Biomimetic simulation permits neuroscientists to better understand the complex neuronal dynamics of the brain. Embedding a biomimetic simulation in a closed-loop neuroprosthesis, which can read and write signals from the brain, will permit applications for amelioration of motor, psychiatric, and memory-related brain disorders. Biomimetic neuroprostheses require real-time adaptation to changes in the external environment, thus constituting an example of a dynamic data-driven application system. As model fidelity increases, so does the number of parameters and the complexity of finding appropriate parameter configurations. Instead of adapting synaptic weights via machine learning, we employed major biological learning methods: spike-timing dependent plasticity and reinforcement learning. We optimized the learning metaparameters using evolutionary algorithms, which were implemented in parallel and which used an island model approach to obtain sufficient speed. We employed these methods to train a cortical spiking model to utilize macaque brain activity, indicating a selected target, to drive a virtual musculoskeletal arm with realistic anatomical and biomechanical properties to reach to that target. The optimized system was able to reproduce macaque data from a comparable experimental motor task. These techniques can be used to efficiently tune the parameters of multiscale systems, linking realistic neuronal dynamics to behavior, and thus providing a useful tool for neuroscience and neuroprosthetics.

NAMD (NAnoscale Molecular Dynamics) is a parallel molecular dynamics application that has been used to make breakthroughs in understanding the structure and dynamics of large biomolecular complexes, such as viruses like HIV and various types of influenza. State-of-the-art biomolecular simulations often require integration of billions of timesteps, computing all interatomic forces for each femtosecond timestep. Molecular dynamics simulation of large biomolecular systems and long-timescale biological phenomena requires tremendous computing power. NAMD harnesses the power of thousands of heterogeneous processors to meet this demand. In this paper, we present algorithm improvements and performance optimizations that enable NAMD to achieve high performance on the IBM Newell platform (with POWER9 processors and NVIDIA Volta V100 GPUs) which underpins the Oak Ridge National Laboratory's Summit and Lawrence Livermore National Laboratory's Sierra supercomputers. The Top-500 supercomputers June 2018 list shows Summit at the number one spot with 187 Petaflop/s peak performance and Sierra third with 119 Petaflop/s. Optimizations for NAMD on Summit include: data layout changes for GPU acceleration and CPU vectorization, improving GPU offload efficiency, increasing performance with PAMI support in Charm++, improving efficiency of FFT calculations, improving load balancing, enabling better CPU vectorization and cache performance, and providing an alternative thermostat through stochastic velocity rescaling. We also present performance scaling results on early Newell systems.

An increasing number of people use mobile devices to monitor their behavior, such as exercise, and record their health status, such as psychological stress. However, these devices rarely provide ongoing support to help users understand how their behavior contributes to changes in their health status. To address this challenge, we aim to develop an interpretable policy for physical activity recommendations that reduce a user's perceived psychological stress, over a given time horizon. We formulate this problem as a sequential decision-making problem and solve it using a new method that we refer to as threshold Q-learning (TQL). The advantage of the TQL method over traditional Q-learning is that it is "doubly robust" and interpretable. This interpretability is achieved by making model assumptions and incorporating threshold selection into the learning process. Our simulation results indicate that the TQL method performs better than the Q-learning method given model misspecification. Our analyses are performed on data collected from 79 healthy adults over a 7 week period, where the data comprise physical activity patterns collected from mobile devices and self-assessed stress levels of the users. This work serves as a first step toward a computational health coaching solution for mobile device users.

Pharmit (http://pharmit.csb.pitt.edu) is an open-source online resource that allows users to interactively search libraries of millions compounds as part of a structure-based drug discovery workflow. Here we describe the systems-level implementation decisions made in designing Pharmit that, when combined with novel sub-linear time search algorithms, allow it to screen millions of molecules in seconds. The key concepts are to maximize parallelism while minimizing intra-thread communication, optimize data layout for sequential processing, and efficiently manage memory allocation. We describe how these concepts are applied to the cheminformatic data inherent to Pharmit and discuss limitations and possible future directions.