WK.03 Serial Crystallography Data Analysis with Cheetah and CrystFEL: Concepts and Tutorials
Time resolved pump-probe SFX studies additionally benefit from the use of tiny crystals because they are smaller than the extinction length of a pump laser, while in current state-of-the-art pump-probe experiments at synchrotrons, typically only 10-40% of molecules in a fixed sample are photoactivated. Non-photoactive molecules with reactions initiated by chemical/physical changes such as temperature, pH, or reactants (e.g. introduced via a mixing jet) also benefit from using microcrystals because of higher diffusion rates in nano- and microcrystals. In SFX, nano/microcrystals are delivered to the pulsed X-ray beam by either a micron-thick liquid jet (most commonly), a recently invented lipidic cubic phase (LCP) jet, electrospray, or by raster scanning fixed target supports. At LCLS, the X-ray pulses arrive at 120 Hz (with much higher rates planned for future XFELs), resulting in huge data sets that need to be efficiently filtered and cleaned. Each diffraction pattern is corrected, cleaned and indexed independently, before the data set is merged. Thousands of diffraction patterns, and hence, crystals, are needed for a full data set.
For time resolved studies, thousands of randomly oriented patterns are needed for each time point. Indeed, a single experiment can result in over 100 terabytes of raw data. These high data collection rates and large data sets have necessitated the development of novel high-throughput, parallelizable data analysis software for "live" feedback during an SFX experiment, as well as for further processing of the clean diffraction patterns. Since the diffraction patterns consist almost entirely of partial reflections, complicated post-refinement/scaling procedures are necessary to reduce the amount of data required to obtain accurate structure factors. A number of post refinement methods are being developed in SFX analysis software, and will be discussed during the workshop.
Synchrotrons and future XFELs - rapidly growing SFX user base.
Technological advances at synchrotron sources, producing smaller, brighter beams, coupled with new detectors (e.g. pixel array detectors running in shutterless mode) and novel crystal delivery devices have enabled the collection of SFX data from microcrystals at synchrotrons. Synchrotron SFX experiments include studies of in vivo grown needle-shaped crystals (helically scanned to avoid radiation damage) and
LCP-SFX of membrane protein crystals in room temperature, in air. Initial LCP-SFX studies at LCLS yielded the first room temperature GPCR structure (from a few hundred micrograms of purified protein), but the recently demonstrated synchrotron based LCP-SFX (still at room temperature) is an even more exciting result for these biomedically important proteins in large part because of the availability of high end MX synchrotron beamlines.
In summary, SFX has yielded several major and unique advances in structural biology previously unattainable with conventional technologies, including the potential for sub-picosecond time-resolved crystallographic studies, probing cyclic or even non-cyclic reactions. With the construction of more than a dozen new XFELs currently under way, and the growing use of serial crystallography at synchrotrons, the potential user base is growing significantly. The development and appropriate use of new software to tackle the unique problems of SFX data analysis is vital to making SFX practicable. It is our desire and responsibility to share the knowledge and experience from early SFX experiments, and train a new generation of scientists in SFX data analysis to make this exciting new technique even more accessible to the crystallographic community. We anticipate that the impact of SFX on (dynamical) structural biology will ultimately be immense.
The workshop will start with an introductory seminar, followed by most of the day dedicated to detailed hands-on tutorials with data sets collected at LCLS using the software suites Cheetah and CrystFEL, two of the most commonly used packages for SFX analysis. All the concepts unique to SFX that we will be discussing are software independent.
The workshop outline is as follows:
1. An introduction to serial femtosecond crystallography and how the data differ from conventional crystallography data.
2. Cheetah tutorial: From raw data to useful diffraction patterns.
The first tutorial session will focus on Cheetah (www.desy.de/~barty/cheetah), and will involve the initial analysis and reduction of raw data to a set of clean, usable diffraction patterns in a facility-independent format (HDF5) for further analysis.
In addition to running the software, a major focus will be optimizing this process to retain the best data. Several data sets (of varying quality) collected at LCLS will be used to detail the steps from data collection to data analysis. Relevant details about LCLS data storage formats and procedures will be described. Detector considerations (geometry calibration and signal corrections) will be discussed and students will be shown how to accurately determine detector geometry, a critical component of successful indexing. Spot finding algorithms will be described and students will have the opportunity to experiment with various parameters to learn the best practices for optimization. Tips and tricks for analyzing and for evaluating the success of a set of parameters will be explained.
We will discuss and demonstrate how to generate virtual powder patterns, radial stacks, histograms of various hit finding statistics, extract the spectrum of each XFEL shot, record pump laser profiles and other pertinent metadata.
Cheetah is open-source and has been released under the GNU GPL v3 license. Though designed for XFEL experiments (both for SFX and single particle / solution scattering), Cheetah can be readily adapted for use where parallelized diffraction data reduction and detector correction is necessary. Its modular, facility-independent library makes Cheetah portable and expandable to use with new detectors, and data type and formats.
3. CrystFEL tutorial: Indexing, integrating, merging, post-refinement and evaluation of serial crystallography data.
The second tutorial session will focus on indexing, integrating and merging the cleaned SFX data using CrystFEL, a software suite created for SFX data analysis (and simulation). The development of CrystFEL is lead by Thomas White (CFEL, DESY), who will be the main instructor for this tutorial, with hands-on support for students from the workshop organizers experienced with CrystFEL: Grant and Zatsepin. The suite is written in C with supporting Perl and shell scripts, and is available as source code under version 3 or later of the GNU General Public License.
At the core of CrystFEL is an automated, high throughput processing pipeline which indexes and integrates each diffraction pattern in a serial crystallography data set. Merging the results yields diffraction data which can be imported into standard crystallographic processing packages for further analysis.
We will describe in what way indexing algorithms differ from conventional crystallography for snapshots from randomly oriented microcrystals, and show students how to optimize indexing parameters to extract the most accurate structure factors from the data. We will show how variations in physical crystal features such as size and unit cell parameters may affect spot shape and intensity and how that influences integration routines. We will discuss how the XFEL spectrum affects SFX data, and how to optimize integration parameters. Methods of evaluating successful integration will also be shown. Detailed descriptions of proper merging procedures will be discussed, as well as how to deal with influencing factors such as scaling, partiality of reflections, detector saturation and more. We will also show students how to determine and evaluate multiple parameters for measuring the quality of the merged data, including signal to noise and multiplicity, and internal consistency quantified by Rsplit and CC1/2.
XFEL pulses generated by self amplified spontaneous emission (SASE) differ in spectrum and intensity from shot to shot. Meanwhile, each crystals, arriving in a random orientation and position relative to the X-ray pulse, may differ in size and shape. Accurate structure factors can be obtained from merging diffraction data from thousands of snapshots collected in this manner by Monte Carlo integration (which we will discuss in the workshop), with accuracy increasing with 1/sqrt(number of patterns). Much development is now dedicated from multiple SFX data analysis developers to post refinement methods that move SFX analysis "beyond Monte Carlo" to a regime where more accurate structures can be obtained from much smaller data sets (significantly reducing sample consumption) by modeling and refining the experimental parameters that fluctuate from shot to shot.
Anomalous signals for de novo phasing (single or multi-wavelength anomalous diffraction) can be enhanced by combining data sets from multiple crystals. De novo SAD phasing from SFX data has also been demonstrated (though not optimized). We will touch upon these topics in the discussions on data quality, post refinement and heterogeneity.
Additional time will be spent discussing how to deal with SFX analysis problems such as resolving the indexing ambiguity arising from the uncorrelated orientations of diffraction snapshots for particular space groups, considering heterogeneity between crystals, dealing with weak diffraction and indexing multiple lattices. Finally, students will be shown how to import their results into standard crystallographic processing packages for further analysis. At each stage students will be encouraged to ask questions and instructors will be available for help.
Computational resources and workshop preparation for students
SFX data reduction and analysis requires high-performance computational resources. Fortunately, SLAC's computational infrastructure has been designed for high-throughput data analysis with the Cheetah and CrystFEL pre-installed. Students will be expected to bring laptops with appropriate software preinstalled for connecting to SLAC via SSH. Students will be sent detailed information prior to the workshop to request access to SLAC computational resources and other instructions for testing software installations and connections. This format worked well for our previous workshop and students were able to quickly connect and use SLAC's computing resources without any appreciable delay. A website will be set up prior to the workshop with useful instructions and publications for workshop preparation. On-site network connectivity will be provided as part of the course. The data used in the workshops will be SFX data collected at LCLS that is publicly available on the Coherent X-ray Imaging Data Bank (cxidb.org). All of the data will be stored and analysis performed on SLAC computers; only text and images will be sent over the network. X-window forwarding will be required to interact with software GUIs, which will require a moderate network connection. Our previous workshop provided ethernet connectivity for individuals to provide the necessary bandwidth for the network traffic, with some students using WiFi, and proved to be sufficient.
Participants will be presumed to have some basic knowledge of conventional crystallographic data analysis such as what indexing, integration, and merging are and will instead focus on the specific features of SFX analysis, as described above.
The workshop will begin at 9:00am and end at approx. 5:00pm. Lunch will be provided. Preregistration and payment is required.
Students and postdocs - $100
Academics others - $150
Corporate - $250
This workshop is organized by the BioXFEL Science and Technology Center established by the National Science Foundation in 2013, committed to the development of XFEL science for the analysis of biological systems.
Hauptman-Woodward Institute, 700 Ellicott Street, Buffalo, NY 14203
Arizona State University, Physics Department, Center for Biological Physics, Tempe, AZ 85287-1504