Google
More docs on the ARB website.
See also index of helppages.
Last update on 25. Nov 2018 .
Main topics:
Related topics:

PT_SERVER: What Why and How

OCCURRENCE

<ARB_NT/ETC/probe*>

<ARB_EDIT4/Edit/aligners>

 

DESCRIPTION

Probe design, probe matching, and searching the nearest relatives require a lot of database searching. Simply scanning the whole database for a specific target takes several tens of seconds. Probe design and searching for nearest relatives start thousands of search operations. Waiting several weeks to get a result from the computer makes every user hate the program, so better algorithms have to be developed.

The basic ARB SEARCH ALGORITHM:

  • When any (calling) program wants to scan a database, it does not do the job itself but calls a special SEARCH_PATTERNS_IN_A_BIG_DATABASE_PROGRAM.
    We named this program PT_SERVER ('Prefix tree server' or synonymously 'Positional tree server'). The PT_SERVER searches for patterns in special database files and sends all matches back to the calling program.
  • Different databases have different PT_SERVERs. You must choose a PT_SERVER to match the database you wish to search. The file $ARBHOME/lib/arb_tcp.dat defines all possible choices.
  • If there is no PT_SERVER running, one is automatically started. A PT_SERVER does not scan the database of the calling program, but the database in $ARBHOME/lib/pts/*.arb.
    That means:
    If you have just entered a new sequence this sequence will not be found by the the nearest-relative search.
    This is normally very useful, as you only want to use old sequences as a reference in the aligning process.
    For generating probes it isn't useful, because new sequences have to be added to the PT_SERVER before designing or matching probes. Remember: By adding 1 new sequence to the database, any previously validated probe may become invalid.
    Re-build your PT-server frequently and re-evaluate any probes you've made before.
    Creating a new PT_SERVER template:
  • Edit the file $ARBHOME/lib/arb_tcp.dat using <ARB_NTREE/Probes/PT_SERVER Admin/CREATE TEMPLATE>
    Note: Make a copy of your changes because a new ARB installation may reinstall and overwrite your changed $ARBHOME/lib/arb_tcp.dat.
  • Restart arb
    Create a database for an existing template:
  • Start ARB with the database you want to send to the PT_SERVER.
  • Open the PT_SERVER ADMIN Window <ARB_NTREE/Probes/PT_SERVER Admin>.
  • Select a template
  • Press <BUILD SERVER>: ARB will save the DB into $ARBHOME/lib/pts/name.arb
  • Wait
    ARB generates an index file $ARBHOME/lib/pts/name.arb.pt
    If any '*.arb' file in $ARBHOME/lib/pts is newer than the corresponding '*.arb.pt' ( == prefix tree) file the '*.arb.pt' file will be updated as soon as the PT_SERVER is started.

Updating a server:

see 'Create a database for an existing template'.
 

NOTES

Once started a PT_SERVER never stops. The only ways to stop a PT_SERVER are:

  • SOFT KILL (everybody) (only idle PT_SERVERS):
    • <ARB_NTREE/Probes/PT_SERVER Admin/KILL SERVER>

  • HARD KILL (supervisor) (all servers):
    • become superuser
    • enter 'ps -auxww |grep pt_server' at any shell
    • enter 'kill -9 PID' ( to get help enter 'man kill' or 'man ps' )


WHY DOESN'T ARB USE BLAST?

The basic idea of blast is to create a finite automaton for all search patterns and do the database search only once. If someone wants to start a fuzzy search ( allowing mismatches) the size of this automata increases exponentially.
For example: Three mismatches in a 20-base pattern would yield about 100,000 states in the automaton per search pattern. If probe design creates 1000 search patterns, and every state needs 10 bytes of computer memory, the search operation will require 1 gigabyte RAM.

You will need a lot of swap space to run multiple PT_SERVER.

 

WARNINGS

Do not modify any databases in $ARBHOME/lib/pts/ except by the <PT_SERVER Admin> tools. The reason is to keep the write protections of those files consistent.

 

BUGS

Sometimes some bugs