Debian Codegraph @

docs/source/saxs/saxs_guinier.rst

raw
1 Guinier analysis
2 ------------------
3 .. _saxs_guinier:
4
5 This tutorial covers basic principles and best practices for doing a
6 Guinier analysis. This is not a tutorial on how to use RAW for this type
7 of analysis. For that, please see the :ref:`RAW tutorial <s1p2>`.
8
9
10 Overview
11 ^^^^^^^^^^^^^^^^^
12
13 Guinier’s approximation states that at low-*q* the scattering profile
14 can be approximated as follows:
15
16 .. math::
17
18 I(q)\approx I(0) e^{-q^2 R_g^2 /3}
19
20 where |Rg| is the radius of gyration and I(0) is the intensity at zero scattering
21 angle (q=0). Because of the exponential in the Guinier approximation, |Rg| and I(0) can be
22 determined by performing a linear fit to a plot of :math:`\ln(I)`
23 vs. :math:`q^2`, called the Guinier plot. The region chosen for the linear
24 fit is called the "Guinier region."
25
26 The Guinier fit provides information on the overall size of the molecule, and
27 about the quality of your data.
28
29 Why do we do a Guinier fit?
30 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
31
32 The Guinier fit is done for two reasons. First, you get the |Rg| and I(0) parameters.
33 The |Rg| tells you about the overall size of the molecule, while I(0) depends
34 on the molecular weight times the concentration. These parameters are useful
35 characterizations of your molecule, and are also needed to calculate
36 other information from the SAXS data, including molecular weight and
37 volume.
38
39 Second, and perhaps most importantly, many of the problems that can affect
40 your SAXS data will show up in the low q region, including:
41
42 * Aggregation
43 * Radiation damage
44 * Interparticle interactions
45 * Buffer mismatch (in some cases)
46
47 These data quality issues cause deviations from linearity in the Guinier region.
48 For this reason, having a good Guinier fit is one of the primary ways we assess
49 the quality of SAXS data. A good Guinier is a strong indicator that your data
50 is from a monodisperse sample and is otherwise free of artifacts. If you cannot
51 obtain a good Guinier fit, or you can only obtain a good Guinier fit by
52 excluding a significant amount of data at the lowest *q* values, then your
53 data probably has one or more of the problems listed above and usually should
54 not be used for further analysis.
55
56
57 How do we do a Guinier fit?
58 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
59
60 The Guinier approximation only holds when the exponential :math:`\exp(-q^2 R_g^2 /3)`
61 is small. This means that in order to do a good Guinier fit, we need
62 |qRg| to be sufficiently small. The |qRg| value at which the Guinier approximation
63 starts to fail for a given scattering profile depends on the overall shape
64 of the scatterer. Below is a figure showing the Guinier approximation (black),
65 and the scattering intensity for a sphere, thin rod, and thin disc (all with
66 the same |Rg|).
67
68 .. figure:: images/guinier_shapes.png
69 :target: ../_images/guinier_shapes.png
70
71 Plot based on Figure 3.3 in [1]. Intensity for the geometric shapes
72 from Table 3.4 in [1]. Dashed lines are at |qRg| of 1.0 and 1.3
73 ((qR\ :sub:`g`) :sup:`2` of 1.0 and 1.69).
74
75 As you can see, the scattering intensity for the rod only agrees with the
76 Guinier approximation until :math:`qR_g\sim 1.0`, the sphere until :math:`qR_g\sim 1.3`,
77 and the disc until :math:`qR_g\sim 1.7`. Thus, depending on the overall particle shape, you
78 should fit different amounts of the low q data to have a good Guinier fit. (Note:
79 the plot's x axis is (qR\ :sub:`g`) :sup:`2`, so :math:`qR_g\sim 1.3` appears
80 as the dashed line at 1.69)
81
82 In practice, we fit both globular (sphere-like and disc-like) objects until :math:`qR_g\sim 1.3`.
83 while we fit highly extended (rod-like) objects until :math:`qR_g\sim 1.0`. These values were chosen to
84 have <10% error resulting from the deviation of actual shape from the Guinier approximation.
85 The reason we accept that much deviation is that you also get uncertainty from
86 fitting fewer points in your data, so there is a trade-off between how well
87 the approximation works (fitting to smaller maximum |qRg|) and how well you can
88 fit your data (fitting to larger maximum |qRg|).
89
90 The range of the Guinier fit is thus ideally from the earliest available *q*
91 value until a maximum |qRg| of 1.0 or 1.3. However, given that |Rg| is derived
92 from the Guinier fit, how do you determine the correct maximum q value for the
93 end of the fit? The answer is that the Guinier fit is done iteratively:
94
95 #. Guess a starting maximum q value for the fit.
96 #. Calculate the Guinier fit and get |Rg|.
97 #. If q\ :sub:`max`\ R\ :sub:`g` > 1.3 (or 1.0), reduce the maximum q. If
98 q\ :sub:`max`\ R\ :sub:`g` < 1.3 (or 1.0), increase the maximum q.
99 #. Repeat steps 2 and 3 until you converge on a final maximum q.
100
101 Most software these days will do this iterative search for you, and for good
102 quality data will provide you with a reasonable maximum q value that may
103 need just a bit of manual refinement.
104
105 The minimum q value of a Guinier fit is usually determined by the minimum available
106 q value in your data, which is set by the instrument on which you make the measurement.
107 However, it is important to have a small enough minimum q to have a reasonable
108 range for the Guinier fit. Typically, the minimum |qRg| value should be
109 :math:`qR_g\leq 0.65`, though for globular systems it can be okay to have
110 :math:`qR_g\leq 1.0`. This means that the minimum q value required depends on the size
111 of the system measured. In some cases, with particularly large systems, you may
112 have to deliberately seek out an instrument that can measure to sufficiently low q.
113
114 If your data has quality issues at low q, which can be caused by the problems
115 listed above, you may find that excluding those data from the fit can improve
116 the quality of the fit. While this can be acceptable, you should proceed
117 with caution when doing that, and always show the full data range on plots.
118 The most acceptable case for this to happen is when the first few points
119 are either too high or too low, but the rest of the range fits perfectly
120 (see below for criteria for a good fit). In that case, those couple of points
121 closest to the beamstop may have poor statistics or higher instrumental
122 background scattering, and can usually be safely ignored.
123
124
125 Criteria for a good Guinier fit
126 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
127
128 You are looking for four essential components in your Guinier fit:
129
130 #. :math:`\mathbf{q_{min}R_g<0.65}` **.**
131
132 * The minimum q of your fit, q\ :sub:`min`, times the |Rg| of your fit
133 should be less than 0.65. This criteria ensures you have enough q range to
134 properly calculate the |Rg| and I(0) values. For globular particles
135 (sphere- or disk-like), you can get away with :math:`q_{min}R_g<1.0`.
136
137 #. :math:`\mathbf{q_{max}R_g \sim 1.3}` **(globular) or**
138 :math:`\mathbf{q_{max}R_g \sim 1.0}` **(extended).**
139
140 * The maximum *q* of your fit, q\ :sub:`max`, times the |Rg| of your fit
141 should be less than 1.3 for globular (sphere- and disc-like) particles
142 and less than 1.0 for extended (rod-like) particles. This ensures you
143 remain in the linear range of the Guinier approximation for the fit.
144
145 #. **The Guinier fit residuals should be flat and randomly distributed about zero.**
146
147 * If your residuals have a ‘smile’ (above zero near start and end of fit,
148 below in the middle), or a ‘frown’ (below zero near start and end
149 of fit, above in the middle), it indicates you have non-ideal data.
150 The ‘smile’ is characteristic of aggregation, the ‘frown’ characteristic
151 of interparticle repulsion.
152
153 #. **The fit extends to the lowest available q point.**
154
155 * You shouldn’t have to excluded very many points at the start of the fit.
156 A few is generally fine, as the points nearest the beamstop can be noisy
157 (depending on the exact details of the measurement). Having to exclude
158 more than 3-5 points at the low *q* may indicate a problem with your data.
159
160 .. figure:: images/guinier_good_gi.png
161 :target: ../_images/guinier_good_gi.png
162
163 A Guinier fit done in RAW for glucose isomerase (available in the RAW
164 Tutorial data). This shows what a good Guinier fit looks like. It has
165 :math:`q_{min}R_g<0.65`, :math:`q_{max}R_g \sim 1.3`, the normalized fit
166 residual (bottom plot) is flat and randomly distributed about zero,
167 and the fit extends to the lowest *q* point available.
168
169 Having a good Guinier fit is a major quality check, and a good sign that
170 your data is from a monodisperse sample with no interparticle interactions.
171
172 What is a bad Guinier fit, and what does it mean?
173 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
174
175 Non-linearities in your Guinier fit are indicative or problems in your sample.
176 The type of non-linearity can indicate what the problem may be. The figure
177 below gives a quick summary off the most common pathologies, more detail is
178 available in the sections below.
179
180 .. figure:: images/guinier_overview_jacques.png
181 :target: ../_images/guinier_overview_jacques.png
182
183 Figure 3 from [2]. A and D show a good (monodisperse) scattering profile and
184 Guinier fit. B and E show scattering profiles with varying degrees of
185 interparticle interference. C and F show scattering profiles with varying
186 degrees of aggregation.
187
188
189 Aggregation
190 *************************
191
192 Aggregation causes a characteristic upturn at low q. This can either be caused by
193 aggregates initially present in your sample, or by radiation induced aggregation
194 (radiation damage). The figure below shows what that might look like in your data.
195
196 .. figure:: images/guinier_aggregation_putnam.png
197 :target: ../_images/guinier_aggregation_putnam.png
198
199 Figure 23 from [3], showing no (left panel), large (middle panel) and small
200 (right panel) amounts of aggregates seen in the Guinier fit.
201
202 The effect of aggregation can also be clearly seen in the fit residual, where it
203 shows up as a 'smile', with the residual above zero near the start and end of the
204 fit and below zero in the middle. The figure below illustrates this.
205
206 .. figure:: images/guinier_residual_aggregated.png
207 :target: ../_images/guinier_residual_aggregated.png
208 :height: 400 px
209
210 For an aggregated sample, even with the worst low *q* values cut off from the
211 fit, the residual (bottom plot) show the characteristic 'smile', rather than
212 being flat and randomly distributed about zero.
213
214
215 Repulsion
216 *************************
217
218 Repulsive interparticle interactions result in a structure factor that causes
219 a downturn in the scattering profile at low q. This is typically caused by
220 electrostatic interactions, and can often be remedied by either reducing the
221 sample concentration or adding more salt to the buffer. These effects are
222 also clearly seen in the fit residual, where it shows up as a 'frown', with the
223 residual below zero near the start and end of the fit and above zero in the
224 middle. This is shown in the figure below.
225
226 .. figure:: images/guinier_residual_repulsion.png
227 :target: ../_images/guinier_residual_repulsion.png
228 :height: 400 px
229
230 For a repulsive sample, the residual (bottom plot) show the characteristic
231 'frown', rather than being flat and randomly distributed about zero.
232
233 Note that the above figure is an extreme example of repulsion, the downward
234 curve may not be that obvious.
235
236 Bad buffer subtraction
237 *************************
238
239 Good SAXS data depends on subtracting away all scattering from the buffer and
240 instrument background. If this subtraction is not good, you can end up with
241 a downturn at low q (over subtraction) or an upturn at low q (under subtraction).
242 This will look similar to aggregation or repulsion in the Guinier fit.
243
244
245 FAQ
246 ^^^^^^^^
247
248 What if I don't know my particle shape, should I fit to a maximum |qRg| 1.3 or 1.0?
249 *************************************************************************************
250
251 Very often, you don't know what your particle shape is before making a SAXS measurement
252 (in fact, this is often one of the purposes of a SAXS measurement). In that case,
253 start out by fitting to a maximum |qRg| of 1.3. If that has a non-flat residual,
254 reduce the fitting range to a maximum |qRg| of 1.0. If the residual becomes
255 flat upon reducing the maximum |qRg|, then your particle is likely more
256 extended than globular, and you should keep the maximum |qRg| at 1.0. If reducing
257 the maximum |qRg| still leaves a non-flat residual, your data is showing signs
258 of aggregation, repulsion, or some other issue.
259
260 My Guinier fit isn't great, can I still use my data?
261 ******************************************************
262
263 If your Guinier fit isn't great, typically you shouldn't use the data. Even small
264 amounts of aggregation (<1%) can affect things like the measured maximum dimension,
265 and three dimensional reconstructions. While data with imperfect Guinier fits
266 can be used in some specific cases, my general recommendation is to collect the
267 data again.
268
269 My Guinier fit is a bit off, how can I fix it?
270 **************************************************
271
272 If your Guinier fit isn't ideal, typically the only way to fix this is to
273 collect the data again, improving your sample/solution conditions. Below are
274 a few things you can try to fix these issues.
275
276 For inherent aggregation:
277
278 #. Spin down your sample in a centrifuge at high speeds (~16000 g) for 5-10
279 minutes before data collection. In some cases, ultracentrifugation may help.
280
281 #. Use size exclusion chromatography coupled to SAXS instead of batch mode SAXS
282 for in-line sample purification.
283
284 #. Reduce the concentration of your sample.
285
286 #. Re-purify your sample using size exclusion or ion exchange chromatography
287 immediately before the SAXS measurement.
288
289 For radiation damage (often aggregation):
290
291 #. Add 1-5% glycerol.
292
293 #. Increase the flow/oscillation speed of your sample.
294
295 #. Reduce the exposure time and/or number of exposures.
296
297 #. Add radical scavengers like DTT.
298
299 #. Attenuate the incident x-ray beam.
300
301 For repulsion:
302
303 #. Add salt to the buffer to reduce repulsion.
304
305 #. Reduce the concentration of your sample.
306
307 #. Change the pH of your buffer.
308
309 For bad buffer subtraction:
310
311 #. Prepare matching buffer using dialysis.
312
313 #. Exchange buffer across a sizing column or desalting column.
314
315
316
317 References
318 ^^^^^^^^^^^^
319 1. `Feigin, L., Svergun, Structure Analysis by Small-Angle X-ray and Neutron Scattering (1987).
320 <https://www.embl-hamburg.de/biosaxs/reprints/feigin_svergun_1987.pdf>`_
321
322 2. Jacques, D. A. & Trewhella, J. (2010). Protein Sci. 19, 642–657.
323 DOI: `10.1002/pro.35 <https://doi.org/10.1002/pro.351>`_
324
325 3. Putnam, C. D., Hammel, M., Hura, G. L. & Tainer, J. a (2007). Q. Rev.
326 Biophys. 40, 191–285. DOI: `10.1017/S0033583507004635
327 <https://doi.org/10.1017/S0033583507004635>`_
328
329 .. |Rg| replace:: R\ :sub:`g`
330
331 .. |qRg| replace:: qR\ :sub:`g`
332