KR-MED 2006 "Biomedical Ontology in Action" November 8, 2006, Baltimore, Maryland, USA “Lmo-2 interacts with Elf-2 ” On the Meaning of Common Statements in Biomedical Literature Stefan Schulz1 , Ludger Jansen2 1 Department of Medical Informatics, Freiburg University Hospital, Germany 2 Department of Philosophy, University of Rostock, Germany Abstract has been repeatedly advocated in order to facili- Statements about the behavior of biological enti- tate the process of automatic processing of domain ties, e.g. about the interaction between two pro- information. teins, abound in the literature on molecular biology Much work in this area has already been done in and are increasingly becoming the targets of infor- the form of ontological investigations on mater- mation extraction and text mining techniques. We ial continuants, such as organs, cells, molecules show that an accurate analysis of the semantics of [1, 2, 3]. However, there has been much less such statements reveals a number of ambiguities emphasis on biologically relevant functions and that is necessary to take into account in the prac- processes. Furthermore, biomedical ontology en- tice of biomedical ontology engineering. Several gineering has been mainly committed to traditions concurring formalizations are proposed. Emphasis of semantic networks, lexical semantics, and cogni- is laid on the discussion of biological dispositions. tive science. Thus rather than being construed as describing real world entities by means of logical Introduction expressions, ontology has been understood as re- The study of so-called protein-protein interactions lating concepts (i.e. representations of word mean- is essential for a better understanding of biological ings) by means of conceptual relations. On this processes, from replication and expression of genes assumption, “Lmo-2 interacts with Elf-2 ” would to the morphogenesis of organisms. Statements simply signify that there is some plausible link- such as “Lmo-2 interacts with Elf-2 ” – with Lmo- age between the concepts (conceived of as men- 2 and Elf-2 being proteins – occur in biomedical tal representations) “Interaction”, “Lmo-2”, and literature abstracts with a very high frequency and “Elf-2”. As much as this approach might be ad- represent, in many cases, the core message of a equate for communicating knowledge about the scientific paper. world by means of natural language or some kind There are several kinds of biomolecular interac- of abstraction (e.g. semantic networks), it fails tions, e.g. binding, inhibition, activation, and where exact statements and reasoning about bi- transport. They all involve (1) at least two bio- ological entities such as molecules, functions, or molecules and (2) the spatial vicinity of these, pathways are required. which leads to (3) a causal influence that they ex- Interestingly enough, scientists and other human ert on each other. agents are perfectly able to communicate by means Text mining, i.e the process of extracting struc- of such sentences, although there is only a vague tured knowledge from unstructured text, primar- consensus about the referents (the entities in the ily targets statements such as these, and there is world) which are denoted by these linguistic ex- a major interest by the text mining community in pressions. Because of the ambiguities of natural obtaining ontological support for their information language, a natural language statement like “Lmo- and knowledge extraction activities. This is one 2 interacts with Elf-2 ” may have more than one of the reasons why so-called bio-ontologies have possible interpretation and thus more than one emerged, and the use of formal ontological criteria formalization in, say, first order predicate logic. In 37 formally representing the meaning of such state- dards for ontologies, especially in the light of the ments, we have thus to make explicit the onto- Semantic Web and the various specifications of De- logical assumptions intended by the speakers or scription Logic (DL) [7], we keep our logic simple. authors of that sentence. So we refrain from higher-order logics, as well as In this paper we will demonstrate that even the temporal or modal logics. We also use a parsimo- formalization of an apparently simple but proto- nious set of relations, following the OBO (Open typical statement about protein interaction like Biological Ontologies) recommendation [8]. As a “Lmo-2 interacts with Elf-2 ” can yield totally dif- primitive formal relation we introduce the irreflex- ferent ontological assumptions. ive, non-transitive and asymmetric instantiation relation inst which relates particular entities to Basic Ontological Assumptions their universal properties. In addition, we need It is widely recognized that the construction of bio- a formal relation for class subsumption between medical ontologies should obey strict logical and universals, expressing scientific findings about re- ontological criteria. To this end, several top-level lations between the kinds of things that are in the ontologies have been devised, such as DOLCE [4], world. As scientific laws are meant to range not BFO [5], and GOL [6]. These ontologies mainly only over all present instances of a given kind, coincide in their fundamental division between but also over all past and future instances and, continuants (endurants, e.g. material objects) and moreover, also over merely possible instances [9], occurrents (perdurants, e.g. events, processes). such a relation is not easily defined. We will here The distinction is that occurrents have temporal follow the OBO standard and introduce, to this parts (they are never fully present at a given time) end, the taxonomic subsumption relation Is-a by and they are existentially dependent on continu- means of the inst 1 relation [8]. We will neglect ants. Continuants are split into independent and the time parameter, which is not important for dependent ones. Examples of independent contin- present purposes. On this basis, we define Is-a as uants are material objects and spaces. Dependent a reflexive, transitive, and antisymmetric relation continuants, on the other hand, are entities which between universals A and B, as follows: inhere in something and are thus ontologically de- pendent on their bearer. Examples of dependent Is-a(A, B) =def (1) continuants are masses, colors, and tendencies: A ∀x : (inst(x, A) → inst(x, B)) particular mass may inhere in a particular mole- cule, a particular color may inhere in a particular Furthermore, we make the following ontological flower. The tendency to divide may inhere in a cell subdivision: When we deal with things of a certain and the tendency to relieve headache may inhere kind, we have to distinguish between individuals in an aspirin tablet. Tendencies are related to oc- belonging to this kind and collectives of individ- currents by the relation of realization. They are uals that belong to the same kind [1]. This very special kinds of dependent entities, in that they natural ontological distinction, which is mirrored need not be realized in order to exist. There are by the singular / plural division in most natural cells which never divide, and aspirin tablets that languages, must be addressed wherever collectives never relieve a headache. or pluralities of individual objects occur. However, Our ontological framework for describing molec- this distinction is often obscured when referring ular interaction patterns includes entities of all to mass entities (e.g. water vs. water molecules). these kinds. For instance, a protein molecule, Given the atomicity of material continuants, we which is a material continuant, has a disposition do not admit material mass entities in our present to perform a certain function, e.g. binding, which framework, but consider them as collectives of par- is a dependent continuant, and an actual realiza- ticles instead. tion of this disposition, viz. the process of binding 1 We use capitalized initial letters for the names of a protein molecule, which is an occurrent. relations between universals as well as for the names Aware of the need to comply with existing stan- of universals. 38 A collective is given, e.g., by all Lmo-2 molecules assumption.) All the possible readings of “Lmo-2 involved in an experiment, as opposed to exactly interacts with Elf-2 ” to be discussed in the re- one individual Lmo-2 molecule. In the follow- mainder of this paper have thus the following as a ing, we will use the subscript ”coll” to refer to common ground: collectives. Thus, for each universal X we prici- pally admit the existence of a corresponding col- Is-a(Lmo-2, P roteinM olecule) ∧ (2) lective XCOLL , the class of collections of instances Is-a(Elf -2, P roteinM olecule) ∧ of X. For instance, “P roteinM oleculeCOLL” de- Is-a(P roteinM olecule, M olecule) ∧ notes the class of collectives of protein molecules as well as “Lmo-2 COLL ” the class of collectives Is-a(M olecule, Continuant) ∧ of Lmo-2 molecules. We also admit collectives of ∃l, e : inst(l, Lmo-2) ∧ inst(e, Elf -2) occurrents, such as Interaction COLL . Despite this common ground, the sentence remains Concurrent interpretations I: highly ambiguous even within a scientific context. Event Readings First we will discuss interpretations of “Lmo-2 in- teracts with Elf-2 ” that interpret it as a report of Let us come back to our example: “Lmo-2 inter- events. Here are some possible interpretations of acts with Elf-2 ”. Such statements are generally the sample statement that belong to this group: formulated by researchers who collect scientific ev- idence by empirical observations. These observa- 1. One individual Lmo-2 molecule interacts with tions are commonly made in an indirect way, since one individual Elf-2 molecule. the objects under scrutiny are below the threshold of visibility. For this reason measurement proce- 2. A collection of Lmo-2 molecules interacts with dures of varying degrees of sophistication are ap- one individual Elf-2 molecule. plied, the results of which can be used to draw con- 3. One individual Lmo-2 molecule interacts with clusions about the significance of an experiment. a collection of Elf-2 molecules. These conclusions may vary in their degrees of cer- tainty. This certainty is affected by measurement 4. A collection of Lmo-2 molecules interacts with errors as well as by errors in the design of the ex- a collection of Elf-2 molecules. periment which then may lead to false conclusions. If a statement like “Lmo-2 interacts with Elf-2 ” Our sample statement appears to describe the fact is being uttered in a laboratory or written in a sci- that exactly one such interaction happened. Alter- entific paper or textbook, the minimal thing that natively, it can describe the fact that a multitude can be inferred is that there are molecules of type of such interactions (as described in 1-4) happens, Lmo-2 and Elf-2. By way of contrast, this in- which would be the normal thing in many bio- ference is not possible if such a sentence appears chemical contexts. This adds up to eight different in a science fiction novel. As we are interested interpretations. But any of these interpretations here in the scientific context only, we assume in is still ambiguous in a very important respect. what follows that there are universals Lmo-2 and With each of these interpretations, the speaker Elf-2 that are kinds of protein molecules. These may mean either that such interaction(s) did ac- types of protein molecules do, of course, belong to tually happen, or the speaker may mean that the the genus of protein molecules, which in turn are molecules in question have the disposition or the molecules, which are a kind of continuants. If one tendency to interact in such a way. This gives has an Aristotelian theory of universals, universals way to even more possible interpretations. Thus, only exist if they are instantiated; that is, the ex- “Lmo-2 interacts with Elf-2 ” turns out to be a istence of the universals Lmo-2 and Elf-2 implies highly ambiguous sentence. We will now discuss that there are individual molecules that instanti- the different possible interpretations of this sen- ate these universals. (Whoever has a different the- tence in turn and suggest methods for representing ory of universals may have to add this as a further them formally. 39 Occurrents involving individual Therefore, we will eliminate the interacts relation, continuants using the technique introduced by Davidson [10] to On the first interpretation, “Lmo-2 interacts with quantify over events. This means that we repre- Elf-2 ” describes the fact that an individual Lmo-2 sent the interaction process as an occurrent entity molecule interacts with an individual Elf-2 mole- in its own right rather than by the relation inter- cule. A standard way to render such a situation acts as in Formulae 3 or 4. This move is made formally would be the use of the existence quanti- possible through our admission of occurrent en- fier of first order predicate logic: tities, and it corresponds to common practice in biomedical ontologies. The relation between the ∃l, e : inst(l, Lmo-2)∧ (3) particular process and the participating particu- inst(e, Elf -2) ∧ interacts(l, e) lar continuants is then given by the relation has- participant [8]. The has-participant relation is a This formalization ensures that there is at least relation between a particular occurrent and a par- one individual Lmo-2 molecule which interacts ticular continuant, in this order. It is irreflexive, with at least one individual Elf-2 molecule at at asymmetric, and non-transitive. For nothing par- least one instant. This interpretation can now be ticipates in a continuant and no occurrent partici- modified in various ways. We could, e.g., add ex- pates in anything. Again, we dispense with a time clusivity postulates like in (4) that ensure that ex- index for sake of simplicity. Within a fully-fledged actly one individual molecule of each kind are in- implementation, a time index should be included, teracting with each other. Though such a solitary as an occurrent may have different participants at event might be rarely observed in experiments, different stages. there may be contexts where this is the intended ∃l, e, i : inst(l, Lmo-2) ∧ inst(e, Elf -2)∧ (5) meaning: inst(i, Interaction) ∧ ∃l, e : inst(l, Lmo-2) ∧ inst(e, Elf -2)∧ (4) has-participant(i, l) ∧ has-participant(i, e) interacts(l, e) ∧ This formalization makes it easier to represent oc- ∀l∗ , e∗ : (inst(l∗ , Lmo-2) ∧ inst(e∗ , Elf -2) ∧ currents with more than two participants, as with interacts(l∗ , e∗ )) → (l∗ = l ∧ e∗ = e) the representation in Formula 3, where we would have to deal with n-ary relations for n partici- Normally, however, this formalization will be pants. According to this formal representation, much too strong an interpretation of our sample “Lmo-2 interacts with Elf-2 ” is to be understood statement. For any statement of this form will be as stating that there is at least one interaction false, if at any other time another Lmo-2 molecule process, in which at least one protein molecule of interacts with an Elf-2 molecule – or if at the very the given kinds is involved. It does not exclude same time another Lmo-2 molecule interacts with that other molecules are involved in this very in- an Elf-2 molecule at any other place. Therefore, teraction process. If we want to secure that Lmo-2 we do not consider it as a useful interpretation of and Elf-2 are the only participants of the mole- our sample sentence. We will, however, refer back cular interaction, we have to employ exclusivity to this formula and the exclusivity clauses used in conditions similar to 4: it in the following discussion. ∃l, e, i : inst(l, Lmo-2) ∧ inst(e, Elf -2)∧ (6) In Formulae 3 and 4 we have expressed the interac- tion event by means of a binary relation interacts inst(i, Interaction) ∧ between individual continuants. This relation on has-participant(i, l) ∧ has-participant(i, e) ∧ the level of instances is irreflexive (nothing ever in- ∀x : (has-participant(i, x) → teracts with itself), symmetric, and non-transitive. inst(x, Lmo-2) ∨ inst(x, Elf -2)) The OBO (Open Biological Ontologies) relation ontologies, however, recommends to restrict our- If we want to keep the requirement of pairwise selves to a parsimonious array of basic relations. interaction, if have to add uniqueness conditions 40 in the fashion of Formula 4 for this purpose: exist in space nor in time. We do not use the has- part relation, because participants in interactions ∃l, e, i : inst(l, Lmo-2) ∧ inst(e, Elf -2)∧ (7) may have parts that do not themselves participate inst(i, Interaction) ∧ in the interaction. A Lmo-2 molecule, e.g., may has-participant(i, l) ∧ has-participant(i, e) ∧ participate in an interaction without every of its electrons being a participant in this interaction. ∀x : (has-participant(i, x) → Whereas has-part is transitive, has-grain is not. It inst(x, Lmo-2) ∨ inst(x, Elf -2)) ∧ is a irreflexive, asymmetric, and intransitive rela- ∀l∗ , e∗ : (inst(l∗ , Lmo-2) ∧ inst(e∗ , Elf -2) ∧ tion that holds between particular collectives and has-participant(i, l∗ ) ∧ has-participant(i, e∗)) individuals. We therefore modify our formalism substituting → (e∗ = e ∧ l∗ = l)) individuals by collectives: In contrast to Formula 4, such a formalization that quantifies over events is still much more realistic, ∃l, e, i : inst(l, Lmo-2COLL )∧ (9) because its truth is compatible with more than one inst(e, Elf -2COLL ) ∧ inst(i, Interaction) ∧ interaction process of the same kind happening at has-participant(i, l) ∧ has-participant(i, e) the same time or at other times. Collectives of occurrents Occurrents involving collectives of continuants Formalism 7 and 9 use the same occurrent type In- As mentioned above, it is important to distinguish teraction for different scenarios: In the first case, a between individuals of a kind and collectives of particular interaction has individual protein mole- individuals of that kind. Rector and Bittner [1] cules as participants, in the second case collectives have accounted for this by introducing the formal of molecules. This ambiguity may be acceptable relation has-grain which relates a collective c to when we talk about such a generic process as in- each of its constituents e. In [3] this account has teraction. It would not be tolerable in the case of a been further developed by introducing a collective more specific one, such as binding. A binding can universal X COLL whose instances are constituted only happen between two individual molecules, by two or more constituents which are instances not between two collectives of molecules. Thus, if of X: we encounter a plurality of bindings within a plu- rality of molecules, it would not be admissible to ∀c : inst(c, XCOLL) → ∃e1 , e2 , ..., en , n > 1 : (8) describe this as a binding between two collectives ^n of molecules but rather a collective of bindings be- inst(eν , X) ∧ has-grain(c, eν ) tween pairs instances of the kinds of molecules in ν=1 question2 . Thus we have to deal with a collective of processes rather than with collectives of contin- Note that has-grain is a subrelation of has-part. uants. As a consequence, we identify a collection as a In order to represent such a situation, let us mereological sum of its constituents (regardless of first introduce the collective interaction univer- their spatiotemporal arrangement), and not as a sal I COLL which is constituted by individual con- mathematical set. The reason for rejecting the set stituents which are instances of I, analogously to approach is two-fold. Firstly, because mathemat- Formula 8. Then we have to determine how each ical sets are extensional and therefore not robust of the grain interactions look like. If they are pair- with regard to the gain and loss of constituents. wise interactions between an Lmo-2 molecule and Secondly, because collectives should be of the same an Elf-2 molecule, each of these interactions fits ontological category as their constituents: A col- lective of material objects should be a material ob- 2 A counterexample is the interaction between ject, and a collective of events should be an event. solutes and solvents in a solution which necessarily Sets, however, are abstract objects that do neither involves colletives of both solvents and solutes. 41 Formula 7. Combining Formulae 7 and 8, we get: 2. What is thought to be the bearer of this prop- erty? ∃p, i1 , i2 , ...in , n > 1 : (10) 3. Which kind of property is in fact intended to be ^n (inst(iν , I) ∧ has-grain(p, iν ) ∧ ascribed? ν=1 The first question can be answered by pointing to ∃lν , eν : inst(lν , Lmo-2) ∧ inst(eν , Elf -2) ∧ one of the many event readings we discussed (and has-participant(iν , lν ) ∧ formalized) thus far. Our answer to the second has-participant(iν , eν ) ∧ question will at least in part depend on our re- ∀x : (has-participant(iν , x) → sponse to question 1. Are all instances of a given universal bearers of the disposition in question? inst(x, Lmo-2) ∨ inst(x, Elf -2)) ∧ Or only some of the instances? Are the individual ∀lν∗ , e∗ν : ((inst(lν∗ , Lmo-2) ∧ molecules the bearers of the disposition, or rather inst(e∗ν , Elf -2) ∧ has-participant(iν , lν∗ ) ∧ collectives of such molecules? The third question, has-participant(iν , e∗ν )) → (e∗ν = eν ∧ lν∗ = lν ))) however, leads us in to the middle of the lively debate going on in philosophy on the ontology of disposition [12, 13, 14]. The dispositional proper- Concurrent Interpretations II: ties most often discussed in the literature are so- Dispositional Readings called surefire dispositions: dispositions to react The above interpretations stated the existence of invariably in a certain way under specific circum- one or more interaction events. However, messages stances. They are one candidate for an answer to of the style “Lmo-2 interacts with Elf-2 ” very of- question 3. From the point of view of knowledge ten do not focus on the accidental occurrence of representation, however, there are some problems an event but are rather meant to express some connected with surefire dispositions. First, things inherent property of the objects under investiga- may react differently in different circumstances. tion. On the one hand it is likely that a biologist Thus to say that Lmo-2 molecules have the dispo- would mean “An interaction between Lmo-2 and sition to interact with Elf-2 molecules still leaves Elf-2 happened” while describing the outcome of it open under which circumstances such an inter- a specific experiment. On the other hand a biol- action will occur. We could account for this by ogy textbook would rather want to communicate explicitly mentioning the conditions of realization something like “Lmo-2 molecules have the disposi- for each disposition. We may, of course, not know tion or tendency to interact with Elf-2 molecules”. all these conditions, but this is an epistemic prob- This ambiguity, of course, matches Aristotle‘s fa- lem only. A more significant problem is that there mous distinction between act and potency, and may be infinitely many causally relevant condi- Aristotle himself observed that “potency” is in it- tions that have to be taken into account, and such self an ambiguous term [11]. Thus the ambigu- an infinite list would be impossible for principled ity of our sample statement increases even more, reason. We could try to circumvent this problem because the dispositional reading of our sample by adding (implicitly or explicitly) quantification sentence is ambiguous in itself. Obviously, such phrases like “In all circumstances” or “In some a reading of “Lmo-2 interacts with Elf-2 ” is in- circumstances”. The all -phrase, however, will not tended to ascribe some causal or statistical prop- do. For if a certain disposition would be realized erty, a disposition or tendency. But even if this under all circumstances, it will never be unreal- is the common ground of the dispositional read- ized. Such cases may exist, but normally a dis- ing, three questions remain open and have to be position will only be realized under certain cir- answered: cumstances and not realized under others. When we use the some-phrase, on the other hand, many 1. Which event is it exactly that the property in statements about dispositions for molecule inter- question is meant to cause? actions will become trivial, since nearly any mole- 42 cule may interact with any other molecule in some • (A) Every instance of U has a tendency to R peculiar way under certain (possibly very extreme) with a probability of 0.5. conditions. A usual way to deal with this prob- • (B) Every second instance of U has a surefire lem is to introduce a set of standard or normal disposition to R; the other instances of U do conditions [15]. In biology, this could mean that not have any disposition to R. the “disposition to interact with Elf-2 molecules” is only ascribed to Lmo-2 if the interaction com- Both of these scenarios would explain the assumed monly occurs under biological conditions, such as observations. Which of these scenarios we choose physiological pH and temperature intervals. But for our account of the observation will depend on the problem is not solved by referring to normal other observations and causal assumptions. If we, conditions. For, first, the problem that infinitely e.g. knew that nearly always the same instances many conditions cannot be described in necessar- of U display R and nearly always the same in- ily finite lists recurs with normal conditions. And, stances of U do not display R, this would prima second, biomedical knowledge may also include facie count as a reason to embrace (B). If, on the the behavior of molecules in non-normal or even other hand, we know that the same instances of U extreme circumstances, like low or high temper- sometimes do display R and sometimes do not dis- atures, exposure to intensive sunlight or atomic play R, this would prima facie count as a reason radiations. One option at this point would be to embrace (A). For such reasoning, however, we to choose a different answer to question 3. In- need background assumptions about the stability stead of ascribing surefire dispositions we could as- of the causal properties in question: how they can cribe probabilistic dispositions, i.e. dispositions to be stable over time, how (if at all) they can be ac- do something (under certain circumstances) with quired and how (if at all) they can get lost. Last a certain probability [16]. Such causal proper- but not least, (B) can indicate that the instances ties are also sometimes called “tendencies” [17] of the universal U differ in certain features, which or “propensities” [18]. While with surefire dis- are crucial to the ability to display R. An impor- positions a certain event will happen invariably tant example of this is the observation of modified in given circumstances, the event in question will proteins produced by mutated genes opposed to only happen with a certain probability when a ten- the observation of normal (wild-type) proteins. dency is ascribed. It will, of course, be crucial to Considering all this, there is quite a long and com- know with which probability the event will hap- plex list of enitities that we implicitly refer to when pen. Following standard procedures in mathemat- ascribing a disposition or tendency to a molecule: ical probability theory, we can represent the quan- tities of the probabilities in question by real num- • (independent) continuants (i.e. the bearer of the disposition), bers between 0 and 1 satisfying the Kolmogorov axioms. In biomedical experiments, the observed • dependent continuants or occurrents (i.e. the result is often such a probability. Tendencies are realization), thus of vital importance for the representation of biomedical knowledge [17]. There can, how- • quantities (of probabilities), and ever, be several ontological groundings for such a • state of affairs (of realization conditions). probability. Suppose that we observed a hundred instances of a given universal U in situations in Conclusion which all conditions necessary for the realization Our deliberations shed light on the need for a more R of a certain disposition were present, but that in principled account of dispositions and processes in only fifty cases R happened, i.e. in only 50 % of all biomedical ontologies. Machine supported infor- cases the disposition realized itself. There are sev- mation extraction and knowledge acquisition tech- eral ontological scenarios that would explain this niques from scientific texts have become a corner- result. Here are two of them: stone in molecular biology and genomics due to the increasing scientific productivity in this field. 43 The necessity of logic based ontologies for this pur- ical work on relevant text corpora may be helpful. pose has been controversially discussed [19]. If we This, however, is already beyond the scope of the subscribe to a formally principled account as a ba- present paper. sis for the semantic representation of the content of scientific texts then we have to take into ac- count that the most common type of statements Acknowledgments: that are of interest in texts describing biochemical This work was supported by the EU Network regularities do not have a clear and unambiguous of Excellence Semantic Interoperability and Data meaning. Assertions of the type “A interacts with Mining in Biomedicine (NoE 507505), the project B” are generally more than accounts of a single Forms of Life sponsored by the Volkswagen Foun- event. Rather they refer to a plurality of events dation, and the Wolfgang Paul Award of the of the same kind, or an event involving pluralities Alexander-von-Humboldt-Foundation. We are in- (collectives) of participants. A universal interpre- debted to Andrew D. Spear and to the anonymous tation such as “For each instance of A there is an referees of KR-MED for valuable comments. interaction with some B” can easily be discarded. The need for universal quantifications can be sat- Address for Correspondence: isfied by introducing dispositions: “Every A has Stefan Schulz, Department of Medical Informat- the disposition to interact with some B.” How- ics, Freiburg University Hospital, Stefan-Meier- ever, not every occurrence of the participation of Str. 26, 79104 Freiburg (Germany), phone: +49 some continuant in some process is proof of the 761 203 6702, e-mail: stschulz@uni-freiburg.de existence of a related disposition. Since interaction is a very general term, it is dif- ficult to express a clear preference in favor of any of the proposed approaches without analyzing the References nature of interaction on a molecular level, as well [1] Alan Rector, Jeremy Rogers, and Thomas as the study of the “normal” behavior of biomole- Bittner. Granularity scale and collectivity: cules. The question when to ascribe a disposition When size does and doesn’t matter. Journal or tendency – and which one – can not be discussed of Biomedical Informatics, 38, 2005. here (but cf. [16] on this). [2] Stefan Schulz and Anand Kumar. Biomedical We demonstrated that sentences like ”A interacts with B” exhibit indeed a wide range of ambigu- ontologies: What part-of is and isn’t. Journal of Biomedical Informatics, 38, 2005. ity. We offered several possible analyses to for- mally represent the different meanings of sentences [3] Stefan Schulz, Elena Beisswanger, Udo Hahn, of this type. Now, which one should we choose? Joachim Wermter, Anand Kumar, and Holger One strategy would be to say: Which strategy Stenzhorn. From GENIA to BioTop towards you choose depends on the intended meaning of a top-level ontology for biology. FOIS 2006 – the particular occurrence of the sentence you deal International Conference on Formal Ontology with. For text mining purposes, however, that in Information Systems, 2006. Accepted for have to digest large amounts of texts in short pe- publication. riods of time and with as much automatization as possible, this strategy would be scarcely feasible. [4] Aldo Gangemi, Nicola Guarino, Claudio Ma- To cope with this situation, several strategies are solo, and Alessandro Oltramari. Sweetening conceivable. One strategy would be to choose the ontologies with Dolce. In Asunción Gómez- highest common factor of all interpretation – that Pérez and V. Richard Benjamins, editors, what is included in all. Another strategy would Proceedings of the 13th International Confer- be to set as a standard interpretation that is most ence – EKAW 2002, volume 2473 of Lecture likely the intended meaning. In order to determine Notes in Artificial Intelligence, pages 166– which interpretation is the best candidate, empir- 181. Berlin: Springer, 2002. 44 [5] Barry Smith and Pierre Grenon. The cornu- Festschrift Franz von Kutschera, pages 89– copia of formal-ontological relations. Dialec- 106. de Gruyter, Berlin/New York, 1997. tica, 58(3):279–296, 2004. [16] Ludger Jansen. Attribuer des dispositions. In [6] Barbara Heller and Heinrich Herre. Ontologi- Bruno Gnassounou and Max Kistler, editors, cal categories in GOL. Axiomathes, 14(1):57– Les dispositions en philosophie et en sciences, 76, 2004. pages 89–106. CNRS Editions, Paris, 2006. [7] Franz Baader, Diego Calvanese, Deborah L. [17] Ludger Jansen. The ontology of tendencies McGuinness, Daniele Nardi, and Peter F. and medical information sciences. In Ing- Patel-Schneider, editors. The Description var Johansson, Bertin Klein, and Thomas Logic Handbook. Theory, Implementation, Roth-Berghofer, editors, WSPI 2006: Contri- and Applications. Cambridge, U.K.: Cam- butions to the Third International Workshop bridge University Press, 2003. on Philosophy and Informatics, volume 14 of IFOMIS Reports, pages 89–106. IFOMIS, [8] Barry Smith, Werner Ceusters, Bert Klagges, Saarbrücken, 2006. Jacob Köhler, Anand Kumar, Jane Lomax, Chris Mungall, Fabian Neuhaus, Alan L. Rec- [18] Karl. R. Popper. A World of Propensities. tor, and Cornelius Rosse. Relations in bio- Thoemmes, Bristol, 1990. medical ontologies. Genome Biology, 6(5), 2005. [19] Sophia Ananiadou and Jun’ichi Tsujii. The- saurus or logical ontology, which one do we [9] Karl. R. Popper. Logic of Scientific Discov- need for text mining? Language Resources ery. Hutchinson, London, 1959. and Evaluation, Springer Science and Busi- ness Media B.V., 39(1):77–90, 2005. [10] Donald Davidson. The logical form of action sentences. In N. Rescher, editor, The Logic of Decision and Action, pages 81–95. Pitts- burgh, PA: University of Pittsburgh Press, 1967. [11] Ludger Jansen. Tun und Können. Ein sys- tematische Kommentar zu Aristoteles’ Theo- rie dispositionaler Eigenschaften im neunten Buch der Metaphysik. Hänsel-Hohenhausen, Frankfurt am Main, 2002. [12] Raimo Tuomela. Dispositions. Reidel, Dor- drecht, 1978. [13] Stephen Mumford. Dispositions. Oxford Uni- versity Press, Oxford, 1998. [14] Bruno Gnassounou and Max Kistler. Les dis- positions en philosophie et en sciences. CNRS Editions, Paris, 2006. An English translation is to be published with Ashgate Publishers. [15] Wolfgang Spohn. Begründungen a priori – oder: ein frischer Blick auf Disposition- sprädikate. In Wolfgang Lenzen, editor, Das weite Spektrum der analytischen Philosophie. 45