Search for: All records

The recent development of spatially resolved transcriptomics (SRT) technologies has facilitated research on gene expression in the spatial context. Annotating cell types is one crucial step for downstream analysis. However, many existing algorithms use an unsupervised strategy to assign cell types for SRT data. They first conduct clustering analysis and then aggregate cluster-level expression based on the clustering results. This workflow fails to leverage the marker gene information efficiently. On the other hand, other cell annotation methods designed for single-cell RNA-seq data utilize the cell-type marker genes information but fail to use spatial information in SRT data.

We introduce a statistical spatial transcriptomics cell assignment model, SPAN, to annotate clusters of cells or spots into known types in SRT data with prior knowledge of predefined marker genes and spatial information. The SPAN model annotates cells or spots from SRT data using predefined overexpressed marker genes and combines a mixture model with a hidden Markov random field to model the spatial dependency between neighboring spots. We demonstrate the effectiveness of SPAN against spatial and nonspatial clustering algorithms through extensive simulation and real data experiments.

https://github.com/ChengZ352/SPAN.

 

The observed winter Barents-Kara Sea (BKS) sea ice concentration (SIC) has shown a close association with the second empirical orthogonal function (EOF) mode of Eurasian winter surface air temperature (SAT) variability, known as Warm Arctic Cold Eurasia (WACE) pattern. However, the potential role of BKS SIC on this WACE pattern of variability and on its long-term trend remains elusive. Here, we show that from 1979 to 2022, the winter BKS SIC and WACE association is most prominent and statistically significant for the variability at the sub-decadal time scale for 5–6 years. We also show the critical role of the multi-decadal trend in the principal component of the WACE mode of variability for explaining the overall Eurasian winter temperature trend over the same period. Furthermore, a large multi-model ensemble of atmosphere-only experiments from 1979 to 2014, with and without the observed Arctic SIC forcing, suggests that the BKS SIC variations induce this observed sub-decadal variability and the multi-decadal trend in the WACE. Additionally, we analyse the model simulated first or the leading EOF mode of Eurasian winter SAT variability, which in observations, closely relates to the Arctic Oscillation (AO). We find a weaker association of this mode to AO and a statistically significant positive trend in our ensemble simulation, opposite to that found in observation. This contrasting nature reflects excessive hemispheric warming in the models, partly contributed by the modelled Arctic Sea ice loss.

 

Key Points The external radiative forcing is the primary driver of the 1979–2013 warming for April–September, with varying decadal warming rates The interdecadal Pacific and Atlantic multidecadal variability intensify/dampen the warming when transitioning to positive/negative phase The combined effects of these factors reproduce the observed varied pace of decadal Arctic troposphere warming during 1979–2013 

Abstract. The main drivers of the continental Northern Hemisphere snow cover are investigated in the 1979–2014 period. Four observational datasets are usedas are two large multi-model ensembles of atmosphere-only simulations with prescribed sea surface temperature (SST) and sea ice concentration (SIC). Afirst ensemble uses observed interannually varying SST and SIC conditions for 1979–2014, while a second ensemble is identical except for SIC witha repeated climatological cycle used. SST and external forcing typically explain 10 % to 25 % of the snow cover variance in modelsimulations, with a dominant forcing from the tropical and North Pacific SST during this period. In terms of the climate influence of the snow coveranomalies, both observations and models show no robust links between the November and April snow cover variability and the atmospheric circulation1 month later. On the other hand, the first mode of Eurasian snow cover variability in January, with more extended snow over western Eurasia, isfound to precede an atmospheric circulation pattern by 1 month, similar to a negative Arctic oscillation (AO). A decomposition of the variabilityin the model simulations shows that this relationship is mainly due to internal climate variability. Detailed outputs from one of the modelsindicate that the western Eurasia snow cover anomalies are preceded by a negative AO phase accompanied by a Ural blocking pattern and astratospheric polar vortex weakening. The link between the AO and the snow cover variability is strongly related to the concomitant role of thestratospheric polar vortex, with the Eurasian snow cover acting as a positive feedback for the AO variability in winter. No robust influence of theSIC variability is found, as the sea ice loss in these simulations only drives an insignificant fraction of the snow cover anomalies, with fewagreements among models. 

Abstract Large ensemble simulations with six atmospheric general circulation models involved are utilized to verify the interdecadal Pacific oscillation (IPO) impacts on the trend of Eurasian winter surface air temperatures (SAT) during 1998–2013, a period characterized by the prominent Eurasia cooling (EC). In our simulations, IPO brings a cooling trend over west-central Eurasia in 1998–2013, about a quarter of the observed EC in that area. The cooling is associated with the phase transition of the IPO to a strong negative. However, the standard deviation of the area-averaged SAT trends in the west EC region among ensembles, driven by internal variability intrinsic due to the atmosphere and land, is more than three times the isolated IPO impacts, which can shadow the modulation of the IPO on the west Eurasia winter climate. 

Abstract To examine the atmospheric responses to Arctic sea-ice variability in the Northern Hemisphere cold season (October to following March), this study uses a coordinated set of large-ensemble experiments of nine atmospheric general circulation models (AGCMs) forced with observed daily-varying sea-ice, sea-surface temperature, and radiative forcings prescribed during the 1979-2014 period, together with a parallel set of experiments where Arctic sea ice is substituted by its climatology. The simulations of the former set reproduce the near-surface temperature trends in reanalysis data, with similar amplitude, and their multi-model ensemble mean (MMEM) shows decreasing sea-level pressure over much of the polar cap and Eurasia in boreal autumn. The MMEM difference between the two experiments allows isolating the effects of Arctic sea-ice loss, which explain a large portion of the Arctic warming trends in the lower troposphere and drives a small but statistically significant weakening of the wintertime Arctic Oscillation. The observed interannual co-variability between sea-ice extent in the Barents-Kara Seas and lagged atmospheric circulation is distinguished from the effects of confounding factors based on multiple regression, and quantitatively compared to the co-variability in MMEMs. The interannual sea-ice decline followed by a negative North Atlantic Oscillation-like anomaly found in observations is also seen in the MMEM differences, with consistent spatial structure but much smaller amplitude. This result suggests that the sea-ice impacts on trends and interannual atmospheric variability simulated by AGCMs could be underestimated, but caution is needed because internal atmospheric variability may have affected the observed relationship. 

Deep learning on graph structures has shown exciting results in various applications. However, few attentions have been paid to the robustness of such models, in contrast to numerous research work for image or text adversarial attack and defense. In this paper, we focus on the adversarial attacks that fool deep learning models by modifying the combinatorial structure of data. We first propose a reinforcement learning based attack method that learns the generalizable attack policy, while only requiring prediction labels from the target classifier. We further propose attack methods based on genetic algorithms and gradient descent in the scenario where additional prediction confidence or gradients are available. We use both synthetic and real-world data to show that, a family of Graph Neural Network models are vulnerable to these attacks, in both graph-level and node-level classification tasks. We also show such attacks can be used to diagnose the learned classifiers. 

This poster paper describes the authors’ single-year National Science Foundation (NSF) project DRL-1825007 titled, “DCL: Synthesis and Design Workshop on Digitally-Mediated Team Learning” which has been conducted as one of nine awards within NSF-18-017: Principles for the Design of Digital STEM Learning Environments. Beginning in September 2018, the project conducted the activities herein to deliver a three-day workshop on Digitally-Mediated Team Learning (DMTL) to convene, invigorate, and task interdisciplinary science and engineering researchers, developers, and educators to coalesce the leading strategies for digital team learning. The deliverable of the workshop is a White Paper composed to identify one-year, three-year, and five-year research and practice roadmaps for highly-adaptable environments for computer-supported collaborative learning within STEM curricula. As subject to the chronology of events, highlights of the White Paper’s outcomes will be showcased within the poster itself.