The document discusses a project named 'Big Mechanism' aimed at enhancing cancer biology understanding through deep reading and text mining. It highlights the challenges of manual annotation in literature mining, the integration of uncertainty measures into event extraction, and introduces tools like LitPathExplorer for visualizing and validating biological pathways. The initiative focuses on increasing the efficiency and accuracy of research by enabling automated processes and improving access to scientific knowledge.

1. Big Mechanism: deep reading
for cancer biology
Sophia Ananiadou
Na-onal Centre for Text Mining
School of Computer Science
Manchester Ins-tute of Biotechnology
The University of Manchester

2. Overview
• From text to knowledge
– Events
– Big Mechanism
• Mining textual (un)certainty
• Visualising and ranking evidence (LitPathExplorer)
2

3. Pathway construc-on
3
mTOR pathway: 964 en--es, 777
reac-ons, 519 papers
Caron, et al. Mol Syst Biol., 6(1)
A MANUAL PROCESS
Inevitable gaps building
models

4. Knowledge
4
• Key to understanding biological systems
• Models need verifica-on and maintenance
(i.e., annota-on/cura-on)
• Scale and speed of literature challenging
• Annota-on/cura-on remains largely a
manual task of incorpora-ng knowledge
from scien-fic publica-ons
Pathways

5. Mo-va-on
5
To support pathway construc-on and design of
experiments
• Extrac-ng evidence from literature
• Events, en--es, contextual interpreta-on

6. 6
The Big Mechanism: reading, assembly, experiments
hZp://nactem.ac.uk/big_mechanism/

7. From concepts to events
1 Concept recogni-on
2 Interac-on recogni-on
3 Concept and interac-on iden-fica-on
DrugBank:DB06712 DrugBank:DB00682 DrugBank:DB04610

8. EventMine
• Machine learning pipeline event extrac-on system
– Rich linguis-c features
• Several parse results: deep parser (Enju), dependency parser
• Dic-onaries
• Coreference resolu-on, domain adapta-on, filtering
8
hZp://www.nactem.ac.uk/EventMine/
Miwa, M. & Ananiadou, S. (2015) Adaptable, high recall, event extrac-on system with minimal
configura-on, BMC Bioinforma,cs, 16(10), S7
Miwa, M., Thompson, P. and Ananiadou, S. (2012) Boos+ng automa+c event extrac+on from
the literature using domain adapta+on and coreference resolu+on. Bioinforma,cs, 28(13)
Miwa, M., Pyysalo, S., Ohta, T. and Ananiadou, S. (2013). Wide coverage biomedical event
extrac+on using mul+ple par+ally overlapping corpora. BMC Bioinforma<cs, 14(175)

9. Event interpreta-on
for
Binding Protein
in binding to MUC1
Theme 1
RAS suggest Results
Event trigger
PKM2
Protein
Theme 2
Event
argument
En-ty
argument
Chemical
is
Regula<on
BRAF required
Cause
not that
SIMPLE EVENT
COMPLEX EVENT
Theme
*Complex events have at
lest one argument that is
an event on its own
Event trigger

10. Event interpreta-on
• Supports users of search systems
– Discovery of new knowledge, research hypotheses
– Detec-on of uncertainty as confidence measure
• Mul-ple dimensions (meta-knowledge)
– Knowledge Type (observa-on, inves-ga-on, analysis,
method, fact)
– Knowledge Source (current, other)
– Polarity (posi-ve, negated)
10
Thompson, P., Nawaz, R., McNaught, J. and Ananiadou, S. 2011. Enriching a biomedical
event corpus with meta-knowledge annota+on. BMC Bioinforma<cs 12, 393

11. (Un)certainty: a measure of confidence
• Is this a fact, a hypothesis, a speculated outcome, a
case under inves-ga-on, a certain or uncertain
interac-on?
• How this informa+on can help pathway construc+on?
11
Zerva, C., Ba-sta-Navarro, R., Day, P. and S. Ananiadou (2017) Using uncertainty to link and
rank evidence from biomedical literature for model reconstruc-on, Bioinforma+cs

12. Uncertainty
Examples from Big Mechanism data
These results indicate that FLCN can interact directly with RagA via its GTPase domain.
Altogether, these results show that cobalt could affect both p53 and HIPK2 ac-vity.
To test if endogenous hPGAM5 interacts with hPINK1, we first generated an an--hPGAM5
an-body
We hypothesize that unphosphorylated cdr2 interacts with c-myc to prevent c-myc
degrada-on
Therefore, AFP may interact with STAT3 in the signal pathway for chemotherapeu-c efficiency
of agents on AFPGC.
These data suggest that PI3K and βARK1 form a macromolecular complex within the cell.
Therefore, LiCl might inhibit GSK3β in different ways
We then examined whether netrin-2 enhances the interac-on between Cdo and S-m1.

13. Uncertainty cues
13
BioNLP-ST, GENIA-MK

14. Hybrid model: Machine learning + Rules
Hybrid model
Machine Learner (Random Forest)
1. Lexical (e.g. cues, POS tags, event-trigger surface form)
2. Syntac-c (e.g. shortest path, dependency cue-trigger)
3. Seman-c (e.g. event type, argument type/role)
Automated Rule Induc<on (from corpus)
1. EventMine (to iden-fy event triggers)
2. Enju (to iden-fy dependencies)
3. Cue lists
BioNLP-ST
GENIA-MK

15. Dependency rela-ons
• Dependency rela-ons between cues and event triggers
• Rule induc-on: generic rule paZerns capturing
dependency rela-ons between cues and trigger
words

16. Rule induc-on
16

17. Dealing with mul-ple event men-ons

18. Results
• Event-annotated corpora:
All results obtained using 10-fold cross valida-on

19. Evalua-on - pathway models
• B-cell acute lymphoblas-c leukemia model
(Pathway studio)
– 72 interac-ons, 260 evidence passages manually
selected
– 12% flagged uncertain by our system

20. Results
Leukemia Pathway (7 annotators) ~ Pathway Studio
• Average accuracy on sentence level: 0.96
• Average accuracy on interac-on level: 0.87
– 1-20 sentences per interac+on

21. Event interpreta-on
• Uncertainty scoring as an expressive confidence
measure
• Hybrid framework
• Value for each event men-oned in a sentence
– Consolidated uncertainty values from different papers
• Effort to decrease manual effort and select more
certain events
21

22. Deep Reading: Integra-ng uncertainty
• LitPathExplorer
– Visual analy-cs tool; maps events from literature to pathway
interac-ons
– Includes uncertainty measure
• Robot Scien-st
– Selec-on of Gene expression and Regula+on events for wet-
lab experiments
– Selec-on of interac-ons (using LitPathExplorer) to assemble
network and predict drug effect on cell-lines

23. LitPathExplorer: a confidence-based
tool for exploring pathway models
1. Enabling flexible search and explora-on of
biomolecular pathway networks
– different views of the data
– various interac-ve func-onali-es
2. Provide a means for making exis-ng evidence in the
scien-fic literature available to support corrobora-on
3. Facilitate the discovery of new interac-ons that are
not yet part of a given model
4. Allow the user to become an ac-ve par-cipant of the
analy-cal process
quan-fy confidence
in the events
23
Video: hZp://nactem.ac.uk/LitPathExplorer_BI/LitPathExplorer.mp4

24. 1. Search
• A pathway model can be searched by
providing:
• event types,
• en--es,
• and/or roles for each en-ty in the
reac-on
• Mul-ple queries can be combined in
a Boolean search
24

25. 2. Network viewer
Reading against the model
25
En--es
Reac-ons/
Events
• Colour encodes event type
• Size encodes confidence

26. 3. Inspector, event confidence
computa-on
26
Mapping IDs for
en--es and events
Overall event
confidence

27. 3. Inspector, quan-fying the confidence
27
Confidence
breakdown

28. Adjus-ng event confidence
28

29. 4. Text Analyzer – Ar-cles & sentences
29
Sentence-level language
confidence
Ar-cle-level language
confidence

30. 4. Word tree visualisa-on: Contrast
event men-ons across the corpus
30
Sentences can be inspected further
upon interac-on
Ver-cal arrangement and gray scale
denotes event confidence

31. Use case
• A pathway model
• contains reac-ons involving the Ras protein
• output of querying PathwayCommons for one- and two-hop
reac-ons centred on Ras
• A corpus of 12,660 full papers
• Retrieved from the PubMed Central Open Access repository
• using as queries “breast cancer” and its synonyms as
keywords, combined with names of breast cancer cell lines,
e.g., “T-47D”, “MCF-7” (and their variants).
• Methods for event extrac-on, model mapping and
confidence computa-on were applied on the events
extracted from the corpus
31

32. Network Viewer: Discovery mode
Extending the model with events found in the
literature
32

33. 33
Discovery mode
Difficult to explore when too many candidate events are found

34. Verifying men-ons in text
34
hZp://nactem.ac.uk/LitPathExplorer/

35. Na#onal Centre for Text Mining
• 1st publicly funded na-onal text
mining centre
• Loca-on: Manchester Ins-tute
of Biotechnology
• Since 2004-
• Fully sustainable 2011-
• Biology, Medicine, Biodiversity,
Humani-es, Social Sciences
www.nactem.ac.uk
BBSRC, AHRC, EPSRC, MRC, JISC, NIH, DARPA, H2020
AZ, Unilever, Pfizer, Elsevier, Nature, BBC, KISTI, AIST