Recent advances in machine learning have leveraged evolutionary information in multiple sequence alignments to predict protein structure. We demonstrate direct inference of full atomic-level protein structure from primary sequence using a large language model. As language models of protein sequences are scaled up to 15 billion parameters, an atomic-resolution picture of protein structure emerges in the learned representations. This results in an order-of-magnitude acceleration of high-resolution structure prediction, which enables large-scale structural characterization of metagenomic proteins. We apply this capability to construct the ESM Metagenomic Atlas by predicting structures for >617 million metagenomic protein sequences, including >225 million that are predicted with high confidence, which gives a view into the vast breadth and diversity of natural proteins.

No abstract

No abstract

Project Loon has been launching superpressure balloons since January 2013 to provide worldwide Internet coverage. These balloons typically fly between 18 and 21 km and provide measurements of winds and pressure fluctuations in the lower stratosphere. We divide 1560 Loon flights into 3405 two‐day segments for gravity wave analysis. We derive the kinetic energy spectrum from the horizontal balloon motion and estimate the temperature perturbation spectrum (proportional to the potential energy spectrum) from the pressure variations. We fit the temperature (and kinetic energy) data to the functional form T′2 = T′o2[ω/ωο)α, where ω is the wave frequency, ωο is daily frequency, T′o is the base temperature amplitude, and α is the spectral slope. Both the kinetic energy and temperature spectra show −1.9 ± 0.2 power‐law dependence in the intrinsic frequency window 3–50 cycles/day. The temperature spectrum slope is weakly anticorrelated with the base temperature amplitude. We also find that the wave base temperature distribution is highly skewed. The tropical modal temperature is 0.77 K. The highest amplitude waves occur over the mountainous regions, the tropics, and the high southern latitudes. Temperature amplitudes show little height variation over our 18–21 km domain. Our results are consistent with other limited superpressure balloon analyses. The modal temperature is higher than the temperature currently used in high‐frequency gravity wave parameterizations.

We address the problem of surveillance in an environment with obstacles. We show that the problem of tracking an evader with one pursuer around one corner is completely decidable. The pursuer and the evader have complete information about each other's instantaneous position. The pursuer has complete information about the instantaneous velocity of the evader. We present a partition of the visibility region of the pursuer where based on the region in which the evader lies, we provide strategies for the evader to escape the visibility region of the pursuer or for the pursuer to track the target for all future time. We also present the solution to the inverse problem: given the position of the evader, the positions of the pursuer for which the evader can escape the visibility region of the target. These results have been provided for varying speeds of the pursuer and the evader. Based on the results of the inverse problem we provide an O(n 3 log n) algorithm that can decide if the evader can escape from the visibility region of a pursuer for some initial pursuer and evader positions. Finally, we extend the result of the target tracking problem around a corner in two dimensions to an edge in three dimensions.