º steps with an ADSC Q4 detector 110 mm from the crystal using 1.54 Å x-rays.
confirm Everything look okay? [Yes]user input -> wavelength is wrong
Conversational User Interface
user input -> wavelength is wrong
Conversational User Interface
user input -> wavelength is wrong
recognition wavelength = -NOT
Conversational User Interface
user input -> wavelength is wrong
recognition wavelength = -NOT
simplify WAVELENGTH
Conversational User Interface
user input -> wavelength is wrong
recognition wavelength = -NOT
query What is the x-ray wavelength? [1.54Å] ->
simplify WAVELENGTH
Conversational User Interface
user input -> wavelength is wrong
recognition wavelength = -NOT
query What is the x-ray wavelength? [1.54Å]user input -> 1
simplify WAVELENGTH
Conversational User Interface
user input -> wavelength is wrong
query What is the x-ray wavelength? [1.54Å]user input -> 1
Conversational User Interface
user input -> wavelength is wrong
report Elves will process from /data/semet/frame001.img to /data/semet/frame100.img Data were collected from 0º to 100º in 1º steps with an ADSC Q4 detector 110 mm from the crystal using 1 Å x-rays.
query What is the x-ray wavelength? [1.54Å]user input -> 1
Conversational User Interface
user input -> wavelength is wrong
report Elves will process from /data/semet/frame001.img to /data/semet/frame100.img Data were collected from 0º to 100º in 1º steps with an ADSC Q4 detector 110 mm from the crystal using 1 Å x-rays.
query What is the x-ray wavelength? [1.54Å]user input -> 1
confirm Everything look okay? [Yes] ->
Conversational User Interface
user input -> wavelength is wrong
report Elves will process from /data/semet/frame001.img to /data/semet/frame100.img Data were collected from 0º to 100º in 1º steps with an ADSC Q4 detector 110 mm from the crystal using 1 Å x-rays.
query What is the x-ray wavelength? [1.54Å]user input -> 1
confirm Everything look okay? [Yes]user input -> Yes
Major Phasing techniques
• Molecular Replacement
• Multiple Isomorphous Replacement
• Multiwavelength Anomalous Diffraction
• Single-wavelength Anomalous Diffraction
?
Molecular Replacementcorrect structure and intensities
http://www.ysbl.york.ac.uk/~cowtan/fourier/coeff.html
Molecular Replacementuse something similar as a starting model
Model Buildingcurrent model is missing something
Model Buildingphases from model
Model Buildingmissing bits show up in “difference map”
Model Buildingmissing bits show up better in FO + (FO - FC) map
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1 2 3 4 5 6 7 8 9 10
data
stru
ctu
re f
acto
r (F
)
spot index (h)
Fitting data
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1 2 3 4 5 6 7 8 9 10
data
1 param
stru
ctu
re f
acto
r (F
)
spot index (h)
Fitting data
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1 2 3 4 5 6 7 8 9 10
data
1 param
3 params
stru
ctu
re f
acto
r (F
)
spot index (h)
Fitting data
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1 2 3 4 5 6 7 8 9 10
data
1 param
3 params
10 params
stru
ctu
re f
acto
r (F
)
spot index (h)
Fitting data
Major Phasing techniques
• Molecular Replacement
• Multiple Isomorphous Replacement
• Multiwavelength Anomalous Diffraction
• Single-wavelength Anomalous Diffraction
inverse Fourier Transformno phase
inverse Fourier Transformno phase
Major Phasing techniques
• Molecular Replacement
• Multiple Isomorphous Replacement
• Multiwavelength Anomalous Diffraction
• Single-wavelength Anomalous Diffraction
dete
ctor
anomalous scattering
sam
ple
x-ray beam
dete
ctoranomalous scattering
sam
ple
x-ray beam
Independent tasks can be performed simultaneously
Multiprocessing Strategy
Multiprocessing Strategy
Multiprocessing Strategy
Multiprocessing Strategy
Multiprocessing Strategy
Multiprocessing Strategy
Multiprocessing Strategy
Multiprocessing Strategy
Multiprocessing Strategy
Multiprocessing Strategy
Multiprocessing Strategy
Multiprocessing Strategy
Multiprocessing Strategy
SOLVE/P212121/doneSOLVE/P21212/doneSOLVE/P21221/doneSOLVE/P22121/doneSOLVE/P2221/doneSOLVE/P2212/doneSOLVE/P2122/busySOLVE/P222/done
Multiprocessing StrategyMultiprocessing Strategy
epmr/P212121/model1/doneepmr/P212121/model2/doneepmr/P21212/model1/doneepmr/P21212/model2/doneepmr/P21221/model1/doneepmr/P21221/model2/doneepmr/P22121/model1/busyepmr/P22121/model2/doneepmr/P2221/model1/epmr/P2221/model2/epmr/P2212/model1/epmr/P2212/model2/epmr/P2122/model1/epmr/P2122/model2/
Multiprocessing Strategy
wARP/P212121/donewARP/P21212/donewARP/P21221/donewARP/P22121/donewARP/P2221/donewARP/P2212/donewARP/P2122/busywARP/P222/done
Multiprocessing Strategy
space group
FOM Rcryst
P3221
P3121
P321
Multiprocessing Advantage
space group
FOM Rcryst
P3221 0.836 0.323
P3121
P321
Multiprocessing Advantage
space group
FOM Rcryst
P3221 0.836 0.323
P3121 0.865 0.196
P321 0.726 0.347
Multiprocessing Advantage
table1.com
table1.com
Elven Automation
How often does it really work?
Apr 6 – 24 at ALS 8.3.1
Elven Automation
27,686 images collected
Apr 6 – 24 at ALS 8.3.1
Elven Automation
27,686 images collected
148 datasets (15 MAD)
Apr 6 – 24 at ALS 8.3.1
Elven Automation
27,686 images collected
148 datasets (15 MAD)
31 investigators
Apr 6 – 24 at ALS 8.3.1
Elven Automation
27,686 images collected
148 datasets (15 MAD)
31 investigators
56 unique cells
Apr 6 – 24 at ALS 8.3.1
Elven Automation
27,686 images collected
148 datasets (15 MAD)
31 investigators
56 unique cells
5 KDa – 23 MDa asymmetric unit
Apr 6 – 24 at ALS 8.3.1
Elven Automation
27,686 images collected
148 datasets (15 MAD)
31 investigators
56 unique cells
5 KDa – 23 MDa asymmetric unit
0.94 – 32 Å resolution (3.2 Å)
Apr 6 – 24 at ALS 8.3.1
Elven Automation
148 datasets
Apr 6 – 24 at ALS 8.3.1
Elven Automation
148 datasets
117 succeded
Apr 6 – 24 at ALS 8.3.1
Elven Automation
148 datasets
117 succeded
~3.5 (0.1-75) hours
Apr 6 – 24 at ALS 8.3.1
Elven Automation
148 datasets
117 succeded
~3.5 (0.1-75) hours
31 failed
Apr 6 – 24 at ALS 8.3.1
Elven Automation
148 datasets
117 succeded
~3.5 (0.1-75) hours
31 failed
~61 (0-231) hours
Apr 6 – 24 at ALS 8.3.1
Elven Automation
148 datasets
117 succeded
~3.5 (0.1-75) hours
31 failed
~61 (0-231) hours
2 / 15 MAD structures
Apr 6 – 24 at ALS 8.3.1
Elven Automation
148 datasets
117 succeded
~3.5 (0.1-75) hours
31 failed
~61 (0-231) hours
2 / 15 MAD structures
Number Description Percent
446028 Images (~7 TB) 33%
2346 Data sets 47%
449 MAD/SAD (1:2) 19%
104 Published 4.4%
8.3.1 in 2003
How many structures get solved?
Why do structures fail?
Overlaps
Why do structures fail?
Overlaps
Signal to noise
Why do structures fail?
Overlaps
Signal to noise
Radiation Damage
Why do structures fail?
Overlaps
Signal to noise
Radiation Damage
Why do structures fail?
Apr 6 – 24 at ALS 8.3.1
Elven Automation
148 datasets
117 succeded
~3.5 (0.1-75) hours
31 failed
~61 (0-231) hours
2 / 15 MAD structures
Apr 6 – 24 at ALS 8.3.1
Elven Automation
148 datasets
117 succeded
~3.5 (0.1-75) hours
31 failed
~61 (0-231) hours
2 / 15 MAD structures
unavoidable overlaps
unavoidable overlaps
dete
ctor
unavoidable overlaps
phi
dete
ctor
unavoidable overlaps
mosaicity
phi
dete
ctor
unavoidable overlaps
mosaicity
phi
dete
ctor
c*
unavoidable overlaps
mosaicity
phi
dete
ctor
c*
Ewald sphere
unavoidable overlaps
mosaicity
phi
dete
ctor
c*
Ewald sphere
unavoidable overlaps
mosaicity
phi
dete
ctor
c*
Ewald sphere
unavoidable overlaps
mosaicity
phi
dete
ctor
c*
Ewald sphere
unavoidable overlaps
mosaicity
phi
dete
ctor
c*
Ewald sphere
unavoidable overlaps
mosaicity
phi
dete
ctor
c*
Ewald sphere
unavoidable overlaps
mosaicity
phi
dete
ctor
c*
Ewald sphere
unavoidable overlaps
mosaicity
phi
dete
ctor
c*
b
c
a
unavoidable overlaps
mosaicity
phi
dete
ctor
c*
b
c
a
unavoidable overlaps
mosaicity
phi
dete
ctor
c*
b
c
a
unavoidable overlaps
mosaicity
phi
dete
ctor
c*
b
c
a
avoidable overlaps
mosaicity
phi
dete
ctor
c*
b
ca
avoidable overlaps
mosaicity
phi
dete
ctor
c*
bc
a
avoiding overlaps
avoiding overlaps
avoiding overlaps
avoiding overlaps
c
c
avoiding overlaps
1000 mm
avoiding overlaps
1000 mm
2 mrad
10 seconds
avoiding overlaps
1000 mm
2 mm
2 mrad
10 seconds
avoiding overlaps
1000 mm
1 mm
1 mrad
20 seconds
avoiding overlaps
1000 mm
300 um
0.3 mrad
60 seconds
Overlaps
Signal to noise
Radiation Damage
Why do structures fail?
Overlaps
Signal to noise
Radiation Damage
Why do structures fail?
Apr 6 – 24 at ALS 8.3.1
Elven Automation
148 datasets
117 succeded
~3.5 (0.1-75) hours
31 failed
~61 (0-231) hours
2 / 15 MAD structures
Apr 6 – 24 at ALS 8.3.1
Elven Automation
148 datasets
117 succeded
~3.5 (0.1-75) hours
31 failed
~61 (0-231) hours
2 / 15 MAD structures
Radiation Damage
why not just avoid it?
Holton & Frankel (2010) Acta D 66 393-408.
B ≈ 4 d2 + 120
20
4
0
60
8
0
100
12
0
aver
age
atom
ic B
fac
tor
1 1.5 2 2.5 3 3.5 4 4.5 5
resolution (Å)
Simulated diffraction imageSimulated diffraction imageMLFSOMMLFSOM
simulatedsimulated realreal
signal vs noise
signal vs noise
signal vs noise
“If you don’t have good data,
then you have no data at all.”
-Sung-Hou Kim
signal vs noiseeasy
hard
impossible
signal vs noiseeasy
hard
impossible
threshold of “solvability”
signal vs noise
“If you don’t have good data,
then you must learn statistics.”
-James Holton
Adding noise
Adding noise
12 + 12 = 1.42
Adding noise
12 + 12 = 1.42
32 + 12 = 3.22
σtotal2 = σ1
2 + σ22
Adding noise
12 + 12 = 1.42
32 + 12 = 3.22
σtotal2 = σ1
2 + σ22
Adding noise
12 + 12 = 1.42
32 + 12 = 3.22
σtotal2 = σ1
2 + σ22
Adding noise
12 + 12 = 1.42
32 + 12 = 3.22
102 + 12 = 10.052
MAD phasing simulation
-0.2
0
0.2
0.4
0.6
0.8
1
0.01 0.1 1 10
Anomalous signal to noise ratio
Cor
rela
tion
coef
ficie
nt t
o co
rrec
t m
odel
mlphare results
SAD phasing simulation
-0.2
0
0.2
0.4
0.6
0.8
1
0.01 0.1 1 10
Anomalous signal to noise ratio
Cor
rela
tion
coef
ficie
nt t
o co
rrec
t m
odel
mlphare results
SAD phasing experiment
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.5 1.0 1.5 2.0 2.5 3.0
Anomalous signal to noise ratio
Cor
rela
tion
coef
ficie
nt t
o pu
blis
hed
mod
el
MR simulation
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0.001 0.01 0.1 1 10 100 1000
Signal to noise ratio
Cor
rela
tion
coef
ficie
nt t
o co
rrec
t de
nsity
corrupted data
“photon counting”
Read-out noise
Shutter jitter
Beam flicker
spot shape
radiation damage
σ(N) = sqrt(N)
rms 11.5 e-/pixel
rms 0.57 ms
0.15 %/√Hz
pixels? mosaicity?
B/Gray?
signal vs noise
Which error dominates?
• Weak spots (high-res)background
• MAD/SAD (small differences)detector calibration
( if not rad dam! )
Holton & Frankel (2010) Acta D 66 393-408.
Holton & Frankel (2010) Acta D 66 393-408.
Background level sets needed photons/spot
Moukhametzianov et al. (2008). Acta Cryst. D 64, 158-166
Holton & Frankel (2010) Acta D 66 393-408.
Optimal exposure time(faint spots)
σtotal2 = σspot
2 + σbg2 + σreadout
2+ σraddam2
too long!
σtotal2 = σspot
2 + σbg2 + σreadout
2+ σraddam2
σtotal2 = σspot
2 + σbg2 + σreadout
2+ σraddam2
Optimal exposure time(faint spots)
σtotal2 = σspot
2 + σbg2 + σreadout
2+ σraddam2
too short!
σtotal2 = σspot
2 + σbg2 + σreadout
2+ σraddam2
Optimal exposure time(faint spots)
σtotal2 = Nphotons
+ σreadout2+ σraddam
2
needlessly long
Nphotons ≈ σdetector2
Optimal exposure time(faint spots)
“optimal”
Nphotons ≈ 10x σdetector2
Optimal exposure time(faint spots)
“buried”
Nspot + Nbg ≈ 10x m
Optimal exposure time(faint spots)
“buried”
2
0
gain
0 + Nbg ≈ 10x m
Optimal exposure time(faint spots)
“buried”
2
0
gain
Optimal exposure time(faint spots)
2
00 10
gain
mgain
bgbghr
Optimal exposure time(faint spots)
2
00 10
gain
mt
t
gain
bgbg
ref
hrref
Optimal exposure time(faint spots)
0
2010
bgbggain
mtt
refrefhr
thr Optimal exposure time for data set (s)tref exposure time of reference image (s)bgref background level near weak spots on
reference image (ADU)bg0 ADC offset of detector (ADU)bghr optimal background level (via thr)σ0 rms read-out noise (ADU)gain ADU/photonm multiplicity of data set (including partials)
Short answer:
bghr ~ 100 ADU
for ADSC Q315r
What error dominates?
• Weak spots (high-res)background
• MAD/SAD (small differences)detector calibration
if not rad dam!
Optimal exposure time(anomalous differences)
I-I+
3%
100
phot
ons
10 photons
100
phot
ons
Optimal exposure time(anomalous differences)
I-I+
3%
100
phot
ons
14 photons
100
phot
ons
Optimal exposure time(anomalous differences)
3%
I-I+
2000
pho
tons
67 photons
Optimal exposure time(anomalous differences)
1%
I-I+
20,0
00 p
hoto
ns
200 photons
Minimum required signal (MAD/SAD)
"#
)(3.1
fsites
DaMW
sd
I
Holton & Frankel (2010) Acta D 66 393-408.
“photon counting”
Read-out noise
Shutter jitter
Beam flicker
spot shape
radiation damage
σ(N) = sqrt(N)
rms 11.5 e-/pixel
rms 0.57 ms
0.15 %/√Hz
pixels? mosaicity?
B/Gray?
signal vs noise
Optimal exposure time(anomalous differences)
σtotal2 = σspot
2 + σbg2 + σreadout
2+ σraddam2
Optimal exposure time(anomalous differences)
σtotal2 = σspot
2 + σbg2 + σreadout
2+ σraddam2
Optimal exposure time(anomalous differences)
0MADt
no detector is perfectly calibrated!
σtotal2 = Nspot + σbg
2 + σreadout
2+ σraddam2
+ (fshutterNspot )2 + (fflickerNspot )2
+ (fcalibNspot )2
σtotal2 = Nspot + σbg
2 + σreadout
2+ σraddam2
Fractional error
mult > (—)2Rmerge
<ΔF/F>
Holton & Frankel (2010) Acta D 66 393-408.
Damage model system
67 consecutive data sets
67 consecutive data sets
Data quality vs exposure
0.5
0.55
0.6
0.65
0.7
0 20 40 60 80 100
model vs warp
Exposure time (min)
Cor
rela
tion
coef
ficie
nt
Data quality vs exposure
579
1113151719212325
0 20 40 60 80 100
3A bin
Exposure time (min)
Data quality vs exposure
15
17
19
21
23
25
27
29
0 20 40 60 80 100
overall
Exposure time (min)
Data quality vs exposure1.7
1.9
2.1
2.3
2.5
2.7
0 20 40 60 80 100
I/sig(I) < 2
Exposure time (min)
Res
olut
ion
limit
Data quality vs exposure0
0.020.040.060.080.1
0.120.140.160.180.2
0 20 40 60 80 100
3A bin
Exposure time (min)
Rsy
m
Experimentally-phased map
Experimentally-phased map
Damage changes absorbance spectrum
0
500
1000
1500
2000
2500
3000
3500
4000
4500
50001
26
40
12
64
5
12
65
0
12
65
5
12
66
0
12
66
5
12
67
0
12
67
5
12
68
0
12
68
5
12
69
0
12
69
5
12
70
0
Photon energy (eV)
coun
ts
Damage changes absorbance spectrum
0
500
1000
1500
2000
2500
3000
3500
4000
4500
50001
26
40
12
64
5
12
65
0
12
65
5
12
66
0
12
66
5
12
67
0
12
67
5
12
68
0
12
68
5
12
69
0
12
69
5
12
70
0
Photon energy (eV)
coun
ts
Damage changes absorbance spectrum
0
500
1000
1500
2000
2500
3000
3500
4000
4500
50001
26
40
12
64
5
12
65
0
12
65
5
12
66
0
12
66
5
12
67
0
12
67
5
12
68
0
12
68
5
12
69
0
12
69
5
12
70
0
beforebeforeburntburnt
Photon energy (eV)
coun
ts
Damage changes absorbance spectrum
0
500
1000
1500
2000
2500
3000
3500
4000
4500
50001
26
40
12
64
5
12
65
0
12
65
5
12
66
0
12
66
5
12
67
0
12
67
5
12
68
0
12
68
5
12
69
0
12
69
5
12
70
0
beforebeforeburntburnt
Photon energy (eV)
coun
ts
1
0
fluorescence probe for damage
Absorbed Dose (MGy)
Fra
ctio
n u
nco
nve
rted
Wide range of decay rates seen
0.
0
0
.2
0.4
0.6
0
.8
1.0
0 50 100 150 200
Half-dose = 41.7 ± 4 MGy“GCN4” in crystal
Half-dose = 5.5 ± 0.6 MGy8 mM SeMet in NaOH
Protection factor: 660% ± 94%
http://ucxray.berkeley.edu/~jamesh/elves
Download Elves from: