Software - Protein Homology Detection Through Alignment of Markov Random Fields - page 32

3. MRFsearch_Edge_Potential.tar.gz ( 1G): this package contains pairwise resi-

due correlation information for the proteins in the database.
4. nr70.tar.gz (

*

1.6G) and nr90.tar.gz ( 2.3G): the formatted NR70 and NR90

* *

protein sequence databases.
5. MRFsearch_Database_List.tar.gz ( 1M): the package contains some database

lists. The database lists will be updated weekly.

*

3.3 Feature Files

For each protein there are two feature file that contains some

basic information about this protein and some predicted local features such as
secondary structure. Notice that not all the features are use in MRFsearch. Below
are some details.

files. One is a *.tgt

Sequence
information

Primal sequence

Pro le
information

NEFF value The average entropy of the protein sequence pro le

PSSM matrix Position-speci c score matrix generated by PSI-blast with

5 iterations and E-value 0.001

PSFM matrix Position-specific frequency matrix generated by PSI-blast

with 5 iterations and E-value 0.001

HMM matrix The Hidden Markov Model generated by buildali2.pl in

the HHpred package

Structure
information

3-class
secondary
structure

The predicted 3-class secondary structure types by
PSIPRED [ 1 ]

8-class
secondary
structure

The predicted 8-class secondary structure types by
RaptorX-SS8 [ 2 ]

3-class solvent
accessibility

The predicted 3-state solvent accessibility, which is
discretized into three equal-frequency states: buried,
intermediate and exposed

Disorder
prediction

The predicted disorder information by DISOPRED [ 3 ]

file contains pairwise residue correlation information produced
by the EPAD package. For a protein of length N, it contains N

The other feature

2 rows,
where each rows contains 9 numbers representing the probability of 13 distance
bins (3

ð N 6 Þ =

4, 4 5, 5 6, … ,14 15, and >15 Å ).

- - - -