An Interaction Pattern Kernel Approach for Protein-Protein Interaction Extraction from Biomedical Literature - Technologies and Applications of Artificial Intelligence - page 39

approach that aims to automatically generate representative patterns from sequences
of expression of protein-protein interactions.
We formulate interaction pattern generation as a frequent pattern mining problem.
First of all, the instances undergo the semantic class labeling process. To illustrate the
process of semantic class labeling, consider the instance I n = " Abolition of the gp130
binding site in hLIF created antagonists of LIF action ", as shown in Fig. 2. First,
" gp130 " and " hLIF " are two given protein names, as tagged PROTEIN1 and
PROTEIN2 respectively. Then, we stem remaining tokens by using porter stemming
algorithm [15]. Finally, trigger words "bind" and "antagonist" are labeled with their
corresponding types by using our compiled trigger word list which extracts from a
BioNLP corpus [8]. Evidently the SCL can group the synonyms together by the same
label. This enables us to find distinctive and prominent semantic classes for PPI
expression in the following stage.
Fig. 2. Semantic class labeling process
After labeling semantic classes, we based on the co-occurrence of semantic classes
to construct a graph to describe the strength of relations between them. Since
semantic classes are of an ordered nature, the graph is directed and can be made with
association rules. In order to avoid the generation of frames with insufficient length,
we empirically set the minimum support of a semantic class as 20 and minimum
confidence as 0.5 in our association rules. Thus, an association rule can be
represented as (1). Fig. 3 is an illustration of a semantic graph. In this graph, vertices
( SC x ) represent semantic classes, and edges represent the co-occurrence of two
classes, SC i and SC j , where SC i precedes SC j . The number on the edge denotes the
confidence of two connecting vertices. After constructing all of the semantic graphs,
we then generate semantic frames by applying the random walk theory [13] in search
of high frequency and representative classes for each topic. Let a semantic graph G be