Databases Reference
In-Depth Information
Tabl e 1 .
Each row shows the template evaluated at one iteration, starting with the
literal template derived from the given seed phrase at iteration zero.
Iteration
Measured
Inherited
Template pattern
matches
matches
|TP|
|FP|
|TP|
|FP|
0
1
0
-
-
[
Λ
: Rad53p] [
Λ
:protein][
Λ
:binds][
Λ
: to]
[
Λ
: Dbf4p]
10
2
0
1
0
[
Γ
: protein, sp] [
Λ
:protein][
Γ
: binding]
[
Ω
:*][
Γ
: protein, sp]
20
86
1
25
1
[
Γ
:protein][
Ω
:?][
Γ
: binding] [
Ω
:*][
Γ
:
protein, sp]
30
286
4
247
4
[
Γ
:protein][
Ω
: ?????] [
Γ
: binding] [
Ω
:
???] [
Γ
: protein, sp]
40
1,202
34
1,202
34
[
Ω
: ??] [
Ω
: ?????] [
Γ
: binding] [
Ω
: ?????]
[
Γ
: protein]
50
2,621
151
2,621
151
[
Ω
: ???] [
Ω
: ?????] [
Γ
: binding] [
Ω
: ?????]
[
Ω
:*]
The number of matches in the positive (TP) and neutral (FP) corpora are shown for
the current template (“measured matches”) and for its parent (“inherited matches”).
Each template element is shown delimited by square brackets in the form [CATE-
GORY: value], with categories literal (
Λ
), gazetteer (
Γ
) and wildcard (
Ω
). A wild-
card element with
n
query symbols matches between 0 and
n
words. The sequence of
templates generated tends to generalise from literals through gazetteers to wildcards
•
We report here that
Grb2 also interacts with tyrosine-phosphorylated IRS-
1
in response to gastrin.
•
...
RII alpha fused to
endonexin II formed dimers but did not bind MAP2
.
•
A protein interaction map
for cell polarity development.
While the first three results correctly show protein-protein interactions,
the last two highlight imperfections in this particular template. The phrase
“did not bind” indicates that no interaction was observed, but the template
still treats this as a positive statement. In the last example shown, the word
“map” is not a protein, although the same word is used to refer to a protein
in other contexts and so it is in the protein gazetteer.
Table 1 also shows that the template evaluated in the 50th iteration
matches 2,621 fragments in the positive corpus, and 151 in the neutral, with
the template pattern: [
Ω
: ???] [
Ω
: ?????] [
Γ
: binding] [
Ω
: ?????] [
Ω
:*]
This might be seen as an example of over-generalisation, in that the tem-
plate consists almost entirely of wildcards. However, the quality of any tem-
plate depends on the requirements of the user, and for some tasks, even this
very general template may still be useful. Note that the total runtime on a
stand-alone PC was 20 min for these 50 iterations.
