Prior files

Prior files are used to include customized or more detailed Dirichlet prior information into an HMM. Three types of priors are available:

• Simple prior (single Dirichlets)
• Mixture prior (counts based, Brown/Haussler)
• Structural mixture prior (structure based, Mitchison/Eddy)

Examples of each sort of file can be found in the Demos directory of the source distribution, in Simple.pri, BrownHaussler.pri, and Structure.pri. These priors are for protein HMMs. Nucleic.pri is an example of a simple prior for DNA or RNA.

In these files, everything is case-insensitive, and anything after a # is an ignored comment. The position of line breaks is unimportant; lines can be broken any way you want, so long as the order of the data is unchanged.

The first non-comment line in the file contains a keyword Simple, Mix, or Struct, stating the type of prior. The second line contains Amino or Nucleic, stating the type of ``alphabet''.

The remainder of the file is arranged as follows:

1. State transition mixtures
   1. Number of mixtures. (1 for simple, >=1 for mixtures)
   2. For each mixture:
      1. A priori probability of this mixture (Mix; 1.0 for Simple)
      2. (or) four input parameters to structural softmax neural net, in order bias, sheet, helix, surface accessibility (Struct only)
      3. Dirichlet alphas for match transition vector (to M, I, D)
      4. Dirichlet alphas for insert transition vector (to M, I, D)
      5. Dirichlet alphas for delete transition vector (to M, I, D)
2. Match emission mixtures
   1. Number of mixtures. (1 for simple, >=1 for mixtures)
   2. For each mixture:
      1. A priori probability of this mixture (Mix; 1.0 for Simple)
      2. (or) four input parameters to structural softmax neural net, in order bias, sheet, helix, surface accessibility (Struct only)
      3. Dirichlet alphas for emission vector (4 or 20 terms)
3. Insert emission mixtures
   1. Number of mixtures. (1 for simple, >=1 for mixtures)
   2. For each mixture:
      1. A priori probability of this mixture (Mix; 1.0 for Simple)
      2. (or) four input parameters to structural softmax neural net, in order bias, sheet, helix, surface accessibility (Struct only)
      3. Dirichlet alphas for emission vector (4 or 20 terms)

For example, here is Structure.pri:

# This is an example structural mixture 

STRUCT       # A structure-based mixture prior
Amino        # for proteins.

# First are the transition mixtures.
#
2   # 2 mixtures

# transition mixture 1 - few gaps
#
1.4129  1.5093  0.9553  0.2306     # perceptron weights: bias, E, H, SA
34.9799    0.00223637  0.0456661   # m->m m->i m->d
2.61166   12.5412      0.00343395  # i->m i->i i->d
0.0842934  0.000110474 0.161144    # d->m d->i d->d
                                   
# transition mixture 2 - lots of gaps
#
-0.4454 -0.0605  -0.6015  0.3357   # perceptron weights: bias, E, H, SA
4.06323  0.251117  0.525515        # m->m m->i m->d
1.83291  3.48332   0.0270406       # i->m i->i i->d
0.746496 0.0485157 0.433166        # d->m d->i d->d


# Match emissions
# Per mixture, first line is perceptron weights bias, E, H, SA
# next line is 20 alphas for A..Y
#
5    # 5 mixtures

# mix 1 likes accessible cysteines 
-0.0978  0.0456  -0.6953 2.0663
13.368 2923.213 0.293 0.332 0.789 1.519 0.473 1.243 0.289 1.551 
0.990 0.433 0.566 0.393 0.310 1.618 2.801 2.796 0.432 0.568 

# Mix 2 likes solvent-exposed residues; Ala, Asp, Glu, Gly, Lys, Asn, Arg, Ser, Thr
0.7453 0.0071  0.5324  6.1060  
0.247 0.013 0.280 0.271 0.039 0.250 0.070 0.061 0.304 0.096 0.035 
0.218 0.148 0.168 0.162 0.279 0.228 0.100 0.020 0.064 

# Mix 3 is buried Ala/Gly/Ser/Thr
-0.1997  -0.2349  1.7820 -8.6385
577.638 0.002 1.281 1.073 0.087 125.160 0.049 0.343 1.053 5.082 
0.014 0.536 0.577 0.724 0.325 92.578 17.997 1.019 0.000 0.182 

# Mix 4 is buried Ala/Ile/Val (fairly rare)
-0.9960  2.6519  0.3427  -2.8808
4.776 0.191 0.252 0.365 0.418 0.310 0.163 75.919 0.245 3.679 0.494 
0.371 0.432 0.247 0.229 0.693 1.767 158.490 0.095 0.263 

# Mix 5 is a common catch-all for everything else.
2.6848  0.5548  0.8311 5.0195
0.092 0.027 0.018 0.016 0.103 0.065 0.028 0.122 0.007 0.190 0.054 
0.021 0.034 0.019 0.026 0.048 0.066 0.137 0.030 0.077 

# Insert emissions
# high-alpha (hardwired) single mixture
1
1.0 0.0 0.0 0.0
17430 6690 9370 9910 7950 17720 6720 7380 16100 17070 2950 8090 
10140 7650 8180 13920 11710 12940 2100 5980