Bioinformatics Solutions Inc.

RAPTOR (RApid Protein Threading predictOR) is a protein threading software package developed by Dr. Jinbo Xu and Dr. Ming Li. It applies novel Linear Programming techniques to the protein threading problem and has achieved great success. RAPTOR minimizes the scoring function (i.e. seeks for the optimal alignment between sequence and template) by integer programming method. The scoring function used by RAPTOR rigorously takes the pair-wise contact potential into account. The threading problem is formulated as a large scale integer programming problem and RAPTOR can find a global optimal alignment. It turns out that RAPTOR can produce high accuracy alignments and is most effective for hard targets.
RAPTOR has been consistently ranked in the top tier in recent CASP’s (CASP5, CASP6, CASP7). In CASP5, RAPTOR was ranked number one and RAPTOR paper was voted as the “most innovative paper” by peers in the research community.

Note that this installation guide is for Windows only.

For that install RAPTOR1.exe first by running Install.sh. Then you can download NR or REFSEQ by yourself from here.

Here are instructions for downloading NR database:
Download nr.00.tar.gz and nr.01.tar.gz to a directory.
Uncompress them in that directory and you will obtain a bunch of files whose names start with “nr.00.” or “nr.01.”.
Move those files to RAPTOR\data\nr
After that, you need to specify the NR database path in the configuration panel. i.e. if the NR database is installed at D:\RAPTOR\data\nr, then the “PSI-BLAST Database” field in the “Advanced” tab of the configuration panel should be set to “D:\RAPTOR\data\nr\nr.”
Note that you need to specify both the path and file prefix for NR database.

Alternatively, you can download REFSEQ database which is much smaller than NR.
Here are instructions for downloading REFSEQ database:
Download refseq_protein.exe to a directory.
Uncompress the file and you will obtain a bunch of files whose names start with “refseq_protein”.
Move those file to RAPTOR\data\REFSEQ
After that, you need to specify the database path in the configuration panel, i.e. if the REFSEQ database is installed in D:\RAPTOR\data\REFSEQ\ then the “PSI-BLAST Database” field in the “Advanced” tab of the configuration panel should be set to D:\RAPTOR\data\REFSEQ\refseq_protein.
Note that you need to specify both the path and file prefix for REFSEQ database.

To test RAPTOR, you can load a test sequence and run it with RAPTOR. To do that, you click “File” in the menu and select “Load Sequence/XML File”. In the file browser, you can go to RAPTOR\data\seq\ and load one test sequence into the work space. After that, you will see an icon on the left panel and the content of the sequence will be displayed in a window on the right. Alternatively, you can paste your fasta-format sequence to notepad, save it as a .seq file and load it into RAPTOR.

Then you can select “Run” in the menu and select “Run Selected” from the dropdown menu. A configuration panel will pop up. The only option that you may need to change is the path of the database used by PSI-BLAST depending on how you install the database. Click “Advanced” tab and find the “Database for PSI-BLAST”. If you have installed NR database, the path should be [home directory]\RAPTOR\data\nr\nr. If you have installed RefSeq database, the path should be [home directory]\RAPTOR\data\RefSeq\refseq_protein. where [home directory] is the path of your home directory.
If you want to build 3D structures, you need to install Modeller by yourself and specify to use Modeller and give its path in the configuration panel.

Click “Run” and RAPTOR will start to run. It will take about one hour to run one sequence depending on the sequence length. After the sequence is finished, a tabbed window will appear on the right. You will find PSP matrix obtained by PSI-BLAST, predicted secondary structure, ranking list of templates and all the alignments.

In RAPTOR\, run RAPTOR_GUI.bat to launch RAPTOR GUI or double click the RAPTOR icon on the Window's desktop.
The navigation panel is on the left and the output display panel is on the right, as shown below.

Click to see full size.

File->Load
You can load a sequence file (.seq) or an output file (.xml).

File->Close Selected
You can close the output windows for the selected sequence

File->Close All
Close the windows for all the sequences in the workspace

File->Delete Output
Delete the XML file for the selected sequence

File->Exit
Exit the GUI

Edit->RAPTOR Config
This will pop up a configuration panel where you can set up the configuration of RAPTOR.

Run->Run Selected
This will pop up the configuration panel and after you press “Run” the sequence will be run.

Run->Run All
This will pop up the configuration panel and after you press “Run” all the sequences in the work space will be run.

This will select different window from the drop down menu.

This will launch a browser to allow you to read this manual or visit BSI website.

Configuration Panel

Basic Options

Click to see full size.

Threading Method
There are three threading methods available in RAPTOR: NoCore, NPCore and IP. You can select to run one, two or all of them in a run.

3D Modeling
You can let RAPTOR call Modeller automatically after performing the threading. Select the check box and locate the Modeller program in the file browser. If you prefer to do 3D modeling with ICM PRO, RAPTOR, you can also output ICM Pro input files. You just select the check box and specify an output path. For example, the path could be C:\modeller8v2\bin\mod8v2 on Windows.
Before you start to use Modeller, you need to set some environmental variables. For details see Using RAPTOR.

Output Path
This is the directory in which RAPTOR will be run and all the output files will be stored.

Output Files
You will need to specify how many templates are saved in the templates. If you save too many in the XML file, the file will take up too much disk space.

Keep raw Files
You can select to keep or remove RAPTOR raw output files.

Advanced Options

Click to see full size.

Template Settings
List Path - The list of the path of the template is a text file which stores the names of all the templates in the template library.
FSSP Path - The directory where all the .fssp files are stored.
PSM Path - The directory where all the .psm files are stored.
PDB Path - The directory where all the trimmed .pdb files are stored.

Database for PSI-BLAST
If you use NR database, it should be [nr path]\nr
If you use RefSeq database, it should be [refseq path]\refseq_protein
Example: if all the NR files are in D:\RAPTOR\data\NR, then this field should read: D:\RAPTOR\data\NR\nr
If the RefSeq files are in D:\RAPTOR\data\RefSeq\, then this field should read: D:\RAPTOR\data\RefSeq\refseq_protein

PDB File Viewer
This is the view that will be called automatically in RAPTOR. A RasMol viewer comes with RAPTOR.

Template Ranking Method
RAPTOR supports two template ranking methods:
Support Vector Machine (SVM) and Z-score. Normally, you should use SVM.
For very long or short sequences, you can use Z-score for possible better result.

Navigation Panel

The left hand side is the navigation panel. Each Sequence is represented by a

icon. After running RAPTOR, the RAPTOR output is represented by

. You can browse different sequences and their outputs by clicking different icons in the navigation panel.

Click to see full size.

PSI-BLAST Profile

The output window is composed of a set of tab windows. The first tab window is PSI-BLAST profile. It is a 20 row matrix, each row corresponding to some amino acid.
The column width is the length of the query sequence. Thus each residue in a query sequence has a 20-element vector with it. Each element represents the occurring frequency of certain amino acid at that position in the multiple sequence alignment obtained from PSI-BLAST output.

The frequency is from 0 to 1. To make it easier for you to read the profile, the frequency is divided into 10 segments. Each segment will be represented by a color. In this way, the matrix can be represented by a rectangle in the window which is composed of many small square cells. The color of cell is determined by the occurring frequency. You can easily find out the conserved residues and non-conserved residues by differentiating colors.

Secondary Structure

Different colors are used to represent helices, beta sheets, loops (add color in html).

Some acronyms
AA          amino acid
PHD        PhiPred predicted secondary structure
E            Beta Strand
H            Helices
Space      Loops
Rel         Confidence of predicted secondary structure type
PrE         Chance of being beta strand (0 to 10)
PrH         Chance of being helix (0 to 10)
PrL         Chance of being loop (0 to 10)

Click to see full size.

Each method is represented by a folder icon. If you double click it, the templates will be displayed, ranked by their E-values. The smaller the E-value, the better. Also displayed are other scored used internally.

Click to see full size.

Table fields:
tName: template name
eValue: E value
tLen: template length
sLen: target length
Score: alignment score
mScore: mutation score
fScore: environmental fitness score
gScore: gap score
ssScore: secondary structure score
pScore: pairwise score
cScore: contact capacity score
SVMout: score output by the Support Vector Machine

If you click a template, its alignment will be displayed in a drop down window. The color of the template is consistent with its actual secondary structure and the color of the target is consistent with its predicted secondary structure.

If you click “View 3D structure with RasMol”, a RasMol window will pop up and the structure will be displayed. If you click “Export pdb file”, a file browser will pop up and you can save the 3D structure in a pdb file.

If you click the “Functional Annotation” tab, a window will drop down and show the functional information extracted from the template pdb file. If you click the template name, a browser will pop up and connect to rcsb PDB website.

Click to see full size.

The left side of the toolbar allows you to select some session(s) and specify how many templates you want to display. The right side of the tool bar allows you to compare any two alignments. To specify an alignment, you can use method name and its rank.

Click to see full size.

This window displays the errors that occurred during the run. Due to incorrectly generated template files or other reasons, the target sequence may not be able to be threaded to some templates. These templates can be ignored and will not have significant influence on the threading results, considering their number is so small.

PSI-BLAST Database

In RAPTOR, PSI-BLAST is used to generate position specific matrix (sequence profile) of a target sequence. By default, PSI-BLAST uses NR database, but the size of NR database is very large (1 G after compression). So an alternative database is RefSeq, which is a curetted non-redundant sequence database of genomes, transcripts and proteins maintained by NCBI. RefSeq is much smaller, about half size of NR. We conducted a comparison of the two. The profiles obtained from them are almost the same. So you can always use RefSeq to replace NR. NR database can be downloaded from ftp://ftp.ncbi.nih.gov/blast/db/nr.00.tar.gz and RefSeq can be downloaded from ftp://ftp.ncbi.nih.gov/blast/db/refseq_protein.tar.gz.
After uncompressing, you can obtain a bunch of index files. You need to put them in some directory and specify the path in the configuration panel (add a hyperlink here).

Threading Methods

Dynamic Programming vs. Integer Programming
RAPTOR has three threading methods available: NoCore, NPCore, and IP. NoCore and NPCore both use dynamic programming to optimize the scoring function. IP uses integer programming to optimize the scoring function. The difference is that if a scoring function considers pair-wise contact, dynamic program can only find a local optimum solution while integer programming can find the global optimal solution. Most of other threading servers are based on dynamic programming and RAPTOR’s integer programming is unique.

NoCore vs. NPCore
NoCore and NPCore are both based on dynamic programming. The difference is that in NPCore, the template and target are first divided into cores before doing threading. A core is a conserved segment of a protein. NoCore and NPCore are very effective for easy targets.

Running One Sequence with Different Methods
IP’s running time is longer than NoCore and NPCore. Thus, given a target sequence, you can run NoCore first. If the prediction is not good, try NPCore. If both cannot give good predictions, you can try IP. This will save you much time. Of course, you can also run more than one methods at one time. RAPTOR can keep up to three methods’ output in the XML file. When you run NPCore after running NoCore, the output will be automatically inserted into the XML file. If you run NoCore for the second time with different configuration, the old result in the XML file will be overwritten by the new result.

The fist step of RAPTOR is to run PSI-BLAST. If you already run NoCore, then when you run NPCore, this step will be skipped, as the PSI-BLAST is stored in PSP/ under the output directory. If the program finds those files, PSI-BLAST will be skipped. This will save running time.

Judging Prediction Quality from Alignment

First, you can compare the actual secondary structure of the template with the predicted secondary structure of the query sequence. As the accuracy of secondary structure is around 80%, this is an important measure of the prediction quality. Then you can look at the gaps in the alignment. The fewer the gaps, the better the prediction quality. The shorter the gaps, the better the prediction quality. Ending gaps normally can be ignored. Sometimes, the ending gaps may be very long. This means the program can only give good prediction for part of the query sequence.

What if the ending gaps are too long? In many cases, for long sequences, they may have more than one domain. Thus the ending gaps may be very long. You can cut them into domains first and run each domain with RAPTOR.

If you are an academic user, you can download Modeller for free from here. And you need to register here to get a license key in order to install Modeller. After you install it, you also need to specify the Modeller path in the configuration panel, i.e. C:\modeller8v2\bin\mod8v2. As Modeller8v2 has used python internally, it may give the follow error message while running, due to a bug in python: 'import site' failed; use -v for traceback” Just ignore it.

To use Modeller8 on Windows, there are several Modeller environmental variables to be defined. First, click Start -> All Programs -> Modeller8v2 -> Modeller, a black command line window will appear. Type "set" and press return, all environmental variables will be displayed. You will find:
KEY MODELLER8v2=XXXXXXX
MODINSTALL8v2=XXXXXXXX
Use Notepad to open RAPTOR_GUI.bat and you will find a line like:
"%RAPTOR_HOME%"\jre\bin\java.exe -Xmx500m -jar "%RAPTOR_HOME%"\RAPTOR_GUI.jar
Copy the previous two environmental variables to the beginning of RAPTOR_GUI.bat and make it look like:
set KEY_MODELLER8v2=XXXXXXX
set MODINSTALL8v2=XXXXXXXX
"%RAPTOR_HOME%"\jre\bin\java.exe -Xmx500m -jar "%RAPTOR_HOME%"\RAPTOR_GUI.jar
Save the file. And now you can specify the Modeller program path in the Configuration panel and let RAPTOR call it automatically.

Customizing Templates

RAPTOR/data/parameters/fssp.list stores the names of all the templates in the template library. If you are interested in a specific template, you can save its name in another file and specify the path in the configuration panel. You can also create your own template library. You need a pdb file and generate PSM and fssp file from it. Then put PSM file in RAPTOR/data/PSM and fssp file in RAPTOR/data/fssp.

The default viewer for pdb files is RasMol. The default display mode is a cartoon. The structure is colored according to the secondary structure. You can rotate the structure by pressing and dragging the left key of the mouse. To move the structure, press the right mouse key and drag. To shrink or enlarge the display, press “shift” key, press the right mouse key and drag. For a full reference of RasMol, you can click here.

If you run RAPTOR on a remote machine, rasmol_32BIT may not work properly with the X server. Instead, you need to configure the configuration panel to run a wrapper shell program “rasmol” which will launch RasMol.
If you want to use some view other than RasMol, please contact us and we can customize it for you.

If you find any problem when you run RAPTOR, you can report the problem to us and we will try to help you out as soon as possible. RAPTOR’s configuration files are in .raptor/ under your home directory. To report a bug, please send us the two .conf files in .raptor/. You can make some snapshots of the RAPTOR GUI and the terminal from which you launched RAPTOR and send them to us.

Feng Jiao, Jinbo Xu, Libo Yu, Dale Schuurmans. Protein Fold Recognition Using Gradient Boost Algorithm. Accepted by CSB 2006.

Jinbo Xu. Protein Fold Recognition by Predicted Alignment Accuracy. ACM/IEEE Transactions on Computational Biology and Bioinformatics, 2(2):157-165. 2005.

Jinbo Xu, Ming Li, Dongsup Kim, Ying Xu. RAPTOR: optimal protein threading by linear programming. Journal of Bioinformatics and Computational Biology 1:1(2003) 95-117.

Jinbo Xu and Ming Li. Assessment of RAPTOR's linear programming approach in CAFASP3. Proteins: Structure, Function, and Genetics, 53(S6): 579--584, Oct. 2003. Invited paper for CASP5, voted by peers as the "most innovative method in CASP5".