Gooey data sets
- 1. 1 Gooey data sets: high throughput structural data from complex amphiphiles Charlotte Broadbent Civil Engineering, Columbia University, New York, NY 10027 Elaine DiMasi Photon Sciences Department, Brookhaven National Lab, Upton, NY 11973
- 2. 2 Abstract Environmental awareness among consumers has prompted many industries, including cosmetic companies, to turn to “green” alternatives for their products. Complex amphiphiles, self-assembled structures of surfactant molecules, give products such as shampoo and liquid hand soap their physical properties; new formulations can be both physically benign and environmentally safe to synthesize. In July 2012, in order to determine the molecular properties of certain complex amphiphiles under varying conditions, small angle x-ray scattering (SAXS) patterns of over 150 samples were obtained at the National Synchrotron Light Source (NSLS) X6B beamline. Current techniques (manually treating one data frame at a time) for analyzing such data are insufficient for the large amount of data that has been acquired; and so, a much faster method is necessary. This project created a method for consistently treating, visualizing, and statistically analyzing these large data sets. Using tools such as Python, we developed large- scale processes in Datasqueeze, a program specifically designed for SAXS images, and MATLAB, whose graphing and imaging capabilities we utilized. This will enable us to finally analyze the July 2012 data in an automatic, time-efficient manner. Such techniques can then be applied not only to studies on complex amphiphiles, but also any other SAXS study at NSLS-II or similar facility.
- 3. 3 I. Introduction Over the past decade, many industries have turned to “green” alternatives for their products in response to increased environmental awareness among consumers. Lubrizol Corp. is no exception to this. They produce chemicals that are sold to companies in the personal care industry, for products such as shampoos, liquid hand soaps, and similar products. Complex amphiphiles, self-assembled structures of surfactant molecules, give these products several physical-chemical properties, such as rheology and clarity, and could reduce skin and eye irritation. In order to test new formulations of complex amphiphiles that are both physically benign and environmentally safe to synthesize, Lubrizol was interested in using SAXS imaging, a fast method of determining the structure, and by extension the physical properties, of the new formulations. In July 2012, representatives from Lubrizol took SAXS images at the NSLS X6B beamline of over 150 samples of complex amphiphiles; this high-throughput data was enabled by well-plates (see Figure 1), which could hold many samples at once during automated scans. Figure 1. Example of a well-plate used in SAXS scans. Holes in the Teflon block are covered by mylar film to contain liquid and gel samples, then plates clamp the assembly together.
- 4. 4 However, the July 2012 experiments produced too much data for current techniques to handle. These current techniques involve manually unwarping (a process to correct from distortion from the detector) each individual frame, extracting the relevant information, and then using that information. Not only were there over 150 samples1 , but the experiments produced over 300 data frames, in part due to the beamline technology; automated scans would scan every well in a well-plate (or row of a well-plate), regardless of content (see Figure 1). Therefore, the data from this experiment has been left largely untouched due to the huge amount of time that would have been necessary to thoroughly analyze it. The main focus of this project was to develop a method to handle the large amount of SAXS images from the July 2012 Lubrizol experiments, a process which can hopefully be applied to other SAXS experiments involving high throughput data. II. Methods Thorough analysis of the data first required thorough knowledge of the data. Before any automated programs were created, I made a spreadsheet so that most of the information was in the same place. The first column contained the file name; the second, a 1 if the file corresponded to a sample, a 0 if the file was a background; and the third, the sample name if the previous column contained a 1. Several other columns were also added, but were to be filled later with the help of Python. Secondly, since the detector that was used to take this data did not unwarp the data automatically, we had to decide whether or not it was necessary. To do this, plots of intensity versus q were made for two different frames of Silver Behenate (used to calibrate detector position in the Datasqueeze2 software; see Figure 2): one that was raw data, and one that had
- 5. 5 been unwarped. Significant differences in the two graphs at higher q ranges told us that unwarping was necessary (See Figure 3). We then used Python to automatically unwarp all of the files, by way of a special script created for the beamline. The next step in the progression was to create a MATLAB program that could plot SAXS images. We decided to create a program that specifically plotted them based on their well-plate position; this decision was made in part because we didn’t know relevant information about the samples (such as surfactant and salt concentration) until late in the experiment. The well-plate position, however, was contained in the file header. Python code was created to extract this information, and then import it into the master spreadsheet. From the spreadsheet, it was imported into MATLAB, where the program I created plotted them. This was done for all eighteen well-plates. After that, the main Python code was written to create a “data dictionary”. This code first used a directory search to find all the file names of the raw data, and for these files extracted the relevant information from the file header: the well-plate position and the x-ray monitor counts (this information was also imported to the master spreadsheet). Then, the code read the master Figure 2. Bragg-rings of Silver Behenate (AgBe). Known d-spacings are used to calibrate detector position in Datasqueeze Software. Figure 3. AgBe raw data (blue) versus unwarped (red). The differences in the peaks at higher Q indicate that all the images must be unwarped in order to maintain accuracy.
- 6. 6 spreadsheet to assign appropriate values to the variables “background” (true or false) and “sample” (the name of the actual sample). Then, all the monitor count values were averaged, and a new variable, “normalized”, was created, which divided the average value by the value of that specific data frame. Lastly, for those data frames that corresponded to samples, the “associated background” was a background frame on that same well plate. For each data frame, all of these variables were saved in a dictionary. This data dictionary could then be used to obtain the desired results in Datasqueeze. Python was used to create a batch file for Datasqueeze which read in each data frame that corresponded to a sample, normalized it, and subtracted its normalized associated background. We were thus left with diffraction data from only the sample itself. Using this data, plots of intensity versus q were made, for the whole data frame as well as “slices” of the pattern, in addition to plots of intensity versus chi (angle) to check for anisotropic samples. Fits of these peaks are currently in the process of being made. III. Results Figure 4. Example of one of the 18 different well- plate scans.
- 7. 7 Figure 4 shows an example of one of the outputs of the MATLAB program. Figure 5 shows the more than 350 SAXS images distributed over eighteen well-plates. This program allows large-scale visualization of the data so that immediate conclusions can be drawn. Figure 5. Five different sets of surfactant solutions distributed across 18 well-plates. Figure 6 shows an example of one of the “data dictionaries”. The dictionaries can be stored in a file which can be read in to any other Python program, thereby allowing the user to utilize any aspect of the data.
- 8. 8 Figure 6. Example of a dictionary for one SAXS image. Figure 7 shows an example of a normalized, background-subtracted isotropic sample. Figure 8 shows the same image for an anisotropic sample. This particular case is one of the few of all samples that showed significant anisotropy (non-uniform scattering patterns). Figure 9 is an example of an intensity versus q graph for an isotropic sample; Figure 10 is the same for an anisotropic sample. Based on these two graphs, one cannot distinguish between isotropic and anisotropic. So, although these are particularly important in the analysis of the data, as they reveal information about the structure of the amphiphiles, another method is needed to identify the anisotropic samples. Figure 11 shows the plot of intensity versus chi, averaged over q, for the isotropic sample; Figure 12, the same for the anisotropic. Notice the peaks in Figure 12 where the sample has higher intensity at that angle, versus the relatively flat line in Figure 11.
- 9. 9 Figure 7. Example of an isotropic sample that has been normalized and had the background subtracted. Figure 8. Example of an anisotropic sample that has been normalized and had the background subtracted. Figure 9 (above). Example of an intensity versus q plot for an isotropic sample. Figure 10 (below). Example of an intensity versus q plot for an anisotropic sample.
- 10. 10 IV. Discussion The development of these programs results in the capability of high throughput SAXS data analysis. The MATLAB program allows visualization of many data frames at once, in our case as many as seventy-two, so that immediate conclusions can be drawn. This program can also be slightly modified so that instead of being plotted by well-plate, the samples are plotted by surfactant concentration versus salt concentration, or other relevant variables, so that the impact on the diffraction patterns is obvious. Figure 11 (above). Example of a plot of intensity versus chi for isotropic sample. No significant peak shows that the sample is isotropic. Figure 12 (below). Example of a plot of intensity versus chi for anisotropic sample. Significant peaks show that the sample is anisotropic.
- 11. 11 While this project has come far, there are still several steps that need to be taken to ensure thorough analysis of the data. The fit parameters of the plots need to be ascertained, and then added to the dictionary. Most importantly, the data needs to actually be analyzed, a task that is made considerably easier by these programs. The SAXS data can be used to study the morphology of these surfactant micelles and phase behavior in aqueous solutions. The contribution to I(q) (see Figures 9 and 10) arising from the micellar electron density (Figure 13 (b)) is termed the form factor; the contribution to I(q) from the variation in electron density in ordered domains (Figure 13 (c)) is referred to as the structure factor. The two are combined in the “interaction peak” (Figure 143 ). While the interaction peak will always contain a form factor peak for any sample of amphiphiles, the degree of the prominence of the structure factor peak is what will vary significantly between samples and what can provide the most insight into the properties of the material for application to industry. V. References 1 All samples were provided by Lubrizol Corporation (Ohio). 2 Heiney, Paul A. Datasqueeze. Computer software. Datasqueeze Software. Vers. 3.0.4. N.p., 7 Feb. 2015. Web. 27 July 2015. Figure 13. (a) Self assembled amphiphiles in water. (b) Across a micelle, the electron density ρ(r) varies in regions dominated by water, denser headgroups, and slightly less dense alkyl chains. (c) Across a domain of ordered micelles, the electron density is periodic on the larger length scale of the micelle spacing. Figure 14. Scattering curve that illustrates typical contributions of form factor (dotted line) and structure factor (dashed line)3 .
- 12. 12 3 Itri, R., and L. Q. Amaral. "Micellar-shape Anisometry near Isotropic–liquid-crystal Phase Transitions." Physical Review E Phys. Rev. E 47.4 (1993): 2551-557. Print. VI. Acknowledgements This project was supported in part by the Brookhaven National Laboratory (BNL) Photon Sciences Department under the BNL Supplemental Undergraduate Research Program (SURP) (U. S. Department of Energy contract numbers DE-AC02-98CH10886 and DE-SC0012704). I would also like to thank Vesna Stanic of LNLS and Ramiro Galleguilos of Lubrizol Corp. for taking the original data and providing me with some essential background information about the project. Lastly, I would like to thank my mentor, Elaine DiMasi, for providing me with the opportunity to work on this project and giving me guidance.