Search

The most straightforward method to increase monoclonal antibody (mAb) product yield is to complete the purification process in less steps. Here, three different fiber chromatographic devices were implemented using a holistic approach to intensify the mAb purification process and increase yield. Fiber protein A (proA) chromatography was first investigated, but traditional depth filtration was not sufficient in reducing the contaminant load as the fiber proA device prematurely fouled. Further experimentation revealed that chromatin aggregates were the most likely reason for the fiber fouling. To reduce levels of chromatin aggregates, a chromatographic clarification device (CCD) was incorporated into the process, resulting in single‐stage clarification of harvested cell culture fluid and reduction of DNA levels. The CCD clarified pool was then successfully processed through the fiber proA device, fully realizing the productivity gains that the fiber technology offers. After the proA and viral inactivation neutralization (VIN) hold step, the purification process was further intensified using a novel single‐use fiber‐based polishing anion exchange (AEX) material that is capable of binding both soluble and insoluble contaminants. The three‐stage fiber chromatographic purification process was compared to a legacy five‐step process of dual‐stage depth filtration, bead‐based proA chromatography, post‐VIN depth filtration, and bead‐based AEX chromatography. The overall yield from the five‐step process was 60%, while the fiber chromatographic‐enabled intensified process had an overall yield of 70%. The impurity clearance of DNA and host cell protein (HCP) for both processes were within the regulatory specification (<100 ppm HCP, <1 ppb DNA). For the harvest of a 2000 L cell culture, the intensified process is expected to increase productivity by 2.5‐fold at clarification, 50‐fold at the proA step, and 1.6‐fold in polishing. Relative to the legacy process, the intensified process would reduce buffer use by 1088 L and decrease overall process product mass intensity by 12.6%. [ABSTRACT FROM AUTHOR]

The use of big data in sport and sport management research is increasing in popularity. Prior research generally includes one of the many characteristics of big data, such as volume or velocity. The present study presents big data in a multidimensional lens by considering the use of sentiment analysis. Specifically focusing on the phenomenon of tanking, the purposeful underperformance in sport competitions, the present study considers the impact that consumers' sentiment regarding tanking has on game attendance in the National Basketball Association. Collecting social media posts for each National Basketball Association team, the authors create an algorithm to measure the volume and sentiment of consumer discussions related to tanking. These measures are included in a predictive model for National Basketball Association home game attendance between the 2013–2014 and 2017–2018 seasons. Our results find that the volume of discussions for the home team and sentiment toward tanking by the away team impact game attendance. [ABSTRACT FROM AUTHOR]

The term "dataflow" generally encompasses three distinct aspects of computation - a data-driven model of computation, a functional/declarative programming language, and a special-purpose multiprocessor architecture. In this paper we decouple the language and architecture issues by demonstrating that declarative programming is a suitable vehicle for the programming of conventional distributed-memory multiprocessors.This is achieved by appling several transformations to the compiled declarative program to achieve iteration-level (rather than instruction-level) parallelism. The transformations first group individual instructions into sequential light-weight processes, and then insert primitives to: (1) cause array allocation to be distributed over multiple processors, (2) cause computation to follow the data distribution by inserting an index filtering mechanism into a given loop and spawning a copy of it on all PEs; the filter causes each instance of that loop to operate on a different subrange of the index variable.The underlying model of computation is a dataflow/von Neumann hybrid in that exection within a process is control-driven while the creation, blocking, and activation of processes is data-driven.The performance of this process-oriented dataflow system (PODS) is demonstrated using the hydrodynamics simulation benchmark called SIMPLE, where a 19-fold speedup on a 32-processor architecture has been achieved.

The Process-Oriented Dataflow System (PODS) is an execution model that combines the von Neumann and dataflow models of computation to gain the benefits of each. Central to PODS is the concept of array distribution and its effects on partitioning and mapping of processes.In PODS arrays are partitioned by simply assigning consecutive elements to each processing element (PE) equally. Since PODS uses single assignment, there will be only one producer of each element. This producing PE owns that element and will perform the necessary computations to assign it. Using this approach the filling loop is distributed across the PEs. This simple partitioning and mapping scheme provides excellent results for executing scientific code on MIMD machines. In this way PODS allows MIMD machines to exploit vector and data parallelism easily while still providing the flexibility of MIMD over SIMD for multi-user systems.In this paper, the classic matrix multiply algorithm, with 1024 data points, is executed on a PODS simulator and the results are presented and discussed. Matrix multiply is a good example because it has several interesting properties: there are multiple code-blocks; a new array must be dynamically allocated and distributed; there is a loop-carried dependency in the innermost loop; the two input arrays have different access patterns; and the sizes of the input arrays are not known at compile time. Matrix multiply also forms the basis for many important scientific algorithms such as: LU decomposition, convolution, and the Fast-Fourier Transform.The results show that PODS is comparable to both Iannucci's Hybrid Architecture and MIT's TTDA in terms of overhead and instruction power. They also show that PODS easily distributes the work load evenly across the PEs. The key result is that PODS can scale matrix multiply in a near linear fashion until there is little or no work to be performed for each PE. Then overhead and message passing become a major component of the execution time. With larger problems (e.g., >/=16k data points) this limit would be reached at around 256 PEs.