Database Scanning

Database scanning within STAMP is unpublished, apart from a brief description in a figure legend [16], but it has been fairly well tested since version 2.0. Indeed, two novel similarities have resulted in publications [9,16].

Immunglobulin domain

One example of a scan is given. The light chain variable domain of the immunoglobulin 2FB4 is used to scan a small database of other protein domains containing both a diverse collection of related folds (greek key folds, including azurin, superoxide dismutase, CD4, etc.), and completely unrelated folds (such as globins). See the directory examples/ig for this example.

The 2FB4 domain is described in 2fb4lv.domain. To scan this against the database type:

stamp -l 2fb4lv.domain -s -n 2 -slide 5 -prefix 2fb4lv_stamp -d some.domains  -cut

`-s' specifies the SCAN mode `-slide' describes how many residues to slide the query sequence (2fb4lv) along each sequence in the file some.domains to provide each initial fit (i.e. the sequence of 2fb4lv is layed on top of each database sequence at postions 1, 6, 11, etc.). `-cut' tells the program to cut down each domain read in from some.domains according to where the similarity is found. If it is not specified, the output will contain domain descriptors identical to those found in `some.domains'. When one is comparing a single-domain query to a database structure having multiple domains, it is desirable to do this. Try running it both ways (with and without -cut) and look at the output to see the difference. (e.g. CHAIN A is converted to A 1 _ to A 60 _ in one descriptor in the SCAN output and A 120 _ to A 175 _ in another, since there are two repeats of the query domain in the database structure).

The above run should write the following to the standard output (again, ignoring the header):

STAMP Structural Alignment of Multiple Proteins

Version 4.4 (May 2010)

 by Robert B. Russell & Geoffrey J. Barton 
 Please cite PROTEINS, v14, 309-323, 1992

Results of scan will be written to file 2fb4lv_stamp.scan
Fits  = no. of fits performed, Sc = STAMP score, RMS = RMS deviation
Align = alignment length, Nfit = residues fitted, Eq. = equivalent residues
Secs  = no. equiv. secondary structures, %I = seq. identity, %S = sec. str. identity
P(m)  = P value (p=1/10) calculated after Murzin (1993), JMB, 230, 689-694
        (NC = P value not calculated - potential FP overflow)

     Domain1         Domain2          Fits  Sc      RMS   Len1 Len2 Align Fit   Eq. Secs    %I    %S     P(m)
Scan 2fb4lv          2fb4lc             1   4.317   2.120  111  105  127   55   46    8  10.87  78.26 1.00e+00
Scan 2fb4lv          2fb4l              1   9.799   0.001  111  166  111  111  111   11 100.00  97.30 0.00e+00
Scan 2fb4lv          1mcplv             1   7.848   1.165  111  113  116   96   95    0  49.47  40.00 2.05e-22
Scan 2fb4lv          1mcphv             1   6.921   1.500  111  122  126   85   81    0  30.86  34.57 1.44e-07
Scan 2fb4lv          1cmsC              1   2.507   1.639  111  148  157   28   24    4   4.17  62.50 1.00e+00
Scan 2fb4lv          3cd4               1   5.939   1.334  111  166  114   78   75   12  20.00  76.00 4.10e-03
Scan 2fb4lv          2hhbb              0   0.000 100.000  111  146    0    0   75    0   0.00   0.00 1.00e+00
Scan 2fb4lv          3dpa               0   0.000 100.000  111  166    0    0   75    0   0.00   0.00 1.00e+00
Scan 2fb4lv          3sgbe              0   1.940   2.313  111  166  204   25   17    3   5.88  88.24 1.00e+00
Scan 2fb4lv          1acx               1   4.152   2.454  111  108  133   57   43    4  16.28  72.09 7.26e-02
Scan 2fb4lv          2abxa              0   0.000 100.000  111   74    0    0   43    0   0.00   0.00 1.00e+00
Scan 2fb4lv          1l01               0   0.000 100.000  111  164    0    0   43    0   0.00   0.00 1.00e+00
Scan 2fb4lv          2azaa              1   4.063   2.463  111  129  134   49   35    5  14.29  82.86 1.00e+00
Scan 2fb4lv          1rnt               0   1.503   2.545  111  104  148   17   13    3  15.38  69.23 1.00e+00
Scan 2fb4lv          2sodo              1   3.611   2.365  111  151  158   42   32    8   9.38  71.88 1.00e+00
Scan 2fb4lv          2pcy               1   3.788   2.052  111   99  125   47   39    6  30.77  79.49 2.27e-04
Scan 2fb4lv          8atca              0   0.000 100.000  111  166    0    0   39    0   0.00   0.00 1.00e+00
See the file 2fb4lv_stamp.scan

where all of the fields are as for the PAIRWISE mode, save for Fits, which indicates the number of fits that were saved to the file `2fb4lv_stamp.scan'. Note that for domain descriptors (see some.domains) containing two Ig type folds (e.g. 2fb4l, 1cd4, etc.) that more than one fit has been saved, since the search found both of the Ig type folds in each of these two proteins. Not also that `Fits' is zero for several of the examples, indicating that the no similarity was found within these proteins. Where more than one Fit is output for a domain in the database, the best $S_{c}$, RMS etc. are reported.

2fbjlv_stamp.scan will contain all the transformations output during the scan. Several of these will be redundant, since it is possible for a particular match to be found twice. To remove repeated transformations, or those not considered interesting, run the program SORTTRANS on the output.

sorttrans -f 2fb4lv_stamp.scan -s Sc 2.0 > 2fb4lv_stamp.sorted

This sorts the input file by $S_{c}$ values, and leaves only those non-redundent domain descriptions having an $S_{c} \geq 2.0$. A cutoff of $2.0$ is generally a good choice pairwise comparisons with a score lower than this tend to produce poor quality alignments.

sorttrans -f 2fb4lv_stamp.scan -s rms 1.5  > 2fb4lv_stamp.sorted

sorts the input file by RMSD values, and leaves only those domain descriptions having an RMSD $\leq 1.5$ Å. Despite its predominance in the literature, RMSD is not a very good means of measuring structural similarity, since low RMSDs can usually be obtained for any two structures if one considers a small enough set of residues.

sorttrans -f 2fb4lv_stamp.scan -s nfit 40 > 2fb4l_stamp.sorted

sorts the input file by the number of atoms used in the final fitting, and leaves only those domain descriptions where nfit $\geq 40$.

sorttrans -f 2fb4lv.scan -s n_sec 6 > 2fb4lv_stamp.sorted

sorts the input file by the number of equivalent secondary structures, and leaves only those having $6$ or more secondary structures equivalent.

Combinations of these can be used to select out interesting domains from a scan output. Probably the best combination involves Sc and nfit (ie. score and nfit), since large structures can give fortuitously large $S_{c}$ values with very few fitted atoms.

The final output is in the file 2fb4lv_stamp.sorted. This is the result of the first example (i.e. -s Sc 2.0). Note that several structures similar to the Ig type domain have been detected, and appear (according to $S_{c}$) in the order one might expect from knowledge of the 3D structures, sequences and functions of these proteins.

The output from scanning can be used as input for other modes of the program Once you have performed a scan, and have sorted the `hits' down to an interesting set, you can then use the output from scan as the input for a multiple alignment. This is discussed in the next section.