NAME

SYNOPSIS

DESCRIPTION

seqfile may contain RNA, DNA, or amino acid sequences (don't mix them, though). They may be in either GCG's MSF multiple alignment format or my SELEX alignment format. See bashford7.slx in the source distribution for an example of SELEX format.

Any of ` `, `.', or `-' are accepted as gap symbols. The output of ClustalV or ClustalW are acceptable SELEX format, if the header lines, consensus line, and coordinate lines (if any) are removed. SELEX format also has optional features for describing consensus structure, a canonical reference coordinate system, and individual structures for each sequence, but these are not described further here. See the HMMER User's Guide for more details.

The algorithm used to build the model is an unpublished maximum likelihood model construction algorithm. It is different from the procedure described by Krogh et al. It is more theoretically satisfying than the ad hoc procedure they describe, but be warned that it tends to favor assigning columns of the alignment to match states instead of insert states rather more than a human's intuition would.

The statistics of the model -- symbol emission probabilities and state transition probabilities -- are set according to a maximum likelihood criterion. We have done some research on alternative parameter estimation methods. Maximum likelihood is sensitive to biased representation in the training data set; a large subfamily of highly related sequences will skew the model's statistics. There are three hmmb options for alleviating this problem. The -d option implements "maximum discrimination", our recommended solution, which uses a rigorous method to find a model that optimally discriminates all the training sequences from random background. The -v and -w options implement two different ad hoc sequence weighting rules, providing the usual solution to the biased representation problem.

OPTIONS

-d

Maximum discrimination parameter estimation, instead of maximum likelihood, to correct for biased representation and improve sensitivity overall. Parameters cannot be calculated analytically using this method, and must be calculated iteratively instead, so it takes a little longer.

-h

Print short usage and help info for the program.

-p <priorfile>

Use the prior probability distributions in <priorfile>, instead of the defaults. The defaults are almost identical to those given in the original Krogh et al. HMM paper. Example files that illustrate the correct prior file format are provided in the distribution for the default nucleic acid prior (Demos/Nucleic.pri) and the default amino acid prior (Demos/Simple.pri). The files are commented and should be self-documenting. More complicated (and experimental) priors based on mixture Dirichlets and Dirichlet mixtures in combination with a softmax neural net taking input from X-ray crystal structure data for a subset of sequences in the alignment are given in (Demos/BrownHaussler.pri) and (Demos/Structure.pri), respectively.

-v

Weight the input sequences to correct for biased representation using the Voronoi weighting rule of Sibbald and Argos (J. Mol. Biol. 216:813-818 1990).

-w

Weight the input sequences to correct for biased representation using the Sonnhammer weighting rule of Gerstein et al. (J. Mol. Biol. 235:1067-1078 1994).

-A <archprior>

Probably useless to you. Sets the "architectural prior", which exerts a force that keeps model lengths under control at the expense of some likelihood. The default is 0.85. To make longer and slightly more high-scoring models, set it higher (to a maximum of 1.0). To make shorter models, set it lower.

-C <cthresh>

Probably useless to you. Sets the convergence threshold for the maximum discrimination rules -d and -M. The default is 0.0001. To continue training longer, set this lower. This only takes effect if -d or -M are also on.

-D <damp>

Probably useless to you. Set the damping factor for making the maximum discrimination rules converge smoothly. Default is 0.99. If -d or -M options seem to be producing oscillatory behavior, increase this to 0.999 or some such number closer to 1.

-E <epsilon>

Probably useless to you. Sets the learning rate of the maximin rule, toggled on by the -M option. Default is 0.1. To learn faster, at the risk of oscillatory behavior, increase this slightly to perhaps 0.2.

-M

Not recommended. Activates the maximin variant of maximum discrimination training.

-P <PAMfile>

(Experimental) Use the scoring matrix in <PAMfile> as a source of prior symbol emission information. This results in behavior that is similar to the behavior of sequence "profiles" (Gribskov et al., Meth. Enzymol. 183:146-159, 1990) when there are only a few sequences in the alignment. Because the default behavior of hmmb is to use a very simple prior, with no information about the chemical similarities of amino acids, we believe the -P option will prove superior.

-R

(Experimental) "Ragged" alignment -- ignore the exterior gaps in the alignment for the purposes of collecting state transition statistics. This is useful if your alignment is known to contain a number of fragments and you don't want the HMM to be counting a number of fictitious delete states to account for the missing ends.

-W <alifile>

Re-saves the alignment to <alifile>. Especially useful with -d, -v, or -w, since the alignment file includes the final weights that were used by the model. The alignment file also contains an #=RF line with x's marking the columns that were assigned to matches by HMM construction. This allows other programs to read the alignment file and readily make exactly the same state assignments as the HMM did.

Individual man pages: hmma(l), hmme(l), hmmfs(l), hmmls(l), hmms(l), hmmsw(l), hmmt(l), hmm-convert(l)

User guide and tutorial: Userguide.ps

BUGS

Not very tolerant of errors on the command line.

NOTES

Sean Eddy
Dept. of Genetics, Washington Univ. School of Medicine 660 S. Euclid Box 8232
St Louis, MO 63110 USA
Phone: 1-314-362-7666
FAX : 1-314-362-2985
Email: eddy@genetics.wustl.edu

Table of Contents

• NAME
• SYNOPSIS
• DESCRIPTION
• OPTIONS
• SEE ALSO
• BUGS
• NOTES