Talk:Implementation Notes

How to add a track to a mirror

I. Introduction.

This describes how to add a track to the UCSC genome browser. The two major steps in adding a track are creating a table containing the track information, and putting a description of the track in trackDd. The browser has one mysql database for each version of each genome that it displays. Both the track table and the track description live in this database. The current human genome database is hg13, while the current mouse database is mm2.

II. MySQL Preliminaries.

Before you get started it is good to look at these databases a little, and make sure that you have update access to them. You could do this directly with the 'mysql' command, but let's do it instead with the 'hgsql' command, which will keep you from having to type your user name and password all the time.

Assuming you've got the browser source already installed in ~/kent/src do the following to create hgsql cd ~/kent/src/lib make This make almost always goes smoothly on Linux. You may need to remove the '-ggdb' flag in the makefile on other systems, and possibly set up a MACHTYPE environment variable, and then mkdir $MACHTYPE on some systems. Next cd ../hg/lib make The main problem that can happen with this make is if the mysql libraries and include files are not found. See kent/src/README for details. The next step is cd ../hgsql make rehash The hgsql program is just a thin wrapper around mysql. It looks for the password and username in the file ~/.hg.conf. Here's the necessary parts of .hg.conf:

1. db.host is the name of the MySQL host to connect to

db.host=localhost

1. db.user is the username is use when connecting to the host

db.user=mysqlUserName

1. this is the password to use with the above hostname

db.password=mysqlPassword

1. this is a database where stuff common to all versions of the genome is stored

central.db=hgcentral

This .hg.conf is similar to the cgi-bin/hg.conf file that the browser uses, but it need not contain everything that file does. Also it's advisable to have a read-only user/password in cgi-bin/hg.conf while you'll want a read-write user/password in ~/.hg.conf. Setting this up can involve doing some 'grants' in mysql. See the documentation at www.mysql.com for how do set up various users.

Assuming your mysql and .hg.conf are set up, and that you already have a mirror site going then the command hgsql database (where database is something like hg13 or mm2) should bring you to the mysql prompt. Do mysql> show tables; and you should see a large list of tables. When you've finished adding a track, the track(s) for your tables will be among them. Also try doing: mysql> describe trackDb; This will list the fields of the trackDb table, which has a row for each track. Then do mysql> select tableName,shortLabel,type from trackDb; This will show you some of the key fields from this table. You won't be updating this table directly, but it can be handy to look at it sometimes for debugging purposes. Some other useful mysql commands are

       mysql> select count(*) from sex;

This will count up all the items in the sex table. You thought there were only two? This table reflects the diversity of the sex fields in genbank. Try mysql> select * from sex; to see the full diversity. Well, enough of that non-normalized nightmare. To get out do:

       mysql> quit

III. Loading the main track table.

The UCSC Genome Browser Database is usually loaded from a text file of some sort. The most popular types of text files are .psl files for blat alignments, .bed files for a wide variety of data, and GTF files for gene predictions. See http://genome.ucsc.edu/goldenPath/help/customTrack.html for further information on these formats. For now we'll assume you have a file in one of these formats that you want to add to the browser.

A) Creating the loader programs cd ~/kent/src/hg/makeDb cd hgLoadPsl make cd ../hgLoadBed make cd ../ldHgGene make cd ../hgPepPred

  The makeDb directory also contains loaders for a number of more
  specialized tables including hgLoadOut for RepeatMasker data.  There
  is also a .doc file describing in detail how we created each database
  in the files named things like makeHg13.doc and makeMm2.doc.

B) Loading a bed file

  Loading a bed file is the most straightforward.  First decide on
  the name you want to call the table.  Then do
      hgLoadBed database tableName file.bed
  type hgLoadBed with no arguments for further information. 
  Database will be something like hg13 or mm2.

C) Loading a psl file

  Loading a psl file is also easy.  Make sure that the psl file
  is sorted by chromosome (tName) and start position (tStart).  Use
  kent/src/hg/pslSort or just plain Unix sort for this if necessary.
  If the number of alignments is somewhat modest (say less than 
  500,000) then do
  	hgLoadPsl database -table=tableName file.psl -tNameIx
  This will load everything into one big table.  For huge numbers of
  alignments the browser will be faster if you first split up the
  data into one file for each chromosome.  Name these files 
  chr1_tableName.psl  chr2_tableName.psl and so forth.  Then do
       hgLoadPsl database chr*_tableName.psl
  This will end up making a separate table for each chromosome.
  Unfortunately it is still a bit complicated to make the details
  pages for a psl format track to include the alignments themselves.
  Please contact us at UCSC if this is a priority for you and we
  will try to make it easier.

D) Loading a GTF (or GFF) file

  Generally GTF is a much more tightly defined standard than GFF, so
  GTF files are more likely to work without tweaking.  However most
  reasonable GFFs will work as well.  To load do
       ldHgGene database tableName file(s).gtf
  This will make a gene-prediction type table. You often will want
  to create an associated predicted peptide table as well. To do this
  do
       hgPepPred database generic tableNamePep file(s).fa
  The first word after the '>' in the fasta files should use the same
  symbol as the 'group' in GFF files or 'transcript_id' in GTF files.

IV. Updating trackDb

Your data will not display in the browser until you load it into trackDb. To do this first cd ~/kent/src/hg/makeDb/trackDb and look at the file trackDb.ra, and read the README file. Then decide whether your new track should be global, organism specific, or assembly specific, and edit the corresponding trackDb.ra file. Generally it's good to find an existing track as similar as possible to the track you want to add, copy and paste it, and modify the copy. Then put any explanitory text you want on the track in trackName.html in the appropriate directory. After this do a make alpha to update the trackDb table, or a

       make

to update trackDb_user (where user is you Unix username). If multiple engineers are working on the project you can set up cgi-bin-user directories with hg.conf files that will tell the browser to use trackDb_user instead of trackDb to avoid conflicts with other engineers code.

After the 'make alpha' the browser should show your track. Congratulations if you've made it this far. See also http://www.soe.ucsc.edu/~sugnet/doc/trackHowto/browserTalk.pdf for Charles Sugnet's description of how to add a track including some code customization.

-Jim Kent Feb 14, 2003.