SA-Search Help

SA-Search

A program to search for 3D fragments similar to (fragments of) a query structure.

Introduction

This program uses the compression of the 3D structures of proteins using a structural alphabet to transpose the classical amino acid sequence methods to the search of 3D similarities in proteins. The score matrix quantifying the
similarities between the letters of the structural alphabet have been derived from the probabilities of observing each letter at each position when encoding series of protein structures. An example of possible scoring matrix is reported here

SA-search is not currently as performant as true 3D similarity search methods, but it provides a very fast way of mining collection of structures, especially for medium range fragments, where true 3D similarity search methods are confronted with a large number of comparisons.
Also, in our experience, the differences between true 3D similarity search methods and SA-Search often come from the lenghts of the structural alignements, more than from the structures identified as "hits".

Some other search services are listed here.

How to use SA-Search ?

SA-search expects on entry a "query" file and an "against" what it will be searched.

The "query" as well as the "against" can be specified using several formats that can be selected using the radio buttons.
Make sure to have the entry specified according to the radio button selected.

If you choose the PDBid format, simply enter a 4 letter PDB code such as "1tim". You can specify a chain by entering "1timA", A denoting the chain selected in the 1tim file.
If you upload a file, please note that the encoding into the SA space cannot be effective if there are missing residues in the structure. Even if each fragment longer than 4 residues will be encoded, the searches will be performed separately. Also, note that the quality of the structure has an impact on the quality of the encoding. SA-Search will check that the geometry is suitable for encoding. This may result in some case to reject structures of poor stereochemical quality.
If you choose a "query" against "bank" search, you can select among several banks extracted from the PDB.
The proteins encontained in each bank were selected among the complete Protein Data Bank using an local version of an algorithm similar to that employed to generate the culledPDB. However, proteins with poor geometry, or proteins with fragments have been discarded from the PDB before defining the lists. Also, the lists are defined in a recursive manner: all the chains in the 30% list are enclosed in the 50% list and so on.

Parameters of the search:

The search parameters allow mostly to control the depth of the search that is achieved.

Search method: currently, search can be achieved using a standard Smith & Waterman algorithm, a faster version of it not allowing gap, or a suffix tree (denoted as "exact words"). For the Smith & Waterman, the gap penalty is an internal parameter of SA-Search. Its value has been adjusted to be compatible with the substitution matrix.
Minimal Match Length: corresponds to a minimal size (letters of the alphabet) for a match to be returned.
RMSd max: It corresponds to an alpha-carbon best fit RMSd. All matches with a RMSd more than this value will be discarded. Note that specifying a value of 10. or more will inhibit the computation of RMSd.
Minimal Score (0. - 1.): corresponds to the minimal score for a match to be kept for further analysis by RMSd computation. This score is normalized as to be of 1 for a perfect match of the fragment.
Maximal number of matches: the matches found beyond this number will not be listed.
New: Search using amino acid sequence (instead of HMM encoding sequence): You can now aternatively apply the different search algorithms to perform a search among amino-acid sequences instead of the HMM encoding of the structures. Using such facility is similar to performing classical amino-acid sequence similarity search.

Results:

The results are returned using the NBRF/PIR format (since it allows to specify the localisation of the matches in the structures, which might be of interest if you want to get the superimposed structures) or in a raw format (one match per line). Note that the returned numbers of the residues currently start from 1 for the first residue of the chain.

For each match, we return the PDB and SCOP Ids, the name of the compound, the amino-acid and HMM-SA sequence identities, the alpha-carbon best fit RMS deviation, and the aligned sequences. For fragmented PDB files, the fragments are numbered from 1, and the PDB id will be on the form: PDBid+PDBChn+_+Fragment number (1qmnA_1 for example).