ABSTRACT It has long been held that the structure of a protein is determined solely by the interactions of the atoms in the sequence of amino acids of which it is composed, and thus the stable, biologically functional conformation should be predictable by ab initio or de novo methods. However, except for small proteins, ab initio predictions have not been successful. We explain why this is the case and argue that the relationship among the different methods, models, and representations of (...) protein structure is one of integrative pluralism. Our defence appeals to specific features of the complexity of the functional protein structure and to the partial character of representation in general. We present examples of integrative strategies in protein science. _1._ Introduction _2._ Partiality of Representation _3._ Protein Functional Complexity _4._ Modelling Protein Structure _4.1_ Integrating ab initio and experimental models _4.2_ Integrating multiple experimental models _5._ Conclusion. (shrink)

Empirical agreement is often used as an important criterion when assessing the validity of scientific models. However, it is by no means a sufficient criterion as a model can be so adjusted as to fit available data even though it is based on hypotheses whose plausibility is known to be questionable. Our aim in this paper is to investigate into the uses of empirical agreement within the process of model validation.

J’analyse dans cet article la valeur explicative que peuvent avoir les simulations numériques. On rencontre en effet souvent l’affirmation selon laquelle les simulations permettent de prédire, de reproduire ou d’imiter des phénomènes, mais guère de les expliquer. Les simulations rendraient aussi possible l’étude du comportement d’un système par la force brute du calcul mais n’apporteraient pas une compréhension réelle de ce système et de son comportement. Dans tous les cas, il semble que, à tort ou à raison, les simulations posent, (...) du point de vue de leur valeur explicative, des problèmes spécifiques qu’il convient de démêler et de décrire précisément. J’essaie dans cet article d’analyser systématiquement ces problèmes en utilisant comme guide les théories existantes de l’explication. J’analyse d’abord le rapport des simulations à la vérité (section 2). J’examine ensuite en quoi les simulations satisfont ou non les exigences de déductivité et de nomicité, qui jouent un rôle central dans le modèle de l’explication de Hempel (section 3). J’étudie dans quelle mesure les simulations sont aptes à véhiculer l’information causale pertinente qu’on attend d’une bonne explication (section 4). Je poursuis en analysant en quoi l’abondance informationnelle et la lourdeur computationnelle des simulations peut sembler problématique par rapport au développement de nos connaissances explicatives et de notre compréhension des phénomènes (section 5). J’analyse enfin en quoi les simulations ont un rôle unificateur comme cela est attendu des bonnes explications (section 6). Au final, cette étude permet de comprendre plus précisément pourquoi les simulations, alors même qu’elles semblent pouvoir satisfaire les conditions que doivent remplir les bonnes explications, semblent spécifiquement problématiques au regard de l’activité explicative. Je suggère que les raisons sont notamment à chercher dans l’épistémologie de l’activité explicative, dans les attentes méthodologiques envers les bonnes explications et dans l’usage spécifique qu’on fait des simulations pour l’étude des cas difficiles – en plus du fait que les simulations constituent une activité qui n’est plus à taille humaine. (shrink)

I develop an account of productive surprise as an epistemic virtue of scientific investigations which does not turn on psychology alone. On my account, a scientific investigation is potentially productively surprising when results can conflict with epistemic expectations, those expectations pertain to a wide set of subjects. I argue that there are two sources of such surprise in science. One source, often identified with experiments, involves bringing our theoretical ideas in contact with new empirical observations. Another, often identified with simulations, (...) involves articulating and bringing together different parts of our knowledge. Both experiments and simulations, then, can surprise. (shrink)

Mesoscale modeling is often considered merely as a practical strategy used when information on lower-scale details is lacking, or when there is a need to make models cognitively or computationally tractable. Without dismissing the importance of practical constraints for modeling choices, we argue that mesoscale models should not just be considered as abbreviations or placeholders for more “complete” models. Because many systems exhibit different behaviors at various spatial and temporal scales, bottom-up approaches are almost always doomed to fail. Mesoscale models (...) capture aspects of multi-scale systems that cannot be parameterized by simple averaging of lower-scale details. To understand the behavior of multi-scale systems, it is essential to identify mesoscale parameters that “code for” lower-scale details in a way that relate phenomena intermediate between microscopic and macroscopic features. We illustrate this point using examples of modeling of multi-scale systems in materials science and biology, where identification of material parameters such as stiffness or strain is a central step. The examples illustrate important aspects of a so-called “middle-out” modeling strategy. Rather than attempting to model the system bottom-up, one starts at intermediate scales where systems exhibit behaviors distinct from those at the atomic and continuum scales. One then seeks to upscale and downscale to gain a more complete understanding of the multi-scale system. The cases highlight how parameterization of lower-scale details not only enables tractable modeling but is also central to understanding functional and organizational features of multi-scale systems. (shrink)