Research

My research generally falls into four areas:

Biological Model & Simulation Development

Obviously the end goal of a project designed to pair traditional experimental approaches with simulations that inform understanding and guide future experimentation is the development of the tool itself. Initially I developed a simulation of the organogenesis of lymphoid tissues in the gut, Peyer’s Patches, that trigger adaptive immune responses to infection, to answer a host of unanswered questions about this tissue (i.e. why we all have a different number, of different sizes, in different positions in the gut). Much of this work has been published in a number of biological and engineering-focused journals now. This work, in detail on the York Computational Immunology Lab website, is all set in the context of developing simulations through using a principled approach that ensures development is transparent and the generated predictions are grounded in the real-world domain. My Frontiers in Immunology paper describes that process. All this work is conducted in close collaboration with biologists with specific expertise in the system being captured.

I have since been involved in the development of numerous other simulation case studies with our PhD students, that capture immune response to Leishmaniasis, mechanisms of Inflammatory Bowel Diseases, and understanding Diabetes, amongst others. All current case studies use the modelling approach detailed in the Frontiers in Immunology paper. Much of this work, and my previous research, focuses on the use of agent-based modelling approaches that permit the modelling of time, space, and differences between individuals, although we also examine the use of hybrid models that include differential equation models to capture some behaviours or attributes of those individuals.

 

Evidencing and Decision Making

In developing the initial simulation of Peyer’s Patches, I became interested in how the design decisions taken while generating as simulation of a biological system were documented and justified. For example, not all the system is understood, thus assumptions need to be made, or a simulation may capture a disease in a specific mouse strain, or simplifications may have to be made to ensure the system can be captured on a computer. This led us to investigate the use of structured argumentation that is applied in the field of safety-critical systems, where software undergoes a rigorous assessment, supported by all available evidence, before that software is approved for use (in say a nuclear power plant or aircraft navigation system). In our Royal Society Interface paper we show how we have adapted this technique such that it can be used to show that a simulation of a biological system is fit for the purpose for which it has been designed. Specifically, the argumentation structure we use is an adapted form of goal-structuring notation (GSN). We released a freely available tool, Artoo, that can be used to develop these arguments. Since then we have published a further tutorial in CPT Systems Pharmacology and Pharmacometrics that describes how argumentation can be used throughout the process of simulation development.

Recently York-based spin-out company SimOmics have developed this approach further, and have now released Artoo Pro for commercial use. I am part of a consultancy agreement with SimOmics and a large multi-national product manufacturer that is rolling this technique out into their model development pipeline.

In addition I am also interested in characterising and quantifying the uncertainties that are introduced in all phases of model and simulation development, such that predictions can be accompanied by a measure of how certain we are in that result.

 

 

Performance & Statistical Analysis

When I came to performing a statistical, or sensitivity, analysis of the Peyer’s Patch simulation, we realised that there did not exist a software package that could perform such analyses for a given set of simulation results. This may explain why some of the analyses can be brought into question, as a reliable result depends on the correct statistical test and the attributes of the simulator (for example, if it is stochastic, do we understand how many replicate runs are needed to ensure a result is gained that is representative of that parameter set?). As such I initiated and continue to lead the development spartan, a freely available R package that aims to provide a generic platform for performing sensitivity analysis of simulation results. There are currently five statistical analysis techniques in spartan, with more planned, all of which are discussed in detail on the York Computational Immunology Lab website, in our paper in PLoS Computational Biology, and tutorial in the R journal. The tool is supported by detailed manuals and example data from the Peyer’s Patch simulator, and has to date attracted around 3,500 downloads.

Spartan is under constant development, with a major release of new functionality and online interface expected for release in 2017.

During her three month PhD rotation at GSK, one of our PhD students, Stephanie Evans, found a key deficiency in tools for performing sensitivity analysis of models developed using a popular model description language (Systems Biology Markup Language), where an intervention such as a drug needs to be introduced to the model. As such Steph and I, in collaboration with GSK, developed ASPASIA, that enables a robust statistical analyses of SBML models where an intervention is required. Our recent publication in PLoS Computational Biology describes the use of this tool and demonstrates application with a key biological case study.

I am also examining the application of evolutionary algorithms in the calibration of simulation parameters (recent paper in Royal Society Interface), and the use of machine learning approaches to examine the issues in performance that can be incurred with complex individual based simulations.

 

Wider Application

Much of my work has been conducted with strong links to the biological and immunological communities. However there is plenty of potential for the application of robust, well analysed, models and simulations outside of these realms. As such I am working to broaden the application of our work, and have recently began working with researchers in the Centre for Health Economics, with the hope of finding areas where we can exploit our model development expertise within studies of cost-effectiveness of healthcare interventions, while sharing expertise that could in turn develop our approaches further.  I have also worked with Social Scientists and would be keen to discuss potential projects in other disciplines.