Fault Detection for Vaccine Refrigeration via Convolutional Neural Networks Trained on Simulated Datasets.

In low-and middle-income countries, the cold chain that supports vaccine storage and distribution is vulnerable due to insufficient infrastructure and interoperable data. To bolster these networks, we developed a convolutional neural network-based fault detection method for vaccine refrigerators using datasets synthetically generated by thermodynamic modelling. We demonstrate that these thermodynamic models can be calibrated to real cooling systems in order to identify system-specific faults under a diverse range of operating conditions. If implemented on a large scale, this portable, flexible approach has the potential to increase the fidelity and lower the cost of vaccine distribution in remote communities.

Ultrafast Polymer Dynamics through a Nanopore.

Ultrathin nanopore sensors allow single-molecule and polymer measurements at sub-microsecond time resolution enabled by high current signals (∼10-30 nA). We demonstrate for the first time the experimental probing of the ultrafast translocation and folded dynamics of double-stranded DNA (dsDNA) through a nanopore at 10 MHz bandwidth with acquisition of data points per 25 ns (150 MB/s). By introducing a rigorous algorithm, we are able to accurately identify each current level present within translocation events and elucidate the dynamic folded and unfolded behaviors. The remarkable sensitivity of this system reveals distortions of short-lived folded states at a lower bandwidth. This work revisits probing of dsDNA as a model polymer and develops broadly applicable methods. The combined improvements in sensor signals, instrumentation, and large data analysis methods uncover biomolecular dynamics at unprecedentedly small time scales.