SPARTA uses the phi, psi and chi1 angles of a given residue to predict secondary shifts for that residue. SPARTA also includes the information from the next and previous residues when making predictions for a given residue. So, in practice, SPARTA uses data for three consecutive residues simultaneously (i.e. 9 torsion angles and 3 residue types) to make predictions for the central residue in a triplet.

The idea behind SPARTA is that if one can find some triplet of residues in a protein of known structure with similar structure and sequence to a triplet in a target protein, then the backbone secondary chemical shifts for this protein will be useful predictors for the backbone secondary chemical shifts in the target.

The similarity is measured with a score based on the weighted sum of squares differences between the torsion angles in the target protein and the database entries, so that lower scores indicated high similarity. In order to take advantage of the correlations between residue type and secondary structure, the score also includes a small, qualitative residue type term which biases the matching towards roughly similar sequences.

The reliability of the SPARTA approach was tested by a cross-validation procedure where each protein was temporarily removed from the database, and its backbone chemical shifts (N, HN, HA, CA, CB and C’) were predicted using the remaining protein data. The RMS deviations between the predicted and experimental shifts are 2.36, 0.46, 0.25, 0.88, 0.97 and 1.01 ppm, respectively. The same shifts prediction accuracies are also obtained for the proteins with known structures which are not contained in the database.

Importantly, it is also found in the test that the standard deviation the shifts from the central residues of the 20 matches are correlated with the shifts prediction errors. By checking the standard deviations in the prediction summary file (pred/pred.tab) will provide an idea of the prediction reliability.

It should be noted that the global structural information, such as ring current shifts and hydrogen bonding, was also carefully considered in SPARTA. The secondary shifts in SPARTA database are actually the corrected shifts using the ring current shifts. As “compensation”, the SPARTA predicted shifts for target protein are also corrected by adding the calculated ring current shifts from target protein. For HA and HN, the predicted secondary shifts are also corrected by using the hydrogen bond length and their relationship with the prediction errors, which were derived from above cross-validation. Therefore, the accuracy of the coordinates of the target protein is critical to obtain the reliable hydrogen bond information and ring current shifts, and the final predicted shifts. The calculated hydrogen bond and ring current shifts information is stored in the input summary file (/pred/protein_in.tab).

The SPARTA system is implemented using C++. The compiled executable files ($SPARTA_DIR/src/SPARTA for Linux, $SPARTA_DIR/src/SPARTA.exe for Windows) or starting script ("$SPARTA_DIR/sparta" for Linux) can be invoked with "TALOS-like" command-line argument. A complete list of options can be invoked and generated with a "-help" command-line argument or simply typing in the executive files or starting script without any command-line arguments.

Running SPARTA requires definition of the environment variables "SPARTA_DIR"; this will be established automatically by the starting script ("$SPARTA_DIR/sparta" in Linux):

   setenv SPARTA_DIR /disk1/SPARTA
   $SPARTA_DIR/src/SPARTA $argv[1-$#argv]

Other files of the SPARTA package include:

$SPARTA_DIR/tab/sparta.tab
The compiled database of residue triplets with their corresponding PHI/PSI/CHI1 angles and secondary shifts.

$SPARTA_DIR/tab/randcoil.tab, rcadj.tab, rcprev.tab, rcnext.tab
The table of random coil shifts, adjustments values from neighboring residues used in the shifts prediction process. (The same tables as used in TALOS, http://spin.niddk.nih.gov/NMRPipe/talos/)

$SPARTA_DIR/tab/homology.tab
The residue type homology factors used in the prediction process, which is similar to the table used by TALOS.

$SPARTA_DIR.tab/weight.tab
The weighting factors of PHI, PSI and CHI1 angles, and residue type homology used in the prediction process.

$SPARTA_DIR.tab/fitting.tab
The fitting parameters between prediction accuracy and precision, which will be used after the prediction process to calculate the estimated prediction error.

$SPARTA_DIR/shifts/*.tab
The files in this directory are only used when compiling a new database. When compiling a new database, only shift tables ending with the ".tab" extension will be used. The files in this directory are the chemical shift tables for the proteins in the database, which are in the same format as the TALOS shifts tables and must be exactly consistent with the corresponding structures in the SPARTA pdb directory.

$SPARTA_DIR/pdb/*.pdb
The PDB coordinates files in this directory are only used along with the files in the SPARTA shifts directory when compiling a new database (e.g. adding new proteins into the database). The sequence and residue numbering must be exactly consistent with the corresponding assignments in the SPARTA shifts directory. Furthermore, the names of these files must be exactly consistent with the corresponding chemical shift tables in the SPARTA shifts directory.

$SPARTA_DIR/test/*
The contents of this "test" directory include the input files and results for a sample SPARTA analysis.

Use of SPARTA to predict backbone chemical shifts involves the following steps:

1. Create a directory for the prediction session; all subsequent commands will be executed from this directory.

2. Prepare the input PDB coordinate file (for example "protein.pdb"), according to the format given above.

3. Run SPARTA ("$SPARTA_DIR/src/SPARTA" or "$SPARTA_DIR/sparta" in Linux, "$SPARTA_DIR/src/SPARTA.exe" in Windows) to perform the database searches. Most commonly, this will simply require a command such as:

      sparta -in protein.pdb

   SPARTA will first generate an input "pred/protein_in.tab" file from PDB coordinates, which contains of the phi, psi, chi1 angles, H-bonding information and ring current shifts. During the database search, a series of files "pred/X/res*.tab" (X = N, H, HA, CA, CB and C) will be created. Each one of these files tallies the 20 best database matches for a given residue in the target protein. Before exiting, a file "pred.tab" will also be created in "pred" directory, which includes a summary of the prediction results. The database search will typically take about 25 sec for a 100-residue protein on a Linux PC with a 2.8GHz CPU.

4. If experimental chemical shifts for target protein are available (with a name "ref.tab", for example, and the same format as typical TALOS shift table file, http://spin.niddk.nih.gov/NMRPipe/talos/), SAELDI prediction can be performed by a command such as:

      sparta -in protein.pdb -ref ref.tab

   SPARTA would compare the predicted chemical shifts and experimental shifts before exiting, and a prediction summary file "pred/pred.tab" will be generated to store the comparison between the reference and predicted shifts, as well as the errors. If the average prediction error larger than 3 times of the expected errors (standard deviation of prediction errors / square root of number of shifts), a warning is printed and a reference correction will be applied to the experimental chemical shifts. The corrected reference chemical shifts are stored into a new file "pred/ref.tab"

The input PDB coordinates should be prepared carefully, so that it has the proper format, naming conventions. SPARTA accept the standard PDB coordinates file, but ONLY the FIRST conformer/chain if more than one exist. For PDB coordinates without hydrogen atoms, the hydrogen atoms are required to be added (by using programs DYNAMO, REDUCE, MOLMOL, or any other similar programs) in order to get the hydrogen bonding information and ring current shifts. For HA atoms of Gly, please use atom names of "HA1/HA2"

New protein chemical shift and structure data can be added to the database. Note well that this should be done with great care and caution, to ensure that only reliable phi/psi/chi1 data with consistently referenced and correct chemical shifts are included. It suggests that

1. The chemical shifts assignments for each candidate protein are better validated by conducting a SPARTA shift prediction using its PDB coordinates.

      sparta -in protein.pdb -ref ref.tab

2. Check the prediction summary table (pred/pred.tab) files, remove the experimental shifts for which the predicted shifts deviated five standard deviations. Notably, HAs, for which ring current shifts are > 1.5ppm and the predicted shifts deviate the three standard deviations, are better removed.

3. Chemical shifts shoule be referenced correctly. A quick check can be conduct by runing above SPARTA prediction for this protein and inspecting the average SPARTA prediction errors, which are listed in the header of prediction summary table (pred/pred.tab). By default, SPARTA will apply a shift referencing correction if the average prediction error is larger than 3 times expected error (i.e., standard deviation of prediction errors / square root of number of shifts), and store the corrected shifts in a file "pred/ref.tab"

Given this, the procedure for adding new proteins to the SPARTA database is simple as:

1. Create a chemical shift table for the new protein according to the TALOS format (http://spin.niddk.nih.gov/NMRPipe/talos/). Copy the table to the "$SPARTA_DIR/shifts" directory; it must have a ".tab" extension in order to be used.

2. Place the corresponding PDB structure file into the "$SPARTA_DIR/pdb" directory; it must have a ".pdb" extension, and its file name, sequence, and residue numbering must correspond exactly with the shift table.

3. Prepare a table file, for example with a name of "list.tab", which only contains the names of proteins to be added into the database. This table must follow the example below:

         VARS   PDB_NAME
         FORMAT %24s
         bpti
         ubiquitin
         profilin
         ...
   

   Note that the "PDB_NAME" in the table file must consistent with the files names (with ".tab" and ".pdb" extension) in the SPARTA pdb and shifts directories.

4. In the "SPARTA" directory, execute the following command to compile a new database:

      sparta -compile -pdbDir ./pdb -pdbList list.tab

5. A new database "$SPARTA_DIR/tab/sparta.tab" will be generated from the files in SPARTA pdb and shifts directories. Please backup the old database, which will be overwritten.

SPARTA was implemented with standard C++ using Standard Template Library (STL). To compile the source codes (in /src directory), your system must have a compatible C++ compiler and STL library. Given this, the compiling of SPARTA executable file is simple as:

   cd $SPARTA_DIR/src
   make