Methodology for identifying alternative solutions in a population based data generation approach applied to synthetic biology

Abstract

Design is an essential component of sustainable development. Computational modelling has become a useful technique that facilitates the design of complex systems. Variables that characterises a complex system are encoded into a computational model using mathematical concepts and through simulation each of these variables alone or in combination are modified to observe the changes in the outcome. This allows the researchers to make predictions on the behaviour of the real system that is being studied in response to the changes. The ultimate goal of any design process is to come up with the best design; as resources are limited, to minimize the cost and resource consumption, and to maximize the performance, profits and efficiency. To optimize means to find the best solution, the best compromise among several conflicting demands subject to predefined requirements. Therefore, computational optimization, modelling and simulation forms an integrated part of the modern design practice. This thesis defines a data analytics driven methodology which enables the identification of alternative solutions of computational design by analysing the generational history of the population based heuristic search used to generate the templates. While optimisation is focused on obtaining the optimal solution this methodology focuses on alternative solutions which are sub optimal by fitness or solutions with similar fitness but different structures. When the optimal design solution is less robust, alternative solutions can offer a sufficiently good accuracy and an achievable resource requirement. The main advantage of the methodology is that it exploits the exploration process of the solution space during a single run, by focusing also on suboptimal solutions, which usually get neglected in the search for an optimal one. The history of the heuristic search is analysed for the emergence of alternative solutions and evolving of a solution. By examining how an initial solution converts to an optimal solution core design patterns are identified, and these were used to improve the design process. Further, this method limits the number of runs of the heuristic search as more solution space is covered. The methodology is generic because it can be used to any instance where a population based heuristic search is applied to generate optimal designs. The applicability of the methodology is demonstrated using three case studies from mathematics (building of a mathematical function for a set target) and biology (obtaining alternative designs for genomic metabolic models [GEM] and DNA walker circuits). In each case a different heuristic search method was used: Gene expression programming (mathematical expressions), genetic algorithms (GEM models) and simulated annealing (DNA walker circuits). Descriptive analytics, visual analytics and clustering was mainly used to build the data analytics driven approach in identifying alternative solutions. This data analytics driven methodology is useful in optimising the computational design of complex systems.