The Parameter Optimisation Problem: Addressing a Key Challenge in Computational Systems Biology

Lead Research Organisation: University of Edinburgh
Department Name: Sch of Biological Sciences


In recent years, advances in both the physical and life sciences have increasingly come from the collaborations of researchers across disciplines, and the development and use of tools from a range of areas. A prototypical example of this interdisciplinary approach to science is systems biology, the field concerned with quantifying how the interaction of individual system components control biological function and behaviour. Systems biology has become increasingly quantitative, with a shift from diagrammatic representations of interaction networks to sets of mathematical equations that model (i.e. simulate) how the concentrations of molecular species vary with time. A key advantage of such models is that they can be used to predict how the networks they represent will respond to specific perturbations, such as changes in environmental conditions (e.g. temperature) or the addition of pharmacological agents. The ability to easily generate such predictions reduces the need for large numbers of expensive and time-consuming experiments.

However, the more complex a biological network is, the more complex the corresponding model needs to be, and the greater the range of possible biological behaviours that can be exhibited. This means that extensive computer simulations are needed to adjust the parameters controlling the model so as to accurately reproduce (i.e. fit) the experimental behaviour observed. For biologically realistic models which can involve hundreds of different molecular species, the number of simulations required to adjust the parameters of a given model to achieve the optimal fit to data can be prohibitively large, far exceeding that which is possible on practical timescales. Thus, for the predictive power of mathematical models to be fully realised in the systems biology domain, methods are required that allow this parameter optimisation procedure to be carried out in a computationally efficient manner.

The proposed project will address this need by bringing state-of-the-art methods from computer science to bear on the problem, which have been successfully applied previously to highly parametrised problems like aircraft conflict alert systems, design optimisation of lightweight materials and routing of mesh sensor networks (amongst others). In addition, we propose to develop new methods specifically engineered for the systems biology domain that can provide insight into model behaviour, beyond simply returning a single estimate of the best fit parametrisation (e.g. methods for identifying parameters yielding equally good fits to data, and also parameters which simultaneously fit the model to data generated in diverse experimental conditions). As part of this, we will develop a package of open source software tools that will be embedded within a software infrastructure designed for systems biologists, enabling the methods developed in this work to be readily applied to problems in the field that are currently computationally intractable.

To test and refine the algorithms developed, they will be applied to the gene network that generates circadian oscillations (the circadian clock) in the key plant species Arabidopsis thaliana, for which high-quality experimental data recorded in a range of genetic and environmental backgrounds is available, together with a suite of mathematical models of varying complexity. As part of this work, biochemically detailed models of the clock will be directly fitted to multiple experimental datasets for the first time, yielding models with greater predictive power. Many processes critical for plant growth and reproduction are regulated by the clock (e.g. photosynthesis and flowering time). In the long term, the ability to optimise plant models of increasing complexity with the class of methods we will develop here may thus help predict how the viability of economically important crop species will be affected by future temperature shifts resulting from climate change.


10 25 50
Description Mathematical research methods were updated by our collaborators in the Exeter part of the award, using rapid prototyping methods in proprietary software.
Our team kept abreast of their progress and provided biological information, example models and explained appropriate modelling conditions.
The Edinburgh team, as proposed, has made the key parts of this software publicly available in open-source, non-proprietary forms that are commonly used in the modelling community, specifically refactoring and benchmarking the NMMSO optimiser in the python language, as "pynmmso", Niching Migratory Multi-Swarm Optimiser for Python.
The tool is packaged for ease of integration with biological modelling environments such as the worked example that we provide, using the python bindings for libRoadRunner ( libRoadRunner is a high performance and portable simulation engine for systems and synthetic biology and can simulate models in SBML format, from the Sauro lab, U. of Washington.
Exploitation Route This mathematical method from Exeter is highly relevant to the increasingly challenging task of calibrating larger biological models to data. However, it will also apply to other models that produce complex optimisation landscapes, in other words where the model works well with several (or many) different parameter sets and it is challenging to find the best set.
Sectors Agriculture, Food and Drink,Digital/Communication/Information Technologies (including Software),Pharmaceuticals and Medical Biotechnology

Description Interdisciplinary projects such as this one between Biology, Computer Science and Mathematics, provide challenging training for already highly-skilled research personnel. In this award, biologist Millar and computer scientist Hume extended their understanding of model optimisation to the novel mathematical methods provided by our collaborators in Exeter, and software developers Neelofer and Wood gained their first exposure to biological modelling. Training highly-skilled people is a key output of discovery research, as they go on to impact society throughout their lives.
First Year Of Impact 2019
Title Niching Migratory Multi-Swarm Optimiser for Python, pynmmso 
Description pynmmso is a python implementation of the Niching Migratory Multi-Swarm Optimser, described in: "Running Up Those Hills: Multi-Modal Search with the Niching Migratory Multi-Swarm Optimiser" by Jonathan E. Fieldsend published in Proceedings of the IEEE Congress on Evolutionary Computation, pages 2593-2600, 2014 ( Examples of the method in action can be obtained from: Code to run benchmark problems using pynmmso can be found at: The Python implementation of NMMSO requires Python 3 and Numpy ( One of the examples illustrates the optimisation of an SBML model of the Neurospora circadian oscillator, using simulations by libRoadRunner from the Sauro group. 
Type Of Technology Software 
Year Produced 2019 
Open Source License? Yes  
Impact The software has just been released.