UC San Diego

Recent developments in the field of quantitative biology have demonstrated that genetic networks rely upon information encoded in their temporal dynamics, rather than beginning and ending steady-states, to govern their behavior. However, until now, there has been no tool with which to continuously observe genome-wide transcriptional dynamics without terminating the subject population. In response to this need, we developed DynOMICS, a total microfluidic and machine learning system that can monitor the state of gene expression across the E. coli genome in real time. We demonstrate its effectiveness as a field-deployable sensor, showing that it can learn the dynamic genomic signatures of heavy metal stress in both actual urban waters from several American cities and in samples from a toxic mining spill. By harnessing the microfluidics to a state-of-the-art deep neural network and an associated explanatory artificial intelligence (XAI) algorithm, we demonstrate its potential as a scientific instrument. We show that, in combination with DynOMICS, we can use deep learning networks to learn and understand bacterial transcriptional dynamics on a genome-scale. The combination of advanced microfluidics and AI-XAI is the first of its kind and is a powerful tool for quantitatively interrogating the E. coli genome.