Automated deep abstractions for stochastic chemical reaction networks

Predicting stochastic cellular dynamics emerging from chemical reaction networks (CRNs) is a long-standing challenge in systems biology. Deep learning was recently used to abstract the CRN by mixture density neural network, trained with traces of original process. Such abstraction dramatically cheaper execute, yet it preserves statistical features training data. However, practice, modeller has take care finding suitable network architecture manually, for each given CRN, through trial-and-error cycle. In this paper, we propose further automatise deep abstractions CRNs, along transition kernel The method applicable any time-saving experts and crucial non-specialists. We demonstrate performance over number CRNs multi-modal phenotypes multi-scale scenario where interact across spatial grid.