Search

The recently proposed Chemical Reaction Neural Network (CRNN) discovers chemical reaction pathways from time resolved species concentration data in a deterministic manner. Since the weights and biases of a CRNN are physically interpretable, the CRNN acts as a digital twin of a classical chemical reaction network. In this study, we employ a Bayesian inference analysis coupled with neural ordinary differential equations (ODEs) on this digital twin to discover chemical reaction pathways in a probabilistic manner. This allows for estimation of the uncertainty surrounding the learned reaction network. To achieve this, we propose an algorithm which combines neural ODEs with a preconditioned stochastic gradient langevin descent (pSGLD) Bayesian framework, and ultimately performs posterior sampling on the neural network weights. We demonstrate the successful implementation of this algorithm on several reaction systems by not only recovering the chemical reaction pathways but also estimating the uncertainty in our predictions. We compare the results of the pSGLD with that of the standard SGLD and show that this optimizer more efficiently and accurately estimates the posterior of the reaction network parameters. Additionally, we demonstrate how the embedding of scientific knowledge improves extrapolation accuracy by comparing results to purely data-driven machine learning methods. Together, this provides a new framework for robust, autonomous Bayesian inference on unknown or complex chemical and biological reaction systems. [ABSTRACT FROM AUTHOR]

Largely unknown just a few decades ago, computational systems biology is now a central methodology for biological and medical research. This amazing ascent raises the question of what the community should do next. The article outlines our personal vision for the future of computational systems biology, suggesting the need to address both mindsets and methodologies. We present this vision by focusing on current and anticipated research goals, the development of strong computational tools, likely prominent applications, education of the nextgeneration of scientists, and outreach to the public. In our opinion, two classes of broad research goals have emerged in recent years and will guide future efforts. The first goal targets computational models of increasing size and complexity, aimed at solving emerging health-related challenges, such as realistic whole-cell and organ models, disease simulators and digital twins, in silico clinical trials, and clinically translational applications in the context of therapeutic drug development. Such large models will also lead us toward solutions to pressing issues in agriculture and environmental sustainability, including sufficient food availability and life in changing habitats. The second goal is a deep understanding of the essence of system designs and strategies with which nature solves problems. This understanding will help us explain observed biological structures and guide forays into synthetic biological systems. Regarding effective methodologies, we suggest efforts toward automated data pipelines from raw biomedical data all the way to spatiotemporal mechanistic model. These will be supported by dynamic methods of statistics, machine learning, artificial intelligence and streamlined strategies of dynamic model design, striking a fine balance between modeling realistic complexity and abstracted simplicity. Finally, we suggest the need for a concerted, community-wide emphasis on effective education in systems biology, implemented as a combination of formal instruction and hands-on mentoring. The educational efforts should furthermore be extended toward the public through books, blogs, social media, and interactive networking opportunities, with the ultimate goal of training in state-of-the-art technology while recapturing the lost art of synthesis. [ABSTRACT FROM AUTHOR]