SolrSherlock's HyperMembrane as an associative fabric component of a machine reading platform. The system entails topic maps, NLP, and a society of agents to support hypothesis formation, experiment planning, and Deep QA
2. Target Benefits
• SolrSherlock will support:
– Hypothesis formation
– Research/Experiment planning
– Deep Question Answering
• Personal medical issues
• …
“Therefore psychologically we must keep all the theories in our heads, and
every theoretical physicist who is any good knows six or seven different
theoretical representations for exactly the same physics.”
―Richard Feynman
“Why, sometimes I've believed as many as six impossible things before
breakfast.”
―The Queen: Through The Looking Glass
3. What We Have Read: HFE
• Human hemochromatosis protein also known
as the HFE protein is a protein which in
humans is encoded by the HFE gene. The HFE
gene is located on short arm of chromosome 6
at location 6p22.2*
– Some mutations which are associated with
Hereditary Hemochromatosis (a genetic
disease)**:
• C282Y
• H63D
*http://en.wikipedia.org/wiki/HFE_%28gene%29
**http://www.genome.gov/10001214
4. What We Have Read: BCR-ABL aka:
Philadelphia Chromosome
• Philadelphia chromosome or Philadelphia
translocation is a specific chromosomal
abnormality that is associated with chronic
myelogenous leukemia (CML). It is the result
of a reciprocal translocation between
chromosome 9 and 22, and is specifically
designated t(9;22)(q34;q11)*
*http://en.wikipedia.org/wiki/Philadelphia_chromosome
5. Are HFE and BCR-ABL Linked?
• One document instance which suggests they
are linked:
– “We found that HFE C282Y might be associated
with a protective role against CMPD. Because
chronic iron deficiency or latent anemia may
trigger disease susceptibility for CMPD, HFE C282Y
positivity may be a genetic factor influencing this
effect.”*
• Note: this response is simply evidence of a link, a
signal; it leaves open many questions
CMPD: Chronic Myeloproliferative Disease
* http://www.ncbi.nlm.nih.gov/pubmed/19258483
6. Where do we go from here?
• We have read about some actors
• We seek evidence for relationships between
those actors
• We have one small piece of evidence
• We turn to Literature-based Discovery (LBD)
– Read and process many papers
– Assemble an evidence field
– Determine answers and confidence levels
10. SolrSherlock’s HyperMembrane
• SolrSherlock Big Picture
– Documents to harvest
– Sentences to parse
• WordGrams from the sentences
– Lenses to interpret the sentences
» NTuples from the WordGrams
– Lenses to interpret whole documents
• HyperMembrane as a fabric woven from the
Ntuples
– Organizes statements read from literature into a kind
of associative fabric, linked into a topic map
13. Sentence Parse
• Salient WordGrams in that sentence:
– C282Y
– might be associated with a
– protective role against
• Transforms to: protect against
– CMPD
We found that HFE C282Y might be associated with a protective role against CMPD
+-----------------MVp-----------------------------------+
| +---------Js------------+ |
+---Cet------+ | | +-------Ds---------+ |
+-Sp-+--TH--+ +--G-+--Ss--+--Ix---+---Pv-----+---MVp--+ | +----A---+ +--Js--+
| | | | | | | | | | | | | |
we found.p that.c HFE C282Y might.v be.v associated.v with a protective.a role.n against CMPD
Parse produced by a Java
implementation of Link
Grammar Parser
14. WordGram instances
created while processing
the sentence
WordGram Example
• Sentence:
– CO2 causes climate change
• WordGrams
– Terminals
• CO2
• causes
• climate
• Change
– Pairs
• CO2 causes
• causes climate
• climate change
– Triples
• CO2 causes climate
• causes climate change
– Quads
• CO2 causes climate change
• Parsed Result—representation of the sentence:
– CO2 (terminal, noun)
– cause (terminal, verb, transformed causescause)
– climate change (pair, noun phrase)
• Resulting NTuple
– {CO2, cause, climate change}
• Where the names are replaced with topic locators from the topic map
These WordGram
instances represent the
sentence; they are wired
into the fabric.
This Ntuple participates
in high-level structure
formation and in
question answering
WordGram instances
created while processing
the sentence
WordGram instances
created while processing
the sentence
WordGram instances
created while processing
the sentence
15. Lenses
• Simple Interpreters
– Based on Canonical Predicates
– Build structures from parsed sentences and
WordGrams
– Examples from biology
• Cause
• Bind
• Augment
• Prevent
• Increase
• Decrease
• Believe
16. Multiple Lenses
• Consider this sentence:
– We believe that A causes B
– Two Lenses in play
• Believe
• Cause
– Result is a nested NTuple
• {We, believe, {A, cause, B}}
17. Canonical Predicate
• Results from transformations on predicates
– E.g.
• A causes B, A can cause B, A will cause B A cause B
• A is caused by B B cause A
18. Actors: Named Entities
• For any given named entity, there will be one and
only one WordGram
– Issue of Ambiguity
• Same name string can serve different topics in the topic map
– Topic map maintains identity for disambiguation
• Thus, a single WordGram might be associated with more
than one individual actor
• This means:
– Fibers (threads) flowing through the fabric must be
maintained in bundles according to their context
(topic)
19. Lens Selection and Action
• The Lens:
– ProtectAgainst
• Selected by the WordGram for “protect against”
– Is a transformation of the WordGram for “protective role
against”
• Lens Action:
– Create an NTuple
• {C282Y, protect against, CMPD}
• We will call that NTuple an Assertion
We found that HFE C282Y might be associated with a protective role against CMPD
20. Weaving an Information Fabric
• Background:
– One and only one
WordGram for each
Actor (named entity)
– One and only one
WordGram for each
canonical Predicate
– One and only one
NTuple for each
Assertion
• WordGrams which form
an NTuple are strung
together as beads on a
string in the fabric.
– Thus, it is the detection
of NTuple structures
(Assertions) which form
the HyperMembrane’s
fabric.
Note: it is next to impossible to diagram the fabric, but it
will likely look like a very tangled knotted structure. https://www.flickr.com/photos/fermicat/27
3539481/in/set-72157601620157588/
21. Fabric Example
• Two NTuples
– {Jack Park, AuthoredBook, The Wind Power Book}
– {Jack Park, AuthoredBook, Ohio State University
Football Vault}
JP101 JP102
Book101
AuthoredBook
Wind Power Book
OSU Football…
Book102
Jack Park
Topic Map organizes fiber bundles
22. Looking Forward
• Lenses, today, are hardwired
– Opportunity for adaptive learning of new lenses
• Fabric, today, is simple
– Opportunity to use cardinalities, frequency counts
in the fabric for:
• Probabilistic modeling
• Topological studies
• Opportunity for a Domain-Specific Language
(DSL) to emerge
23. Completed Representation
antioxidants
kill
free radicals
Contraindicates
macrophages use
free radicals to
kill bacteria
Bacterial Infection Antioxidants
Because
Appropriate For
Compromised Host
Let us co-create Cognitive Agents for Discovery
jackpark@topicquests.org
Thanks to Mei Lin Fung, David Alexander Price, and Patrick Durusau for
valuable comments
SolrSherlock at:
http://debategraph.org/SolrSherlock and https://github.com/SolrSherlock