Matter evolved under the influence of gravity from minuscule density fluctuations. Nonperturbative structure formed hierarchically over all scales and developed non-Gaussian features in the Universe, known as the cosmic web. To fully understand the structure formation of the Universe is one of the holy grails of modern astrophysics. Astrophysicists survey large volumes of the Universe and use a large ensemble of computer simulations to compare with the observed data to extract the full information of our own Universe. However, to evolve billions of particles over billions of years, even with the simplest physics, is a daunting task. We build a deep neural network, the Deep Density Displacement Model ([Formula: see text]), which learns from a set of prerun numerical simulations, to predict the nonlinear large-scale structure of the Universe with the Zel'dovich Approximation (ZA), an analytical approximation based on perturbation theory, as the input. Our extensive analysis demonstrates that [Formula: see text] outperforms the second-order perturbation theory (2LPT), the commonly used fast-approximate simulation method, in predicting cosmic structure in the nonlinear regime. We also show that [Formula: see text] is able to accurately extrapolate far beyond its training data and predict structure formation for significantly different cosmological parameters. Our study proves that deep learning is a practical and accurate alternative to approximate 3D simulations of the gravitational structure formation of the Universe.

Galaxy redshift surveys, such as the 2-Degree-Field Survey (2dF) 1, Sloan Digital Sky Survey (SDSS) 2, 6-Degree-Field Survey (6dF) 3, Galaxy And Mass Assembly survey (GAMA) 4 and VIMOS Public Extragalactic Redshift Survey (VIPERS) 5, have shown that the spatial distribution of matter forms a rich web, known as the cosmic web 6 . Most galaxy survey analyses measure the amplitude of galaxy clustering as a function of scale, ignoring information beyond a small number of summary statistics. Because the matter density field becomes highly non-Gaussian as structure evolves under gravity, we expect other statistical descriptions of the field to provide us with additional information. One way to study the non-Gaussianity is to study filaments, which evolve non-linearly from the initial density fluctuations produced in the primordial Universe. In our study, we report the detection of lensing of the cosmic microwave background (CMB) by filaments, and we apply a null test to confirm our detection. Furthermore, we propose a phenomenological model to interpret the detected signal, and we measure how filaments trace the matter distribution on large scales through filament bias, which we measure to be around 1.5. Our study provides new scope to understand the environmental dependence of galaxy formation. In the future, the joint analysis of lensing and Sunyaev-Zel'dovich observations might reveal the properties of 'missing baryons', the vast majority of the gas that resides in the intergalactic medium, which has so far evaded most observations.

Deep learning is a promising tool to determine the physical model that describes our universe. To handle the considerable computational cost of this problem, we present CosmoFlow: a highly scalable deep learning application built on top of the TensorFlow framework. CosmoFlow uses efficient implementations of 3D convolution and pooling primitives, together with improvements in threading for many element-wise operations, to improve training performance on Intel® Xeon Phi™ processors. We also utilize the Cray PE Machine Learning Plugin for efficient scaling to multiple nodes. We demonstrate fully synchronous data-parallel training on 8192 nodes of Cori with 77% parallel efficiency, achieving 3.5 Pflop/s sustained performance. To our knowledge, this is the first large-scale science application of the TensorFlow framework at supercomputer scale with fully-synchronous training. These enhancements enable us to process large 3D dark matter distribution and predict the cosmological parameters ΩsubM/sub, σsub8/sub and nsubs/sub with unprecedented accuracy.

Single-cell genomics enables dissection of tumor heterogeneity and molecular underpinnings of drug response at an unprecedented resolution^1-11. However, broad clinical application of these methods remains challenging, due to several practical and preanalytical challenges that are incompatible with typical clinical care workflows, namely the need for relatively large, fresh tissue inputs. In the present study, we show that multimodal, single-nucleus (sn)RNA/T cell receptor (TCR) sequencing, spatial transcriptomics and whole-genome sequencing (WGS) are feasible from small, frozen tissues that approximate routinely collected clinical specimens (for example, core needle biopsies). Compared with data from sample-matched fresh tissue, we find a similar quality in the biological outputs of snRNA/TCR-seq data, while reducing artifactual signals and compositional biases introduced by fresh tissue processing. Profiling sequentially collected melanoma samples from a patient treated in the KEYNOTE-001 trial¹², we resolved cellular, genomic, spatial and clonotype dynamics that represent molecular patterns of heterogeneous intralesional evolution during anti-programmed cell death protein 1 therapy. To demonstrate applicability to banked biospecimens of rare diseases¹³, we generated a single-cell atlas of uveal melanoma liver metastasis with matched WGS data. These results show that single-cell genomics from archival, clinical specimens is feasible and provides a framework for translating these methods more broadly to the clinical arena.