This file and all files in this release of the Glimmer system are
    OPEN SOURCE software.  Feel free to redistribute, and please cite
    our publications.

Glimmer 2.0 is described in:
  A.L. Delcher, D. Harmon, S. Kasif, O. White, and S.L. Salzberg.
  Improved Microbial Gene Identification with Glimmer.  
  Nucleic Acids Research, 27 (1999), 4636-4641.
Please reference this paper if you use the system as part of any
published research.  Note that Glimmer 1.0 is described in
  S. Salzberg, A. Delcher, S. Kasif, and O. White.
  Microbial Gene Identification using Interpolated Markov Models.
  Nucleic Acids Research, 26:2 (1998), 544-548.

Quickstart: if you just want to run Glimmer 2.0 on your genome
and you don't want to adjust any parameters (although we don't
recommend this), you can simply compile this system and run
it with the included run-glimmer2 script.  E.g.:
unix-prompt> make
[various compilation messages appear]
unix-prompt> run-glimmer2 mygenome

run-glimmer2 will create an Interpolated Markov Model of your genome
and store it in a binary file called tmp.model.  It will store
the predicted gene coordinates in g2.coord.  Along the way
it will extract long ORFs and store them and their coordinates
in tmp.train and tmp.coord.

Recommended: read the readmes.

Glimmer 1.0 had 4 readme files, and Glimmer 2.0 maintains that
structure.  The four main programs are:
  1. long-orfs
  2. extract
  3. build-icm
  4. glimmer2
There are files called *.readme for each of these programs.  Please
read these first before emailing the authors with any questions.

Art Delcher, adelcher@umiacs.umd.edu, is the primary programmer for
most of the Glimmer code, and he can answer most technical questions.

CHANGELOG, 7/31/00:
   - Weak scores are now only invoked with the -w option.  Any weak-score
     gene is rejected automatically by an overlap with a regular gene.
   - Weak-scores genes and "voted" genes are now annotated by [Weak] and
     [Vote] in the final listing.  Voted genes are those which have a
     significant number of relatively high-scoring subregions.  Voted
     genes also are rejected automatically by overlaps with regular genes.
   - Weak scores are computed to be more independent of architecture-dependent
     floating-point features.  (Previously, 64-bit machines would sometimes
     generate different results from 32-bit machines.)
   - Fixed bug in RNABin function that occurred when the gene
     started on the very last base of the genome.  This function is
     now not called at all if the Choose_First_Start_Codon option is
     selected (which is the default).
   - Fixed problem that occurred on short pieces of genome when one
     frame (or more) had no stop codons.  
   - An ignore option (-i) to specify a list of regions in which no predictions
     will be made, such as ribosomal RNAs.  This feature has not yet been
     thoroughly tested.

CHANGELOG, 9 December 2002
   - Raw scores are now printed in the main listing and in []'s in
     the final list of putatative genes
   - Add +S option to us a "stricter" independent (intergenic) model
     that discounts stop codons.  Since only orfs (which have no stop
     codons) are ever scored, the independent model is at a disadvantage
     unless it also assumes that it is only scoring orfs.  Thus, with the
     +S option, the independent score is done codon by codon.
     The probabilities of codons are intially set to what the
     previous independent model would be:
     The probability of a codon "atg", for example is:
         Pr[a] * Pr[t] * Pr[g]
     Then each of these is divide by the sum of the probabilities of the
     non-stop codons.
   - Add -L option to specify the name of a file containing a list
     of coordinates.  The genes in these lists are scored separately by
     the ICM, output, and then the program stops (i.e., no
     overlapping/voting rules).

CHANGELOG, 5 February 2003
   - The strict independent (intergenic) model is now the only mode.
     The +S option is tolerated but has no effect.

CHANGELOG, 18 April 2003 
   - Compute the optimal length for minimum "long" orfs, so that the 
     program will return the largest number of orfs possible.  The -g 
     switch still works if specified, but I don't know why anyone would 
     want to use that for a training set.  
   - Change minimum overlap by default to be 0.  This means that genes
     that overlap even by 1 base will be considered in conflict by Glimmer,
     and the program will try to adjust their start codons to remove the
     conflict or else delete one of the genes.

CHANGELOG, 7 October 2003
   - Fix bug on long-orfs.cc to avoid occasional array out-of-bounds
     error (detected on Mac OS X).