Continuous-time temporal logic specification and verification for nonlinear biological systems in uncertain contexts

Abstract

In this thesis we introduce a complete framework for modelling and verification of biological systems in uncertain contexts based on the bond-calculus process algebra and the LBUC spatio-temporal logic. The bond-calculus is a biological process algebra which captures complex patterns of interaction based on affinity patterns, a novel communication mechanism using pattern matching to express multiway interaction affinities and general kinetic laws, whilst retaining an agent-centric modelling style for biomolecular species. The bond-calculus is equipped with a novel continuous semantics which maps models to systems of Ordinary Differential Equations (ODEs) in a compositional way.

We then extend the bond-calculus to handle uncertain models, featuring interval uncertainties in their species concentrations and reaction rate parameters. Our semantics is also extended to handle uncertainty in every aspect of a model, producing non-deterministic continuous systems whose behaviour depends either on time-independent uncertain parameters and initial conditions, corresponding to our partial knowledge of the system at hand, or time-varying uncertain inputs, corresponding to genuine variability in a system’s behaviour based on environmental factors.

This language is then coupled with the LBUC spatio-temporal logic which combines Signal Temporal Logic (STL) temporal operators with an uncertain context operator which quantifies over an uncertain context model describing the range of environments over which a property must hold. We develop model-checking procedures for STL and LBUC properties based on verified signal monitoring over flowpipes produced by the Flow* verified integrator, including the technique of masking which directs monitoring for atomic propositions to time regions relevant to the overall verification problem at hand. This allows us to monitor many interesting nested contextual properties and frequently reduces monitoring costs by an order of magnitude. Finally, we explore the technique of contextual signal monitoring which can use a single Flow* flowpipe representing a functional dependency to complete a whole tree of signals corresponding to different uncertain contexts. This allows us to produce refined monitoring results over the whole space and to explore the variation in system behaviour in different contexts.