Ten Years and Change

the MX data archive at ALS 8.3.1

ALS 8.3.1 data collection history

0

10

20

30

40

50

60

70

2001200220032004200520062007200820092010201120122013

actual

doubling = 2.8 years

tera

byte

s (u

ncom

pres

sed)

ALS 8.3.1 data collection history

0

10

20

30

40

50

60

70

2001200220032004200520062007200820092010201120122013

Proteum 300

Q210

Q315 (907)

Q315r (926)

tera

byte

s (u

ncom

pres

sed)

ALS 8.3.1 data collection history

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

2001200220032004200520062007200820092010201120122013

Proteum 300

Q210

Q315 (907)

Q315r (926)

imag

es x

106

ALS 8.3.1 data collection historyim

ages

x 1

06

ALS 8.3.1 data collection history

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5images collected

PDB data collection

PDB deposition

imag

es x

106

1250

1000

750

500

250

0

PD

B e

ntr

ies

ALS 8.3.1 data collection historyim

ages

x 1

06

1250

1000

750

500

250

0

PD

B e

ntr

ies

imag

es x

106

1250

1000

750

500

250

0

PD

B e

ntr

ies

DVD data archive: 82 TB

Which data go with which PDB?

• 260,000 images are called “test”

• cell: 48 62 84 90 101 104– is within 5 Å and 5° of 16,000 PDBs

focusing on 2001-2006

• 490 PDBs credit ALS 8.3.1 with data

• 44 of these didn’t actually collect data

• 64 collected data, but no credit

1. images from 2001-2006

2. collected “near” edges

3. find “runs” of >10 images

4. unify multi-wedge sets

5. run labelit & XDS

6. >70% complete?

7. I/σ > 10

8. reduced cell vs PDB

1,604,031

682,712

3602

3331

2524

1479

1054

1 to 200+

Which data go with which PDB?

Responses to inquiries

“I have to find my old note book as I have no idea what that is.”

“I have changed jobs a few times since and am really far away from crystallography now.”

“Will see what I can find.”

“We solved it but never published it. Sorry!”

DVD data archive

Primary failure mode of DVDs

dataset identification protocol

1. images from 2001-2006

2. collected “near” edges

3. find “runs” of >10 images

4. sort out multi-wedge sets

5. run XDS

6. >70% complete?

7. I/σ > 10

8. reduced cell vs PDB

1,604,031

682,712

3602

3331

2524

1479

1054

1 to 200+

Unit Cell: 90.9 90.9 46.8 90 90 120

0.3

0.35

0.4

0.45

0.5

0.55

0.6

0.00 0.50 1.00 1.50 2.00

best

Rcr

yst a

fter

rig

id-b

ody

refin

emen

t

RMS unit cell length deviation (Å)

1hh7 M. TB CSOR

1rb5

myoglobin

MAD/SAD datasets

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0.20 0.30 0.40 0.50 0.60

Ris

o vs

PD

B d

epos

it

best Rcryst after rigid-body refinement

Published

non-isomorphous

Unsolved?

EGDADec 01 19:45:12 2001 egda46_*1_E#_###.img (1112 images, Se MAD)Dec 02 15:10:06 2001 egda27_*1_###.img (180, 1A, native?)Dec 02 19:21:55 2001 egdau1_*1_###.img (427, 8000eV (U?) SAD)Dec 02 20:58:26 2001 egdau1_*2_###.img (360, 8000eV (U?) SAD)Jun 01 14:07:43 2002 egda60_*1_###.img (360, Lutetium SAD)

“I think that these EGDA data sets are very likely some of xxx’s data sets, he was working on E.coli guanine deaminase, something he brought from yyy. No structure was ever published James, xxx was unable to solve the structure from these data.”

~2.9 ÅP21212

R = 0.32Rfree = 0.39

PDB ID: ????

E. coliguaninedeaminase

Summary

• saving data could double productivity

• unit cell is not a good score

• lossy compression: rallying cry?

• backup vs archive

• metadata: what do we really know?

Brief Summary

• this is a lot of work.

• who is going to pay for it?

backblaze.com “pod” server

backblaze.com offers “unlimited storage” data backup for $5/month.

backblaze offers

“unlimited storage” data backup for

$5/month.

backblazedoes not sellthese “pods”,but “protocase.com” will.

compresses 4.2x

compresses 337x

compresses 5x, but only one per dataset!

compresses 3.5x

compressed ~50x

compresses 5.2x

Lossy compression vs R/Rfree

0.18

0.2

0.22

0.24

0.26

0.28

0.3

0.32

0.34

0.36

0.38

1 10 100

R_cryst

R_free

R f

acto

r

compression ratio