Bio-Medical Interaction Extractor Syed Toufeeq Ahmed ASU

Matching with BioMedical Ontology Gene List (508,477) from Locus Link ( Interaction List (1500) from UMLS ( A noun phrases is tagged as a GENE (G) if it matches a gene from the Gene list Any word is tagged as an INTERACTION (I) if it matches an interaction from the interaction list (after stemming)

Syntactic Roles with Link Grammar “HMBA could inhibit the MEC-1 cell proliferation by down-regulation of PCNA expression.” Subject Verb Object Modifying Phrase

Scopes Various syntactic roles (such as Subject, Object and Modifying phrase) and their linguistically significant combinations makes up SCOPES A SCOPE MATCHING is: Elementary (E) : If the scope contains a Gene /Protein (G) name or an interaction word (I). Partial (P) : If the scope has a Gene/Protein (G) name and an interaction word (I). Complete (C) : If the scope has at least two Gene /Protein (G) names and an interaction word (I).

Scopes Elementary (Subject) Elementary (Object) Partial (Modifying Phrase) “HMBA could inhibit the MEC-1 cell proliferation by down-regulation of PCNA expression.” Interaction (Verb)

Scopes & Matches “The kinase phosphorylation of Gene1 by Gene2 could inhibit Gene3. ” Complete (Subject)

Algorithm of Interaction Extractor: SOMPMP S-OS- M SubjectModifying Phrase Object complete (G,I,G)  interact: {G,I,G} complete (G,I,G)  interact: {G,I,G} complete (G,I,G)  interact: {G,I,G} Elementary (G1) Elementary (G2) Is Main Verb an Interaction (I) ? Interaction : { G1, I, G2 } Partial (I,G2) Interaction : { G1, I, G2 }

Algorithm 1) Using the linkage given by the Link Grammar parser, the Subject, the Object and the modifying phrase scopes are obtained (S, O and MP respectively). 2) If S, O or MP has a complete interaction, then we use Preposition- based approach to find agent, theme and action to extract the interaction. 3) a) Identify the main verb of the sentence and extract interaction from the combination of Subject and Object scopes. b) If the above step gives a complete interaction from subject-object combination (S4 = C) and the scope of modifying phrase is Elementary then skip STEP 4. 4) Extract interaction from the combination of Subject and modifying scopes.

Different possible cases for subject-object combination when main verb is not an interaction word. When main verb is NOT an Interaction wordExtracted Interaction a)S = E (G1)O = P (I1,G2){G1,I1,G2} b)S = P (G1,I1)O = E (G2){G1,I1,G2} c)S = C (G1,I1,G2)O = P (I2,G3){(G1,I1,G2), I2,G3} d)S = P (G1,I1)O = C (G2,I2,G3){G1,I1,(G2,I2,G3)}

When main verb is an Interaction word (I1)Extracted Interaction a)S = E (G1)O = E (G2) {G1,I1,G2} b)S = E (G1)O = P (I2,G2) {G1,I1,(I2/G2)} c)S = P (G1,I2)O = E (G2) {(G1/I2), I1, G2} d)S = P (G1,I2)O = P (I3,G2){(G1/I2), I1, (I3/G2)} e)S = C (G1,I2,G2)O = E (G3){(G1,I2,G2), I1,G3) f)S = E (G1)O = C (G2,I2,G3){G1,I1,(G2,I2,G3)} Different possible cases for subject-object combination when main verb is an interaction word.

Different possible cases for subject-modifying phrase combination when main verb is not an interaction word. When main verb is NOT an Interaction wordExtracted Interaction a)S = E (G1)MP = P (I1,G2){G1,I1,G2} b)S = P (G1,I1)MP = E (G2){G1,I1,G2} c)S = C (G1,I1,G2)MP = P (I2,G3){(G1,I1,G2), I2,G3} d)S = P (G1,I1)MP= C (G2,I2,G3){G1,I1,(G2,I2,G3)}

Different possible cases for subject- modifying phrase combination when main verb is an interaction word. When main verb is an Interaction word (I1)Extracted Interaction a)S = E (G1)MP = E (G2) {G1,I1,G2} b)S = E (G1)MP = P (I2,G2) {G1,I1,(I2/G2)} c)S = P (G1,I2)MP = E (G2) {(G1/I2), I1, G2} d)S = P (G1,I2)MP = P (I3,G2){(G1/I2), I1, (I3/G2)} e)S = C (G1,I2,G2)MP= E (G3){(G1,I2,G2), I1,G3) f)S = E (G1)MP = C (G2,I2,G3){G1,I1,(G2,I2,G3)}

Example Elementary (G) Elementary (G) Partial “HMBA could inhibit the MEC-1 cell proliferation by down-regulation of PCNA expression.” Main Verb (I) { “HMBA”, “inhibit”, “the MEC-1 cell proliferation” } { “HMBA”, “down-regulation”, “PCNA expression”}

Example 1) The main verb ( “inhibit” ) is identified: Subject: “HMBA” (Elementary) Object: “the MEC-1 cell proliferation” (Elementary) Modifying Phrase: “by down-regulation of PCNA expression” (Partial) “HMBA could inhibit the MEC-1 cell proliferation by down-regulation of PCNA expression.”

{ “HMBA”, “inhibit”, “the MEC-1 cell proliferation” } a)S = E (G1)O = E (G2) {G1,I1,G2} b)S = E (G1)O = P (I2,G2) {G1,I1,(I2/G2)} c)S = P (G1,I2)O = E (G2) {(G1/I2), I1, G2} d)S = P (G1,I2)O = P (I3,G2){(G1/I2), I1, (I3/G2)} e)S = C (G1,I2,G2)O = E (G3){(G1,I2,G2), I1,G3) f)S = E (G1)O = C (G2,I2,G3){G1,I1,(G2,I2,G3)} 3) Interaction between subject and object is extracted.

4) Now we extract interaction between Subject and modifying phrase. a)S = E (G1)MP = P (I1,G2){G1,I1,G2} b)S = P (G1,I1)MP = E (G2){G1,I1,G2} c)S = C (G1,I1,G2)MP = P (I2,G3){(G1,I1,G2), I2,G3} d)S = P (G1,I1)MP= C (G2,I2,G3){G1,I1,(G2,I2,G3)} {“HMBA”, “down-regulation”, “PCNA expression”}

Example 2 “The kinase phosphorylation of Gene1 by Gene2 could inhibit Gene3. ” Complete (Subject) Elementary (Object) Main verb

Preposition-based patterns Subject is Complete and Of / by pattern: ….. Of.. … by.. …. “The kinase phosphorylation of Gene1 by Gene2” { “Gene2”, “phosphorylation”, “Gene1” }

Sub: “The kinase phosphorylation of Gene1 by Gene2” Obj: “Gene3” Verb: “inhibit” Nested Interaction: { { “Gene2”, “phosphorylation”, “Gene1” }, “inhibit”, “Gene3” } “The kinase phosphorylation of Gene1 by Gene2 could inhibit Gene3. ” { “Gene2”, “phosphorylation”, “Gene1” }

Next Steps Handling negations in the sentences (such as “not interact”, “fails to induce”, “does not inhibit”). Extraction of detailed contextual attributes of interactions (such as bio-chemical context or location) by interpreting modifiers: Location/Position modifiers (in, at, on, into, up, over…) Agent/Accompaniment modifiers (by, with…) Purpose modifiers( for…) Theme/association modifiers ( of..) Extraction of relationships between interactions from among multiple sentences in abstracts (signaling pathways)

Next Steps Visualization of Signaling Pathways

Preliminary Results DatasetPrecision %Recall % Curated text 95.4 % Abstracts %89.18 %

References Link Grammar: LocusLink: UMLS: