Attribute Exploration with Proper Premises and Incomplete Knowledge Applied to the Free Radical
Theory of Ageing
Johannes Wollbold, R¨udiger K¨ohling and Daniel Borchmann
Systems Biology and Bioinformatics, University of Rostock [email protected]
Oscar Langendorff Institute of Physiology, University Medicine, Rostock, Germany Institute of Theoretical Computer Science, Technische Universitt Dresden, Germany
ICFCA 2014, 10 - 13 June 2014
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 1 / 25
Outline
1 The free radical theory of ageing
2 Assembling a knowledge base of ripple down rules
3 Validation and completion by attribute exploration
4 Attribute exploration with proper premises and incomplete knowledge
5 Outlook
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 2 / 25
Free radicals
Free radicals have unpaired electrons.
One subclass of reactive oxygen species (ROS), highly oxidative small molecules capable of damaging organic molecules.
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 3 / 25
ROS generation in the mitochondrial respiratory chain
Figure: [Tim Vickers. In: en.wikipedia.org, Electron transport chain.]
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 4 / 25
Mitochondria get damaged with age.
Figure: Liver mitochondria from young and old rats. [Jose Vina. Antioxidant &
Redox Signaling, 19 (8), 2013, Figure 2]
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 5 / 25
Vicious cycle of ROS generation and molecular damage
[Aubrey de Grey. The Mitochondrial Free Radical Theory of Aging, 1999, Fig. 6.1]
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 6 / 25
Outline
1 The free radical theory of ageing
2 Assembling a knowledge base of ripple down rules
3 Validation and completion by attribute exploration
4 Attribute exploration with proper premises and incomplete knowledge
5 Outlook
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 7 / 25
The workflow
A collection of Ripple Down Rules (RDR) and cornerstone cases is converted to a complete knowledge base of implications.
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 8 / 25
Rules from literature
Knowledge1 is collected in a tree of general and exceptional rules (Ripple Down Rules, RDR).
Observed cases, defined by attributes m∈M\C, are classified by classes C ⊆ {ROS.old.+,ROS.old.−,Lifespan.+,Lifespan.−} ⊆M.
Iterative process of knowledge base construction: 1. >→ ROS.old.+, Lifespan.–
1.1. AntiOx1.+→ROS.old.–, Lifespan.+
1.1.1. AntiOx1.+, AntiOx2.–→ROS.old.+
1.3. AntiOx2.–, Mouse→ROS.old.+ 1.3. AntiOx2.–, CElegans→ ROS.old.+
Background knowledge for later exploration: AntiOx2.– →ROS.old.+
1Kirkwood, TBL and Kowald, A. The free-radical theory of ageing – older, wiser and still alive: Modelling positional effects of the primary targets of ROS reveals new support. BioEssays 2012, p. 1f.
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 9 / 25
Rules from literature
Knowledge1 is collected in a tree of general and exceptional rules (Ripple Down Rules, RDR).
Observed cases, defined by attributes m∈M\C, are classified by classes C ⊆ {ROS.old.+,ROS.old.−,Lifespan.+,Lifespan.−} ⊆M.
Iterative process of knowledge base construction:
1. >→ ROS.old.+, Lifespan.–
1.1. AntiOx1.+→ROS.old.–, Lifespan.+
1.1.1. AntiOx1.+, AntiOx2.–→ROS.old.+
1.3. AntiOx2.–, Mouse→ROS.old.+ 1.3. AntiOx2.–, CElegans→ ROS.old.+
Background knowledge for later exploration: AntiOx2.– →ROS.old.+
1Kirkwood, TBL and Kowald, A. The free-radical theory of ageing – older, wiser and still alive: Modelling positional effects of the primary targets of ROS reveals new support. BioEssays 2012, p. 1f.
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 9 / 25
Rules from literature
Knowledge1 is collected in a tree of general and exceptional rules (Ripple Down Rules, RDR).
Observed cases, defined by attributes m∈M\C, are classified by classes C ⊆ {ROS.old.+,ROS.old.−,Lifespan.+,Lifespan.−} ⊆M.
Iterative process of knowledge base construction:
1. >→ ROS.old.+, Lifespan.–
1.1. AntiOx1.+→ROS.old.–, Lifespan.+
1.1.1. AntiOx1.+, AntiOx2.–→ROS.old.+
1.3. AntiOx2.–, Mouse→ROS.old.+ 1.3. AntiOx2.–, CElegans→ ROS.old.+
Background knowledge for later exploration: AntiOx2.– →ROS.old.+
1Kirkwood, TBL and Kowald, A. The free-radical theory of ageing – older, wiser and still alive: Modelling positional effects of the primary targets of ROS reveals new support. BioEssays 2012, p. 1f.
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 9 / 25
Rules from literature
Knowledge1 is collected in a tree of general and exceptional rules (Ripple Down Rules, RDR).
Observed cases, defined by attributes m∈M\C, are classified by classes C ⊆ {ROS.old.+,ROS.old.−,Lifespan.+,Lifespan.−} ⊆M.
Iterative process of knowledge base construction:
1. >→ ROS.old.+, Lifespan.–
1.1. AntiOx1.+→ROS.old.–, Lifespan.+
1.1.1. AntiOx1.+, AntiOx2.–→ROS.old.+ 1.3. AntiOx2.–, Mouse→ROS.old.+ 1.3. AntiOx2.–, CElegans→ ROS.old.+
Background knowledge for later exploration: AntiOx2.– →ROS.old.+
1Kirkwood, TBL and Kowald, A. The free-radical theory of ageing – older, wiser and still alive: Modelling positional effects of the primary targets of ROS reveals new support. BioEssays 2012, p. 1f.
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 9 / 25
Rules from literature
Knowledge1 is collected in a tree of general and exceptional rules (Ripple Down Rules, RDR).
Observed cases, defined by attributes m∈M\C, are classified by classes C ⊆ {ROS.old.+,ROS.old.−,Lifespan.+,Lifespan.−} ⊆M.
Iterative process of knowledge base construction:
1. >→ ROS.old.+, Lifespan.–
1.1. AntiOx1.+→ROS.old.–, Lifespan.+
1.1.1. AntiOx1.+, AntiOx2.–→ROS.old.+
1.3. AntiOx2.–, Mouse→ROS.old.+ 1.3. AntiOx2.–, CElegans→ ROS.old.+
Background knowledge for later exploration: AntiOx2.– →ROS.old.+
1Kirkwood, TBL and Kowald, A. The free-radical theory of ageing – older, wiser and still alive: Modelling positional effects of the primary targets of ROS reveals new support. BioEssays 2012, p. 1f.
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 9 / 25
Rules from literature
Knowledge1 is collected in a tree of general and exceptional rules (Ripple Down Rules, RDR).
Observed cases, defined by attributes m∈M\C, are classified by classes C ⊆ {ROS.old.+,ROS.old.−,Lifespan.+,Lifespan.−} ⊆M.
Iterative process of knowledge base construction:
1. >→ ROS.old.+, Lifespan.–
1.1. AntiOx1.+→ROS.old.–, Lifespan.+
1.1.1. AntiOx1.+, AntiOx2.–→ROS.old.+
1.3. AntiOx2.–, Mouse→ROS.old.+
1.3. AntiOx2.–, CElegans→ ROS.old.+
Background knowledge for later exploration: AntiOx2.– →ROS.old.+
1Kirkwood, TBL and Kowald, A. The free-radical theory of ageing – older, wiser and still alive: Modelling positional effects of the primary targets of ROS reveals new support. BioEssays 2012, p. 1f.
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 9 / 25
Rules from literature
Knowledge1 is collected in a tree of general and exceptional rules (Ripple Down Rules, RDR).
Observed cases, defined by attributes m∈M\C, are classified by classes C ⊆ {ROS.old.+,ROS.old.−,Lifespan.+,Lifespan.−} ⊆M.
Iterative process of knowledge base construction:
1. >→ ROS.old.+, Lifespan.–
1.1. AntiOx1.+→ROS.old.–, Lifespan.+
1.1.1. AntiOx1.+, AntiOx2.–→ROS.old.+
1.3. AntiOx2.–, Mouse→ROS.old.+
1.3. AntiOx2.–, CElegans→ ROS.old.+
Background knowledge for later exploration:
AntiOx2.– →ROS.old.+
1Kirkwood, TBL and Kowald, A. The free-radical theory of ageing – older, wiser and still alive: Modelling positional effects of the primary targets of ROS reveals new support. BioEssays 2012, p. 1f.
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 9 / 25
Mutations of respiratory chain molecules are studied in the ROSAge project
Figure: [Tim Vickers. In: en.wikipedia.org, Electron transport chain.]
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 10 / 25
A rule derived from ROSAge data
Figure: Basal ROS measurements by Dept. of Hematology Rostock.
Mouse, Mut-ETC,→∅
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 11 / 25
A conflicting rule derived from ROSAge data
Figure: Basal ROS measurements by Institute of Physiology Rostock for the mouse strain NOD (mutation in complex IV of the respiratory chain). A significant increase of free radicals in 24 month old mice is measured, compared to 3 month.
Mouse, Mut-ETC →ROS.old.+
⇒ Mut-ETC, Mouse →∅ not accepted as background knowledge.
Both cases are stored in the formal context of cornerstone cases.
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 12 / 25
The tree of rules and exceptions
1. >→ ROS.old.+, Lifespan.–
1.1. AntiOx1.+→ROS.old.–, Lifespan.+
1.1.1. AntiOx1.+, AntiOx2.–→ROS.old.+
1.2. Mouse, AntiOx2.–→ROS.old.+
1.3. CElegans, AntiOx2.–→ ROS.old.+
1.4. Mouse, Mut-ETC→ ⊥
2. OxStress →... 2.1. ...
3. ATP.old.–→... 3.1. ...
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 13 / 25
The tree of rules and exceptions
1. >→ ROS.old.+, Lifespan.–
1.1. AntiOx1.+→ROS.old.–, Lifespan.+
1.1.1. AntiOx1.+, AntiOx2.–→ROS.old.+
1.2. Mouse, AntiOx2.–→ROS.old.+
1.3. CElegans, AntiOx2.–→ ROS.old.+
1.4. Mouse, Mut-ETC→ ⊥ 2. OxStress →...
2.1. ...
3. ATP.old.–→... 3.1. ...
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 13 / 25
The tree of rules and exceptions
1. >→ ROS.old.+, Lifespan.–
1.1. AntiOx1.+→ROS.old.–, Lifespan.+
1.1.1. AntiOx1.+, AntiOx2.–→ROS.old.+
1.2. Mouse, AntiOx2.–→ROS.old.+
1.3. CElegans, AntiOx2.–→ ROS.old.+
1.4. Mouse, Mut-ETC→ ⊥ 2. OxStress →...
2.1. ...
3. ATP.old.–→...
3.1. ...
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 13 / 25
The incomplete formal context of cornerstone cases
AntiOx1.+ AntiOx1.− AntiOx2.+ AntiOx2.− CElegans Mouse Mut-ETC ROS.old.+ ROS.old.− Lifespan.+ Lifespan.−
1. ? ? ? ? × × ? × ×
1.1. × ? ? × × × ×
1.1.1. × × × × × ? ?
1.2.-1.3. ? ? × × × × ? ?
1.4a ? ? ? ? ? × × ? ?
1.4b ? ? ? ? ? × × × ? ?
Table: Examples (cornerstone cases) related to the RDR knowledge base: Certain context K+ and possible contextK?.
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 14 / 25
Outline
1 The free radical theory of ageing
2 Assembling a knowledge base of ripple down rules
3 Validation and completion by attribute exploration
4 Attribute exploration with proper premises and incomplete knowledge
5 Outlook
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 15 / 25
Confirmed rules (some logical juggling)
AntiOx1.– →ROS.old.+
AntiOx1.–, AntiOx2.+ →⊥(background) implies that AntiOx2.+ does not hold.
As RDR, we had already the stronger rule AntiOx1.+, AntiOx2.– → ROS.old.+.
AntiOx2.+ →ROS.old.–
AntiOx1.–, AntiOx2.+ →⊥(background) implies that AntiOx1.– does not hold.
Hence, AntiOx2.+, AntiOx1.– →ROS.old.+ (parallel to RDR 1.1.1) is the only exception.
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 16 / 25
Confirmed rules (some logical juggling)
AntiOx1.– →ROS.old.+
AntiOx1.–, AntiOx2.+ →⊥(background) implies that AntiOx2.+ does not hold.
As RDR, we had already the stronger rule AntiOx1.+, AntiOx2.– → ROS.old.+.
AntiOx2.+ →ROS.old.–
AntiOx1.–, AntiOx2.+ →⊥(background) implies that AntiOx1.– does not hold.
Hence, AntiOx2.+, AntiOx1.– →ROS.old.+ (parallel to RDR 1.1.1) is the only exception.
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 16 / 25
Confirmed rules
AntiOx1.+, AntiOx2.+, CElegans →Lifespan.+
The strong conclusion can be assumed for the short living worm C.
elegans.2
AntiOx1.–, Mouse, Mut-ETC →Lifespan.–
Mutations (deletions) of the mitochondrial DNA can cause lifespan reducing damage for long-lived animals like mice.3
3Sohal R., Orr, W.Free Radical Biology and Medicine52(3), 539-555 (2012)
3Kirkwood TBL, Kowald A.BioEssays 2012, p. 6
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 17 / 25
Confirmed rules
AntiOx1.+, AntiOx2.+, CElegans →Lifespan.+
The strong conclusion can be assumed for the short living worm C.
elegans.2
AntiOx1.–, Mouse, Mut-ETC →Lifespan.–
Mutations (deletions) of the mitochondrial DNA can cause lifespan reducing damage for long-lived animals like mice.3
3Sohal R., Orr, W.Free Radical Biology and Medicine52(3), 539-555 (2012)
3Kirkwood TBL, Kowald A.BioEssays 2012, p. 6
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 17 / 25
Counterexamples
Mouse, CElegans, Mut-ETC, AntiOx1.+ →ROS.old.–
Counterexample:
AntiOx2.+ →ROS.old.– was accepted, but here AntiOx2.– is possible.
AntiOx2.+ →Lifespan.+
ROS are reduced, but this is not sufficient to extend lifespan. Counterexample:
(AntiOx1.+,) AntiOx2.+, (Mouse,) ROS.old.–, NOT Lifespan.+
3Fitzenberger E, Boll M, Wenzel U. Impairment of the proteasome is crucial for glucose-induced lifespan reduction in the mev-1 mutant of Caenorhabditis elegans. Biochim Biophys Acta, 2013 Apr, 1832(4), p. 565-73.
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 18 / 25
Counterexamples
Mouse, CElegans, Mut-ETC, AntiOx1.+ →ROS.old.–
Counterexample:
AntiOx2.+ →ROS.old.– was accepted, but here AntiOx2.– is possible.
AntiOx2.+ →Lifespan.+
ROS are reduced, but this is not sufficient to extend lifespan.
Counterexample:
(AntiOx1.+,) AntiOx2.+, (Mouse,) ROS.old.–, NOT Lifespan.+
3Fitzenberger E, Boll M, Wenzel U. Impairment of the proteasome is crucial for glucose-induced lifespan reduction in the mev-1 mutant of Caenorhabditis elegans. Biochim Biophys Acta, 2013 Apr, 1832(4), p. 565-73.
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 18 / 25
Outline
1 The free radical theory of ageing
2 Assembling a knowledge base of ripple down rules
3 Validation and completion by attribute exploration
4 Attribute exploration with proper premises and incomplete knowledge
5 Outlook
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 19 / 25
Implications with proper premises
Definition
For a given formal context (G,M,I) and a set of attributesP ⊆M, define P• to be the set of those attributes inM\P that follow fromP but not from a strict subset of P, i.e.
P• =P00\ P ∪ [
S(P
S00
P is called aproper premise ifP• is not empty. It is called a proper premise for m ifm∈P•.
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 20 / 25
Exploration with incomplete counterexamples
PP-Implications are suited for disjoint basic sets for the premises and conclusions:
Decision of an implicationP →C by closure under all implications of the base: C ⊆ L(P)?
PP base is iteration free: Closure is reached in one step.
For disjoint implications, no iteration is possible.
⇒ Standard algorithm for the base construction of the whole context can be used, with iteration through the interesting m∈C (M. Implications have to be valid for any realizer of K+ andK?.
Proposition (Proposition 30 from Ganter/Obiedkov 20134) A set P ⊆M possibly entails m∈M if and only if m∈P+?.
4Ganter, B., Obiedkov, S.: Conceptual Exploration. Preprint, Dresden (2013)
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 21 / 25
A new algorithm
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 22 / 25
Outline
1 The free radical theory of ageing
2 Assembling a knowledge base of ripple down rules
3 Validation and completion by attribute exploration
4 Attribute exploration with proper premises and incomplete knowledge
5 Outlook
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 23 / 25
Result and open questions
The new method gives a structured overview on facts and open questions of the free radical theory of ageing.
”True“ exploration of RDR:
^
i∈I
αi ∧^
j∈J
¬αj →β
Easier thanrule exploration of general clauses? Two contexts with positive and negated attributes? Integration of insecure data - Fuzzy FCA?
Larger, biologically more relevant application.
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 24 / 25
Result and open questions
The new method gives a structured overview on facts and open questions of the free radical theory of ageing.
”True“ exploration of RDR:
^
i∈I
αi ∧^
j∈J
¬αj →β
Easier thanrule exploration of general clauses?
Two contexts with positive and negated attributes?
Integration of insecure data - Fuzzy FCA? Larger, biologically more relevant application.
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 24 / 25
Result and open questions
The new method gives a structured overview on facts and open questions of the free radical theory of ageing.
”True“ exploration of RDR:
^
i∈I
αi ∧^
j∈J
¬αj →β
Easier thanrule exploration of general clauses?
Two contexts with positive and negated attributes?
Integration of insecure data - Fuzzy FCA?
Larger, biologically more relevant application.
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 24 / 25
Result and open questions
The new method gives a structured overview on facts and open questions of the free radical theory of ageing.
”True“ exploration of RDR:
^
i∈I
αi ∧^
j∈J
¬αj →β
Easier thanrule exploration of general clauses?
Two contexts with positive and negated attributes?
Integration of insecure data - Fuzzy FCA?
Larger, biologically more relevant application.
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 24 / 25
Acknowledgements
Christin Kretzschmar and Catrin Roolf Dept. of Hematology, Rostock
Gesine Reichart and Johannes Mayer Dept. of Phyisology, Rostock
Britta Burkhardt
Siegfried Weller Institut, Eberhard-Karls-Universit¨at T¨ubingen
Johannes Wollbold (SBI Rostock) PP-Exploration of the FRTA June 2014 25 / 25