| 1 |
Advanced Series processing – Regularized Alternating Least Squares (REGALS) |
| 2 |
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ |
| 3 |
.. _raw_regals: |
| 4 |
|
| 5 |
REGALS is an algorithm for deconvolution of mixtures in small angle scattering data. |
| 6 |
It can be applied to deconvolving overlapping peaks in SEC-SAXS, changing baseline |
| 7 |
and elution peaks in ion exchange chromatography SAXS, mixtures in time resolved |
| 8 |
SAXS data and equilibrium titration series data, and likely other cases. In |
| 9 |
practical application it can be thought of as an extension of the :ref:`evolving |
| 10 |
factor analysis (EFA) <raw_efa>` technique to more complex conditions where |
| 11 |
components are not necessarily entering and exiting the dataset in a strict |
| 12 |
first-in-first-out approach like in SEC-SAXS. EFA is recommended for |
| 13 |
standard SEC-SAXS data, but for more complex data, as listed above, you should |
| 14 |
use REGALS. REGALS can also handle deconvolution of SEC-SAXS data with a |
| 15 |
sloping baseline. |
| 16 |
|
| 17 |
To learn more about REGALS we recommend these resources: |
| 18 |
|
| 19 |
* *REGALS: a general method to deconvolve X-ray scattering data from evolving |
| 20 |
mixtures* S. P. Meisburger, D. Xu, and N. Ando. IUCrJ (2021). 8(2), 225-237. |
| 21 |
DOI: `10.1107/S2052252521000555 <https://doi.org/10.1107/S2052252521000555>`_ |
| 22 |
|
| 23 |
* The `source code for the REGALS algorithm on github. <https://github.com/ando-lab/regals>`_ |
| 24 |
|
| 25 |
* This `talk by Dr. Steve Meisburger <https://youtu.be/AO2kru097Wc>`_ |
| 26 |
|
| 27 |
If you use REGALS in RAW, in addition to citing the RAW paper, please cite the |
| 28 |
REGALS paper: S. P. Meisburger, D. Xu, and N. Ando. IUCrJ (2021). 8(2), 225-237. |
| 29 |
DOI: `10.1107/S2052252521000555 <https://doi.org/10.1107/S2052252521000555>`_ |
| 30 |
|
| 31 |
The written version of the tutorial follows. |
| 32 |
|
| 33 |
Deconvolving ion exchange chromatograph coupled SAXS (IEC-SAXS) data |
| 34 |
************************************************************************ |
| 35 |
|
| 36 |
Ion exchange chromatography is similar to size exclusion chromatography, except |
| 37 |
that it uses a charged column media and a salt gradient to separate samples by |
| 38 |
charge rather than size. IEC-SAXS (sometimes called anion exchange or AEX-SAXS) |
| 39 |
can be useful in cases where there are multiple species in solution that are |
| 40 |
not separable by size. However, the use of a salt gradient causes a changing |
| 41 |
buffer background for the sample, making accurate buffer subtraction challenging. |
| 42 |
REGALS can be used to deconvolve the scattering of the eluted macromolecules |
| 43 |
from the changing scattering of the buffer. |
| 44 |
|
| 45 |
|
| 46 |
#. Clear all of the data in RAW. Load the **nrde_iec.hdf5** file in the |
| 47 |
**series_data** folder. |
| 48 |
|
| 49 |
* *Note:* The data were provided by the Ando group at Cornell University, |
| 50 |
and were originally published in the paper: Parker, M. J. et al. (2018). |
| 51 |
*An endogenous dAMP ligand in Bacillus subtilis class Ib RNR promotes |
| 52 |
assembly of a noncanonical dimer for regulation by dATP.* Proc Natl |
| 53 |
Acad Sci USA 115, E4594–E4603. |
| 54 |
DOI: `10.1073/pnas.1800356115 <https://doi.org/10.1073/pnas.1800356115>`_. |
| 55 |
It is also available from the `REGALS github repository <https://github.com/ando-lab/regals>`_. |
| 56 |
|
| 57 |
|regals_series_plot_png| |
| 58 |
|
| 59 |
#. We will use REGALS to extract out the scattering components in both peaks |
| 60 |
of the data. Right click on the **nrde_iec.hdf5** item in the Series list. |
| 61 |
Select the "REGALS" option. |
| 62 |
|
| 63 |
#. The REGALS window will be displayed. On the left are controls, on the right are plots of |
| 64 |
the value of the singular values and the first autocorrelation of the left and right |
| 65 |
singular vectors. |
| 66 |
|
| 67 |
* *Note:* Large singular values indicate significant components. What |
| 68 |
matters is the relative magnitude, that is, whether the value is large |
| 69 |
relative to the mostly flat/unchanging value of high index singular values. |
| 70 |
|
| 71 |
* *Note:* A large autocorrelation indicates that the singular vector is |
| 72 |
varying smoothly, while a low autocorrelation indicates the vector is |
| 73 |
very noisy. Vectors corresponding to significant components will tend |
| 74 |
to have autocorrelations near 1 (roughly, >0.6-0.7) and vectors |
| 75 |
corresponding to insignificant components will tend to have |
| 76 |
autocorrelations near 0. |
| 77 |
|
| 78 |
|regals_panel_png| |
| 79 |
|
| 80 |
#. Nominally REGALS should work with either subtracted or unsubtracted data. |
| 81 |
Here, a buffer measured before the start of the dataset has been |
| 82 |
subtracted off of all the profiles in the dataset. This helps reduce the |
| 83 |
magnitude of the buffer components in the data, which can make it easier |
| 84 |
to refine the macromolecular components. We will use the full range |
| 85 |
of the dataset, but REGALS can also be done on restricted ranges. |
| 86 |
|
| 87 |
* *Note:* Using a restricted range for REGALS might be useful to exclude |
| 88 |
one or more components of a complicated elution from the deconvolution. |
| 89 |
The more components you have the harder it is to do REGALS. There is a |
| 90 |
trade off in the amount of data used (more is better) and the number of |
| 91 |
components in the deconvolution (fewer is better) that can require |
| 92 |
some experimentation to find the right balance for a given dataset. |
| 93 |
|
| 94 |
#. RAW attempts to automatically determine how many significant singular values |
| 95 |
(SVs) there are in the selected range. This corresponds to the number of |
| 96 |
significant scattering components in solution that REGALS will attempt to |
| 97 |
deconvolve. At the bottom of the control panel, you should see that RAW |
| 98 |
thinks there are four significant SVs (scattering components) in our data. |
| 99 |
For this data set, that is accurate. We evaluate the number of significant |
| 100 |
components by how many singular values are above the baseline, and how many |
| 101 |
components have both left and right singular vectors with autocorrelations |
| 102 |
near one. For this data there are four singular values above baseline, and |
| 103 |
four singular vectors with autocorrelations near 1 (see step 3). |
| 104 |
|
| 105 |
|regals_components_png| |
| 106 |
|
| 107 |
* *Note:* Typically you expect the number of significant singular values and |
| 108 |
the number of singular vectors with autocorrelations near 1 to be equal. |
| 109 |
If they aren't, it likely indicates a weak or otherwise poorly resolved |
| 110 |
component in the dataset. Try the deconvolution first with the lower then |
| 111 |
the higher number of components. |
| 112 |
|
| 113 |
* *Note:* RAW can find the wrong number of components automatically. You will |
| 114 |
always want to double check this automatic determination against the SVD results in |
| 115 |
the plots. If you change the data range used (or data type), the number |
| 116 |
of components will not automatically update so you should check and update |
| 117 |
it if necessary. |
| 118 |
|
| 119 |
#. For REGALS there are two other settings you should check. First, set |
| 120 |
the experiment type. In this case, the default of 'IEC/SEC-SAXS' is |
| 121 |
correct. This sets default values later on in the deconvolution. Second |
| 122 |
specify whether you want to use the evolving factor plots to find |
| 123 |
initial guesses for the range of components in solution. This is useful in |
| 124 |
highly sampled datasets like IEC-SAXS, but may not be very accurate for |
| 125 |
more sparsely sampled data like a titration series, and so can be skipped. |
| 126 |
For this dataset leave it checked. |
| 127 |
|
| 128 |
|regals_exp_type_png| |
| 129 |
|
| 130 |
#. Click the "Next" button in the lower right-hand corner of the window to |
| 131 |
advance to the second stage of the REGALS analysis |
| 132 |
|
| 133 |
* *Note:* If, as in this case, you are going to use the EFA plots |
| 134 |
to find the component ranges, it will take some time to compute |
| 135 |
the necessary values for this next step, so be patient. |
| 136 |
|
| 137 |
|regals_bkg_efa_png| |
| 138 |
|
| 139 |
#. A new window will open on top of the current one when you click "Next". The |
| 140 |
bottom window is the main REGALS window, with the forward and backward |
| 141 |
evolving factor plots. On top of that is the REGALS Background Components |
| 142 |
window. We will first use the Background Components window to estimate |
| 143 |
how many of our components are background components. |
| 144 |
|
| 145 |
* *Note:* For the purposes of REGALS, a background component is a component |
| 146 |
that spans the full range of the dataset. |
| 147 |
|
| 148 |
* *Note:* If you select an experiment mode other than 'IEC/SEC-SAXS' |
| 149 |
the Background Components window will not open automatically, but |
| 150 |
can still be opened by clicking the "Find background components" button |
| 151 |
if desired. |
| 152 |
|
| 153 |
#. In the Background Components window, click "Add Region" to add a region to |
| 154 |
the plot. The SVD of the data in that region is shown in the plots on the right. |
| 155 |
Set the range of that region to 0 to 100. The SVD plots show that there is |
| 156 |
only one strong component in the dataset in this range. |
| 157 |
|
| 158 |
|regals_bkg_1_png| |
| 159 |
|
| 160 |
#. Add a region corresponding to the last 100 frames of the dataset. This will |
| 161 |
appear as a second colored range on the series plot, with corresponding |
| 162 |
second colored sets of lines in the SVD plots on the right. Notice that |
| 163 |
there is just one component in the last 100 frames of the dataset. |
| 164 |
|
| 165 |
* *Note:* The left autocorrelation is shown with the solid line, the right |
| 166 |
autocorrelation with the dashed line. |
| 167 |
|
| 168 |
|regals_bkg_2_png| |
| 169 |
|
| 170 |
#. Add several more similar regions through the initial upward sloping buffer |
| 171 |
region. Notice that there appears to be just one strong value throughout |
| 172 |
most of the buffer region, both before and after the eluted peaks. So as a |
| 173 |
starting point we will set the number of significant background components |
| 174 |
to 1, by setting the "# Significant SVs" input to 1. Once that's done, |
| 175 |
click the "Done" button. |
| 176 |
|
| 177 |
* *Tip:* You can remove regions if they are not longer useful. To do so, |
| 178 |
select the region by clicking to the right of the pick button and then |
| 179 |
clicking the "Remove region" button. |
| 180 |
|
| 181 |
|regals_bkg_3_png| |
| 182 |
|
| 183 |
#. The Background Components window will close and the number of background |
| 184 |
components shown in the main REGALS window will update. The main REGALS |
| 185 |
window will now be fully visible. |
| 186 |
|
| 187 |
* *Note:* You can reopen the Background Components window using the "Find |
| 188 |
background components" button if desired. |
| 189 |
|
| 190 |
|regals_efa_1_png| |
| 191 |
|
| 192 |
#. In the User Input panel, tweak the "Forward" value start frames so that the |
| 193 |
frame number, as indicated by the open circle on the plot, aligns with |
| 194 |
where the singular value first starts to rise above the baseline. This |
| 195 |
should be around 0, 350, 750, and 1195. |
| 196 |
|
| 197 |
* *Note:* For the Forward EFA plot, SVD is run on just the first two |
| 198 |
frames, then the first three, and so on, until all frames in the range |
| 199 |
are included. As more frames are added, the singular values change, as |
| 200 |
shown on the plot. When a singular value starts rising above the |
| 201 |
baseline, it indicates that there is a new scattering component in the |
| 202 |
scattering profile measured at that point. So, for the first ~350 |
| 203 |
frames, there is only one scattering components (i.e. just buffer |
| 204 |
scattering). At frame ~350, we see the second singular value (the |
| 205 |
singular value with index 1, labeled SV 1 on the plot) start to |
| 206 |
increase, showing that we have gained a scattering component. |
| 207 |
|
| 208 |
* *Note:* One component starts above baseline, indicating there is already |
| 209 |
a significant scattering component at the start of our dataset. In this |
| 210 |
case that is the sloping buffer gradient. |
| 211 |
|
| 212 |
#. In the User Input panel, tweak the "Backward" value start frames so that the |
| 213 |
frame number, as indicated by the open circle on the plot, aligns with where |
| 214 |
the singular value drops back to near the baseline. This should be around |
| 215 |
700, 1325, 1600, and 1736. |
| 216 |
|
| 217 |
* *Note:* For the Backward EFA plot, SVD is run on just the last two |
| 218 |
frames, then the last three, and so on, until all frames in the range |
| 219 |
are included. As more frames are added, the singular values change, as |
| 220 |
shown on the plot. When a singular value drops back to baseline, it |
| 221 |
indicates that a scattering component is leaving the dataset at that |
| 222 |
point. |
| 223 |
|
| 224 |
* *Note:* One component ends above baseline, indicating there is still |
| 225 |
a significant scattering component at the end of our dataset. In this |
| 226 |
case that is the sloping buffer gradient. |
| 227 |
|
| 228 |
|regals_efa_2_png| |
| 229 |
|
| 230 |
#. Click the "Next" button in the bottom right corner to move to the last |
| 231 |
stage of the REGALS analysis. |
| 232 |
|
| 233 |
|regals_regals_1_png| |
| 234 |
|
| 235 |
#. This window shows controls on the left and results on the right. In the |
| 236 |
controls area at the top are general controls. You can adjust the number |
| 237 |
of components, calibrate the X axis, and set the convergence criteria for |
| 238 |
the REGALS algorithm. A plot of the ranges is also shown. On the bottom |
| 239 |
are the controls for each individual component. We won't go through all |
| 240 |
the possible permutations for each setting, so if you want to know more |
| 241 |
check out the links :ref:`at the top <raw_regals>` of this tutorial. |
| 242 |
|
| 243 |
* *Note:* The ranges are automatically assigned based on the start and |
| 244 |
end points for components found in the EFA plots. Background components |
| 245 |
ranges are assigned on the principle of first-in last-out, whereas all the |
| 246 |
other component ranges are assigned by first-in first-out. |
| 247 |
|
| 248 |
* *Note:* Components are only shown three across. Scroll down in the |
| 249 |
component area to see more component settings. |
| 250 |
|
| 251 |
#. Now we need to start refining our component ranges, and tweaking other settings |
| 252 |
in the components. Component 0 looks good for now, so we will start with component 1. |
| 253 |
On the concentration plot on the right, notice that despite the component 1 |
| 254 |
range being defined from only 350-700, the component shows features at a |
| 255 |
wide range of frame numbers (index). Looking at the scattergram with the |
| 256 |
ranges plotted, there's no obvious elution component in the range where |
| 257 |
component 1 is defined. This means that component 1 is likely a background |
| 258 |
component as well. Since the SVD for the background components and the EFA |
| 259 |
plots were not clear on where the component should end, we will fit it |
| 260 |
from the start point we found, 350, until the end of the dataset. To do |
| 261 |
this, change the endpoint of the concentration range to 1736. Also uncheck |
| 262 |
the 'Zero at Xmax' box for this component, as it may contribute to the |
| 263 |
buffer scattering at the end of the dataset. |
| 264 |
|
| 265 |
* *Tip:* If you return to the Background Components window, if you add |
| 266 |
a single range from 0 to 700 you'll see two components, indicating |
| 267 |
that there are likely multiple background components. However, if you |
| 268 |
look at the end region, assuming you avoid the tail of the peak (~1600-1736) |
| 269 |
there's only one component. So it's not clear where the first component might |
| 270 |
end and the second one start, and how much coexistence there is between |
| 271 |
these two background components. In this case, fitting both to end at the |
| 272 |
end of the dataset works well. |
| 273 |
|
| 274 |
|regals_regals_2_png| |
| 275 |
|
| 276 |
#. Because it can take a while to run, REGALS does not automatically update |
| 277 |
the results. To see how changing the range changed the deconvolution, |
| 278 |
click the "Run REGALS" button. |
| 279 |
|
| 280 |
* *Note:* If lambda is automatically updated for a component, this value |
| 281 |
in the GUI will not be updated until you run REGALS. |
| 282 |
|
| 283 |
* *Note:* When you have changes to your deconvolution settings and REGALS |
| 284 |
hasn't been run with those settings the "Run REGALS" button will have |
| 285 |
a yellow background. |
| 286 |
|
| 287 |
|regals_regals_3_png| |
| 288 |
|
| 289 |
#. There is a definite improvement in the REGALS results after rerunning |
| 290 |
the data. Next we will refine the protein components. We will first refine the |
| 291 |
ranges. On the concentration plot, notice that the component 3 (red) |
| 292 |
concentration is pulled down to zero pretty sharply on the left side. At |
| 293 |
the same point there's a peak in the component 2 (green) concentration right |
| 294 |
where the red one starts. There's also a small spike in the chi^2 plot around |
| 295 |
frame 1200, which is where component 3 starts. This indicates that we've |
| 296 |
restricted component 3 excessively and some of the scattering is going into |
| 297 |
component 2. We will change the range for component 3 to start at an |
| 298 |
earlier point and see how that affects the deconvolution. To do this, set |
| 299 |
the start of component 3 to 1150 and run REGALS. |
| 300 |
|
| 301 |
|regals_regals_4_png| |
| 302 |
|
| 303 |
#. After running REGALS, notice that the chi^2 spike is completely gone, |
| 304 |
and that the range 3 concentration is coming back down to zero in a more |
| 305 |
natural way. Also notice that the spike in component 2 concentration |
| 306 |
is reduced. If you look closely at the component 2 concentration you can |
| 307 |
see a bit of a double peak around ~1150. There could be a little bit more |
| 308 |
component 3 there, so we will start the component 3 range earlier. Try 1125 |
| 309 |
and 1100 for the component 3 start. |
| 310 |
|
| 311 |
#. As there's minimal change between 1100 and 1125, set the component 3 start |
| 312 |
back to 1125. |
| 313 |
|
| 314 |
#. At this point the deconvolution is starting to look reasonable. There's nothing |
| 315 |
obviously wrong with the component ranges, so next we will adjust the lambda |
| 316 |
values for the components. These control the degree of smoothing in the |
| 317 |
deconvolution. For strong components with lots of measured profiles, like |
| 318 |
the peaks in this dataset, we don't need a large lambda. In fact, we might |
| 319 |
not need any lambda. For component 2 and 3 concentrations turn off "Auto |
| 320 |
lambda", and set the lambda value to 0. Then run REGALS again. |
| 321 |
|
| 322 |
* *Tip:* Make sure you're setting lambda for the concentration, not |
| 323 |
the profile, for each component. |
| 324 |
|
| 325 |
|regals_regals_5_png| |
| 326 |
|
| 327 |
#. After running REGALS with the peak component concentration lambdas |
| 328 |
set to 0, the component 2 concentration (green) has a small negative dip |
| 329 |
at the end of the concentration range. This indicates that we should adjust |
| 330 |
the range for that component. In particular, we'll reduce the end point to |
| 331 |
try to eliminate that dip. Try 1300 and 1275 as endpoints for component |
| 332 |
2 concentration. |
| 333 |
|
| 334 |
* *Note:* If we had set the lambdas particularly poorly, we would start to see |
| 335 |
the chi^2 plot deviate from ~1. Since we don't, we can conclude |
| 336 |
that our lambda values are probably okay. |
| 337 |
|
| 338 |
#. Notice that 1275 essentially eliminates the dip in the concentration. |
| 339 |
Set that as the endpoint for the component 2 concentration. |
| 340 |
|
| 341 |
* *Try:* You can try small tweaks near 1275 if you like, but you |
| 342 |
shouldn't see significant changes in the concentration shape. So we'll |
| 343 |
leave it set at 1275 for now. |
| 344 |
|
| 345 |
#. The final thing we will adjust is the lambdas for the background component |
| 346 |
concentrations. Those concentration profiles are a bit wavy. We would |
| 347 |
expect the concentrations of the buffer to increase linearly, since a |
| 348 |
linear salt gradient was applied during elution. To increase the smoothness |
| 349 |
of the concentration, we will increase lambda. Turn off "Auto lambda" for |
| 350 |
components 0 and 1. Then use the up arrow next to the lambda box to |
| 351 |
increase each lambda by an order of magnitude. Run REGALS to see how this |
| 352 |
affects the deconvolution. |
| 353 |
|
| 354 |
* *Note:* Generally you want to adjust lambda by an order of magnitude |
| 355 |
or more. Smaller adjustments will have minimal effect on the deconvolution. |
| 356 |
|
| 357 |
* *Tip:* You can also type a new lambda value directly into the box. |
| 358 |
|
| 359 |
|regals_regals_6_png| |
| 360 |
|
| 361 |
#. There are two major changes with the new lambdas. First, the high q of |
| 362 |
the scattering profiles for ranges 2 and 3 will get more similar. Second, the |
| 363 |
concentration for ranges 0 and 1 will get smoother. The change in high q |
| 364 |
comes from a decrease in the high q values component 2, while component 3 |
| 365 |
stays mostly the same. This implies that more of the buffer scattering is |
| 366 |
getting picked out component 2. Keep increasing both buffer component |
| 367 |
lambdas by an order of magnitude until this stops changing. You will |
| 368 |
also see a sudden and dramatic change in the component 1 profile at |
| 369 |
some point. |
| 370 |
|
| 371 |
#. Once you see the high q backgrounds match and the large change in the |
| 372 |
scattering profile for component 1, it means you're oversmoothing |
| 373 |
the buffer components. Reduce the lambda for both buffer components to |
| 374 |
the last good value, which would be ~4e8. |
| 375 |
|
| 376 |
|regals_regals_7_png| |
| 377 |
|
| 378 |
#. This is a more or less optimum solution for REGALS deconvolution for this |
| 379 |
dataset. Since we're satisfied, we can now save the results. First, |
| 380 |
click on the "Save REGALS data (not profiles)" button in the bottom |
| 381 |
right corner and save. This saves a spreadsheet (.csv) file with all |
| 382 |
of the information from REGALS that isn't the scattering profiles, |
| 383 |
including the component settings and concentration and chi^2 vs. frame |
| 384 |
data. |
| 385 |
|
| 386 |
#. Finally, click the "Done" button to close the REGALS window and send the |
| 387 |
deconvolved scattering profiles to the Profiles Plot. |
| 388 |
|
| 389 |
#. In the main RAW window, go to the Profiles control tab and the Profiles |
| 390 |
plot. There you should see the deconvolved profiles. The labels _0, _1, |
| 391 |
_2, and _3 correspond to the 0, 1, 2, and 3 components from REGALS. |
| 392 |
|
| 393 |
|regals_profiles_png| |
| 394 |
|
| 395 |
|
| 396 |
|
| 397 |
Deconvolving equilibrium titration series SAXS data |
| 398 |
************************************************************************ |
| 399 |
|
| 400 |
Titration series SAXS data looks at the change in scattering profile of a |
| 401 |
macromolecule as a substance (salt, substrate, another macromolecule, etc.) |
| 402 |
is titrated in or our of the solution. It is done in equilibrium, so buffers |
| 403 |
are prepared ahead of time and samples are equilibrated in the buffer, then |
| 404 |
measured. These measurements are typically done in batch mode SAXS (without |
| 405 |
in-line separation from SEC), and often involve transitions between different |
| 406 |
conformations or oligomeric states, and so the measured scattering at a |
| 407 |
given titration point is not from a homogeneous and monodisperse sample. |
| 408 |
We can use REGALS to deconvolve the different scattering components in the |
| 409 |
titration series to get pure profiles for each component. |
| 410 |
|
| 411 |
#. Clear all of the data in RAW. Load the **pheh_titration.hdf5** file in the |
| 412 |
**series_data** folder. |
| 413 |
|
| 414 |
* *Note:* This data are a titration series of phenylalanine (L-phe) into a |
| 415 |
sample of phenylalanine hydroxylase (PheH). 16 different concentration |
| 416 |
points were collected, ranging from 0 to 80 mM L-phe. The data are |
| 417 |
already background subtracted, using buffers containing a matching |
| 418 |
concentration of L-phe. A conformational change has previously been |
| 419 |
observed on PheH binding L-phe. A small amount of aggregation was also |
| 420 |
observed at all concentrations, preventing the use of saturated |
| 421 |
endpoints in the titration series to completely determine each conformational |
| 422 |
state in batch mode experiments. We will deconvolve both conformations |
| 423 |
and the aggregate scattering from the titration series. `Prior analysis |
| 424 |
of this data without the use of REGALS is published |
| 425 |
<http://dx.doi.org/10.1021/jacs.6b01563>`_. |
| 426 |
|
| 427 |
* *Note:* The data were provided by the Ando group at Cornell University, |
| 428 |
and were originally published in the paper: Meisburger, S. P. et al. (2016). |
| 429 |
*Domain Movements upon Activation of Phenylalanine Hydroxylase |
| 430 |
Characterized by Crystallography and Chromatography-Coupled |
| 431 |
Small-Angle X-ray Scattering.* J Am Chem Soc 138, 6506–6516. |
| 432 |
DOI: `10.1021/jacs.6b01563 <https://doi.org/10.1021/jacs.6b01563>`_. |
| 433 |
It is also available from the `REGALS github repository |
| 434 |
<https://github.com/ando-lab/regals>`_. |
| 435 |
|
| 436 |
|regals_pheh_series_plot_png| |
| 437 |
|
| 438 |
#. We will use REGALS to extract out the scattering of the individual |
| 439 |
conformers and the aggregate in the titration series. Right click on the |
| 440 |
**pheh_titration.hdf5** item in the Series list. Select the "REGALS" option. |
| 441 |
|
| 442 |
#. Normally we use the plots of singular values and autocorrelations to determine |
| 443 |
the number of significant singular values (i.e. scattering components) in |
| 444 |
the dataset. Based on the plots of this titration series, there are |
| 445 |
~4-5 significant singular values (SVs), for reference RAW's automated |
| 446 |
method found 4 such values. However, prior knowledge indicates there |
| 447 |
are only two conformations we care about, and we suspect the other components |
| 448 |
are various aggregates. So we will fit the data to three components: |
| 449 |
each conformation and a single aggregate scattering component. Set the |
| 450 |
"# Significant SVs" to 3. |
| 451 |
|
| 452 |
#. Set the experiment type to "Titration". This affects the default settings |
| 453 |
later in the deconvolution. |
| 454 |
|
| 455 |
#. Because we have so few data points, and we're using fewer components than |
| 456 |
the actual number of significant SVs, trying to find the component ranges |
| 457 |
using the EFA plots will not be useful. Uncheck the "Use EFA to find |
| 458 |
component ranges" box. |
| 459 |
|
| 460 |
|regals_pheh_exp_settings_png| |
| 461 |
|
| 462 |
#. Click the "Next" button in the lower right-hand corner of the window to |
| 463 |
advance to the final stage of the REGALS analysis. |
| 464 |
|
| 465 |
* *Note:* In this case, since we aren't determining the component ranges |
| 466 |
with EFA, REGALS is not automatically run initially. We will have to |
| 467 |
set the component settings, then run REGALS. |
| 468 |
|
| 469 |
|regals_pheh_regals_1_png| |
| 470 |
|
| 471 |
#. Notice that the profile part of each component has a "realspace" regularizer. |
| 472 |
This means that instead of constraining the data in q space via the |
| 473 |
scattering profile we will be constraining the data in real space using |
| 474 |
the P(r) function. So in addition to setting component ranges for the |
| 475 |
concentration and the lambda values, we will also set the |Dmax| value for |
| 476 |
the P(r) function. |
| 477 |
|
| 478 |
|regals_pheh_regals_2_png| |
| 479 |
|
| 480 |
#. The titration concentrations are not equally spaced from 0 to 80 mM, so we |
| 481 |
will calibrate the X axis appropriately. Click the "Calibrate X axis" button. |
| 482 |
|
| 483 |
|regals_pheh_regals_3_png| |
| 484 |
|
| 485 |
#. In the window that opens click the "Load X values from file" button. |
| 486 |
|
| 487 |
|regals_pheh_regals_4_png| |
| 488 |
|
| 489 |
#. Select the **pheh_titration_conc.txt** file in the **series_data** folder and |
| 490 |
load it in. |
| 491 |
|
| 492 |
* *Note:* A calibration file should consist of a single column of the |
| 493 |
concentration values, in the order that the profiles were loaded into |
| 494 |
the series. So the first line of the file has the concentration for |
| 495 |
the first profile in the series, and so on. |
| 496 |
|
| 497 |
#. Upon loading you should see the values in the X column update. |
| 498 |
|
| 499 |
* *Note:* The concentrations are in uM, so you will see concentrations |
| 500 |
from 0 to 80000. |
| 501 |
|
| 502 |
#. The concentration values are not linearly spaced, as is typical of a |
| 503 |
titration series. It is also most common to work with titration series |
| 504 |
on a logarithmic concentration scale. To do this, we need to define |
| 505 |
the first concentration as not zero (as log(0) is undefined). For this |
| 506 |
data we will set the concentration to 10 uM. Double click in the first box |
| 507 |
in the X column and enter 10.0. |
| 508 |
|
| 509 |
* *Note:* You can enter all the concentrations manually if you want, you |
| 510 |
don't have to create the concentration file we loaded in the previous |
| 511 |
steps. However, for a long series this can get a bit tedious. |
| 512 |
|
| 513 |
|regals_pheh_regals_5_png| |
| 514 |
|
| 515 |
#. As we want to work with our data in log space, select "Log10(X)" in the |
| 516 |
"Use for X axis" list. Then click the OK button to exit the window and |
| 517 |
save the X calibration. |
| 518 |
|
| 519 |
|regals_pheh_regals_6_png| |
| 520 |
|
| 521 |
#. In the main REGALS window, notice that the X axis is now calibrated as |
| 522 |
Log(concentration), and that the ranges for the concentration components |
| 523 |
have updated accordingly. |
| 524 |
|
| 525 |
#. Next we will set the ranges for our components. We will use components 0 |
| 526 |
and 1 as the different conformations, resting and active respectively, |
| 527 |
and component 2 as the aggregate. Prior analysis of the system showed that |
| 528 |
of the two conformations, only the resting conformation, component 0 is |
| 529 |
present at 0 mM L-phe. So set the "Zero at Xmin" to True for component 1. |
| 530 |
|
| 531 |
#. Prior analysis showed that features of the scattering profile associated |
| 532 |
with the active state saturated above 3 mM. So we will assume that there |
| 533 |
is no contribution to the scattering from the resting state at/above 3 mM |
| 534 |
(3.48 on the log10(X) axis). Set the end range for component 0 to 3.48, |
| 535 |
and apply a Zero at Xmax boundary condition to it. |
| 536 |
|
| 537 |
|regals_pheh_regals_7_png| |
| 538 |
|
| 539 |
#. We will apply not constraints to the concentration range of the aggregate, |
| 540 |
so we are done with the concentration ranges. Next we need to set the |
| 541 |
|Dmax| values for the components. We'll start with the aggregate. Since we |
| 542 |
have no prior knowledge about the aggregate, we will assume it is a non-specific |
| 543 |
size. Because of the q range of the data, the largest dimension of an object |
| 544 |
that can be measured is ~300 Å, based on the Shannon limit of |
| 545 |
:math:`D_{max}<\pi/q_{min}`. So we will set the |Dmax| value for the |
| 546 |
aggregate, component 2, to 300. |
| 547 |
|
| 548 |
|regals_pheh_regals_8_png| |
| 549 |
|
| 550 |
#. We will now run REGALS to get an initial look at the deconvolution. |
| 551 |
Click the "Run REGALS" button. |
| 552 |
|
| 553 |
|regals_pheh_regals_9_png| |
| 554 |
|
| 555 |
#. Looking at the results, it's already a reasonable deconvolution. The |
| 556 |
concentrations make sense with previous knowledge, e.g. that there's |
| 557 |
a transition from the resting (component 0) to active (component 1) |
| 558 |
with increasing L-phe concentration, and that there's a low level |
| 559 |
of aggregate throughout that increases significantly at higher |
| 560 |
concentrations of L-phe. The profiles and P(r) functions, while not |
| 561 |
completely correct, are at least reasonable shapes, and the chi^2 |
| 562 |
is mostly relatively low. |
| 563 |
|
| 564 |
* *Note:* On the concentration plot, the markers are the concentrations |
| 565 |
calculated at the titration points. The smooth lines are the |
| 566 |
concentration calculated at the regularlizer grid points, so it is |
| 567 |
effectively interpolated between the measured titration points. |
| 568 |
When you have more than 40 profiles in the series, only the concentration |
| 569 |
at the actual measured points is shown, and there it is shown as |
| 570 |
a continuous line, not individual points (as in the above IEC-SAXS |
| 571 |
example). |
| 572 |
|
| 573 |
#. Next we will refine the |Dmax| values for the resting and active states. |
| 574 |
On the P(r) plot, notice that for both components the P(r) function |
| 575 |
is forced to zero sharply, indicating that an underestimated |Dmax| value. |
| 576 |
Increase both |Dmax| values to 110 and run REGALS again. |
| 577 |
|
| 578 |
|regals_pheh_regals_10_png| |
| 579 |
|
| 580 |
#. Notice that at a |Dmax| of 110 the P(r) functions are still somewhat |
| 581 |
forced to zero, but the chi^2 is lower, so this has improved the |
| 582 |
deconvolution. Continue increasing |Dmax| in steps of 10-20 until you |
| 583 |
reach 160. You should notice several things. First, there's a range |
| 584 |
from ~130-150 where the chi^2 is relatively stable, indicating |Dmax| |
| 585 |
values in those ranges all provide relatively good fits to the data. |
| 586 |
Second, as you increase |Dmax| the concentration of the aggregate |
| 587 |
decreases, particularly in the lower titration concentrations. This implies |
| 588 |
that having a larger |Dmax| is letting that component take up some of the |
| 589 |
aggregate scattering. Third, after ~120 the P(r) functions stop looking |
| 590 |
as forced to zero. Fourth at 160 the chi^2 starts noticeably increasing, |
| 591 |
indicating that's too large for the |Dmax| value. Based on this, we want |
| 592 |
to pick a |Dmax| value near 130, to exclude as much of the aggregate as we |
| 593 |
can while still getting good P(r) functions and chi^2 values. |
| 594 |
|
| 595 |
#. Set the |Dmax| of both components to 130 and run REGALS. |
| 596 |
|
| 597 |
* *Note:* The |Dmax| values found by previous analysis was ~130, so this |
| 598 |
validates our choice of |Dmax|. |
| 599 |
|
| 600 |
* *Note:* The |Dmax| values for different components won't necessarily |
| 601 |
agree. In this case they happen to. |
| 602 |
|
| 603 |
* *Tip:* If it takes a while to run REGALS every time you change a component, |
| 604 |
you can speed up the convergence by starting with the previous results. |
| 605 |
To do so, you would check the "Start with previous results" box. Then |
| 606 |
change the convergence criteria to "Iterations" and set the number of |
| 607 |
iterations to 10. This will allow you to quickly iterate on changes like |
| 608 |
|Dmax|, as long as the magnitude of the change is relatively small. |
| 609 |
Just be sure to set the convergence criteria back to the default (not |
| 610 |
using previous results, and Chi^2 with 1000 iterations) to do your |
| 611 |
final REGALS run. |
| 612 |
|
| 613 |
#. The REGALS deconvolution is now as optimized as we can make it. Notice |
| 614 |
that there's still a relatively high chi^2 value for the last frame. This |
| 615 |
indicates that the aggregate may be changing shape as the amount increases, |
| 616 |
and so a single component cannot fit all the data well. You can redo the |
| 617 |
deconvolution without the last data point if you want, the results are very similar |
| 618 |
albeit without the chi^2 spike for the last point. Since we're satisfied |
| 619 |
with these results we can now save them. First, click on "Save REGALS data |
| 620 |
(not profile)" and save. |
| 621 |
|
| 622 |
* *Note:* Among other things, this .csv file contains the P(r) functions, |
| 623 |
and the smoothed concentration curves (lines on the concentration plot). |
| 624 |
In this case the smoothed concentration curves are particularly useful |
| 625 |
because the protein changes shape but not size. As the P(r) functions |
| 626 |
are normalized to I(0), the concentration curves for both components |
| 627 |
are on the same overall scale. This means they differ by simply a |
| 628 |
uniform scale factor from the true concentrations (e.g. in mg/ml), |
| 629 |
and so could be useful to help characterize the two state transition. |
| 630 |
This will not always be true, such as if you're characterizing an |
| 631 |
oligomerization reaction. |
| 632 |
|
| 633 |
|regals_pheh_regals_11_png| |
| 634 |
|
| 635 |
#. Finally, click the "Done" button to close the REGALS window and send the |
| 636 |
deconvolved scattering profiles to the profiles plot. |
| 637 |
|
| 638 |
|
| 639 |
|
| 640 |
General notes |
| 641 |
**************** |
| 642 |
|
| 643 |
#. It is okay to mix and match different types of regularizers (e.g. have |
| 644 |
both smooth and real space profile regularizers) for the same series. |
| 645 |
|
| 646 |
#. You can change the regularizers away from the defaults for a given |
| 647 |
experiment type. |
| 648 |
|
| 649 |
#. The REGALS examples shown here were chosen to demonstrate the features |
| 650 |
of the GUI. REGALS is not restricted in application to just IEC-SAXS |
| 651 |
and titration data. It has been successfully applied to time resolved |
| 652 |
SAXS data (similar to the titration series example), and we expect it will |
| 653 |
be applied to a range of other types of experiments as well. |
| 654 |
|
| 655 |
#. As you saw in the PheH titration series example, it is quite useful, and |
| 656 |
sometimes necessary, to have additional information to input to the |
| 657 |
deconvolution, such as the range of the components, or a known maximum |
| 658 |
dimension. One of the advantages of REGALS is that it can incorporate |
| 659 |
these additional pieces of information to improve the deconvolution. |
| 660 |
|
| 661 |
#. By default, RAW bins the profiles before doing an SVD and calculating the |
| 662 |
evolving factor plots, in order to speed up the process. The final REGALS |
| 663 |
rotation is done on the full unbinned dataset. You can turn binning on and |
| 664 |
off and adjust the binning parameters in the Series options panel in the |
| 665 |
Advanced Options window. |
| 666 |
|
| 667 |
.. |regals_series_plot_png| image:: images/regals_series_plot.png |
| 668 |
:target: ../_images/regals_series_plot.png |
| 669 |
|
| 670 |
.. |regals_panel_png| image:: images/regals_panel.png |
| 671 |
:target: ../_images/regals_panel.png |
| 672 |
|
| 673 |
.. |regals_components_png| image:: images/regals_components.png |
| 674 |
:target: ../_images/regals_components.png |
| 675 |
|
| 676 |
.. |regals_exp_type_png| image:: images/regals_exp_type.png |
| 677 |
:target: ../_images/regals_exp_type.png |
| 678 |
:width: 300 px |
| 679 |
|
| 680 |
.. |regals_bkg_efa_png| image:: images/regals_bkg_efa.png |
| 681 |
:target: ../_images/regals_bkg_efa.png |
| 682 |
|
| 683 |
.. |regals_bkg_1_png| image:: images/regals_bkg_1.png |
| 684 |
:target: ../_images/regals_bkg_1.png |
| 685 |
|
| 686 |
.. |regals_bkg_2_png| image:: images/regals_bkg_2.png |
| 687 |
:target: ../_images/regals_bkg_2.png |
| 688 |
|
| 689 |
.. |regals_bkg_3_png| image:: images/regals_bkg_3.png |
| 690 |
:target: ../_images/regals_bkg_3.png |
| 691 |
|
| 692 |
.. |regals_efa_1_png| image:: images/regals_efa_1.png |
| 693 |
:target: ../_images/regals_efa_1.png |
| 694 |
|
| 695 |
.. |regals_efa_2_png| image:: images/regals_efa_2.png |
| 696 |
:target: ../_images/regals_efa_2.png |
| 697 |
:width: 175 px |
| 698 |
|
| 699 |
.. |regals_regals_1_png| image:: images/regals_regals_1.png |
| 700 |
:target: ../_images/regals_regals_1.png |
| 701 |
|
| 702 |
.. |regals_regals_2_png| image:: images/regals_regals_2.png |
| 703 |
:target: ../_images/regals_regals_2.png |
| 704 |
:width: 250 px |
| 705 |
|
| 706 |
.. |regals_regals_3_png| image:: images/regals_regals_3.png |
| 707 |
:target: ../_images/regals_regals_3.png |
| 708 |
:width: 250 px |
| 709 |
|
| 710 |
.. |regals_regals_4_png| image:: images/regals_regals_4.png |
| 711 |
:target: ../_images/regals_regals_4.png |
| 712 |
|
| 713 |
.. |regals_regals_5_png| image:: images/regals_regals_5.png |
| 714 |
:target: ../_images/regals_regals_5.png |
| 715 |
:width: 250 px |
| 716 |
|
| 717 |
.. |regals_regals_6_png| image:: images/regals_regals_6.png |
| 718 |
:target: ../_images/regals_regals_6.png |
| 719 |
:width: 250 px |
| 720 |
|
| 721 |
.. |regals_regals_7_png| image:: images/regals_regals_7.png |
| 722 |
:target: ../_images/regals_regals_7.png |
| 723 |
|
| 724 |
.. |regals_profiles_png| image:: images/regals_profiles.png |
| 725 |
:target: ../_images/regals_profiles.png |
| 726 |
|
| 727 |
.. |regals_pheh_series_plot_png| image:: images/regals_pheh_series_plot.png |
| 728 |
:target: ../_images/regals_pheh_series_plot.png |
| 729 |
|
| 730 |
.. |regals_pheh_exp_settings_png| image:: images/regals_pheh_exp_settings.png |
| 731 |
:target: ../_images/regals_pheh_exp_settings.png |
| 732 |
:width: 300 px |
| 733 |
|
| 734 |
.. |regals_pheh_regals_1_png| image:: images/regals_pheh_regals_1.png |
| 735 |
:target: ../_images/regals_pheh_regals_1.png |
| 736 |
|
| 737 |
.. |regals_pheh_regals_2_png| image:: images/regals_pheh_regals_2.png |
| 738 |
:target: ../_images/regals_pheh_regals_2.png |
| 739 |
:width: 250 px |
| 740 |
|
| 741 |
.. |regals_pheh_regals_3_png| image:: images/regals_pheh_regals_3.png |
| 742 |
:target: ../_images/regals_pheh_regals_3.png |
| 743 |
:width: 250 px |
| 744 |
|
| 745 |
.. |regals_pheh_regals_4_png| image:: images/regals_pheh_regals_4.png |
| 746 |
:target: ../_images/regals_pheh_regals_4.png |
| 747 |
|
| 748 |
.. |regals_pheh_regals_5_png| image:: images/regals_pheh_regals_5.png |
| 749 |
:target: ../_images/regals_pheh_regals_5.png |
| 750 |
|
| 751 |
.. |regals_pheh_regals_6_png| image:: images/regals_pheh_regals_6.png |
| 752 |
:target: ../_images/regals_pheh_regals_6.png |
| 753 |
|
| 754 |
.. |regals_pheh_regals_7_png| image:: images/regals_pheh_regals_7.png |
| 755 |
:target: ../_images/regals_pheh_regals_7.png |
| 756 |
|
| 757 |
.. |regals_pheh_regals_8_png| image:: images/regals_pheh_regals_8.png |
| 758 |
:target: ../_images/regals_pheh_regals_8.png |
| 759 |
:width: 250 px |
| 760 |
|
| 761 |
.. |regals_pheh_regals_9_png| image:: images/regals_pheh_regals_9.png |
| 762 |
:target: ../_images/regals_pheh_regals_9.png |
| 763 |
|
| 764 |
.. |regals_pheh_regals_10_png| image:: images/regals_pheh_regals_10.png |
| 765 |
:target: ../_images/regals_pheh_regals_10.png |
| 766 |
:width: 300 px |
| 767 |
|
| 768 |
.. |regals_pheh_regals_11_png| image:: images/regals_pheh_regals_11.png |
| 769 |
:target: ../_images/regals_pheh_regals_11.png |
| 770 |
|
| 771 |
.. |Dmax| replace:: D\ :sub:`max` |