Development of large scale in silico models to explore biological mechanisms in research and teaching

Abstract

Computational models are increasingly valued as a method to help with both describing and explaining biological phenomena. Models work by collating information (both qualitative and quantitative) which has been discovered experimentally to relate to a biological output of interest. Models can then be used to test the validity of hypotheses or findings about this output, as well as making predictions of future experimental findings. Models come in many flavours, usually focussing on a particular level of biological inquiry (molecular, cellular, systems etc.). Biochemical modelling aims to offer explanations of biological phenomena through simulating the actions of biochemical networks, chiefly by defining molecular concentrations, interaction rates, and localisation. To date however, this type of model has been restricted to only truncated models representing only partial elements of biochemical networks. This has limited their utility, as well as the perception amongst non-modellers as being a useful parallel method of analysis.

In the first half of this thesis I present a reappraisal of biochemical modelling, highlighting its strengths as well as its underutilisation. In particular, I note how the development of these models requires collaboration with experimenters, both in the acquisition of data to improve models, as well as in better explanations of how models work currently when kinetic interaction data is scarce. I show some of these principles in the development of a merged model that collates pre-existing synaptic plasticity models into a single “maximal” model, the AMPAR-centric Model of Postsynaptic Signalling (AMPS). I use this experience to develop a sketch of how “maximalist modelling” should be, defining its core principles and how it may be used in both improving collaborative practice and biological understanding.

One large restriction of models like AMPS is that many of the molecules associated with synaptic plasticity may adopt multiple functional states. In traditional modelling approaches, these states must all be explicitly defined, leading to an exponential rise in model size as the numbers of components increases. However, thanks to recent advances in computational power and algorithms, one can generate rule-based models, which are better suited to explain complex phenomena such as synaptic plasticity, and that align more closely to maximalist modelling principles. I adapt AMPS to capitalise on this and show how I have iteratively constructed and developed the model to explore how AMPARs are trafficked at the synapse during plasticity events. I show that the stability of elevated levels of AMPARs in the early phase of long-term potentiation (LTP) requires depression of phosphatase activity, which can be achieved through high levels of CaMKII autophosphorylation. Moreover, I show that this stabilisation does not require any new protein synthesis of AMPARs, or any change to other molecular concentrations. I confirm that trafficking is a complex mixture of receptor-modification, synaptic stabilisation and localisation that is best suited to being explored in silico. I also confirm that mGluR-mediated pathways are crucial for a complete understanding of synaptic plasticity, with their absence leading to potentiation when paired-pulse low-frequency stimulation is applied.

The second half of the thesis arises from one defined aspect of maximal models, which is that they should be educationally valuable tools as well as being valuable in research. I have developed both a typical and atypical approach to biological modelling that can be used in secondary through tertiary pedagogy. The typical approach is to make use of common, well characterised model software such as COPASI to allow learners to construct their own model (and subsequently their own understanding) of complex biochemical systems. Adopting a constructivist stance, I show how learners with no prior knowledge of these systems can generate accurate, academically useful models that are shaped by the learners’ interests developed over the course of model construction. I also provide evidence for the utility of this approach in improving theoretical understanding and exposing experimental gaps in the literature for the field as a whole, one which benefits from the models being constructed by non-specialists.

The atypical approach focusses more on how complex biological concepts can be abstracted into more visual and tactile environments in a way that does not result in any meaningful loss of their complexity. I present an extended case study in using the sandbox game Minecraft, where I have developed applications for its use in undergraduate learning, multi-audience outreach and academic experimentation. I show how the use of such an environment allows for creativity in a learner’s approach to biology, which is often difficult to include in more traditional pedagogical approaches. I also present a pipeline for how to incorporate actual biological data into the game world, and how this may help to reduce oversimplification of concepts in gamified settings.

Together, this work demonstrates that the development process of models provides a hereto underexploited method of theoretical collation, critique and understanding which has notable benefits for both academics and educators.