A machine learning approach for generating temporal logic classifications of complex model behaviours

Systems biology aims to facilitate the understanding of complex interactions between components in biological systems. Petri nets (PN), and in particular Coloured Petri Nets (CPN) have been demonstrated to be a suitable formalism for modelling biological systems and building computational models over multiple spatial and temporal scales. To explore the complex and high-dimensional solution space over the behaviours generated by such models, we propose a clustering methodology which combines principal component analysis (PCA), distance similarity and density factors through the application of DBScan. To facilitate the interpretation of clustering results and enable further analysis using model checking we apply a pattern mining approach aimed at generating high-level classificatory descriptions of the clusters' behaviour in temporal logic. We illustrate the power of our approach through the analysis of two case studies: multiple knockdown of the Mitogen-activated protein-kinase (MAPK) pathway, and selective knockout of Planar Cell Polarity (PCP) signalling in Drosophila wing.