Volume 1

  • No. 8 August 2021

    Machine learning with scalable optical computing

    Optical computing offers advantages such as high-speed calculations and relatively low energy consumption. However, nonlinear information processing with optics still remains a challenging task. In this issue, Uğur Teğin et al. demonstrates a scalable and energy-efficient optical computing framework to perform machine learning tasks with optical fibers. The reported optical computing method substantially reduces the energy cost while maintaining comparable accuracy with its digital counterparts.

    See Uğur Teğin et al.

  • No. 7 July 2021

    Efficient protein model refinement with deep learning

    Protein refinement methods, which are used to improve the quality of protein structural models, commonly rely on extensive conformational sampling, and therefore, are very time-consuming. In this issue, Xiaoyang Jing and Jinbo Xu propose a method that uses graph neural networks to substantially reduce the time taken to refine protein models, from hours to minutes on a single CPU, while having comparable accuracy with the leading approaches in the field.

    SeeXiaoyang Jing and Jinbo XuandPhilip Kim

  • No. 6 June 2021

    Quantum advantage for neural networks

    Quantum computing offers promises of increasing computational efficiency and capability. However, there are still debates on whether and how this quantum advantage can be achieved. In this issue, Amira Abbas et al. showcases that a well-designed quantum neural network can learn a broader class of functions without compromising accuracy when compared to its classical counterparts.

    See Amira Abbas et al. and Patrick J. Coles

  • No. 5 May 2021

    Implementing digital twins at scale

    A digital twin is a complex computational model (or a set of coupled computational models) that continually receives and integrates data from a physical entity (for instance, an aircraft) to provide an up-to-date digital representation of that entity. The digital twin paradigm has seen significant interest across a range of application areas as a way to support data-driven decision making, but most implementations are custom-based, which makes it challenging to deploy them at scale. In this issue, Niederer et al. discuss challenges and opportunities for scaling digital twins, and Kapteyn et al. propose a mathematical representation of asset-twin systems as a first step to enable digital twins at scale.

    See Niederer et al. and Kapteyn et al.

  • No. 4 April 2021

    Protecting against racial profiling in DNA databases

    Criminal DNA databases at national, state and city level are used to search for crime perpetrators, and DNA profiles from as many plausible suspects as possible are often collected to look for a match. However, these profiles may be retained in these databases, and because crime correlates with socio-economic status, geographic location and ethnicity, this results in racial genetic profiling. In this issue, Blindenbach et al. propose a solution to protect the privacy of these DNA profiles, ensuring that criminal databases learn nothing about them.

    See Blindenbach et al. and Syndercombe Court

  • No. 3 March 2021

    Colocalization for high-resolution microscopy

    The cover depicts an analysis of spatial proximity, or colocalization, of the mitochondrial outer membrane protein Tom40 (blue) and the mitochondrial ribosomal protein Mrpl4 (red) in yeast cells. It displays the geodesic (shortest path) transport to match both protein distributions in the spatially most efficient way. The identification of spatial colocalization is a powerful approach to characterize the complex interactions of biological macromolecules.

    See Tameling et al. and Wang and Yuan

  • No. 2 February 2021

    The future of Earth-system modeling

    Weather and climate prediction has achieved steady progress over the past few decades thanks to advances in computational science. Nevertheless, our Earth-system models must adapt, and fast, to the explosion of data challenges and to future computing architectures. In this issue, Bauer et al. discuss the current limitations in the field and potential solutions to best exploit what new digital technologies have to offer.

    See Bauer et al.

  • No. 1 January 2021

    Predicting properties of materials using graph networks

    While predicting the properties of materials is an important goal in materials science, the scarcity of high-fidelity models has made this goal challenging. Chi Chen et al. propose a graph network model that can accurately predict high-fidelity properties of ordered and disordered materials by using data with varying fidelities. The current issue also features additional content in computational materials science, including the work by Rahul Singh et al., which implements an optimization algorithm to efficiently model high-entropy alloys, and a Comment by Matthew Horton et al. on the current challenges and opportunities in the development of computational materials databases.

    See Chen et al., Singh et al. and Horton et al.