Author(s)

• plan_sad_experiment: Tom Terwilliger

Purpose

plan_sad_experiment is a tool for estimating the anomalous signal that you might get from your SAD experiment and for predicting whether this signal would be sufficient to solve the structure. plan_sad_experiment is normally used along with scale_and_merge and anomalous_signal to plan a SAD experiment, scale the data, and analyze the anomalous signal before solving the structure.

Usage

How plan_sad_experiment works:

• You supply plan_sad_experiment with a sequence file, the anomalously-scattering atom you plan to use for the experiment, and the wavelength for data collection.
• plan_sad_experiment will estimate the necessary I/sigI of your dataset to provide enough anomalous signal to solve the structure.
• plan_sad_experiment will try various values of I/sigI for your dataset at each of several resolutions. For each I/sigI it will estimate the half-dataset anomalous correlation that would result along with the likely true correlation between your anomalous differences and those that would be calculated from a final model of your structure (cc*_ano). From this anomalous correlation (cc*_ano), plan_sad_experiment will estimate the anomalous signal (related to cc*_ano by the square root of the number of reflections divided by the square root of the number of sites). Then plan_sad_experiment will choose a value of I/sigI that gives an anomalous signal of about 30 (if achievable with the maximum I/sigI you specify).
• The way that plan_sad_experiment and anomalous_signal estimate the probability that you can solve your dataset is to compare the anomalous signal in this dataset with the anomalous signal in other datasets at the same resolution. Then the fraction of similar datasets that can be solved by HySS is used as the probability that the anomalous substructure for your dataset will also be found.
• Similarly, the mean figure of merit for datasets with an estimated anomalous correlation (cc*_ano) similar to that for your data is used as an estimate of the figure of merit that you would obtain if the substructure is found for your crystal.

Output from plan_sad_experiment

plan_sad_experiment provides a summary of the scattering expected from your crystal and a summary of the anomalous signal expected if you are able to measure your data with the suggested overall I/sigI. You can set the maximum I/sigI to look for. Here is an example setting max_i_over_sigma=30:

   ----------Dataset overall I/sigma required to solve a structure----------


Dataset characteristics:
  Target anomalous signal:    30.0
  Residues: 325
  Chain-type: PROTEIN
  Solvent\_fraction:    0.50
  Atoms: 2642
  Anomalously-scattering atom: se
  Wavelength:  0.9792 A
  Sites: 7
  f-double-prime:    3.84

Target anomalous scatterer:
  Atom: se  f": 3.84  n:    7   rmsF:   10.2

Other anomalous scatterers in the structure:
  Atom:  C  f": 0.00  n: 1674   rmsF:    0.1
  Atom:  N  f": 0.01  n:  445   rmsF:    0.1
  Atom:  O  f": 0.01  n:  514   rmsF:    0.3
  Atom:  S  f": 0.23  n:   10   rmsF:    0.7

Normalized anomalous scattering:
  From target anomalous atoms rms(x**2)/rms(F**2):     2.97
  From other anomalous atoms rms(e**2)/rms(F**2):      0.24
  Correlation of useful to total anomalous scattering: 1.00


  ----------Dataset <I>/<sigI> needed for anomalous signal of 15-30----------


-------Targets for entire dataset-------  ----------Likely outcome-----------

                              Anomalous    Useful    Useful
                            Half-dataset  Anom CC   Anomalous
 Dmin   N     I/sigI sigF/F     CC       (cc*\_anom)  Signal   P(Substr)   FOM
                      (%)                                        (%)

 6.00    852    29    3.0      0.58        0.64        7         51       0.22
 5.00   1473    29    3.0      0.62        0.66        9         79       0.15
 3.00   6821    29    3.0      0.64        0.66       19         89       0.22
 2.50  11787    29    3.0      0.70        0.68       25         96       0.19
 2.00  23021    28    3.2      0.62        0.66       29         97       0.17
 1.50  54569    13    6.7      0.18        0.42       29         97       0.15

Note: Target anomalous signal not achievable with tested I/sigma (up to 30 )
for resolutions of  2.50 A and lower. I/sigma shown is value
of max\_i\_over\_sigma.


This table says that if you collect your data to a resolution of   2.0 A with
an overall <I>/<sigma> of about  28 then the half-dataset anomalous
correlation should be about  0.62 (typically within a factor of 2).  This
should lead to a correlation of your anomalous data to true anomalous
differences (CC*\_ano) of about  0.66, and a useful anomalous signal around
 29 (again within a factor of about two). With this value of estimated
anomalous signal the probability of finding the anomalous substructure is
about  96% (based on estimated anomalous signal and actual outcomes for
real structures.), and the estimated figure of merit of phasing is 0.17.

The value of sigF/F (actually rms(sigF)/rms(F)) is approximately the inverse
of I/sigma. The calculations are based on rms(sigF)/rms(F).

Note that these values assume data measured with little radiation damage or at
least with anomalous pairs measured close in time. The values also assume that
the anomalously-scattering atoms are nearly as well-ordered as other atoms.
If your crystal does not fit these assumptions it may be necessary to collect
data with even higher I/sigma than indicated here.

Note also that anomalous signal is roughly proportional to the anomalous
structure factors at a given resolution. That means that if you have 50%
occupancy of your anomalous atoms, the signal will be 50% of what it otherwise
would be.  Also it means that if your anomalously scattering atoms only
contribute to 5 A, you should only consider data to 5 A in this analysis.

What to do next:

1. Collect your data, trying to obtain a value of I/sigma for the whole
   dataset at least as high as your target.
2. Scale and analyze your unmerged data with phenix.scale\_and\_merge to get
   accurate scaled and merged data as well as two half-dataset data files
   that can be used to estimate the quality of your data.
3. Analyze your anomalous data (the scaled merged data and the two
   half-dataset data files) with phenix.anomalous\_signal to estimate
   the anomalous signal in your data. This tool will again guess the
   fraction of the substructure that can be obtained with your data,
   this time with knowledge of the actual anomalous signal.  It will also
   estimate the figure of merit of phasing that you can obtain once you
   solve the substruture.
4. Compare the anomalous signal in your measured data with the
   estimated values in the table above. If they are lower than expected
   you may need to collect more data to obtain the target anomalous signal.

Literature

List of all available keywords

• i_over_sigma = None Optional I/sigI. If supplied, the expected values of half-dataset correlation and cc*_ano based on this I/sigI be calculated.
• max_i_over_sigma = 100 Limit search of necessary I/sigI to less than this value. You might increase this if you plan to do a very careful or very high-multiplicity experiment.
• target_signal = 30. The anomalous signal that you would like to obtain. The value of I/sigma will be adjusted to obtain this signal. Typically you will need a signal of 15-30 so solve the substructure.
• min_cc_ano = 0.15 You can set the target minimum (true) anomalous correlation (CC*_ano). This value affects the phasing accuracy after the substructure is determined.
• ideal_cc_anom = 0.75 The ideal_cc_anom is the expected anomalous correlation between an accurate model with isotropic anomalous scatterers and perfectly-measured data. The ideal_cc_anom is determined empirically. It is typically not unity because anomalous scatterers may have multiple locations with low occupancy or may be non-isotropic. A value of about 0.75 is a reasonable guess.
• include_weak_anomalous_scattering = Auto At longer wavelengths the scattering of C, N, and O become significant relative to S. Default is to consider the scattering from C, N, O as noise. Additionally, (see intrinsic_scatterers_as_noise) if intrinsic anomalous scatterers (P and S) are weak, they will be counted as noise. This weak anomalous scattering is effectively noise and has the same effect as the ideal_cc_anom but it can be calculated from the composition. Its effects are added to those modeled by the ideal_cc_anom parameter. Default is to include weak anomalous scattering if a sequence file or the number of sulfurs is provided
• intrinsic_scatterers_as_noise = None Applies if include_weak_anomalous_scattering=True. You can choose to treat any intrinsic scatterers (S for protein, P for nucleic acid) as noise, just like any contributions from C, N, or O atoms. This is default if anomalous scattering (f-double-prime) from these atoms is less than half that of your specified anomalous scatterer. Alternatively these atoms are excluded from the noise calculation and are assumed to be included in the number of sites you specify.
• bayesian_updates = False Use Bayesian updates of half-dataset CC and signal. First predict these values using standard approach, then use empirical half-dataset CC and signal for a training set of datasets to re-estimate these values. This helps correct for typical errors in measurement and typical resolution-dependent effects. Note that if you use bayesian_updates=True then the predictions may not vary smoothly with resolution or changes in parameters.
• input_files
  • data = None Data file (I or I+ and I- or F or F+ and F-). Any standard format is fine.
  • data_labels = None Optional label specifying which columns of anomalous data to use. Not necessary if your input file has only one set of anomalous data.
• crystal_info
  • resolution = None High-resolution limit. Either a high-resolution limit or a (Wilson) b_value or both is required
  • b_value = None Estimated Wilson B-value for the dataset. Either a high-resolution limit or a (Wilson) b_value or both is required
  • b_value_anomalous = None Estimated Wilson B-value for the anomalously-scattering atoms. Normally leave as None and it will be estimated from b_value.
  • seq_file = None Optional sequence file (1-letter code). Separate chains with a blank line or line starting with >.
  • chain_type = *PROTEIN RNA DNA Chain type (PROTEIN RNA DNA). This is used to estimate the number of atoms from the number of residues
  • ncs_copies = None Optional estimate of NCS copies in your crystals (only used if a data file is supplied).
  • solvent_fraction = None Optional estimate of solvent fraction in your crystals
  • residues = None The number of residues in the molecule or asymmetric unit. Note that it is the ratio of residues to anomalously-scattering atoms that matters.
  • atom_type = None Optional name of anomalously-scattering atom. If supplied, you also need to supply the wavelength for X-ray data collection. If not supplied, then you need to supply a value for f_double_prime.
  • number_of_s = None You can specify the number of S atoms in the asymmetric unit. Only used if include_weak_anomalous_scattering=True. If not set, the number is guessed from the sequence file if present.
  • f_double_prime = None F-double-prime value for the anomalously-scattering atom. Alternatively you can specify the atom type and wavelength.
  • wavelength = None Wavelength for X-ray data collection. If supplied, also specify the atom_type. Alternatively you can specify the value of f_double_prime
  • sites = None The number of anomalously-scattering atoms in the molecule or asymmetric unit. Note that it is the ratio of residues to anomalously-scattering atoms that matters.
  • occupancy = 1 Estimate of occupancy of anomalously-scattering atoms
• control
  • fixed_resolution = False Only run calculation at high_resolution limit
  • show_summary = False Show summary only
  • verbose = False Verbose output