SHARP/autoSHARP Tutorials	previous next
Chapter 1

Step-by-step explanation for running autoSHARP

(MAD/SAD or SIR(AS)/MIR(AS) experiment without known sites)

Preparation
Input files
Setting the job up
Analysing the log file
What to watch out for
Fine-tuning

1. Preparation

Collect all available information about the data and the project. This can/should include:

number of residues per molecule
The "molecule" is more precisely the "unique entity" in your asymmetric unit: so everything of which there might be more than one copy of.
number of molecules in asymmetric unit
Unless you are sure about this it is not necessary to know this value before starting to run the program: autoSHARP will try to detect this automatically.
amino-acid sequence of molecule
This is the most accurate information about the contents of the asymmetric unit (since the number of residues and molecular weight can be calculated from that).
space-group name/number
Make sure to check all extinctions to know the correct space group - or maybe a list of possible space groups (if the extinctions are misleading or not measured).
type of heavy atom in each dataset
Even if several different types of heavy atoms might be available for phasing (e.g. S and Ca), there is usually one type that is most visible within a dataset.
f'/f'' values for each dataset
Unless you are sure that the dataset was collected far enough away from the edge for the given heavy atom element, you should have a fluorescence scan that gives you these values. If you are far enough away from the edge, the calculated values should be fine. autoSHARP will use these calculated values if no f'/f'' values are known - but it then relies on an accurate wavelength (see below).
wavelength for each dataset
autoSHARP uses these in Ångstrom - so you might need to convert energies into Å.
resolution limits of each dataset
Usually, any reflection file should have only valid data in it. However, if e.g. data was processed to the high resolution limit of the detector (rather than to the resolution limit of the data) you might want to remember the limits for which the data follows the standard quality criteria (I/sig(I), Rmerge, completeness ... or whatever you prefer).

Only two types of files need to be prepared (one of these is optional). All of these files need to be in your sharpfiles/datafiles directory and need to be readable by the server (e.g. running the UNIX command 'chmod 0666'). Also, some naming conventions need to be followed.

reflection data
This can be in merged MTZ format (i.e. usually from the CCP4 program TRUNCATE) or merged SCALEPACK format (i.e. using the keyword 'MERGE' in scalepack). Ideally, the files should have the correct spacegroup, cell parameters and resolution limits (e.g. in the header).
(optional) sequence file
A simple ASCII file with the one-letter amino-acid sequence can be used to describe the contents of the asymmetric unit. If you are sure about the number of molecules in the asymmetric unit, you can cut-and-paste the same sequence several times into this file. The complete file needs to follow the format for sequence files.

3. Setting the job up

Use one of the supported browsers (Netscape 4.7X, Mozilla or Firefox) to connect to your local SHARP/autoSHARP server (something like http://localhost:8080/ - but ask the person responsible for installing the software - a role we call 'BDG administrator'). You should then give your SHARP/autoSHARP user name and password (which have nothing to do with your UNIX account/password). Again, if you are unsure about these, please ask your BDG administrator.

You can then start the autoSHARP input interface (using the 'Start' autoSHARP button at the top). Normally, you would leave the selection at the default value of 'None' - since you are probably starting a completely new autoSHARP run.

First page

On the first page specify the type of experiment (MAD/SAD or MIR/SIR) and the number of wavelengths or derivatives you have. For a MIR experiment it is advisable to first use only datasets of similar quality and resolution (and where you have a reasonable hope to have heavy atom bound in the crystal). So you might want to actually start with SIR(AS) of your best derivative and afterwards start adding additional derivatives to your phasing. However, it might also be interesting to throw all datasets into a huge MIR experiment: autoSHARP will do a lot of analysis between the various datasets. And this can be helpful in deciding which to start with.

Note : the hyperlinks explanation can be used to bring up the manual page with the relevant information for any given item (so please make use of it!).

Second page

The second page will have several sections: a General one and several dataset related ones. What needs to filled in should be self-explanatory (but if in doubt, use the explanation link to see the manual section).

Some related information can be given as several items (e.g. content of asymmetric unit or f'/f'' values). Only one of these need to be input (leave the others at the default values). It is recommended to use a sequence file and the f'/f'' values when available.

The least known value might be the number of heavy atoms to search for: it shouldn't matter too much if this value is overestimated - but not by a factor of 5 or 10.

It is usually a god idea to let autoSHARP run through the whole sequence of steps (it will not try automatic building if only 5Å data is available). Especially the density modification step will (hopefully) show a clear difference in correlation between the two hands - a good sign that things are on the right track.

Submitting the job

Once all the necessary items are filled in, you can submit the job. Most likely, the only method available will be 'Interactive' - so the actual autoSHARP will run on the same machine that runs the httpd (you know which machine that is, since the http://machine:8080/ URL usually contains the machine name). One thing to keep in mind is, that it is fairly easy to accidentally re-submit the same job several times (which is obviously pointless and just slows everything down - it is even possible that the program gets confused). So when looking at the log file and going back: if you are asked if you want to re-submit the data make sure you know that this is exactly what you want to do. If you are trying to go back to a job submission page and re-submit it, the job gets re-submitted to.

4. Analysing the log file

Apart from the warning about re-submitting a job again (see above) you can freely click your way through the autoSHARP log file. Remember that all of these files are written to while you are looking at them. So from time to time you might need to hit the reload button on your browser.

The main log file should have a header that says 'autoSHARP run'. This log file will give all crucial information: which steps the program performed and what messages each step generated. For each step there should be a details and an explanation link to give you more information.

5. What to watch out for

The main things to look out for are Error and Warning messages: there might be good reasons for some warning messages. But it is recommended to follow the hyperlink proviede with them to make sure you understand these. Error messages will make the program stop (hopefully with a sensible explanation about what went wrong).

If you are not happy with the results of an autoSHARP run try following these warning messages: maybe you are able to correct the reasons for these. Especially if the warnings are related to data quality or how well different datasets relate to eac other, it can be helpful to start with fewer (but better and better matching) datasets than trying to do everything at once.

Furthermore, some warnings might be fixed when using a stricter resolution cut (either by changing the reflection file or by setting the limits directly - after specifying yourself as Expert in the Preferences page).

6. Fine-tuning

Try to get rid of as much warning messages as possible in the initial data analysis stages. Especially problems in scaling of different datasets or suspicious data items in a dataset can make life very complicated further down the automated pipeline.
Check the heavy atom detection:
- what does the Patterson maps look like?
  A weird looking difference or anomalous Patterson map can point to scaling problems or data outliers. These should have been detected (to some degree) during the initial data analysis stage - so going back to any warning messages there is advisable.
- does a set of potential sites (directly out of RANTAN) has a weird combination of coordinates (e.g. all Z coordinates of the first 10 positions are the same, or one value is always zero, or positions are on special positions ...)?
  Any solution used based on a set of such sites is hardly trustworthy (the same applies for solutions based on 'funny' or poor looking Patterson maps).
- did the program find a solution with the type of data you expected the best signal anyway (e.g. peak wavelength in MAD, isomorphous difference in SIR/MIR when native and derivative scale very well, ...)?
  If the peak wavelength in a MAD experiment doesn't give a sensible solution it is very unlikely that the remote anomalous difference will. But maybe there were things that happened during data collection that explain this?
Does the density modification of the two hands show a significant difference in correlation? Is the starting correlation very low?
The starting correlation for the correct hand should ideally be above 0.20 (or higher). The wrong hand should then have a correlation of maybe 0.10 to 0.15. With very good phases this might also be 0.40 (correct hand) versus 0.20 (wrong hand).
A very low starting correlation (e.g. below 0.15) for both hands nearly always shows that either the heavy atoms sites are wrong (most likely) or that there isn't enough signal in the data.
Improving the heavy atom model.
Either use the SHARP run that represents the correct hand (if the density modification statistics enable you to decide) or just the latest one to start the SHARP Input Editor. This allows you to change the exact parameters for the heavy atom refinement and phasing step (i.e. what SHARP does).
A visual inspection of the various residual maps (as opposed to the automatic analysis done inside the autoSHARP pipeline) can help getting a more complete or corrected heavy atom model.

Last modification: 16.11.04