Exploratory strategies: experiments and simulations

Po Garcia / Marisa Velasco Universidad Nacional de Crdoba, Argentina

Introduction
The use of computer simulations in different scientific activities has grown considerably in the last few years. However, philosophical reflection on computer simulations is rather scarce, partly due to the relative novelty of this type of activity and also probably because of its comple ity. Among the crucial elements necessary to analy!e simulations philosophically, the decision as to what aspects will be ta"en into account or what perspective should be adopted to underta"e the investigation is most important. #n the present wor" we suggest that an analysis based on e ploratory strategies can be used to illuminate and characteri!e epistemic and methodological aspects of computer simulations. $utting the focus on the e ploratory strategies implies adopting a perspective which although is relevant to the e periments and simulation analysis, is not limited or circumscribed to any of these areas. % ploratory strategies can be found in very diverse scientific practices, not withstanding they can be used to underline philosophically relevant aspects of e perimental practices as well as of computer simulation practices. #n this way, our proposal goes e plicitly beyond the comparison between e periments and simulations, although we are convinced that analy!ing e ploratory strategies in e periments and simulations helps substantially in the elucidation of their methodological and epistemological roles. #n this paper we will try to contribute to the philosophy of computer simulations and to the philosophy of e perimentation beginning with the elucidation of the role of the e ploratory strategies in both activities. &ecause of the fact that the relation between computer simulation and e perimentation has been understood in a very different way by the philosophical literature, a brief revision of some of the central ideas of these differences will be re'uired. Also, philosophical literature on e ploratory e periments supposes, although it rarely ma"es it e plicit, an e ploration notion, and so we will briefly refer to this discussion as well. #t is important to note that in their original conte ts, both discussions have the pretention of drawing limits ( between simulations and e periments or between e ploratory e periments and other e periments. )n the contrary, as we repeatedly say in this paper, rather than stipulating limits for each type of practice we propose to establish a sort of *geography+ of the ways science tests, searches and e plores. ,e will organi!e our wor" in the following way. -irstly we will schematically describe the way in which we try to account for simulations from their comparison to e periments so as to firmly establish the differences with the perspective we defend in this paper. .econdly we refer to the distinction between e ploratory e periments and other types of e periments. This second aspect will allow us to present our proposal as a different way to approach epistemic and methodical aspects of scientific practices. After an initial characteri!ation of e ploratory strategies we will consider some cases of e perimental and simulative activity. .eparation of e ploratory strategies into e periments and simulations is done with the purpose of ma"ing our argument clearer, and also to "eep a certain similarity to the way this matter has been 'uestioned in philosophy. Nevertheless, in most scientific practices boundaries are not so clear. /ore than being a difficulty in the selected cases, this seems to be the way in which contemporary scientific activity is organi!ed. ,e thin" this could be an additional advantage of adopting a perspective 0an e ploratory strategy1, which is not centered on the distinction between e periment and simulation, neither is it restricted only to e perimental practices.
2

Philosophy of Computer Simulations: experiments and simulations

% periments have been compared to simulations ta"ing into account ontological and inferential aspects on one hand, and representative or interventional aspects on the other. .o, although there are many ways of arranging a simulation with a computer, they can be characteri!ed, as $ar"er would say, as a se'uence of time ordered states which represent another se'uence of time ordered states. 3uala 045541, Hartmann 026671 and $ar"er 045561 propose a very similar characteri!ation. 8epresentation and imitation appear to be the principal concepts. )n the other hand an e periment is usually characteri!ed as an interventional activity. A central preoccupation in reflections on these sub9ects is usually the 'uestion of the validity of the simulations, or in more general terms, their epistemic credentials. At the same time, the discussion of ontological and epistemic problems usually assumes a basic scheme: simulations are seen as systems characteri!ed principally by some model that typically refers to another system 0usually called target system1. #n this way the 'uestion about the validity is set forth in terms of the relation between a given system and the target system, as well as the possibility of generali!ing these results or connecting them to other systems. Accordingly, -rancesco 3uala 045541 has proposed that the differences between an e periment and a simulation can be understood appealing to the type of 9ustification of the inferences which relate the different systems. The inferential lin" that can relate the e perimental result of a system A with the result of another system & can be very strongly 9ustified, according to 3uala, if both systems are operating under the same causes. Here one must suppose that there is only an abstract and formal correspondence between a computer simulation and a simulated system while between an e periment and a target system there usually is a correspondence at a deep level. #n this second case the same causes should be operating in the e periment as well as in the target system 2. 3ilbert y Troit!sch 026661 also seem to implicitly suppose the relation between an e perimental or computer system and a target system to outstand the difference between a simulation and an e periment, although they focus on the interventional aspects. These authors point out that in the case of an e periment what one is handling is a real ob9ect, while in a simulation if anything is to be handled it is a model : *while in an e periment one is controlling the actual ob9ect of interest 0for e ample, in a chemistry e periment, the chemicals under investigation1, in a simulation one is e perimenting with a model rather than the phenomenon itself+ 03ilbert ; Troit!sch, 2666, p. 2< 1. #t is probable that the same type of intuition that is sustaining the notion of materiality, or the correspondence at a deep level, is operating behind the notion of the real ob9ect. $ar"er 045561 suggests that the notion that should be ta"en into account is the one of relevant similarity more than the one of materiality. #f, as we have already said above, a simulation is understood as a representative system, an e periment is characteri!ed as *an investigative activity that involves intervening on a system in order to see how properties of interest of the system change, if at all, in light of that intervention+ 0$ar"er, 4556, p. <=>1. ,hen a simulation is arranged on a digital device we have a *computer simulation+. Although a computer simulation is characteri!ed in terms of representation, a *computer simulation study+ can be seen as an interventional activity in as much as one re'uires *setting the state of the digital computer from which a simulation will evolve, and triggering that evolution by starting the computer program that generates the simulation+ 0$ar"er, 4556, p. <==1. $ar"er believes that this way of characteri!ing a *computer simulation study+ allows us to tal" of an interventional activity and therefore of an e periment, because the focus is not on the model but on a programmed digital computer. Although $ar"er?s proposal seems interesting because it tries to account for the difference between mere
$ar"er argues that 3uala@s proposal is too restrictive in his e perimental notion, even though he accepts that 3uala doesn@t mean to say all material causes are relevant in this case, but only the ones that are Acloser?. ,ell then, scientists that are trying new drugs on mice are e perimenting, even though the results on humans later turn out to be different. Also, 3uala?s proposal is very restrictive when it refers to simulations, because it is too strong to say thet there are Anever? common material causes. 0$ar"er, 45561 4
2

modeling and computer simulation, it does not appear that the sole act of pointing out an interventional aspect is enough to 'ualify a *computer simulation study+ as an e periment in a relevant sense. There are other philosophers such as ,insberg 045561 that suggest that instead of considering ontological aspects to distinguish e periments from simulations, one must pay attention to the type of inference made and the role ta"en by the bac"ground "nowledge. #n the case of simulations, this bac"ground "nowledge allows us to build models that can later be ta"en as the ob9ect of investigation. Conse'uently:
,hen an investigation fundamentally re'uires, by way of relevant bac"ground "nowledge, possession of principles deemed reliable for building models of the target systems, and the purported reliability of those principles, such as it is, is used to 9ustify using the ob9ect to stand in for the target, when a belief in the ade'uacy of those principles is used to sanction the e ternal validity of the study, then the activity in 'uestion is a simulation. )therwise, it is an e periment. 0,insberg, 4556, p. B=71

-ollowing a very similar way of thin"ing to that of the philosophers cited above, /organ 0455BC 455D1 argues not only about the material but also the inferential aspects of simulations. % periments and simulations should have different *epistemic power+: the inferences on e perimental systems can be better 9ustified when the e periment and the target system the e periment refers to are made of the same *stuff+. /organ supposes that traditional e periments have greater epistemic power than simulations because as the later depend completely on a model they can?t confuse or surprise scientists. Here the notion of confusion is not ta"en in a psychological but in an epistemic sense: the phenomenon in 'uestion is somewhat *surprising+ as it can?t be accounted for with the available theoretical resources. /organ 0455D1 also points out to the material aspect as discriminatory criteria, but now between material and nonEmaterial e periments. .imulations would be among the later. #n fact one of the most interesting suggestions that this philosophy ma"es, related to this paper, is that starting from the analysis of the cases, it seems that usually hybrid situations are found, and in these it is not easy 0and probably not convenient1 to ma"e a distinction between e periments and simulations. Ta"ing a different perspective to the one we have 'uoted up to this point, /orrison 045561 emphasi!es the functional dimension of simulations and e periments. Here we must ta"e the functional dimension to be simply those aspects which allow the role played by a certain scientific practice to be understood. .omehow he also tries to clarify the conception of e periment by describing e perimental activities such as measurements. This change in perspective allows him to argue about some ways of creating a model with similar functions to a measuring instrument. #n this sense, in some particular cases, we could spea" of simulations as e perimental measuring instruments. Usually, by comparing e periments and simulations, one tries to build a philosophy of computer simulations based on ontological, inferential, representational, interventional or functional aspects. They could serve to establish limits between e perimental and simulative activities. According to their relative weight in the argumentation, it seems obvious that the aspects which could be considered ontological are considered decisive in the comprehension of simulations. &eyond the intrinsic virtues of this way of approaching simulations, our proposal can be ta"en as a move from ontological towards methodological and epistemological problems through the consideration of e ploratory strategies. #n this sense, most of the ideas 'uoted in this section allow us to see differences clearer when using our e ploratory strategies proposal. Considering that the notion of e ploration has been mainly used in the philosophy literature on e periments, in the following section we will begin with this discussion to continue later with the notion of e ploratory strategy.

From exploratory experiments to exploratory strategies

The notion of e ploration in science has not received a great deal of treatment in the literature on philosophy. Nevertheless, among the diverse proposals to classify scientific e periments it is possible to find reflections about the role of e ploration in science. #t is 'uite common for the category of *e ploratory e periment+ to appear among the various e periment classifications. #n this way some science philosophers 0for e ample, cf. &urian, 455>C %lliott, 455>C -ran"lin, 455BC )?/alley, 455>C $eschard, 4556C .teinle, 266>, 4554C ,aters, 455>1 have attempted to elucidate e ploratory e periments presuming e ploration characteri!ations. Although we will show that the diverse characteri!ations of e ploratory e periments don?t achieve ade'uate discrimination of these e periments, the analysis of the different proposals will allow us to show some of the characteristics of e ploration. Although these e ploration characteristics won?t be useful to do an ade'uate ta onomy of the e periments, they will be the basis of our characteri!ation of *e ploratory strategies+. #n this sense it is important to remember that e ploratory strategies are not a specific patrimony of e ploratory e periments, but they do constitute a relevant aspect of the best part of scientific activity, including e perimentation and scientific simulation. #t has been suggested that e ploratory e periments can be understood based on the contrast with e periments where the relevance of theory is substantive. This is an outstanding aspect in the ma9ority of the characteri!ations that have a distinctive feature. Thus, for e ample, -ran"lin 0455B1 sustains that e ploratory e periments are the ones scientists do without considering a particular theory concerning the effects of their intervention on the values they are measuring. #n a similar way .teinle suggests there are e perimental designs that are guided by theory 0theory driven1 and others which are not, the latter being the e ploratory e periments. Thus, e ploratory e periments could be characteri!ed by their relative independence to strong theoretical restrictions 0Cf. -ran"lin, 455B, p. ===C .teinle, 266>, p. .76, 4554, p. <2=1. #n the same sense, .teinle 0266>1 sustains that the standard perspective on e perimentation usually considers those which are theoryEdriven as the only type of e perimentation. Accordingly, e perimental activity is understood in terms of: *there was a theory that led to e pecting a certain effectC the e pectation led to designing and conducting an e perimentC and the success of e periment counted as support for the theory+ 0.teinle, 4554, p. <2=1. Nevertheless, from this author?s perspective, the e ploratory e perimentation: *#t typically ta"es place in those periods of scientific development in which ( for whatever reasonsE no wellEformed theory or even no conceptual framewor" is available or regarded as reliable+0.teinle, 266>, p. .>21. %ven those authors that don?t e plicitly uphold this idea seem to "eep this way of understanding e ploratory e periments in mind, as they choose e amples from the initial stages of scientific disciplines. Nevertheless, the related literature, that sees the discriminatory criteria in the theoretical guide, also underlines other aspects that are important for the characteri!ation such as the purpose of these e periments, their use or the e perimenter?s e pectations. .o, for e ample, the variety of epistemic goals present in the e ploratory e periments becomes outstanding:
The contrast of e ploratory e perimentation to the theoryEdriven type, as understood as the standard view, is not only visible in the different epistemic goals 0search for regularities vs. test of e pectations1, but also in the character of the guidelines of the e perimental activity. 0.teinle, 4554, p. <441

Fi"ewise, we must remar" that the e ploratory e periments are not circumscribed to a particular historical or theoretical conte t:
G e ploratory e perimentation is not so much bound to certain historical periods, fields of research, or scientific traditions, but first and foremost to specific epistemic situations: those situations namely in which, for reasons whatsoever, the very concepts by which a certain field is treated have been destabili!ed and become open for revision. .ituations in which theories and wellEformed e pectations are tested, in contrast, re'uire a wellEelaborated conceptuali!ation, a stable language by which the e pectation can be e pressed in the first place. % ploratory and

<

theoryEdriven e perimentation are connected to different constellations and situations of our "nowledge, to different regimes of stability on a conceptual level. (Steinle, 2002, pp. 425426)

Fast of all some philosophers underline the relation between the e perimental activity@s purposes and their results:
8oughly spea"ing, the aim of e ploratory e periments is to generate significant findings about phenomena without appealing to a theory about these phenomena for the purpose of focusing e perimental attention on a limited range of possible findings. The findings might be significant with respect to a variety of goals ranging from the practical goal to learn how to manipulate a phenomenon to the theoretical goal to develop a conceptual framewor" that will help focus future e perimental attention. 0,aters, 455>, p. B1

#t is important to say that the previous three 'uotations correspond to authors that defend the theoretical bond criteria for the distinction of e ploratory e periments. However, in all three one can notice that other characteristics are the ones that stand out in this type of e periments. ,hen e ploration is ta"en as a distinctive note of some e periments, it doesn@t seem then to be ade'uately characteri!ed by its bondage to the theory. #n fact, whoever tries to defend this idea should be able to e plain the different levels of theory involved in an e periment, and determine which of these levels is or are relevant to the *theoretical guidance+. This is, to say that an e periment is guided by a theory would mean that the e pectations regarding its results are theoretical, or that the design of the e periment depends on a theory, or that the instruments used are theoretically dependent, etc. Then possibly great part of the difficulty of the characteri!ation, in terms of theoretical dependence of the e ploratory e periments, is due to the lac" of a sophisticated notion of theory and of levels of theory involved in the design, e ecution and analysis of e perimental results4. &ut the perspective of the bondage to theory, even when attention has been paid to the previous observation, does not reflect important epistemological and methodological aspects of e ploration in science. #n this sense, .teinle himself states that in e ploratory e periments we can distinguish the following activities as typically methodological practices:
Harying a large number of different e perimental parameters, Ietermining which of the different e perimental conditions are indispensable, which

are only modifying,

Foo"ing for stable empirical rules, -inding appropriate representations by means of which those rules can be formulated, -orming e perimental arrangements which involve only the indispensable conditions,

thus presenting the rule in particular clarity. Those e periments are typically characteri!ed as *simple+, *elementary+, or *pure+ case+ (Steinle, 1997, p. S70)

)ne of the activities pointed out as typical in an e ploratory e periment is the variation of parameters. This is usually a regular procedure in e periments, not only e ploratory but also in general. The variation of parameters can be done with different ob9ectives: from the calibration of a measuring instrument to the pretention of obtaining some empiric regularity. Now, the purpose of an e periment can turn the variation of parameters into an e ploratory e periment even though it is carried out under strong theoretical restrictions. This ma"es one believe that an e ploratory e periment is best characteri!ed from its purpose than from its particular relation to theoretical bodies. This relationship with theory can only be understood if no 'ualification appears as an inade'uate characteri!ation. Also, putting emphasis on the aims and the roles of the e periments helps us to distinguish between e ploratory strategies that not only form part of e ploratory e periments but also of e periments with different purposes, as for e ample, hypothesis confirmation. This point seems
4 A first approach to the different types and levels of theory involved in an e periment can be found in 0Hac"ing, 26641.

relevant for different reasons. #n the analysis of cases that usually accompanies the characteri!ation of e ploratory e periments there is confusion between what we could call the e ploratory e periment typology and the searching and e ploring strategies. &ut, it seems clear that there are searching and e ploring strategies that are part of nonEe ploratory e periments, as they are also part of other nonEe perimental scientific practices. None the less, it is necessary to point out that search strategies or e ploratory restrictions do not seem to be what is definite to understand e ploratory e periments, as do the aims and purposes of the e periments. However their consideration allows a more ade'uate description of certain aspects of scientific and e perimental practices in particular.

Exploratory strategies in scientific practices

,ith the aim of developing our proposal we will organi!e this section in the following way: firstly we will e plicit what we understand as e ploratory strategyC then we will show some e amples that will permit us to conte tuali!e our proposal. -inally we will analy!e some cases about e ploratory e periments presented by the literature to show the advantages of our proposal. #n particular we will show how adopting the e ploratory strategy@s point of view leads to a more direct lin" to computer simulations. #n the conte t of the present paper we will consider any activity or resource which allows us to do an e ploration as an *e ploratory strategy+. These resources or activities can be very different but as long as they perform the function of allowing us to 'uestion, search, probe or e plore we will call them *e ploratory strategies+. #t is important not to restrict the notion of a strategy by setting rules or any other structure implying a procedure. The relevance of this point will become evident in the following paragraphs as we present the ways in which e ploratory strategies can be instantiated. A first way of characteri!ing e ploratory strategies is in terms of what they are loo"ing for 0what they e plore1, the way in which they do this operation 0how they e plore1 and to what end they are doing it 0what they e plore for1. A second way of characteri!ing e ploratory strategies would be paying attention to the structure of the search ( or if you prefer to the restrictions in the searcher ( and the structure of the media in which one is searching ( the restrictions in the search space. .o, we could have searches with few restrictions ( for them to be as ample as possible ( or searches with strongly selective criteria. This way of understanding e ploratory strategies can be instantiated as searching rules or heuristics. The structure of the searching rule will let us account for part of the restrictive capacity of the e ploratory strategy. ,e can name this way of understanding e ploratory strategies as Atype 2?. ,e can also e plore using 9udicious construction of an e ploration space or by limiting the si!e of this space. This way of understanding e ploratory strategies we will call Atype4? strategies. ,e are using, although in an indirect way, the metaphor that distinguishes between the search strategy and the space where the strategy ta"es place. #n the same way that we tal" of restrictions in the type2 strategies, we can suppose that the space structuration involves, among other things, graduating the restrictions. )ne could argument that the distinction between type 2 and 4 strategies is not genuine because type 4 strategies can be reEwritten in terms of type 2. Although in principle it seems possible to do this reEwriting tas" and eventually reduce one type to the other, this perspective doesn@t seem ade'uate when we are trying to account for epistemic and methodical aspects associated to the description of scientific practices. #n this Areduction? tas" the specific ways in which the e ploration is being done, in the area of simulation or the e periments, wouldn?t be ade'uately shown. &efore continuing, let us see some e amples that help put our discussion in conte t. The first case involves the area of the so called combinatory chemistry and highEthroughput screening. Combination chemistry can be described, in very general and schematic terms, by methods

associated with combinatory synthesis and highEthroughput screening. According to Halerie 3illet, a researcher in this area, combinatory chemistry *refers to the synthesis of large numbers of compounds in parallel where product molecules are formed as combinations of available reagents or buildings bloc"s+ 0Andrew 8. Feach ; 3illet, 455>, p. 72>1. HighEthroughput screening *is an automated process whereby a large number of compounds 025< ( 25B1 are rapidly screened for biological activity+ 0Andrew 8. Feach ; 3illet, 455>, p. 72>1. This focus on synthesis and screening, more than a new methodology, represents an automation of traditional methods with the help of new instrumental resources. Combinatory synthesis can be seen as a procedure to construct, in different ways and suppositions, a search area. ,hereas it seems highEthroughput screening can be seen, not only by its design but also by its application, as a search strategy. Fet?s ta"e a loo" at this type of analysis now. .tarting with highEthroughput screening, one can understand the types of search involved here from their historical evolution. Iuring the =5@s in the last century it was ta"en as an advantage for this search to be as least restricted as possible, so as to be able to obtain the largest number of candidates for a new drug 0leads1, by the sole record of the force of automation. ,hat@s more, the term used to describe the way the search was done at this time was *serendipity+ 0Cf. 3arcJa, 45561. ,ith time it became clear, from the poor results obtained, that a more selective search was re'uired. .o, restrictions were added to the type of mechanism that was desirable in a drug precursor. #n this way we can find, from Fipins"i?s rule of B 0Fipins"i 266B1 E this is a set of desirable characteristics in a drug, such as solubility or permeabilityE to the superposition or ad9ustment between molecular structures. This way of understanding highEthroughput screening is sometimes called *rational+ or of *design+ to mar" its differences to a search without too many restrictions. The increase of restrictions in the search wasn@t the only resource used to improve the performance of these systems. The use of combinatory chemistry to design libraries of compounds became more sophisticated as from the 65@s. This aspect can be seen as an instantiation of the second criteria to classify e ploratory strategies. #n fact, the construction of libraries of substances seems li"e a way to restrict or guide the search by means of a rational structuring of the problem@s space. .o, this started the discussion as to in what way one could obtain libraries of substances with a sufficient degree of variation to ma"e the search interesting. This supposes a relevant notion of the meaning of *variation+. Although this notion can be instantiated differently according to the case, the construction of a more general notion has been attempted. A standard way of understanding this notion is to lin" the concept of *diversity+ to that of similarity. #n the conte t of the construction of libraries what one needs is some codification of the substances in 'uestion 0molecular descriptors are usually used1 to be able to later define a measure that can be ta"en as a degree of similitude. #n this way one can determine if a sample of substances is more *diverse+ than another in terms of some 'uantifiable criteria 0Cf. &leicher, &ohm, /uller, ; Alanine, 455D1. #n several of the more interesting stages of wor" in combinatory chemistry, simulation, e perimentation and e ploration strategy practices are involved. This ma"es it not simple ( and probably not convenient ( to ma"e differences between e periments and simulations when what are under consideration are the investigation processes rather than their constituent parts. As an e ample one can cite not only the construction but also the search in virtual libraries 0an appro imation sometimes called in silico1. #n this case libraries of aspects and properties of the substances with descriptions that are assumed relevant are constructed and the search is done in these virtual spaces. Thus *in silico+ screening *refers to the use of computational techni'ues to select compounds, either from e isting librariesGor from virtual libraries that represent the compounds that could potentially be made via combinatorial synthesis+ 0Andrew 8. Feach ; 3illet, 455>, p. 72=1. #f what interests the researchers involves the "nown lin"s 0KligandsK1 for a compound with a certain KtargetK, then you can build a KpharmacophoreK model to account for the compounds structural aspects. #t is important to point out that *a pharmacophore does not represent a real molecule or a real association of functional groups, but a purely abstract concept that accounts for the common molecular interaction capacities of a group of compounds towards their target structure+ #U$AC 8ecommendation cited by 0A. 8 Feach, 3illet, Fewis, ; Taylor, 45251 p.
>

BD61. -or this reason, this type of model is usually used for virtual searches 0virtual screening1 and predictive Kdoc"ingK models in general, that is, rational design models that in some way allow to have e pectations on candidate ligand receptors 0Andrew 8. Feach ; 3illet, 455>, p. 277ss1. )ne way to systemati!e the virtual search types is to consider the amount of structural information and biological activity available 0Andrew 8. Feach ; 3illet, 455>, p. 2B=ss1. -irst, if what is "nown is only a single active molecule, virtual search will generally be based on the possibility of binding. .econd, if there are various active molecules, you can build a pharmacophore model and then search using DI properties. Third, a neural networ" can be used when there is sufficient information about active and inactive structures. -inally, when the DI structure of a protein is "nown, KproteinEligand doc"ing can be employedK in the search 0Andrew 8. Feach ; 3illet, 455>, p. 2751. Thus, the virtual search involved in this type of computer simulation depends on the type and the amount of information available on the chemical compound@s space. Any results that may be of interest can become part of e periments in real libraries. The virtual library search can hardly be seen as an e ploratory e periment, in this case however, it is clear that e ploratory strategies are used. /oreover, it is possible that similar strategies are used in both, the virtual and real e perimental spaces. Considering the discussion in the previous section, here we would have e periments using e ploratory and search strategies in conte ts where the goals can be confirmatory. An e ample of this would be the search process that is triggered when one estimates one has achieved a precursor or a more or less reliable candidate for a drug, but this time see"ing to confirm an initially promising result. This is a good e ample of situations that /orrison called KhybridK, so common in current scientific practices. .hown in this schematic presentation are some of the advantages of our proposed characteri!ations. However, it is clear that such characteri!ations re'uire a more careful elucidation. 8eturning to the first characteri!ation of e ploratory strategies, it could be affirmed that the functional aspects of these strategies are only lin"ed with the third item 0what they e plore for1 of the first characteri!ation. However, to account properly for the functional aspects re'uires an e planation of the ways in which the e ploration is performed and certainly of its goal. #n turn, the first aspect of the first characteri!ation of e ploratory strategies 0what are they loo"ing forL1, could be understood as referring to the particular scientific discipline in which they are used E chemical, physical or otherwiseEor the main purpose of the e periment or activity performed. &ut this does not seem to be its primary meaning. #f an e perimental design uses a variation of parameters in order to calibrate an instrument, then the alleged e ploratory strategy here seems to be aimed at the instrument itself 0or at least at some relevant conte ts to increase the precision or accuracy of the instrument1. -rom this point of view we can see that e ploration activities are also involved in the different aspects of design and e ecution of an e periment or other scientific activities. Among the latter we can point out the situation in which models are e plored, that is, for e ample, we e plore the limits of a model, the empirical appro imation of a model, the ad9ustment of a model to other models, etc. This aspect of e ploratory strategies can also have a more sophisticated networ". -or e ample, when we consider the field of computer simulations, we have, and typically this is the case, e ploratory strategies at different levels, in this sense, the 'uestion about what these strategies loo" for, has a different response at each of these levels, although we have one overall goal at the KhigherK level of computer simulation. Thus, we can have a computer simulation whose main ob9ective is to e plore a model in a certain way, but has different e ploratory strategies in its MlowerM levels. This comple ity is not necessarily a difficulty, since it allows us to e plain the relative independence E at least in a certain sense E between each of these e ploration strategies. Iepending on the type of problem in which we are interested, we can distinguish different strategies at different levels. 8egarding the second aspect of the first characteri!ation of e ploratory strategies E how to perform the e ploration E here one can consider different types of search, e ploration and in'uiry. Hirtually all we have considered in type 2 and 4 strategy, could come into this point, ta"ing into account the restrictions on the search and the structuring of the space where the search is made. %ither way, it is possible that this approach can be developed in the future by chec"ing other aspects
=

through which the types of search can be seen, or by specifying the types of restrictions that are ta"en into account. -or e ample, returning to the case of high throughput screening, one could ta"e into account the differences between the e ploration based on structural aspects 0such as KhostEbased couplingK1 and those based on properties. The latter could include the aforementioned rule of Fipins"i. A compound can be KdrugEli"nessK if it has certain properties, at least, properties such as absorption or permeability 0which is unli"ely if the compound in 'uestion has a molecular weight greater than B551. ,hile what we might call structural searches also have the goal of finding compounds that may be plausible candidates for a new drug, the type of restriction that guides the search is very different. The methods called Kmolecular doc"ingK would fall into this category. #n general, what is intended to be found in the *doc"ing+ e periments is the DI structure resulting from two or more molecules. Computational methods involved here are concentrated on two tas"s: e ploring the space of possible KproteinEligand geometriesK 0Andrew 8. Feach ; 3illet, 455>, p. 2721 and the weighted evaluation of these geometries to enable them to Kidentify the most li"ely binding model for each compound and to assign a priority order to the molecules. K0Andrew 8. Feach ; 3illet, 455>, p. 2721. To a large e tent, the difficulty of this tas" lies in the degrees of freedom in rotation and translation of molecules E in relation to other molecules E which affect their geometrical configuration. -or this reason the biggest effort is often in the design and implementation of algorithms that account for this problem. The third aspect of the first characteri!ation of e ploratory strategies E what is the e ploration performed for E is important because it allows us to highlight another angle of the relationship between e ploratory strategies and other scientific activities and e periments. )ne might suppose that this appearance coincides with the one that allowed us to distinguish between e ploratory e periments and confirmatory e periments. &ut as we saw above, the goal of an overall activity is not the reason of the e ploratory strategies involved. Continuing with the e ample we mentioned above, we might have an e periment whose design or calibration assumed e ploratory strategies, but whose ultimate goal is to test or confirm a given hypothesis. At least in principle, given the comple ity of current e perimental designs and the different levels sometimes involved, there may be e ploratory strategies at an e ecution level, although the aim of the e periment itself is another. -rom this characteri!ation, ta"ing some cases from the literature on e ploratory e periments, e ploratory strategies can be analy!ed in a new light. As noted in the previous section, the concern about drawing a line between e ploratory e periments and other interventional practices has left in the bac"ground other interesting aspects of these scientific practices. #n order to show the advantages of our proposal we discuss some of these e amples. #n our analysis we will focus on what we called type 4 strategies, that is in the form in which the e ploration space is selected and configured. #n a complementary way, we highlight the importance of instruments in these cases. -ran"lin 0455B1 noted that the adoption of certain instruments is often accompanied by an increase in e perimental Ke ploratory practicesK. The impact of the instruments is similarly highlighted by several of the philosophers who analy!e e ploratory e periments. However, it can be argued, that not only are there methodological strategies associated with the mere adoption of an instrument to carry out a tas", but also with its configuration and use. /oreover, the notion of instrument, understood from the methodological resources, can serve to understand the role of computer simulations. )M/alley 0455>1 presents a case within the scope of what is often called Kmetagenomics.K This case is related to the discovery of proteorhodopsin in marine bacteria gene 0in the /onterey &ay1. /etagenomics can be characteri!ed by the manner in which the samples to be e amined are selected. Under the assumption that genomic diversity is not sufficiently represented by the crop cloning methods used to se'uence the genetic information, samples ta"en from some KnaturalK environment that is relevant to the investigation are used. #n this way of selecting the samples we can see a first sense of e ploration lin"ed to the selection of the space where the investigation is to be performed. This consideration is reinforced when comparing the sample *area+ to the cultivated samples space where e perimental protocols tend to favor their KisolationK. This aspect is also highlighted by )M/alley as it allows him to describe a form of
6

e ploratory e periment. &ut beyond this first e ploratory direction presented by )M/alley, one might suggest a second direction associated with the se'uencing of the samples DD. -or several years a parallel se'uencing techni'ue 0KshotgunK .anger se'uencing or massively parallel pyrose'uencing1 is being used, and is often seen as a suitable complement to the ob9ectives of metagenomics. #t seems clear that there is a sense of e ploration involved in the parallelism in this type of method 0%isen, 455>1. Another case considered in the literature on e ploratory e periments is 'uoted by -ran"lin concerning the use of what is called KmicroarraysK 0INA chips1. -ran"lin underlines the use of this tool in the investigating of the role of certain proteins in the cell cycle. This instrument consists of a KplateK of a material that can serve as a KgridK for samples to be analy!ed. These plates must be constructed of a material that allows INA binding. Iue the parallel nature of this techni'ue it is often used to investigate Kthe differential e pression of the genes.K The level of gene e pression is measured by a KprobeK 0probe1 that is added to the sample to be investigated and that has been Kmar"edK 0with a fluorescent or radioactive tracer1. These mar"s are analy!ed in an image that indicates the level of gene e pression. The differential aspect of gene e pression can be investigated by changing the conditions in cells 0with or without a particular alteration1. The first e ploratory aspect can be noted in this conte t has to do with the configuration of the space in which the se'uencing will be done. This point becomes clearer when comparing this techni'ue with that which gave rise to it: Northern and .outhern blot. #n principle, the only difference between the two techni'ues is that the latter only allows the study of genes Kone at a time.K &ut there is another aspect of the e ploration that can be pointed out here related to the design of the configuration of the INA chip samples. &ecause of the amount of samples available, the design of these early stages is usually done with the help of speciali!ed software that allows putting into practice strategies that are considered appropriate for the e ploration. Thus, one can ta"e any diversity criterion E for e ample by homology E and apply it to construct a sample. #n this sense, it seems that there is a relationship between sample design and the ways to e plore.

Final words.
,e have tried to show how an insight on e ploratory strategies allows us to show different methodological and epistemological aspects of computer simulations in science and scientific e perimentation. )ur intention has been to collaborate in the construction of theoretical tools that help analy!e and highlight the epistemological and methodological richness of certain aspects of these practices that have been overloo"ed in philosophical research. #n this sense, we believe that the activities of e ploring and searching have not received sufficient attention in the field of the philosophy of science. The analysis of e ploratory strategies has allowed us, through the presentation of some cases, to show the different levels at which e ploration is relevant in scientific practices. However, this is only a first approach to the characteri!ation of a concept that we believe is promising, as we introduce ourselves into some of the problems that the role of computer simulation and scientific e perimentation poses to the philosophy of science in these days. 045521

References
The use of shotgun se'uencing is 'uoted in )?/alley@s wor", although its methodological importance isn@t highlighted in terms of e ploratory strategies
25
D

&leicher, N. H., &ohm, H. O., /uller, N., ; Alanine, A. #. 0455D1. Hit and lead generation: beyond highEthroughput screening. Nature Reviews Drug Discovery, 20B1, D76(D>=. &urian, 8. /. 0455>1. )n /icro8NA and the Need for % ploratory % perimentation in $ostE3enomic /olecular &iology. History & Philosophy of the Life Sciences, 290D1, 4=B(D24. %lliott, N. C. 0455>1. Harieties of e ploratory e perimentation in nanoto icology. History And Philosophy f !he Life Sciences, 290D1, D2D(DD7. -ran"lin, F. 8. 0455B1. % ploratory e periments. Philosophy of Science, "20B1, ===(=66. 3arcJa, $. 045561. Iiscovery by serendipity: A new conte t for an old 8iddle. #oundations of $he%istry, &&021. 3ilbert, 3. N., ; Troit!sch, N. 3. 026661. Si%ulation for the social scientist . /aidenhead, %nglandC New Por", NP: )pen University $ress. 3uala, -. 045541. /odels, simulations, and e periments. 'odel()ased reasoning* Science+ technology+ values, B6(><. Hac"ing, #. 026641. The selfEvindication of the laboratory sciences. #n A. $ic"ering 0%d.1, Science as practice and culture 0Hol. D51. University )f Chicago $ress. Hartmann, .. 026671. The world as a process: .imulations in the natural and social sciences. 'odelling and si%ulation in the social sciences fro% the philosophy of science point of view, >>(255. Feach, A. 8, 3illet, H. O., Fewis, 8. A., ; Taylor, 8. 045251. ThreeEdimensional pharmacophore methods in drug discovery. ,- 'ed- $he%, ./041, BD6(BB=. Feach, Andrew 8., ; 3illet, H. O. 0455>1. An 0ntroduction to $he%oinfor%atics. .pringer. Fipins"i, C. A., Fombardo, -., Iominy, &. ,., ; -eeney, $. O. 045521. % perimental and computational approaches to estimate solubility and permeability in drug discovery and development settings2. Advanced drug delivery reviews, 1202ED1, D(47.

22

/organ, /. 0455B1. % periments versus models: New phenomena, inference and surprise. ,ournal of 3cono%ic 'ethodology, &2041, D2>(D46. doi:25.25=5Q2DB52>=5B555=7D2D /organ, /. .. 0455D1. % periments without material intervention: model e periments, virtual e periments and virtually e periments. #n H. 8adder 0%d.1, !he philosophy of scientific e4peri%entation. University of $ittsburgh $ress. )?/alley, /. 0455>1. % ploratory e perimentation and scientific practice: /etagenomics and the proteorhodopsin case. History and philosophy of the life sciences, 290D1, DD>(D75. $ar"er, ,. 045561. Ioes matter really matterL Computer simulations, e periments, and materiality. Synthese, &290D1, <=D(<67. doi:25.255>Qs22446E55=E6<D<ED $eschard, #. 045561. /odeling and % perimenting. 8etrieved from

http:QQphilsciEarchive.pitt.eduQB222Q .teinle, -. 0266>1. %ntering new fields: % ploratory uses of e perimentation. Philosophy of Science, 210<1, .7B. doi:Article .teinle, -. 045541. % periments in history and philosophy of science. Perspectives on Science, &50<1, <5=(<D4. ,aters, C. N. 0455>1. The nature and conte t of e ploratory e perimentation: An introduction to three case studies of e ploratory research. Hist- Phil Life Sci-, 290D1, 4>B(4=<. ,insberg, %. 045561. A tale of two methods. Synthese, &290D1, B>B(B64.

doi:25.255>Qs22446E55=E6<D>E5

24