"Building Better Batteries from the Ground Up"
Michael Counihan, chemistry, works with the Rodríguez-López Group
Storing energy from grid-scale renewable sources is critical for moving away from fossil fuels and addressing climate change. However, lithium ion batteries (LIBs) are not viable to use at large scales, and the processes involved in making LIBs present many safety and socioeconomic issues. To this end, the Rodríguez-López group focuses on new types of energy storage materials and studies them at the fundamental level to build smarter, safer, and more efficient batteries. I will focus on the non-aqueous redox flow battery (NRFB) as a new system for large scale energy storage. This battery, which stores charge in a liquid form instead of a solid form, has advantages over other batteries commercialized today because of the greater voltage and easier tailoring of the battery chemistry. However, charge stability, electrode degradation, and higher cost are big setbacks that need to be addressed for NRFBs to contribute to grid-scale energy storage. This talk will provide an overview of the techniques to study these limitations and the nature-inspired chemistries being developed to overcome them.
"Capturing the Dynamics of Proteins on Film Using 'Nanoaquariums' and Electron Microscopy"
Johnny Smith, materials science and engineering, works with the Qian Chen Group
Proteins — the molecular “machinery” of life — accomplish tasks ranging from chemical catalysis, to signal transduction, to cell locomotion through a complex interplay between their structure and structural dynamics at the nanoscale. However, it is challenging to investigate both aspects simultaneously, at the single-protein level, using currently available imaging techniques. For example, super-resolution optical microscopy can localize individual proteins with extreme precision in liquid media but generally does not reveal protein structure, whereas cryogenic electron microscopy can achieve even atomic resolution structural information, but requires immobilization of specimens in ice. Meanwhile, a new imaging tool known as liquid-phase electron microscopy has opened opportunities to image real-time, continuous dynamics in a liquid environment, and with the resolution of an electron microscope. Our goal, therefore, is to establish liquid-phase electron microscopy methods that enable imaging not just “hard,” inorganic materials, but “soft,” low-contrast, and even electron-sensitive biomolecules. I will discuss how we are using liquid-phase electron microscopy with graphene-based chamber designs, low electron dose imaging, and advanced image/video processing to investigate “single-particle” dynamics of biomolecules, such as membrane protein nanodiscs and biomolecular motors. We hope that these tools can soon help provide new insight into functionality in diverse biomolecular systems.
"Deep Learning for Optimizing Medical Imaging Systems"
Weimin Zhou, bioengineering, works with the Anastasio Group
The objective optimization of modern medical imaging systems and data-acquisition designs is often guided by task-based measures of image quality (IQ). Task-based measures of IQ quantify the performance of an observer on specific tasks (e.g., detection of tumors) for an ensemble of objects being imaged. When optimizing imaging systems and data-acquisition designs for signal detection tasks, the Bayesian Ideal Observer (IO) that sets an upper performance limit has been advocated. However, computation of the IO test statistic generally is analytically intractable and requires complete knowledge of the statistical information in the measurement images, including object variability. In order for imaging systems and data-acquisition designs to be optimized for signal detection tasks, it is desirable to develop methods to approximate the IO test statistic and establish a stochastic object model (SOM) that can produce numerous images that describe object variability. Our work explored deep learning methods that employ convolutional neural networks to approximate the IO test statistic for signal detection tasks and investigated the ability of an augmented generative adversarial network architecture named AmbientGAN to learn realistic SOMs from noisy and/or indirect experimental data. These deep learning methods enable the objective optimization of imaging systems and data-acquisition designs.