1. Towards high-throughput structure determination at SSRL
Ashley Deacon Stanford Synchrotron Radiation Laboratory

2. Motivation for high-throughput structure determination
SMB user program
Structural genomics
Five macromolecular
crystallography beamlines in operation (including 11-1).
BL 11-1 Stanford/TSRI/SSRL Monochromatic
BL 9-1 Monochromatic
BL 9-2 Multi-wavelength
BL 1-5 Multi-wavelength
BL 7-1
Monochromatic
SPEAR 3 upgrade of storage ring
to 3rd generation capabilities
by 2003
Solve hundreds of structure per year without relying on many crystallographers

3. High-throughput goals
Automate the crystallography experiment
New hardware (e.g. crystal mounting robot)
Rapid crystal characterization
Optimal data collection from best crystals
Paul Phizackerley P34,
Ana Gonzalez P30, Aina Cohen P27
Automate crystallographic computations
Include latest crystallographic techniques
Route data through an analysis pipeline
Evaluate progress of structure determination
Integrate the experiment and the analysis
Feedback to the diffraction experiment.
Feedback to the other core groups.
Paul Ellis P12
ASAP is a system designed to take frozen crystals and push them as far through the crystallographic structure determination process as possible.
We’re not going to realize a fully automated system in one easy step, so its important to break the development down into stages and integrate each stage into the current SMB program at SSRL.
Develop the Automated Structural Analysis of Proteins (ASAP) system
Integrate SDC with the SMB program.
Staged delivery of useful components.
Aina Cohen P27, Hsiu-Ju Chiu P33
Thomas Eriksson P29, Scott McPhillips P32

4. The world of crystallography according to Ashley (pre-JCSG) I
SSRL
Mosflm
SnB
Mlphare
warpNtrace
XPLOR
???
Data Collection
Data Processing
Locate
Heavy
Atoms
Solve Structure
Model Building
Model Refinement
What do I do if this approach fails?
Often there are too many options to try in a truly systematic fashion and may not be aware of all useful parameters.
Can use our crystallography knowledge to make smart scripts for individual programs with some feedback from the output of a previous job.
Re-run programs with modified parameters
Slow trial-and-error process
Not very systematic
What if that fails?

5. The world of crystallography according to Ashley (pre-JCSG) II
Consult the literature
Discover the “Golden Bullet”.
Learn new applications.
Energy Barrier
Consult colleagues
Borrow scripts.
Try out suggestions.
Problems and bottlenecks…
Slow learning process.
Cannot systematically try all applications / possibilities.
Rely on hearsay.
Often rely on hearsay and word of mouth in some cases.
May resort to repeating the experiment rather than systematically try out all the options.

6. The world of crystallography according to Ashley (pre-JCSG) III
SnB
Locate
Heavy
Atoms
Mosflm
Mlphare
warpNtrace
XPLOR
Data Processing
Solve Structure
Model Building
Model Refinement
SnB
Locate
Heavy
Atoms
SSRL
Data Collection
Still have problems
Limited experience
Not systematic.
This is the knowledge accumulated after several months or years of iteration around the crystallography process.
Any one crystallographer only has the chance to solve a few structures per year at the most.
Even this knowledge is therefore quite limited.
DENZO
SHELX
SHARP
Data Processing
Locate
Heavy
Atoms
Solve Structure

7. The world of crystallography according to ASAP I
Ashley
Frank
Tassos
Duncan
Gerard
Glen
Maybe for ASAP we just take the knowledge from all our collaborators and try each of their structure determination protocols.
What we really want to do is take the best bits from each persons knowledge.
We want to take advantage of crystallographic software developers knowledge and use their protocols, but also include the possibility of including scripts from regular users so that their knowledge is also incorporated.
We want many possibilities for each step and then a way to distill out the most useful components.
Work out which algorithms work best and under what circumstances.

8. The world of crystallography according to ASAP II
JCSG staff
The Operation Manager allows
Single-click execution of Operations.
Standardized file input and output to all Operations.
A common communication protocol to Operations for developers via an API and Library.
Operation Manager
Operations
Operations Manager will start out as human driven and then later become automated.
Operation Managers won’t all have the same set of Operations available e.g. the “data collection” Operation will only be available via a limited number of Operation Managers and not all computers have the same set of crystallography applications installed.
An Operation Manager will control the execution of Operations.
The Operation Manager will deal with all file handling, so the user doesn’t need to worry about where the input files are read from nor where the output files are written.
The Operation Manager will store all relevant information about the locations in a database. The crystallographic data and the bulk of the additional crystallographic information will remain in these files.
The Operation Manager and associated API and library provide an easy way for non-JCSG developers to contribute their crystallographic knowledge to the ASAP system.

9. The world of crystallography according to ASAP III
JCSG Staff and Scripted Operations
The Scheduler supports
Multiple Operation Managers.
Distribution of resources to multiple projects.
Efficient use of all resources.
Scheduler
Market-based resource allocation
Operation Manager
Operation Manager
Support multiple Operation Managers – each can have a different set of Operations available to it.
May start off with a simple less efficient algorithm until the Market-based resource allocation algorithm has been fully developed.
Will provide distributed operation and efficient use of resources.

10. The world of crystallography according to ASAP IV
Static rules-based Solver
An “if…then…else…” approach.
All decisions must be preprogrammed.
Hard to take all factors into account.
Nothing learnt from past operations.
Dynamic rules-based Solver
Modify rules on the fly to reflect knowledge accumulated from all projects.
Take all characteristics of the current project into account when interpreting rules.
Solver
Rules-based execution
of a project
Scheduler
Market-based resource allocation
Operation Manager
Operation Manager
Two possible approaches. The static version may be easy to implement at first, but a dynamic rules-based system will be more powerful.
The ASAP system is going to have hundreds of projects to learn and train from, which will make the dynamic version become more powerful with time.
The static version incorporates crystallographic knowledge. We can envision a dynamic rules based system that does not need any specific crystallographic knowledge to carry out its task.

11. The world of crystallography according to ASAP V
Operations
Can be connected together as defined by the inputs they require and the outputs they produce.
Can incorporate some internal feedback and intelligence to make them smart.
An ASAP Operation
Inputs
Outputs
Attributes
Describe the inputs and outputs of an operation.
Correlations between the attributes can be used to generate rules, which can guide the Solvers.
Input Attributes
Output Attributes
Solver can connect up Operations according to their inputs and outputs. Thus the Solver knows which programs it can run.
It then needs to decide which its best options are amongst all the possible options.
For this it must use rules.
Attributes describe the inputs and outputs of each Operation.
Attributes can be used to derive rules based on correlations between input and output attributes observed in all previous projects.
Give easy example : one particular solvent flattening algorithm may be particularly good for cases with low solvent content (because it does more than just density modification), the correlation between low solvent content (input attribute) and good phases (output attribute) will be recognized over time….

12. The world of crystallography according to ASAP VI
Build a graph of Operations
Traverse the graph by the most efficient route or try many routes and choose the best results

13. The world of crystallography according to ASAP VII
Feedback to Solver
Solver
Rules-based execution of a project
Data Miner
Derives rules for the Solver
Scheduler
Market-based resource allocation
Operation Manager

14. A production ASAP system
Solver
Rules-based execution of a project
Data Miner
Derives rules for the Solver
System State Database
Stores past operations
Scheduler
Market-based resource allocation
System State Database will
Store file locations and all Attributes derived from past operations for Data Miner.
Track progress of all crystals relating to each target protein.
Operation Manager

15. ASAP – Summary The ASAP architecture is
Capable of parallel operation on multiple samples within a project
Capable of parallel operation on multiple projects
Flexible and modular in design
Scalable in both hardware and software
Maintainable
Testable
The ASAP staged-delivery will
Provide a series of useful systems that gradually improve throughput.
Ultimately lead to a fully automated production system.

16. Acknowledgements The entire SMB group at SSRL Fred Bertsch
Tim McPhillips
Peter Kuhn
GNF
Glen Spraggon
The Scripps Research Institute
Frank von Delft
Syrrx
Duncan McRee
SSRL is funded by:
Department of Energy, Office of Basic Energy Sciences,
The Structural Molecular Biology Program is supported by:
National Institutes of Health, National Center for Research Resources, Biomedical Technology Program and Department of Energy, Office of Biological and Environmental Research.