Information Technology Reference
In-Depth Information
[ Positive_regulation ]->[ Regulation ]-> [ Gene_expression ]->[ PROTEIN1 ]. Moreover,
the reduction of the semantic classes space provided by pattern selection is critical. It
allows the execution of more sophisticated text classification algorithms, which lead
to improved results. Those algorithms cannot be executed on the original semantic
classes space because their execution time would be excessively high, making them
impractical [1]. Therefore, to select patterns closely associated with an interaction
would improve the performance of PPI extraction. We use the pointwise mutual
information (PMI) [1], a popular statistical approach used in feature selection, to
discriminate semantic classes for PPI instances. Given a training dataset comprised of
positive instances, the PMI calculates the likelihood of the occurrence of a semantic
class in the expressions of PPI. A semantic class with a large PMI value is thought to
be closely associated with the interaction. Lastly, we rank the interaction patterns in
the training dataset based on a sum of semantic classes PMI values and retain the top
20 for representing protein-protein interactions.
4
Interaction Pattern Tree Construction
A PPI instance is represented by the interaction pattern tree (IPT) structure, which is the
shortest path-enclosed tree (SPT) of the instance enhanced by following steps. To
facilitate comprehension of the construction process, the positive instance shown in Fig.
4(a), which mentions the interaction between "AVP" and "PKC", serves as an example.
(a). Full Parse Tree (FPT)
(b). Shortest Path-enclosed Tree (SPT)
(c). IPT Pruning
(d). IPT Ornamenting
Fig. 4. The interaction pattern tree construction procedure for a PPI instance “The inhibitory
action of AVP involves both the activation of PKC and the transcription of iNOS mRNA in
cultured rat GMC”
Search WWH ::




Custom Search