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Database scanning within STAMP is unpublished, apart from a brief description in a figure legend [16], but it has been fairly well tested since version 2.0.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 through 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 mulitple domains, it is desirable to do this. Try running it both ways (with and without -cut) and look at the output: you will see what I mean. (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 may take a couple of minutes, and should write the following to the standard output (again, ignoring the header):

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

     Domain1  Domain2   Fits  Sc      RMS   Len1 Len2 Align Fit   Eq. Secs    %I    %S     P(m)
Scan 2fb4lv   2fb4lc      1   3.819   7.880  111  105  127   45   45    7   6.31  26.13 8.28e-02 
Scan 2fb4lv   2fb4l       8   9.799  10.383  111  166  111  111  110   11  66.27  64.46 0.00e+00 
Scan 2fb4lv   1mcplv      1   7.802   9.561  111  113  118   94   90   10  39.82  69.03 9.06e-22 
Scan 2fb4lv   1mcphv      1   5.637   8.513  111  122  125   71   68    9  18.85  47.54 7.00e-08 
Scan 2fb4lv   1cmsC       1   2.152   6.421  111  148  153   27   21    4   1.35   8.78 1.00e+00 
                                     <etc.>
Scan 2fb4lv          1rnt               1   2.210   4.679  111  104  145   28   28    4   1.80  18.92 1.00e+00 
Scan 2fb4lv          2sodo              1   3.586   7.776  111  151  158   39   32    7   1.32  15.23 1.00e+00 
Scan 2fb4lv          8rubs            skipped domain -  sequence is too short
Scan 2fb4lv          2pcy               1   3.331   7.604  111   99  124   41   33    6   5.41  19.82 6.44e-02 
Scan 2fb4lv          8atca              0   0.000 100.000  111  166    0    0   33    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, one must run the program SORTTRANS on the output.

sorttrans -f 2fb4lv_stamp.scan -s Sc 2.5 > 2fb4lv_stamp.sorted

sorts the input file by S_c values, and leaves only those non-redundent domain descriptions having an $S_{c} \geq 2.5$. In practice, I tend to use a value of 2.0, and then sort through the output to look for interesting similarities.

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 collection 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 (ie. -s Sc 2.5). 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 is totally compatable with the 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. E.g.,

transform -f 2fb4lv_stamp.sorted -g -o ig_like.pdb

will read in the files, transform the coordinates and save them to the file ig_like.pdb (with each chain labelled starting with a different letter). This program is explained in one of the next sections.

stamp -l 2fb4lv_stamp.sorted -prefix ig_like > ig_like.log

will read in the transformations, and run PAIRWISE and TREEWISE comparisons to generate a multiple alignment of these structures. The results of this run are in the examples/ig directory. Note that there are several `LOW SCORE' warnings in the output (stored in ig_like.log). Note that one would normally edit the output from a scan before performaing a multiple alignment (i.e. to include only those domains one wants to consider further).