Models and modeling have pride of place in contemporary philosophy of science. However, in discussion of modeling methods and choices scant attention is paid to the way in which these are motivated in practice by particular cognitive constraints. Discussions on model-based reasoning that track the role of mental models, visualization and analogy in the model-building process provide notable exceptions (see Nersessian, Thagard, Giere, Magnani). In part, the neglect of cognitive constraints in philosophy stems from a lack of understanding of the intricate role these and other cognitive processes play in reducing complexity during model construction. In cases where researchers are up against significant complexity without well-structured theoretical or methodological protocols for handling it these processes carry a significant load. Satisficing limits the kinds of epistemic goals and aims researchers take on, promotes abstraction and idealization techniques, as well as promoting the distribution of cognition where possible to computational simulation and technologies.
In this talk we provide a case study from integrative systems biology using data from our 4 year ethnographic study of model-building practices in two systems biology labs. Complexity is a dominant concern for modelers in systems biology for a variety of reasons, not least the actual complexity of the biological systems they address. We show that distributed model-based reasoning plays a central role in the model-building process and researchers consistently rely on mental modeling to infer network structure and simplify their problems in order to make them tractable. Cognitive constraints such as working memory constraints may well limit the scale of networks that can be tackled and the epistemic goals that can be formulated with respect to them. Researchers in these field in fact acknowledge that cognition plays a decisive role in what they can and can't do, and in turn the need to formulate cognitively manageable problem-solving strategies. One instance of this is the strategy of mesoscopic modeling (Voit et al., 2012, Voit, 2013). Mesoscopic modeling works to simplify and abstract both higher systems-level and lower molecular-level relationships in the form of models of medium size. This enables researchers in practice to work with relatively simplified systems that facilitate the inferences and model-based reasoning they need to produce accurate system models. However with these models in place, hierarchical learning can be applied through experimenting and simulating a model to understand important causal relations within the system and more reliably expand and correct these models despite the complexity of the systems. Given the importance of such strategies for handling complex systems, we have good grounds for asserting the strong role cognitive constraints play in methodological choice in this context.
Systems biology encountered a recent boost with extension of technologies that are able to efficiently generate and handle large amounts of reliable data. However in theoretical sciences, the questions of complex systems’ structure and behavior, of their functionality and adaptability in the face of large number of interacting parts, has long been central. Integrating the newly available data with this knowledge is essential in order to facilitate the questions asked on data on the one side, as well as to meaningfully focus the development of theory. While integration of theory and data is an agreed upon goal, it is by no means a trivial task.
The common idea of systemic approaches is that systems are more than the sum of their parts and hence cannot be deduced from information about the latter. This emergent quality is often attributed to specific relationships or interactions between parts, and between parts and environment, which only come into view when the system is considered as whole, or in different contexts. The rationale behind such description is that these relationships structure the system in a constructive way that constrains how parts contribute to a common function (e.g. hierarchy or modularity are such structures). Yet what are the specific features of interactions that contribute to emergent system behavior?
Systems biology shares the problem of complexity with social studies, as well as ecology. In addition to evolutionary genetics, we draw on insights from theoretical and experimental work in social dynamics and ecological systems with the aim to identify the properties of interactions that are at the center of complex systems function. Our ultimate goal is to determine the character of pivotal interactions. This will enable us to develop an approach to identify the pivotal interactions in systems-biological problems.
Developments within the life sciences increasingly reflect the need for new interdisciplinary strategies to deal with the ‘grand challenges’ in society. An example is the current expansion of systems biology to systems medicine: a biomedical research field that aims to provide a better understanding of complex diseases and to account for variation among individual patients by developing personalized models of ‘digital patients’. The proponents envision that computational integration of new data types will revolutionize biomedical research and health practice through prediction and prevention of a number of diseases. Yet, the emerging literature on systems medicine reveals strong disagreements on the best way to advance systems medicine research and on the wider implications of these strategies for science and society. This paper analyzes the basis for these disagreements and examines the methodological and theoretical challenges highlighted by those skeptical of the strategies of systems medicine.
I focus in particular on challenges of integrating models and data in comprehensible large-scale models, and the associated controversies regarding the role of genomics for understanding complex diseases. The development of systems medicine, as it is currently conceived, is dependent on access to and collection of new data types such as whole-genome sequencing and continuous measurement of disease-related variables such as blood pressure, heart rate, blood sugar levels, protein markers etc. Accordingly, data-collection and model development will be conducted by different groups of basic science researchers, clinicians, practicing health service personnel, and individual patients. Patients are expected to actively engage in data collection as ‘consumers’ of health-technologies for self-monitoring of health-relevant information. The integration of scientific research and disease-preventing health strategies is a potent resource for understanding disease from a perspective that accounts for interdependencies between different diseases and for conditions specific for individual patients. But several researchers have raised concerns about the possibility of turning the vast amount of information into medically useful models. While some see genomics as a powerful tool for a new medicine, tailored to individuals with specific genetic profiles, others argue that making sense of the effects of gene mutations requires a better understanding of the organization of higher levels such as tissues and the whole human body. I demonstrate how such disagreements are often rooted in different views on the ontology of diseases, and in differences in epistemic standards for modeling. For instance, it is currently debated whether cancer is a cell-based disease (caused by mutations) or a tissue-based disease (caused by failure of higher-level organization). Furthermore, it is unclear how patient-specific information can be adequately related to relevant reference classes and to what extent large-scale models will provide evidence needed for health professional to make an informed decision. Analyzing the debate on the prospects of systems medicine, I examine how scientific disagreements relate to different epistemic and ontological standards for how we may best approach and perceive the functioning and malfunctioning of living systems.