Independent Component Analysis for the Aggregation of Stochastic Simulation Output

Computational models and simulation algorithms are commonly applied tools in biological sciences. Among those, discrete stochastic models and stochastic simulation proved to be able to effectively capture the effects of intrinsic noise at molecular level, improving over deterministic approaches when system dynamics is driven by a limited amount of molecules. A challenging task that is offered to researchers is then the analysis and ultimately the inference of knowledge from a set of multiple, noisy, simulated trajectories. We propose in this paper a method, based on Independent Component Analysis (ICA), to automatically analyze multiple output traces of stochastic simulation runs. ICA is a statistical technique for revealing hidden factors that underlie sets of signals. Its applications span from digital image processing, to audio signal reconstruction and economic indicators analysis. Here we propose the application of ICA to identify and describe the noise in time-dependent evolution of biochemical species and to extract aggregate knowledge on simulated biological systems. We present the results obtained with the application of the proposed methodology on the well-known MAPK cascade system, which demonstrate the ability of the proposed methodology to decompose and identify the noisy components of the evolution. Quantitative descriptions of the noise component can be further analytically characterized by a simple first order autoregressive model.