PROSPERO User Manual



PROSPERO (PRediction of Outcome using Sequence & Protein Experimental Results Online) can help you:

The the basic steps are:

This manual goes through these steps in detail.

You can also use PROSPERO's tools independently of the predictor to fit differential scanning fluorimetry (DFS, aks Tm) or size exclusion chromatogram curves:

[ Go to top ]


Menu Bar

Each page of the site has a menu bar just below the logo at the top, in white letters on black. Options include:

When you first go to the site, you also get these options:

Once you have started using the site by selecting one of these three entry options, you instead see these options instead:

[ Go to top ]

Guest Access, Registration and Login

When you first go to the PROSPERO home page, you have the option to use the site as a guest, to login to an existing account, or to register a new account. All users can securely store data between logins, though guest data is subject to deletion after several days unless the guest user registers. Registration makes it easier for users to log in again and access data added in previous sessions. Registered users can also more easily share data with other users on the same project.


To use PROSPERO as a guest, click use as a guest on the home page or Guest Access in the menu bar on any page. You will get a welcome page:

with options to:

The URL link takes you to get a "Welcome back" page with options to register or go to your projects. You can register at any time during this session by editting your using the edit your account link on the welcome page or use the Account menu tab then click Edit. You get a form:

Enter your email address and password to register.


When you click Register from the home page or menu bar, you get a similar short form:

Only the first 4 lines are required. Your email address is used as your account name. Registration information is private and will not be shared with anyone; we will only send e-mail to the address you supply if you forget your password. Your institutional affiliation information will assist us in helping you maintain shared data access, but is not required.

Once you click Submit you are automatically logged in to your new account. The next time you use the site, you only need to enter your e-mail address and password to log in.

[ Go to top ]

Enter Sequence and Experimental Data

When you click on "projects" as a guest user, register as a new user, or log back in to PROSPERO, you start out at the "All Projects" page. The first time, you'll start with an empty "Projects" table:

This page is the starting point for viewing and adding to the PROSPERO data hierarchy:

User is at the top level.

In version 1.0 of PROSPERO, you can store and share results for sequence and for six types of experiments (though only the first four types are used in the original prediction model, HyGX1):

Once you have added projects, this page also allows you to grant other users access to those projects through the Access Control tab.

[ Go to top ]

Add a Project

To add samples and results, start by adding a project with the Add new Project button above the "Projects" table. This brings up a form where you should give your project an easily identifiable short name. You can also enter a longer description of up to 255 characters.

The project name is displayed in bold letters above the table for this project (see next figure). The description is written in smaller type below the name on that project page. The name will also used in the "Project" column of the All Projects table;

this will be more readable if you keep the name fairly short. Once you've drilled down into this project, the name also shows up in the navigation outline at the top of the page to identify which project you're in.

[ Go to top ]

Once you enter project name and description, click "Submit" and you'll see an empty table for this project:

You can now:

[ Go to top ]

Access Control (Project Membership)

The creator of a project can add other users to the project. Those users can see the data for all samples in that project, and can add samples and sample data to the project. Clicking Access Control from the Project page gives you a form:

The Allowed Users tab shows any users you have already added to the list of those who have access to this project. You can select and remove users from the list with checkboxes and the Remove button.

The Add User tab (or the "Add user to project" link) lets you give access to more users by entering their e-mail address. If you use this tab to invite a user who is not already registered in PROSPERO, or if you directly use the Invite New User tab, you get a form to enter the user's name as well as e-mail address. When you Submit, PROSPERO uses that address to send and e-mail invitation containing a link to the "Welcome" page as for guests. The invited user can then save the link or register to access your project.

All users with access to a project can read or add samples and sample data to that project. Only the user who created a project (the "owner") can grant other users access to that project. If you have access to projects owned by other users, the owner's name will show up in the "All Projects" table ("Projects for " in the nav outline), in the column between "Sample" and "Created".

[ Go to top ]

Add a Sample

To add a sample, click Add next to the "Sample" header in the table. You get a form where you can fill in a relatively short sample name to be used in table columns and navigation outlines, and a longer description (up to 255 characters) displayed above the table for this sample.

Below the description, you can also specify whether this sample is from a sequence of eukaryotic or prokaryotic origin (the origin of the gene, not the type of heterologous expression system used). This helps in the interpretation of sequence analysis and experimental results: PROSPERO's HyGX1 prediction algorithm was based on eukaryotic proteins and in our tests is the best method for predicting outcome for these proteins. Pxs was based on, and makes better predictions for, bacterial and archeal proteins; PROSPERO includes the Pxs results in its report, when possible.

[ Go to top ]

When you submit this form, you get a blank table for this sample:

The navigation outline above the table shows which project you're in and lets you navigate up to the table of all samples for this project, or to the table of all projects. On this Sample page, you can:

The colored bars in the left column show what data you still need:

          Red bars mean you have not yet added results.
          Green bars show up as you add results for each experiment type.
          Yellow bars mean you have not finished loading the results for an experiment
Once you've entered sequence and results, you can

[ Go to top ]

Add Sequence

When you click Add Sequence, you can upload or paste in the sequence as one-character IUPAC symbols:

Enter a name and description for the sequence, e.g. the gene identifier and the start and end points of truncatation (or indicate it's full length). Sequence names that are distinct from sample names are especially important if your sample contains a complex of different macromolecules.

PROSPERO can store and share data for nucleic acids, though it will not be able to accurately predict their crystallization with the current HyXG1 predictor. For nucleic acids, change the sequence type to DNA or RNA.

When you submit the sequence, you get a new sequence table on the Sample page:


[ Go to top ]

You can now go on to add experimental results, or click on the sequence name to see details of the sequence analysis such as estimated MW, hydropathy and predicted disorder:


[ Go to top ]

Add Experimental Results

To upload or enter results, click the "Add" button for that type an experiment: DSF aka Tm, SEC, DLS, yield, SDS PAGE or limited proteolysis. Each type of experiment has its own type of data. In most cases each brand of equipment and each version of software will export that data in different formats. PROSPERO uses input modules to convert exported data into a standard XML format and analyze that data to derive summary values for each experiment. It then reports those summary values and uses them to predict crystallization outcome. In most cases, you have several options:

The first option is most convenient: other options require some expertise in either programming or analysis. If your data is in a form that our input module cannot handle, you can download our input module and modify it, or contact us. We may be able to help by modifying our general input module, or by sending you a specially modified module, to handle your data. If you have a better method for analyzing the data, we'd also like to hear about that.

Differential Scanning Fluorimetry (DSF, aka Tm)

Differential scanning fluorimetry uses the dequenching of a fluorescent dye on binding to hydrophobic pockets to monitor proteins unfolding. The temperature at the midpoint of the unfolding transition, Tm, is a measure of protein stability and has been reported as predictive of crystallization success (Dupeux et al. 2011 Acta Cryst. D67, 915-919). In our predictor, the most important characteristic of the melting curve is the relative fluorescent intensity at low temperatures. We quantify this as R30, the ratio of fluorescence at 30°C to fluorescence at Tm. For proteins of less than 36kDa, we found that almost all of those with low R30 (<0.1) produced well-diffracting crystals, in sharp contrast to those with high R30 values.

Click "Add DSF" to load or enter results from (Tm) experiments. This gives you a form with several options for loading data:

[ Go to top ]

You can now:

If you do not have the data file available, just the summary analysis or image of the Fluorescence vs Temperature curve, then enter:

Estimates of Tm and FWHM are usually available from the DSF machine, but may not be reliable. You will need to calculate R30 from the intensity values at 30 °C and Tm yourself.

If you have the data file available, click Upload data file to load it and select wells for analysis. You'll get a simple upload form:


This will send your data to our input module, which can take a variety of formats as long as the file has:

See DSF (Tm input format details for more information on what formats you can load and how to convert among formats.

If you have a file you've already analyzed with your own input module, and you want to keep that analysis rather than having it redone by our site, click Upload XML file to get a similar upload form for files in the XML format produced by our input module.

Note: The XML format includes the original data as well as the analysis and summary, so you can also upload XML files and have them re-analyzed using the Upload data file button. This would only be useful if we make a major improvement in our analysis program and you no longer have the original data file. Then you can download a .tmc.csv file in our XML format from PROSPERO and re-upload it.

[ Go to top ]

Select Wells for Analysis

After you upload a file using either Upload data file or Upload XML file, you get a form to select which wells you want analyzed. You can select any well for analysis, but for predicting crystallization the most useful wells are those with pH of about 6 or above and no added ligand. The original HyXG1 predictor was based on proteins in a standard "SGPP" buffer of 25 mM HEPES pH 7.25, 500 mM NaCl and 5% glycerol. We found in buffer optimization tests that other salt and glycerol conditions with medium or high pH gave similar curves for most proteins. If you're doing buffer optimization, you can pick two or three wells with conditions most similar to the SGPP buffer.

Also, since HyXG1 used only samples with no ligands added after purification, the predictor may not be accurate using wells with added ligand. You can analyze all the wells and then later choose not to use some of them for prediction. If you're doing ligand screening, you probably have several control wells with no added ligand. Those control wells are the ones you should use for prediction.

To start with, you can select from a list of the wells by the well labels from the file you uploaded:

      . . .

Check boxes for one or more samples, then hit Submit. Or you can choose a different view of the sample list using the buttons in the upper right corner. Thumbnails gives you a small fluorescence vs. temperature graph for each sample:

      . . .

[ Go to top ]

If you have to choose from many samples, e.g. from a 96-well plate, you would have to scroll through several pages to select by thumbnail or even by list. If you know which well positions in the plate you want to use, then you may be able to use Grid to select wells from an 8x12 grid:

This option will only work if your labels can be interpreted as well positions, i.e. if they start with "A1" or "A01" to "H12", possibly followed by a colon or space and well composition e.g.
    A01: 25mM HEPES ph 7.25, 500 mM NaCl, 5% glycerol
    A1 Buffer

Once you have selected the wells you want using any of the selection forms, hit the Submit button on that form. List and thumbnail selection forms may have Submit buttons on both top and bottom if you have many samples; use either button.

You will then get a warning that analysis is in process and might take some time. This is usually only a few seconds per curve (faster for XML upload), but could take several minutes if the curve fitter is having trouble finding a good fit or if you selected many wells. See the tm_calc manual for details on the curve fitting analysis.

[ Go to top ]

Select which wells to use in predictor

When the curve fitter is finished, you'll get a page showing a summary of the analysis for all of your DSF samples:

This page shows the name of the file you uploaded and the date you uploaded it in a row across the whole table. If you add more files, the most recent ones show up on the top of this table. For each well you selected from each file, you'll see:

From this page, in addition to using the navigation outline you can:

[ Go to top ]

View model details and select major transition

In addition to showing you the fitting results, if there are multiple transitions the detail page allows you to alter the choice of major transition. The program chooses the model curve with the steepest slope as the major transition, which is usually the same as the curve with the largest overall intensity change. But there may be cases where you want to choose a different transition as the major one, e.g. if a more gradual transition has a greater intensity change, or if the model does not match the observed data well.

The top of the curve fitting detail page has overall summary values, values for each transition, and graphs of Fluorescence vs Temperature and the derivative, dF/dT, vs T:

In theory a single transition is steepest at its midpoint, at the temperature where there are equal concentrations of folded and unfolded protein. Well-separated multiple transitions should also have roughly equal concentrations of proteins in the "before" and "after" state at the midpoint of each transition. For nearby transitions, e.g. those with Tms closer than the sum of their FWHM, midpoints become more difficult to determine and to interpret.

[ Go to top ]

The tables at the top are summary values and values for each transition:

Summary values are:

Below these are values derived from fitting one or more Boltzmann transitions to the observed data, if any could be fit (see tm_calc documentation for details). Derived values are:

The table also includes buttons for selecting the major transition, if there are multiple transitions and if you are the owner of this sample. To start with, the transition with the steepest slope is selected as the major transition, highlighted with a green background in this table. In some cases there may be another transition with a lower slope but more overall intensity change, or the curve fitter may have fit multiple transitions improperly; in these cases you can use this button to select a different transition as the major one. This will move the green bar, and recalculate the value of R30 in the summary at the top of this page and on all other pages.

NOTE: After changing the major transition, DO NOT use the Back button on your browser.
Use the navigation outline instead. Otherwise you would see the wrong value of R30 unless you reload the page.

[ Go to top ]

In each of the graphs,

Transitions are marked with horizontal and vertical lines:


In the first plot, above, the total fitted curve is in red and a small orange circle shows the intensity at 30° C. Horizontal bars are at height FTm. Each vertical bar spans the range of fluorescence covered by that transition, starting from the top of the previous transition (if any) and going up by ΔF.


The middle graph, "Individual Transition Curves," will look like the first one unless you have multiple transitions or a non-zero background (especially if you have uploaded an xml file with an exponential background). This graph shows the observed data with the background subtracted, along with each of the transitions that make up the total fitted curve shown in the first plot. As in the first plot, vertical bars are ΔF high. But in the second plot these bars all start from 0 fluorescence so you can directly compare the intensity changes of the transitions. The bars cross at the midpoint of the fitted transition, i.e. at ΔF/2.

[ Go to top ]


The bottom curve shows the raw observed derivative dF/dT in gray, smoothed observed derivative in black, total fitted derivative in red. The height of the vertical bars here show the maximum slope of the fitted curve. Horizontal bars span the model derivative peak at half that maximum slope.

In such a plot, an ideal transition has its midpoint at the center of a symmetric peak, rising from and falling to zero. The ends of the green lines should touch the black observed curve on both sides. In reality, for many proteins the peak tends to fall off faster above the highest Tm than it rose below that point, and the derivative curve falls below zero at higher temperatures. The decay of the fluorescence signal may be due to protein aggregation followed by either precipitation or exclusion of dye. The curve fitter attempts to avoid modelling these high temperatures where the signal decays.

In addition to high-temperature fluorescence decay, the smoothing required to give a reasonable signal on subtracting two noisy signals may distort dF/dT peaks. As a result, the curve fitter may not put Tm exactly at the center or at the highest point of the peak, and the horizontal bar may not line up exactly with the half-maximum of the observed derivative peak on this plot. However, the fitter generally produces good fits to the observed data (rather than to the derivative) resulting in reasonable estimates for Tm and thus R30.

[ Go to top ]

Zoom in on plot

To zoom in on a region of any graph, select the region of interest on the plot or on the thumbnail to the right by holding the left mouse button down while dragging it from one corner to the opposite corner of the rectangle you want to view. This will highlight the selected region with a pale pink background as you drag:

When you release the button it replots using the selected rectangle (the upper and lower limits of the smoothed derivative, in this case) as the plot limits:

You can zoom in further by selecting again on this zoomed plot. The thumbnail on the right shows which part of the original range is being displayed:

You can zoom to a different region by selecting on the thumbnail. To zoom back out to the full plot, click Reset under the thumbmail.

[ Go to top ]

Once you have selected the major transitions and the runs to use in the predictor, return to the Sample page e.g. by clicking the "Sample: ..." line in the navigation outline. The iaverage R30 value will be displayed next to the TM (aka DFS) box in the list of experiments you have already loaded (last row of this table):

   . . .
   . . .

[ Go to top ]

Dynamic Light Scattering (DLS)

Dynamic light scattering indicates the abundance of particles within each range of hydrodynamic radii, from small molecules to large aggregates. Monodisperse samples, having all particles of similar size, tend to crystallize and diffract more readily. In our initial predictor, polydispersity (broad or multiple peaks) was not as predictive as oligomerization state, quantified as the molecular weight estimated from the radius of the major peak divided by the weight of a monomer calculated from the sequence.

To enter or upload dynamic light scattering results, click DLS in the "Add ..." row at the top of the "Experiments" table on the Sample page, or click Add DLS within the table. You get a form where you can either load a file or enter values by hand:

[ Go to top ]

If you have a DynaPro DLS machine with Dymanics software, and you have a DLS experiment (".exp") file, you should be able to export a text (".txt") file which you can upload to get the best analysis. (If not, see below). It is simplest to export everything from the ".exp" file; you should at a minimum export "Results Summary" and "Regularization Results." You can then click Upload DLS Dynamics Txt to upload that .txt file. This gives you a page with summary scores, peak analysis and a histogram of intensities, and allows you to alter the selection of the main peak:

  . . .

[ Go to top ]

The table at the top summarizes the results. The key value used in predictions in HyGX1 is DLSMR, the ratio of the major peak's molecular weight to the MW calculated for the monomer from the sequence.

Next is a table of values for each peak showing calculated MW and values as in the Dynamics table: radius, absolute and percent polydispersity, percent intensity and percent mass. The last column allows you to change the selection of the major peak. The currently selected major peak is highlighted with a green background, and is plotted in green in the histogram below. To change the major peak, click the Change radio button for the new major peak row, then push Set Major Peak. This recalculates the values in the summary table and changes which peak is colored in the histogram below.

At the bottom of the page is the histogram of percent intensity versus radius. This gives you a graphic view of the data to allow better evaluation of the results, and to help in selection of the major peak.

PROSPERO stores peaks by both intensity and mass, but currently uses only the intensity peaks. A choice to use mass peaks instead may be implemented later.

You can now click on Experiment: DLS in the navigation outline to see the list of DLS results (just this one, so far), from which you can add more DLS runs for this sample:

[ Go to top ]

Entering DLS data by hand

If you have a DLS machine with different software you should still be able to find and enter at least the radius for the major peak. Even if you at one point had the right software and .exp file, you may now only have summary data: radius and possibly polydispersity, percent intensity and percent mass. This may be in the form of a screen image from Dynamics, e.g.:

Given such data, you can enter the values for the major peak by hand. The major peak's radius is the only value used in the original predictor, and the only one currently required on the form. But if you have the other values available, you should enter them for the major peak so that you can store and share the data with other group members.

[ Go to top ]

The major peak is the second row in this case: the peak with most of the intensity, and the only one with a radius in the ~2 to 10 nm range. Leave the peak type as "Intensity", unless you're really sure you want to use mass peaks. (In this case, peak 1 has most of the mass but little intensity, and is much lower MW than a protein, so it's not a suitable major peak). In this case, you'd enter:

Manual entry of values for other peaks for data storage and sharing may be implemented in later.

When you hit Submit, the molecular weight of the peak will be calculated from the radius. You get a DLS detail page with only the details you entered showing:

Only one peak and no histogram is shown, since that's all the data you've entered.

[ Go to top ]

NOTE: If you want to be able to upload data extracted from other machines, it may be possible to write a converter to transform your data into a form PROSPERO can read. But this would only be worth while if your DLS software does not provide similar analysis, e.g. peak tables and a histogram.

Click on Experiment: DLS in the navigation outline to return to the list of DLS results:

As for the list of DFS results, you can choose which results to use in the predictor: click "Yes" or "No" under "Use in Predictor" - then hit Update. You can now click on Sample: ... in navigation outline to return to the sample summary page. The Experiments summary section shows the average DLSMR and DLSSC of the DLS runs you have selected for use in prediction:

. . .

This table also shows the types of experimental results not yet loaded: SEC, SDS PAGE, yield, and limited proteolysis.

[ Go to top ]

Size Exclusion Chromatography (SEC, aka Gel Chromatograph)

In addition to its use in protein purification, size exclusion chromatography provides information on the composition and oligimerization or aggregation state of the sample. As expected, among our samples those absorbance curves showing single Gaussian peaks and little else tended to have better outcomes (though this only applied in our predictor for high-molecular weight proteins with less than "extremely strong" yields.)

To add data from an analytical or preparative SEC run, go to the Sample page (shown above) and click on SEC in the "Add" row above the table, or on Add SEC within the "Need" section of the table. You get this form:

You have 3 options for adding SEC data:

  1. Enter SEC values SECR1 and SECPP on this form. Only do this if you can't do either of the others.
  2. Upload SEC curve & doc files exported from AKTA software. If you have the .res file and the software, this is the best option.
  3. Upload the SEC XML file created by an input module modified to handle data from other chromatography software (see downloads for documentation and source).

[ Go to top ]

Enter SEC values

The values PROSPERO wants are:

The residual after fitting one Gaussian peak to the absorbance curve over the whole range of fractions collected:

| Observed A280 - 1 Gaussian peak | / (Observed A280 - background)

integrated over all fractions.

The purity of the fractions pooled to make this sample:

(A280 of main Gaussian - background) / (Observed A280 - background)

integrated over the fractions pooled for this sample, where "main Gaussian" means the Gaussian peak with the largest area within the pooled fractions, not necessarily the main peak in the whole curve.

If you have the curve data in numerical form but you cannot convert it into a form PROSPERO can use, you may be able to use other software to fit Gaussian peaks to the SEC curve and calculate the values yourself. Otherwise, if you have an image of the SEC curve you can make a rough estimate of the values by comparing the image to the examples below. To fit Gaussians to SEC data, you can follow the procedure used in

[ Go to top ]

Estimating SEC values by eye

For PROSPERO's prediction, the key cutoff is SECR1 < 0.2 - this is a nearly symmetric peak with possibly a few small additional bumps. For example, this curve has SECR1 = 0.18, just under the cutoff:

Curves with less symmetrical main peaks or with a higher fraction of absorbance in other peaks will have SECR1 > 0.2 . Curves with many large peaks or with irregular backgrounds may have values above 1, since a single Gaussian is a poor model of such curves.

SECPP is the fraction of the pooled peak which is not under any other peak in a suitable multiple-Gaussian model. For example, in the following curve the blue peak has most but not all of the area under the pooled fractions, 19 to 25; SECPP = 0.92:

SECPP is not used in the original version of PROSPERO's predictor, but may be a valuable number for comparison of different samples of the same target protein.

[ Go to top ]

Upload SEC curve & doc files from AKTA software

Click on Upload SEC curve & doc files to get the upload form:

The main data from SEC is the absorbance curve, usually A280 vs time or volume. But such data need additional documentation, or metadata, for proper interpretation: start and end times (or volumes) of fraction collection, flow rater, fraction size and the range of fractions pooled to make the sample. The software that comes with AKTA columns, PrimeView Evaluation or Unicorn, allows you to export two files: one with the absorbance curve (curve.asc) and one with all but the last piece of metadata (doc.asc) - see export instructions below. (If you have other column software and you want to put it into a form that you can load here, see the sample SEC curve and doc files to see what data is expected.) You can then use the above form to:

You'll have another chance to enter or select the pooled range based on the curve, in case you can't get the range from your notebook or LIMS.

[ Go to top ]

Export curve and doc files from AKTA software

To name your files, you will need a unique identifier (SEC ID) that you can remember or easily retrieve. This ID must containing only letters and numbers, e.g. "12345", because PrimeView will read, but won't let you write, file names with special characters such as space, dash or underscore. If you have only one SEC per sample, you can use the sample number from your LIMS if you have one, or the sample # from PROSPERO shown to the right of the sample name in the top navigation outline. If you have more than one SEC curve, e.g. from analytical SEC, you may need to add distinguishing letters to the sample ID to make your SEC ID unique, e.g. 12345a, 12345b, etc.

Once you have a SEC ID:

Now export the documentation (metadata) file:

[ Go to top ]

You now have absorbance vs time (or volume) in e.g.12345curve.asc, and a log by time (or volume) in 12345doc.asc . PROSPERO should be able to handle any combination of time or volue in either file, but you might be the one to discover an odd case where it has problems. In that case, you may need to export both the curve and doc files in units of time and upload again:

Other troubleshooting tips on using AKTA software:

[ Go to top ]

Once you submit the curve and doc files, you get a preliminary image of the curve which allows you to numerically or graphically select the range of pooled fractions:

You can enter new values for the first and last pooled fractions, or use the mouse to click on one end of the range and drag to the other. When you click Submit, it will fit Gaussians to the curve (which may take a few minutes) using the new ranges in calculating the pool purity.

[ Go to top ]

SEC Curve Fitting Results

Once the curve-fitting is done, you get a page with a a table summarizing the results at the top:

Values of Residual for 1 Gaussian (SECR1 below 0.2 are good, at least in our initial predictor. Higher residuals may indicate a higher fraction of proteins other than the target in the original purification, possibly due to low yield for the target protein. or a range of oligomerization states,

Sample purity (SECPP) is based on the model the program determined was best, but this choice may be wrong. If you get a purity value closer to 0.5 than 1.0, look at the plot of individual fitted peaks at the bottom of the page to see if the model is reasonable. You may find that the program has picked a model with too many peaks, e.g. 2 similar Gaussians under what appears to be a single peak. Users who want more accurate purity estimates can download the SEC input module and run it locally using the "-n" option to specify the maximum number of peaks to find, then upload the resulting XML file.

Other values in the table come from the metadata: flow rate, pooled fraction range, fraction volume, and column date.

[ Go to top ]

Below the table are plots of the observed absorbance curve (black) and the overall model of fitted absorbance (red):

Also shown are the pooled fractions (green). You can zoom in by selecting a region on the main plot or on the overview to the right.

At the bottom is a plot showing the individual peaks fitted to the data,

with obeserved absorbance in black dots and each Gaussian peak in a different colors as shown in the legend on the right.

[ Go to top ]

Upload SEC XML files

If you have chromatography data from non-AKTA software, or if you have AKTA software but you want to adjust the parameters used in curve fitting (e.g. change peakmin or npeaks to avoid excess peak fitting), you can download the SEC input module and run it on your own computer with modified parameters or modified code to create SEC XML files (.sec.csv files). You can then upload those files PROSPERO by clicking Upload SEC XML file in either of the other two "add SEC" pages. You get this form:

which allows you to select and upload the XML file. You then get the SEC details page as for uploaded curve and doc files.

Modifying the SEC input module for other chromatography software

SEC XML files contain both absorbance curves and metadata, including pooled fraction range if it's available. If you have non-AKTA chromatography software, you will probably find it easiest to generate the proper XML files by modifying or using parts of the existing input module , rather than starting from scratch. This program carries out several functions sequentially in separate subroutines, which you can modify or use independently:

  1. Parse the input absorbance curve: handles files with "min" or "ml" and "mAu" data in delimited columns; would need modification for files with data in rows, or with different column headers.
  2. Parse the input documentation (log file): needs careful tailoring to each version of the log file format, or replacement with manual input or fixed values of metadata, so that the program can convert the X axis units of the curve file (time or volume) into fractions, determine the range to fit, and calculate pooled purity.
  3. Fit the curve: make initial estimates, pass estimates and data to gnuplot, retrieve the results, calculate residual, and repeat.

The key XML tags used by PROSPERO are <DERIVED_VALUES><R1> and <DERIVED_VALUES><BEST_MODEL><PP> for SECR1 and SECPP values, respectively. The <MODEL> values for background and peaks and the fraction, absorbance and pool columns from the <MODEL_VALUES> table are used in displaying the observed data and fitted model on the SEC details page.

[ Go to top ]

If you enter or upload more than one SEC result, e.g. from sequential preparative runs or an analytical run after a preparative column, you can see the list of SEC results by clicking on SEC in the navigation outline, or on List SEC in the Sample page experiment table. You can then select which run(s) to use in outcome predition, similar to the selection of DLS or TM runs.

Once you have entered or uploaded all your SEC results, when you go to the sample view page using the navigation outline you'll see the SEC scores (averaged for all the SEC runs you've selected to use in the predictor) summarized in the Experiments table:

. . .

[ Go to top ]

Expression Yield

Proteins with high yields in heterologous expression systems (or in high abundance in natural sources) provide sufficient material for purification away from other proteins; high yield may also indicate stability of the properly folded protein in the expression system. In our initial samples, for large proteins (>36kDa) those with extremely high yields in high-throughput screening (equivalent to ~100 mg or more per liter of culture) were much more likely to produce well-diffracting crystals than those with lower yields.

To add screening or large-scale expression yield results, click Yield in the "Add" line above the sample Experiments table, or Add Yield within the table. You get a form:

. . .

where you can enter yield from large scale expression, gel score or both.

[ Go to top ]

For large scale expression, enter the total mass of purified protein in milligrams for Yield (mg). To be consistent with the screening gel yield, this should ideally be the mass after the first affinity column (e.g. immobilized metal), but protein concentration and volume are more commonly measured and recorded after the final step. Then enter the volume in liters of cell culture used. For cell-free expression systems, use the reaction volume for documentation purposes. (Cell-free systems were not used in training the predictor, but this will not be an issue in most cases since such systems are most commonly used for small proteins, and the predictor does not use yield for proteins under 36 kDa).

For gel scoring, e.g. of samples from high-throughput screening or from large scale expression, compare your gel band to the soluble protein lanes, the right lane of each pair in examples shown to the right of the form. These lanes have protein from about 8% of a 600 μL culture = 48 μL. Each score unit is roughly a factor of 3 difference in yield, so the mass of protein band on the gel and the corresponding large scale expression yield are approximately:

Visual Yield Score Band Appearance Mass in band (μg) Yield from Large Scale (mg/L Culture)
5: Extremely high heavily overloaded ≥ 5 ≥ 100
4: High yield strong but not smeared out 1.5 to < 5 30 to < 100
3: Medium easily detectable but not strong 0.5 to < 1.5 10 to < 30
2: Weak barely detectable 0.15 to < 0.5 3 to < 10
1: Not detectable no visible band < 0.15 < 3

If your gels have the protein from a similar volume of culture, choose the score for the soluble band that is most similar in intensity to your band.

For gels using protein from very different volumes of culture, adjust the score by about 1 for each factor of 3. For example:

The cutoff level for the initial PROSPERO predictor is 5, so if you use large volumes for your screening gels you should use the large scale expression mass and culture volume instead if possible.

In addition to the gel score, you can select and upload a gel image file. You should either crop the image to include only the lanes of interest, or mark the relevant lanes on the copy of the image file you upload - preferably with a mark above or below the body of the gel or at least far from the band of interest. Later versions of PROSPERO may include automated analysis of such gel images for automatic scoring; this would require lanes without extraneous marks to avoid complications in the estimation of background - usually a trivial task for humans but not for software.

Once you click Submit, you get a Yield details screen:

(followed by another copy of the scoring example gel)

You can review and revise the visual score here. Note that uncropped high-resolution gel images may have the lane of interest off the side of the screen, so you need to scroll right to see the lane, then back left to change the score.

If you have added more than one yield result (more than one pair of visual score and large scale yield), you can see the list and select which one(s) to use in the predictor by clicking on Yield in the navigation outline, as for other experimental results. Click on Sample in that navigator to go back to the Sample page and add other types of experiments, or make a prediction.

[ Go to top ]

Limited Proteolysis (LP)

Super-stable proteins, those that show little change in intensity or molecular weight after 24 hours of treatment by a variety of proteases, tend to crystallize and diffract well in our experience. Unfortunately, few proteins fall into this category. We have developed a scoring system to rate proteins based on the intensity and position of the bands seen after 1 or 24 hours of proteolysis. can store the image of the resulting gel and the average score for a battery of proteases, though it cannot yet use this score in predicting outcome.

Click on LP in the Add row above the sample Experiment table, or on Add LP within the table, to add limited proteolysis results. You get a form for entering the LP score,

followed by a table of rules for scoring limited proteolysis:

Score Band Intensity Compared to 0h Band Position Compared to 0h Description
1h 24h
5 ≥ 90% ≥ 90% ≥ 95% Little change in intensity or position in 24h
4 ≥ 90% ≥ 50% ≥ 85% Little change in intensity at 1h; at least half of intensity remains at 24h; ≤15% of molecular weight lost
3 ≥ 90% < 50% ≥ 85% Little change in intensity at 1h but half lost by 24h OR at least half of intensity remains at 1h; ≤15% of molecular weight lost
≥ 50% NA
2 >0 but <50% NA ≥ 85% Most but not all of intensity lost at 1h; ≤15% of molecular weight lost.
1 0% NA none ≥ 85% No visible bands within 15% of original molecular weight.

The relative intensity and position of the major band closest to the position of the original major band are tested by the criteria for each score, in descending order. Samples are assigned the first score for which they meet all criteria.

NA, not applicable: no criterion applied to intensity at 24h.

[ Go to top ]

Below that are example gel lanes applying the scoring method:

Score: 1 2 3(a) 3(b) 4 5
No visible bands ≥85% of original MW Half lost in 1h at ≥85% MW Little change in 1h but half lost by 24h at ≥85% MW 50% to 90% remains at 1h ≥85% MW Little change at 1h, half remains at 24h ≥85% MW Little change in intensity or MW in 24h

For these Coomassie Blue-stained gels, purified protein at 1 mg/ml in HEPES pH 7.5 buffer with 500mM NaCl, 5mM CaCl2 and 10% glycerol was exposed to 20 μg/mL trypsin, chymotrypsin, subtilisin A or endoproteinase Glu C for 0, 1 and 24 hours; reactions were then stopped with 0.17M acetic acid.

Score the group of lanes for each protease (e.g. 0, 1h, 24h) separately, then take the average score and enter it here.

[ Go to top ]

Once you hit Submit, you get a page showing the score and gel:

You can now Use the navigation outline to see and select from the list of LP results (if you have more than one) with the LP link, or go back to the Sample page with Sample to add other experiments, or predict.

[ Go to top ]


Electrophoresis is a standard method for assaying protein purity. But since SDS PAGE is often used to determine if proteins are pure enough to attempt crystallization, we have little data on proteins which show major impurities on gels. Therefore we did not find the quality of SDS gels for purified proteins to be predictive of outcome (if they're called "purified", they mostly have good gels). PROSPERO allows storage and sharing of images and scores of these gels for confirmation of purity.

To detect both minor impurities and impurities near the same molecular weight as the target, gels can be run with two or more lanes for each sample containing total protein mass differing by a factor of 2 or more, e.g. 1.5 and 3.0 μg for Coomassie Blue staining.

To upload a gel image, click SDS_PAGE in the Add line above the Experiment table on the Sample page, or click Add SDS_PAGE within that table. You get this form:

Browse to and select the image file for the gel. If you have gels with more than one sample, you may want to use a graphics program to label the lanes with sample identifiers or crop a copy of the image to include only the sample you're interested in. Use the examples on the right to determine which most closely resembles your gel, and check the appropriate score button. Note that all of the examples are for scores of 3 or above. If your sample has a lower score, further purification is probably in order.

[ Go to top ]

Once you hit Submit, you get a page showing the image and score:

Note that gel images may be very high resolution, so if you did not crop the image you may need to scroll to the right to see the lane(s) of interest. Scroll back to the left to update the score after review.

Later versions of PROSPERO may feature numerically analysis of such gel images to estimate the puritiy of the major band, assuming that staining and scanning are linear with mass.

You can now Use the navigation outline to see and select from the list of SDS gels (if you have more than one) with the SDS_PAGE link, or go back to the Sample page with Sample.

[ Go to top ]

Summary of Sample Results

Once you have added all available results, go back to the Sample page using the navigation outline and you'll see the Experiments table with green bars for each experimental type:

. . .

[ Go to top ]

Predict Crystallization Outcome

You can now click the Predict button and PROSPERO will use your data to estimate the likelihood of crystallization and diffraction for your sample. If you have some data missing, you can still submit your sample for prediction - but the prediction may not be as accurate as it would be with more results (see HyGX-1 to see which results are most valuable for your sample).

NOTE: for proteins of prokaryotic origin, the Pxs server predicts crystallographic success of purified protein from sequence alone at least as accurately as PROSPERO, when it is available.

The prediction page has 5 parts:

Diffraction Score

The Diffraction Score (DS) is a value from 0 to 6 meaning:

[ Go to top ]

Input Values and Predicted Outcome

The input experimental and sequence results are presented on top, with a pale blue (or red) background for criteria with values favoring (or disfavoring) crystallization, followed by the prediction:

The prediction is the mean Diffraction Score of samples in our training set which had input values similar to your sample. See the decision tree for details on how the input values are used to categorize samples. Below the numerical prediction is a brief text description of the likely outcome, then a link to the distribution of outcomes for similar samples. NOTE: Some categories with with low mean DS include a few successful targets, so check the distribution, not just the mean.

[ Go to top ]

Recommendations for Difficult Targets

Next comes suggestions for further steps you might take if your initial crystal trials produce no hits:

These suggestions are based on the assumption that you have already set up initial trials, and you are looking at this prediction because those trials failed. For samples with poor predicted outcomes, the suggested changes in sequence, vector or protocol are intended to move your sample to a category with a better predicted outcome. We do not have extensive data on the effects of such changes on of individual targets, but we do have several cases where a small change (e.g. deletion of a few amino acids) greatly improved both the predicted diffraction score and the actual outcome.

[ Go to top ]

Distribution of Outcomes

Below the suggestions is a plot of the distribution of outcomes:

The distribution shows how often samples with properties similar to yours produced crystals with each range of diffraction, from better than 2 Å (top bar) down to no diffraction or no crystals. (In all cases, salt crystals were excluded.) It is important to consider both the mean and the distribution, since some categories of proteins with moderate or low means produced a number of well-diffracting crystals - they just aren't as likely to do so as categories with higher means.

The number of samples in our training and test sets are shown with closed and open bars, respectively; these bars are color coded from blue (best diffraction) to red (no crystals). The mean of the category is shown as a black dot.

[ Go to top ]

Decision Tree

Currently, the predition is made using the HyGX-1 predictor. The tree used for your sample is shown at the bottom of the prediction page:

This shows the decision path, from left to right. At each branch point, the path taken is controlled by the value for the criterion shown in the blue box (or if the value is missing, the box is shown in red and no further path is taken). The predicted Diffraction Score is shown on the right, colored green, yellow or red for good, intermediate or bad predicted outcome.

The HyGX-1 Predictor

The initial PROSPERO predictor uses a simple binary tree to categorize samples, with one value used as a criterion at each branch point as shown above. The criteria, mean diffraction scores (and probability of getting crystals with 2.8 Å diffraction or better, based on 107 eukaryotic samples) are:

This means that DSF (Tm) is used in the predictor only for proteins under 36 kDa, yield is only used for larger proteins, and SEC is used only for larger proteins which do not have extremely high yields.

Disorder is only used for large proteins with less than extreme yields and SEC curves with more than 21.5% residual from one Gaussian. DLS is only used for large proteins with less than extreme yield, bad SEC and no stretches of predicted disorder longer than 18 residues. But even when they are used, disorder and DLS were not very useful in predicting outcome amoung the proteins we tested: none of those with high MW, low yield and bad SEC gave crystals with 2.8 Å diffraction or better, regardless of disorder or DLS.

[ Go to top ]

Expanding the Predictor

The original HyGX1 predictor was based on 6 types of experiments run on 107 samples of protozoan protein. These proteins may be more representative of eukaryotic proteins in general than the mostly prokaryotic proteins usually used to train sequence-based predictors, but the predictor, the data storage and sharing system, could still be improved in several ways:

If you have experimental results for DSF (Tm) or SEC in a format PROSPERO currently does not handle, you may be able to download and modify the existing input module to read your data. If you need help in making such modifications, or if you are successful and are willing to share your code with other users, please contact Dr. Ethan Merritt.

Also, please contact us if you have results from other types of experiments you would like to store, e.g. mass spectroscopy, NMR data or static light scattering. Given samples of the experimental results, we can determine the level of effort required to add a new experimental type to those stored in PROSPERO.

If you have samples of proteins similar to our original set (eukaryotes, especially protozoa) with consistently collected experimental results also similar to our original set (especially DSF (Tm), SEC, yield and DLS), and with final crystallization and diffraction outcomes, we would be interested in collaborating on expanding the sample size for our current general predictor.

We would be particularly interested if you have a substantial number of samples (for any class of proteins) with consistently collected experimental results (not necessarily similar to our set) and outcomes, and you want to work with us on developing a new predictor specifically for your class of samples.

[ Go to top ]


The Navigation Outline

The overall organization of data in PROSPERO is:

PROSPERO automatically builds and allows you to navigate this hierarchy:

-- At all levels below "All Projects", an outline near the top of the page shows where you are and lets you move to higher levels. The outline shows whether you're looking at your own projects or "Group Projects", the name of project, the sample, and the sequence or experimental type if you're down that far. Each line is a link so you can jump up one or more levels - e.g. from an experimental result page to the Project page. NOTE: if you use the navigation outline, rather than the back button, then you won't have to use your browser's "reload" button to update the page to include all current data.

-- At each level there is a page with a table that lists what's at and below that level. This table lets you navigate to or add items below the current level. E.g. from the Project page you can jump to any sample in the project, or any experimental type within the sample; you can also add samples to the project or experiments to an existing sample.

[ Go to top ]

Handling Multiple Experimental Results

Researchers will often have multiple results for DSF (Tm, from multiple wells in a single run or from multiple runs. Repeated DLS runs are also common. If you prepare each sample from a single large scale expression and purification, then you will probably only have one preparative SEC curve and one yield; PROSPERO allows you to enter multiple values in case e.g you combine multiple batches into one sample, or use analytical SEC to produce more than one SEC result. For all experimental types, you can select which values to use in the prediction in the that type's list of results. The mean of all selected values is used.

[ Go to top ]


PROSPERO allows you to enter multiple sequences for samples with macromolecular complexes. The HyGX1 crystallization predictor was based on single proteins, and therefore it is not known how well it works on complexes. Currently, predictions for complexes are based on the maximum estimated molecular weight and stretch of disordered residues, under the assumption that the largest macromolecule and the largest stretch of disorder are most likely to interfere with formation of well-ordered crystals. More data is needed to assess the validity of this assumption and to make accurate predictions for complexes.