Software - Protein Homology Detection Through Alignment of Markov Random Fields - page 34

Column 6 ' Edge ' The accumulative edge alignment potential.

Column 7 ' qRange ' Range of aligned region in the query protein.

Column 8 ' tRange ' Range of aligned region in the subject protein.

Column 9 ' tLength ' The length of the subject protein.

Column 10 ' Cols ' The number of aligned positions in the pairwise alignment.

Column 11 ' #tGaps ' The number of gaps in the subject protein (Insertions).

Column 12 ' #qGaps ' The number of gaps in the query protein (Deletions).

Column 13 ' #seqID ' The number of identical residues in the alignment.

3.5 Interpreting P-Value

In the ranking dence score indicating the

relative quality of the top-ranked proteins and (corresponding) alignments. To cal-
culate the P-value, we employs a set of

file, P-value can be interpreted as a con

“

reference proteins (in databases/

”

CAL_TGT) 1,800 single-domain proteins belonging to dif-

ferent SCOP folds. Given a query protein, we

), which consists of *

first align it to this reference protein
database and then estimate an extreme value distribution from the 1,800 alignment

scores. Based upon this distribution, we calculate the P-value of each alignment
when aligning the query protein to the subject protein database. The P-value actually
measures the likelihood of each subject protein being homologous to the query
protein by comparing it to the reference proteins.
To see the relationship between the P-value and the closeness of the

*

rst-ranked
protein by MRFsearch to a query protein, we conduct an experiment on the
368 CAMEO target proteins. For each CAMEO target, the

first-ranked protein in
the database is treated as the homolog of this target. To measure the quality of an
alignment, we use un-normalized Global Distance Test (GDT). GDT has been
employed as an of cial measure of a protein model quality by CASP for many

years. When applied to alignments, uGDT can be interpreted as the number of
correctly-aligned positions in an alignment, but weighted by alignment quality at
each position. We say one alignment is good when its uGDT is larger than 50. We
use 50 as a cutoff because that many proteins similar at only the fold level have
uGDT around 50. Figure 3.2 shows the relationship between P-value and uGDT on
the 368 CAMEO targets. Figure 3.3 is a zoom-in graph of Fig. 3.2 , showing
relationship between P-value and uGDT on the 132 CAMEO targets with

− log(P-

value) < 20. As shown in Fig. 3.3 , when P-value is small (i.e. <10e 10), most

alignments have uGDT greater than or equal to 50. That is, when P-value is less
than 10e

−

first-ranked protein is very likely to share a similar fold as the
query protein. When P-value is between 10e

− 10, the

5 and 10e 10, more than half of the

− −

alignments have uGDT > 50.