Showing papers by "University of Tübingen published in 2020"
••
German Cancer Research Center1, Helmholtz-Zentrum Dresden-Rossendorf2, McGill University3, Moffitt Cancer Center4, Harvard University5, Brigham and Women's Hospital6, Kettering University7, Johns Hopkins University8, University of Pennsylvania9, University Medical Center Groningen10, University of Zurich11, King's College London12, University of Lausanne13, Netherlands Cancer Institute14, Stanford University15, University of Michigan16, Maastricht University Medical Centre17, University of Tübingen18, University of Bergen19, University of California, San Francisco20, University of Geneva21, University of British Columbia22, Cardiff University23, Leiden University Medical Center24
TL;DR: A set of 169 radiomics features was standardized, which enabled verification and calibration of different radiomics software and could be excellently reproduced.
Abstract: Background Radiomic features may quantify characteristics present in medical imaging. However, the lack of standardized definitions and validated reference values have hampered clinical use. Purpose To standardize a set of 174 radiomic features. Materials and Methods Radiomic features were assessed in three phases. In phase I, 487 features were derived from the basic set of 174 features. Twenty-five research teams with unique radiomics software implementations computed feature values directly from a digital phantom, without any additional image processing. In phase II, 15 teams computed values for 1347 derived features using a CT image of a patient with lung cancer and predefined image processing configurations. In both phases, consensus among the teams on the validity of tentative reference values was measured through the frequency of the modal value and classified as follows: less than three matches, weak; three to five matches, moderate; six to nine matches, strong; 10 or more matches, very strong. In the final phase (phase III), a public data set of multimodality images (CT, fluorine 18 fluorodeoxyglucose PET, and T1-weighted MRI) from 51 patients with soft-tissue sarcoma was used to prospectively assess reproducibility of standardized features. Results Consensus on reference values was initially weak for 232 of 302 features (76.8%) at phase I and 703 of 1075 features (65.4%) at phase II. At the final iteration, weak consensus remained for only two of 487 features (0.4%) at phase I and 19 of 1347 features (1.4%) at phase II. Strong or better consensus was achieved for 463 of 487 features (95.1%) at phase I and 1220 of 1347 features (90.6%) at phase II. Overall, 169 of 174 features were standardized in the first two phases. In the final validation phase (phase III), most of the 169 standardized features could be excellently reproduced (166 with CT; 164 with PET; and 164 with MRI). Conclusion A set of 169 radiomics features was standardized, which enabled verification and calibration of different radiomics software. © RSNA, 2020 Online supplemental material is available for this article. See also the editorial by Kuhl and Truhn in this issue.
1,563 citations
••
TL;DR: A set of recommendations for model interpretation and benchmarking is developed, highlighting recent advances in machine learning to improve robustness and transferability from the lab to real-world applications.
Abstract: Deep learning has triggered the current rise of artificial intelligence and is the workhorse of today’s machine intelligence. Numerous success stories have rapidly spread all over science, industry and society, but its limitations have only recently come into focus. In this Perspective we seek to distil how many of deep learning’s failures can be seen as different symptoms of the same underlying problem: shortcut learning. Shortcuts are decision rules that perform well on standard benchmarks but fail to transfer to more challenging testing conditions, such as real-world scenarios. Related issues are known in comparative psychology, education and linguistics, suggesting that shortcut learning may be a common characteristic of learning systems, biological and artificial alike. Based on these observations, we develop a set of recommendations for model interpretation and benchmarking, highlighting recent advances in machine learning to improve robustness and transferability from the lab to real-world applications. Deep learning has resulted in impressive achievements, but under what circumstances does it fail, and why? The authors propose that its failures are a consequence of shortcut learning, a common characteristic across biological and artificial systems in which strategies that appear to have solved a problem fail unexpectedly under different circumstances.
924 citations
••
Valley Hospital1, University College London2, University of Liverpool3, UCL Institute of Neurology4, Wexham Park Hospital5, University of Tübingen6, Mahosot Hospital7, University College London Hospitals NHS Foundation Trust8, University of Oxford9, Watford General Hospital10, University of Cambridge11, John Radcliffe Hospital12, Northwick Park Hospital13, University of Hertfordshire14, Imperial College Healthcare15, Queen Mary University of London16, Guy's and St Thomas' NHS Foundation Trust17
TL;DR: A case series of 43 patients with neurological complications of SARS-CoV-2 infection includes encephalopathies, encephalitis, acute disseminated encephalomyelitis with haemorrhagic change, transverse myelitis, ischaemic stroke, and Guillain-Barré syndrome.
Abstract: Preliminary clinical data indicate that severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection is associated with neurological and neuropsychiatric illness. Responding to this, a weekly virtual coronavirus disease 19 (COVID-19) neurology multi-disciplinary meeting was established at the National Hospital, Queen Square, in early March 2020 in order to discuss and begin to understand neurological presentations in patients with suspected COVID-19-related neurological disorders. Detailed clinical and paraclinical data were collected from cases where the diagnosis of COVID-19 was confirmed through RNA PCR, or where the diagnosis was probable/possible according to World Health Organization criteria. Of 43 patients, 29 were SARS-CoV-2 PCR positive and definite, eight probable and six possible. Five major categories emerged: (i) encephalopathies (n = 10) with delirium/psychosis and no distinct MRI or CSF abnormalities, and with 9/10 making a full or partial recovery with supportive care only; (ii) inflammatory CNS syndromes (n = 12) including encephalitis (n = 2, para- or post-infectious), acute disseminated encephalomyelitis (n = 9), with haemorrhage in five, necrosis in one, and myelitis in two, and isolated myelitis (n = 1). Of these, 10 were treated with corticosteroids, and three of these patients also received intravenous immunoglobulin; one made a full recovery, 10 of 12 made a partial recovery, and one patient died; (iii) ischaemic strokes (n = 8) associated with a pro-thrombotic state (four with pulmonary thromboembolism), one of whom died; (iv) peripheral neurological disorders (n = 8), seven with Guillain-Barre syndrome, one with brachial plexopathy, six of eight making a partial and ongoing recovery; and (v) five patients with miscellaneous central disorders who did not fit these categories. SARS-CoV-2 infection is associated with a wide spectrum of neurological syndromes affecting the whole neuraxis, including the cerebral vasculature and, in some cases, responding to immunotherapies. The high incidence of acute disseminated encephalomyelitis, particularly with haemorrhagic change, is striking. This complication was not related to the severity of the respiratory COVID-19 disease. Early recognition, investigation and management of COVID-19-related neurological disease is challenging. Further clinical, neuroradiological, biomarker and neuropathological studies are essential to determine the underlying pathobiological mechanisms that will guide treatment. Longitudinal follow-up studies will be necessary to ascertain the long-term neurological and neuropsychological consequences of this pandemic.
839 citations
••
Paris 12 Val de Marne University1, University Medical Center Groningen2, Eindhoven University of Technology3, University Hospital of Lausanne4, French Institute of Health and Medical Research5, Università Campus Bio-Medico6, University of Belgrade7, University of Cologne8, Ludwig Maximilian University of Munich9, École Polytechnique Fédérale de Lausanne10, Turku University Hospital11, University of Regensburg12, Università telematica San Raffaele13, Paris Descartes University14, Paracelsus Private Medical University of Salzburg15, University of Bern16, Universidade Nova de Lisboa17, Medical Park18, University of Göttingen19, University of Messina20, Central European Institute of Technology21, University of Siena22, University of Turku23, University of Tübingen24
TL;DR: These updated recommendations take into account all rTMS publications, including data prior to 2014, as well as currently reviewed literature until the end of 2018, and are based on the differences reached in therapeutic efficacy of real vs. sham rT MS protocols.
822 citations
••
TL;DR: This review evaluates the dynamic interactions of cancer cells with their microenvironment consisting of stromal cells and extracellular matrix components in various advanced cancer models, with a focus on 3D systems as well as lab-on-chip devices.
Abstract: The dynamic interactions of cancer cells with their microenvironment consisting of stromal cells (cellular part) and extracellular matrix (ECM) components (non-cellular) is essential to stimulate the heterogeneity of cancer cell, clonal evolution and to increase the multidrug resistance ending in cancer cell progression and metastasis. The reciprocal cell-cell/ECM interaction and tumor cell hijacking of non-malignant cells force stromal cells to lose their function and acquire new phenotypes that promote development and invasion of tumor cells. Understanding the underlying cellular and molecular mechanisms governing these interactions can be used as a novel strategy to indirectly disrupt cancer cell interplay and contribute to the development of efficient and safe therapeutic strategies to fight cancer. Furthermore, the tumor-derived circulating materials can also be used as cancer diagnostic tools to precisely predict and monitor the outcome of therapy. This review evaluates such potentials in various advanced cancer models, with a focus on 3D systems as well as lab-on-chip devices.
693 citations
•
TL;DR: Two extensions of the PGD-attack overcoming failures due to suboptimal step size and problems of the objective function are proposed and combined with two complementary existing ones to form a parameter-free, computationally affordable and user-independent ensemble of attacks to test adversarial robustness.
Abstract: The field of defense strategies against adversarial attacks has significantly grown over the last years, but progress is hampered as the evaluation of adversarial defenses is often insufficient and thus gives a wrong impression of robustness. Many promising defenses could be broken later on, making it difficult to identify the state-of-the-art. Frequent pitfalls in the evaluation are improper tuning of hyperparameters of the attacks, gradient obfuscation or masking. In this paper we first propose two extensions of the PGD-attack overcoming failures due to suboptimal step size and problems of the objective function. We then combine our novel attacks with two complementary existing ones to form a parameter-free, computationally affordable and user-independent ensemble of attacks to test adversarial robustness. We apply our ensemble to over 50 models from papers published at recent top machine learning and computer vision venues. In all except one of the cases we achieve lower robust test accuracy than reported in these papers, often by more than $10\%$, identifying several broken defenses.
667 citations
••
TL;DR: The nf-core framework is introduced as a means for the development of collaborative, peerreviewed, best-practice analysis pipelines that can be used across all institutions and research facilities and introduces a higher degree of portability as compared to custom in-house scripts.
Abstract: To the Editor — The standardization, portability and reproducibility of analysis pipelines are key issues within the bioinformatics community. Most bioinformatics pipelines are designed for use on-premises; as a result, the associated software dependencies and execution logic are likely to be tightly coupled with proprietary computing environments. This can make it difficult or even impossible for others to reproduce the ensuing results, which is a fundamental requirement for the validation of scientific findings. Here, we introduce the nf-core framework as a means for the development of collaborative, peerreviewed, best-practice analysis pipelines (Fig. 1). All nf-core pipelines are written in Nextflow and so inherit the ability to be executed on most computational infrastructures, as well as having native support for container technologies such as Docker and Singularity. The nf-core community (Supplementary Fig. 1) has developed a suite of tools that automate pipeline creation, testing, deployment and synchronization. Our goal is to provide a framework for high-quality bioinformatics pipelines that can be used across all institutions and research facilities. Being able to reproduce scientific results is the central tenet of the scientific method. However, moving toward FAIR (findable, accessible, interoperable and reusable) research methods1 in data-driven science is complex2,3. Central repositories, such as bio. tools4, omictools5 and the Galaxy toolshed6, make it possible to find existing pipelines and their associated tools. However, it is still notoriously challenging to develop analysis pipelines that are fully reproducible and interoperable across multiple systems and institutions — primarily because of differences in hardware, operating systems and software versions. Although the recommended guidelines for some analysis pipelines have become standardized (for example, GATK best practices7), the actual implementations are usually developed on a case-by-case basis. As such, there is often little incentive to test, document and implement pipelines in a way that permits their reuse by other researchers. This can hamper sustainable sharing of data and tools, and results in a proliferation of heterogeneous analysis pipelines, making it difficult for newcomers to find what they need to address a specific analysis question. As the scale of -omics data and their associated analytical tools has grown, the scientific community is increasingly moving toward the use of specialized workflow management systems to build analysis pipelines8. They separate the requirements of the underlying compute infrastructure from the analysis and workflow description, introducing a higher degree of portability as compared to custom in-house scripts. One such popular tool is Nextflow9. Using Nextflow, software packages can be bundled with analysis pipelines using built-in integration for package managers, such as Conda, and containerization platforms, such as Docker and Singularity. Moreover, support for most common highperformance-computing batch schedulers and cloud providers allows simple deployment of analysis pipelines on almost any infrastructure. The opportunity to run pipelines locally during initial development and then to proceed seamlessly to largescale computational resources in highperformance-computing or cloud settings provides users and developers with great flexibility. The nf-core community project collects a curated set of best-practice analysis pipelines built using Nextflow. Similar projects Participate
663 citations
••
University of Tübingen1, University of Pavia2, Charité3, University of Leicester4, University of Barcelona5, University of Graz6, Istanbul University7, Paris Diderot University8, University of Birmingham9, Norfolk and Norwich University Hospital10, Instituto de Medicina Molecular11, Peking Union Medical College Hospital12, University of Iceland13, University of East Anglia14, University of Oxford15
TL;DR: The recommendations for the management of LVV have been updated to facilitate the translation of current scientific evidence and expert opinion into better management and improved outcome of patients in clinical practice.
Abstract: BACKGROUND
Since the publication of the European League Against Rheumatism (EULAR) recommendations for the management of large vessel vasculitis (LVV) in 2009, several relevant randomised clinical trials and cohort analyses have been published, which have the potential to change clinical care and therefore supporting the need to update the original recommendations.
METHODS
Using EULAR standardised operating procedures for EULAR-endorsed recommendations, the EULAR task force undertook a systematic literature review and sought opinion from 20 experts from 13 countries. We modified existing recommendations and created new recommendations.
RESULTS
Three overarching principles and 10 recommendations were formulated. We recommend that a suspected diagnosis of LVV should be confirmed by imaging or histology. High dose glucocorticoid therapy (40-60 mg/day prednisone-equivalent) should be initiated immediately for induction of remission in active giant cell arteritis (GCA) or Takayasu arteritis (TAK). We recommend adjunctive therapy in selected patients with GCA (refractory or relapsing disease, presence of an increased risk for glucocorticoid-related adverse events or complications) using tocilizumab. Methotrexate may be used as an alternative. Non-biological glucocorticoid-sparing agents should be given in combination with glucocorticoids in all patients with TAK and biological agents may be used in refractory or relapsing patients. We no longer recommend the routine use of antiplatelet or anticoagulant therapy for treatment of LVV unless it is indicated for other reasons.
CONCLUSIONS
We have updated the recommendations for the management of LVV to facilitate the translation of current scientific evidence and expert opinion into better management and improved outcome of patients in clinical practice.
564 citations
••
TL;DR: The sound of a 3,000 year old mummified individual has been accurately reproduced as a vowel-like sound based on measurements of the precise dimensions of his extant vocal tract following Computed Tomography (CT) scanning, enabling the creation of a3-D printed vocal tract.
Abstract: The sound of a 3,000 year old mummified individual has been accurately reproduced as a vowel-like sound based on measurements of the precise dimensions of his extant vocal tract following Computed Tomography (CT) scanning, enabling the creation of a 3-D printed vocal tract. By using the Vocal Tract Organ, which provides a user-controllable artificial larynx sound source, a vowel sound is synthesised which compares favourably with vowels of modern individuals.
518 citations
••
Stanford University1, Kyoto University2, Universitaire Ziekenhuizen Leuven3, Yonsei University4, French Institute of Health and Medical Research5, University of Vermont6, Kaiser Permanente7, University of Tübingen8, University of Padua9, Eli Lilly and Company10, The Royal Marsden NHS Foundation Trust11
TL;DR: The addition of abemaciclib to fulvestrant provided a clinically meaningful median OS benefit of 9.4 months for patients with HR-positive, ERBB2-negative advanced breast cancer that had progressed on endocrine therapy.
Abstract: Importance Statistically significant overall survival (OS) benefits of CDK4 and CDK6 inhibitors in combination with fulvestrant for hormone receptor (HR)–positive, ERBB2 (formerly HER2)-negative advanced breast cancer (ABC) in patients regardless of menopausal status after prior endocrine therapy (ET) has not yet been demonstrated. Objective To compare the effect of abemaciclib plus fulvestrant vs placebo plus fulvestrant on OS at the prespecified interim of MONARCH 2 (338 events) in patients with HR-positive, ERBB2-negative advanced breast cancer that progressed during prior ET. Design, Setting, and Participants MONARCH 2 was a global, randomized, placebo-controlled, double-blind phase 3 trial of abemaciclib plus fulvestrant vs placebo plus fulvestrant for treatment of premenopausal or perimenopausal women (with ovarian suppression) and postmenopausal women with HR-positive, ERBB2-negative ABC that progressed during ET. Patients were enrolled between August 7, 2014, and December 29, 2015. Analyses for this report were conducted at the time of database lock on June 20, 2019. Interventions Patients were randomized 2:1 to receive abemaciclib or placebo, 150 mg, every 12 hours on a continuous schedule plus fulvestrant, 500 mg, per label. Randomization was stratified based on site of metastasis (visceral, bone only, or other) and resistance to prior ET (primary vs secondary). Main Outcomes and Measures The primary end point was investigator-assessed progression-free survival. Overall survival was a gated key secondary end point. The boundaryPvalue for the interim analysis was .02. Results Of 669 women enrolled, 446 (median [range] age, 59 [32-91] years) were randomized to the abemaciclib plus fulvestrant arm and 223 (median [range] age, 62 [32-87] years) were randomized to the placebo plus fulvestrant arm. At the prespecified interim, 338 deaths (77% of the planned 441 at the final analysis) were observed in the intent-to-treat population, with a median OS of 46.7 months for abemaciclib plus fulvestrant and 37.3 months for placebo plus fulvestrant (hazard ratio [HR], 0.757; 95% CI, 0.606-0.945;P = .01). Improvement in OS was consistent across all stratification factors. Among stratification factors, more pronounced effects were observed in patients with visceral disease (HR, 0.675; 95% CI, 0.511-0.891) and primary resistance to prior ET (HR, 0.686; 95% CI, 0.451-1.043). Time to second disease progression (median, 23.1 months vs 20.6 months), time to chemotherapy (median, 50.2 months vs 22.1 months), and chemotherapy-free survival (median, 25.5 months vs 18.2 months) were also statistically significantly improved in the abemaciclib arm vs placebo arm. No new safety signals were observed for abemaciclib. Conclusions and Relevance Treatment with abemaciclib plus fulvestrant resulted in a statistically significant and clinically meaningful median OS improvement of 9.4 months for patients with HR-positive, ERBB2-negative ABC who progressed after prior ET regardless of menopausal status. Abemaciclib substantially delayed the receipt of subsequent chemotherapy. Trial Registration ClinicalTrials.gov identifier:NCT02107703
502 citations
••
University of Montana1, University of California, San Diego2, University of Münster3, University of Jena4, University of Lübeck5, Statens Serum Institut6, University of Tübingen7, University of Geneva8, Bruker9, Paris Descartes University10, University of São Paulo11, Technical University of Berlin12, Georgia Institute of Technology13, Saint Petersburg State University14, Waters Corporation15, Academy of Sciences of the Czech Republic16, Sookmyung Women's University17, University of Grenoble18, University of Oklahoma19, Carnegie Mellon University20, University of West Alabama21, Leibniz Association22, University of Corsica Pascal Paoli23, Massachusetts Institute of Technology24, Michigan State University25, University of Glasgow26, Wageningen University and Research Centre27, Kangwon National University28
TL;DR: Feature-based molecular networking (FBMN) as discussed by the authors is an analysis method in the Global Natural Products Social Molecular Networking (GNPS) infrastructure that builds on chromatographic feature detection and alignment tools.
Abstract: Molecular networking has become a key method to visualize and annotate the chemical space in non-targeted mass spectrometry data. We present feature-based molecular networking (FBMN) as an analysis method in the Global Natural Products Social Molecular Networking (GNPS) infrastructure that builds on chromatographic feature detection and alignment tools. FBMN enables quantitative analysis and resolution of isomers, including from ion mobility spectrometry.
•
30 Apr 2020TL;DR: Experiments show that BERT pre-training achieves a new state of the art on TIMIT phoneme classification and WSJ speech recognition and the algorithm uses a gumbel softmax or online k-means clustering to quantize the dense representations.
Abstract: We propose vq-wav2vec to learn discrete representations of audio segments through a wav2vec-style self-supervised context prediction task. The algorithm uses either a gumbel softmax or online k-means clustering to quantize the dense representations. Discretization enables the direct application of algorithms from the NLP community which require discrete inputs. Experiments show that BERT pre-training achieves a new state of the art on TIMIT phoneme classification and WSJ speech recognition.
••
Katrina L. Grasby1, Neda Jahanshad2, Jodie N. Painter1, Lucía Colodro-Conde3 +356 more•Institutions (115)
TL;DR: Results support the radial unit hypothesis that different developmental mechanisms promote surface area expansion and increases in thickness and find evidence that brain structure is a key phenotype along the causal pathway that leads from genetic variation to differences in general cognitive function.
Abstract: The cerebral cortex underlies our complex cognitive capabilities, yet little is known about the specific genetic loci that influence human cortical structure. To identify genetic variants that affect cortical structure, we conducted a genome-wide association meta-analysis of brain magnetic resonance imaging data from 51,665 individuals. We analyzed the surface area and average thickness of the whole cortex and 34 regions with known functional specializations. We identified 199 significant loci and found significant enrichment for loci influencing total surface area within regulatory elements that are active during prenatal cortical development, supporting the radial unit hypothesis. Loci that affect regional surface area cluster near genes in Wnt signaling pathways, which influence progenitor expansion and areal identity. Variation in cortical structure is genetically correlated with cognitive function, Parkinson's disease, insomnia, depression, neuroticism, and attention deficit hyperactivity disorder.
••
University of California, Irvine1, Alfred Wegener Institute for Polar and Marine Research2, University of Texas at Austin3, University of Tübingen4, University of Bremen5, British Antarctic Survey6, National Space Institute7, Northumbria University8, Polar Research Institute of China9, Ohio State University10, Norwegian Polar Institute11, University of Kansas12, Université libre de Bruxelles13, University of Tasmania14, Institute for Geosciences and Natural Resources15, California Institute of Technology16, Utrecht University17
TL;DR: In this paper, a high-resolution and physically based description of Antarctica bed topography using mass conservation is presented, revealing previously unknown basal features with major implications for glacier response to climate change.
Abstract: The Antarctic ice sheet has been losing mass over past decades through the accelerated flow of its glaciers, conditioned by ocean temperature and bed topography. Glaciers retreating along retrograde slopes (that is, the bed elevation drops in the inland direction) are potentially unstable, while subglacial ridges slow down the glacial retreat. Despite major advances in the mapping of subglacial bed topography, significant sectors of Antarctica remain poorly resolved and critical spatial details are missing. Here we present a novel, high-resolution and physically based description of Antarctic bed topography using mass conservation. Our results reveal previously unknown basal features with major implications for glacier response to climate change. For example, glaciers flowing across the Transantarctic Mountains are protected by broad, stabilizing ridges. Conversely, in the marine basin of Wilkes Land, East Antarctica, we find retrograde slopes along Ninnis and Denman glaciers, with stabilizing slopes beneath Moscow University, Totten and Lambert glacier system, despite corrections in bed elevation of up to 1 km for the latter. This transformative description of bed topography redefines the high- and lower-risk sectors for rapid sea level rise from Antarctica; it will also significantly impact model projections of sea level rise from Antarctica in the coming centuries.
••
TL;DR: There exists a number of candidate drugs that may inhibit infection with and replication of SARS-CoV-2, and chloroquine and hydroxychloroquine, and off-label antiviral drugs, such as the nucleotide analogue remdesivir, HIV protease inhibitors lopinavir and ritonavir, broad-spectrum antiviral Drugs arbidol and favipiravir as well as antiviral phytochemicals available to date may prevent spread.
••
TL;DR: In hospitalized patients, clinicians should consider low lymphocyte count as well as the serum levels of CRP, D-dimers, ferritin and IL-6 which may be used in risk stratification to predict severe and fatal COVID-19.
•
TL;DR: In this article, the authors demonstrate that adaptive attacks can be circumvented despite attempting to perform evaluations using adaptive attacks, and provide guidance on how to properly perform adaptive attacks against defenses to adversarial examples, and thus allow the community to make further progress in building more robust models.
Abstract: Adaptive attacks have (rightfully) become the de facto standard for evaluating defenses to adversarial examples. We find, however, that typical adaptive evaluations are incomplete. We demonstrate that thirteen defenses recently published at ICLR, ICML and NeurIPS---and chosen for illustrative and pedagogical purposes---can be circumvented despite attempting to perform evaluations using adaptive attacks. While prior evaluation papers focused mainly on the end result---showing that a defense was ineffective---this paper focuses on laying out the methodology and the approach necessary to perform an adaptive attack. We hope that these analyses will serve as guidance on how to properly perform adaptive attacks against defenses to adversarial examples, and thus will allow the community to make further progress in building more robust models.
•
TL;DR: D-NeRF is introduced, a method that extends neural radiance fields to a dynamic domain, allowing to reconstruct and render novel images of objects under rigid and non-rigid motions from a single camera moving around the scene.
Abstract: Neural rendering techniques combining machine learning with geometric reasoning have arisen as one of the most promising approaches for synthesizing novel views of a scene from a sparse set of images. Among these, stands out the Neural radiance fields (NeRF), which trains a deep network to map 5D input coordinates (representing spatial location and viewing direction) into a volume density and view-dependent emitted radiance. However, despite achieving an unprecedented level of photorealism on the generated images, NeRF is only applicable to static scenes, where the same spatial location can be queried from different images. In this paper we introduce D-NeRF, a method that extends neural radiance fields to a dynamic domain, allowing to reconstruct and render novel images of objects under rigid and non-rigid motions from a \emph{single} camera moving around the scene. For this purpose we consider time as an additional input to the system, and split the learning process in two main stages: one that encodes the scene into a canonical space and another that maps this canonical representation into the deformed scene at a particular time. Both mappings are simultaneously learned using fully-connected networks. Once the networks are trained, D-NeRF can render novel images, controlling both the camera view and the time variable, and thus, the object movement. We demonstrate the effectiveness of our approach on scenes with objects under rigid, articulated and non-rigid motions. Code, model weights and the dynamic scenes dataset will be released.
••
TL;DR: A new framework, named MedGAN, is proposed for medical image-to-image translation which operates on the image level in an end- to-end manner and outperforms other existing translation approaches.
••
TL;DR: With extended study follow-up, results from KEYNOTE-426 show that pembrolizumab plus axitinib continues to have superior clinical outcomes over sunit inib, and these results continue to support the first-line treatment with pembroizumAB plus ax itinib as the standard of care of advanced renal cell carcinoma.
Abstract: Summary Background The first interim analysis of the KEYNOTE-426 study showed superior efficacy of pembrolizumab plus axitinib over sunitinib monotherapy in treatment-naive, advanced renal cell carcinoma. The exploratory analysis with extended follow-up reported here aims to assess long-term efficacy and safety of pembrolizumab plus axitinib versus sunitinib monotherapy in patients with advanced renal cell carcinoma. Methods In the ongoing, randomised, open-label, phase 3 KEYNOTE-426 study, adults (≥18 years old) with treatment-naive, advanced renal cell carcinoma with clear cell histology were enrolled in 129 sites (hospitals and cancer centres) across 16 countries. Patients were randomly assigned (1:1) to receive 200 mg pembrolizumab intravenously every 3 weeks for up to 35 cycles plus 5 mg axitinib orally twice daily or 50 mg sunitinib monotherapy orally once daily for 4 weeks per 6-week cycle. Randomisation was done using an interactive voice response system or integrated web response system, and was stratified by International Metastatic Renal Cell Carcinoma Database Consortium risk status and geographical region. Primary endpoints were overall survival and progression-free survival in the intention-to-treat population. Since the primary endpoints were met at the first interim analysis, updated data are reported with nominal p values. This study is registered with ClinicalTrials.gov, NCT02853331. Findings Between Oct 24, 2016, and Jan 24, 2018, 861 patients were randomly assigned to receive pembrolizumab plus axitinib (n=432) or sunitinib monotherapy (n=429). With a median follow-up of 30·6 months (IQR 27·2–34·2), continued clinical benefit was observed with pembrolizumab plus axitinib over sunitinib in terms of overall survival (median not reached with pembrolizumab and axitinib vs 35·7 months [95% CI 33·3–not reached] with sunitinib); hazard ratio [HR] 0·68 [95% CI 0·55–0·85], p=0·0003) and progression-free survival (median 15·4 months [12·7–18·9] vs 11·1 months [9·1–12·5]; 0·71 [0·60–0·84], p Interpretation With extended study follow-up, results from KEYNOTE-426 show that pembrolizumab plus axitinib continues to have superior clinical outcomes over sunitinib. These results continue to support the first-line treatment with pembrolizumab plus axitinib as the standard of care of advanced renal cell carcinoma. Funding Merck Sharp & Dohme Corp, a subsidiary of Merck & Co, Inc.
••
Max Planck Society1, West Chester University of Pennsylvania2, Wildlife Conservation Society3, Kwame Nkrumah University of Science and Technology4, Cornell University5, University of Stirling6, University of Cambridge7, Arizona State University8, Jane Goodall Institute9, Lincoln Park Zoo10, Liverpool John Moores University11, University College London12, Kyoto University13, Washington University in St. Louis14, University of Tübingen15, Harvard University16, University of St Andrews17
TL;DR: It is shown that chimpanzees exhibit greater behavioural diversity in environments with more variability — in both recent and historical timescales, suggesting that environmental variability was a critical evolutionary force promoting the behavioural, as well as cultural diversification of great apes.
Abstract: Large brains and behavioural innovation are positively correlated, species-specific traits, associated with the behavioural flexibility animals need for adapting to seasonal and unpredictable habitats. Similar ecological challenges would have been important drivers throughout human evolution. However, studies examining the influence of environmental variability on within-species behavioural diversity are lacking despite the critical assumption that population diversification precedes genetic divergence and speciation. Here, using a dataset of 144 wild chimpanzee (Pan troglodytes) communities, we show that chimpanzees exhibit greater behavioural diversity in environments with more variability — in both recent and historical timescales. Notably, distance from Pleistocene forest refugia is associated with the presence of a larger number of behavioural traits, including both tool and non-tool use behaviours. Since more than half of the behaviours investigated are also likely to be cultural, we suggest that environmental variability was a critical evolutionary force promoting the behavioural, as well as cultural diversification of great apes. Environmental variability is one potential driver of behavioural and cultural diversity in humans and other animals. Here, the authors show that chimpanzee behavioural diversity is higher in habitats that are more seasonal and historically unstable, and in savannah woodland relative to forested sites.
••
TL;DR: It is demonstrated that SARS-CoV-2-neutralizing antibodies are readily generated from a diverse pool of precursors, fostering hope for rapid induction of a protective immune response upon vaccination.
••
TL;DR: In this paper, a set of recommendations for model interpretation and benchmarking, highlighting recent advances in machine learning to improve robustness and transferability from the lab to real-world applications, are presented.
Abstract: Deep learning has triggered the current rise of artificial intelligence and is the workhorse of today's machine intelligence. Numerous success stories have rapidly spread all over science, industry and society, but its limitations have only recently come into focus. In this perspective we seek to distil how many of deep learning's problem can be seen as different symptoms of the same underlying problem: shortcut learning. Shortcuts are decision rules that perform well on standard benchmarks but fail to transfer to more challenging testing conditions, such as real-world scenarios. Related issues are known in Comparative Psychology, Education and Linguistics, suggesting that shortcut learning may be a common characteristic of learning systems, biological and artificial alike. Based on these observations, we develop a set of recommendations for model interpretation and benchmarking, highlighting recent advances in machine learning to improve robustness and transferability from the lab to real-world applications.
••
Washington University in St. Louis1, Icahn School of Medicine at Mount Sinai2, University of Pittsburgh3, Niigata University4, Osaka City University5, Hirosaki University6, University of Tokyo7, Columbia University8, Indiana University – Purdue University Indianapolis9, Harvard University10, Mayo Clinic11, Butler Hospital12, Brown University13, University of New South Wales14, Neuroscience Research Australia15, Florey Institute of Neuroscience and Mental Health16, University of Melbourne17, Edith Cowan University18, University College London19, German Center for Neurodegenerative Diseases20, Ludwig Maximilian University of Munich21, University of Tübingen22, French Institute of Health and Medical Research23
TL;DR: A pattern of tau staging is identified where site-specific phosphorylation changes occur at different periods of disease progression and follow distinct trajectories over time, providing insights into the pathways linking tau, amyloid-β and neurodegeneration in dominantly inherited Alzheimer's disease.
Abstract: Development of tau-based therapies for Alzheimer's disease requires an understanding of the timing of disease-related changes in tau. We quantified the phosphorylation state at multiple sites of the tau protein in cerebrospinal fluid markers across four decades of disease progression in dominantly inherited Alzheimer's disease. We identified a pattern of tau staging where site-specific phosphorylation changes occur at different periods of disease progression and follow distinct trajectories over time. These tau phosphorylation state changes are uniquely associated with structural, metabolic, neurodegenerative and clinical markers of disease, and some (p-tau217 and p-tau181) begin with the initial increases in aggregate amyloid-β as early as two decades before the development of aggregated tau pathology. Others (p-tau205 and t-tau) increase with atrophy and hypometabolism closer to symptom onset. These findings provide insights into the pathways linking tau, amyloid-β and neurodegeneration, and may facilitate clinical trials of tau-based treatments.
••
TL;DR: This work highlights analytical and bioanalytical approaches to isolating, characterizing, and tracking groups of chemicals of concern in complex matrices and proposes techniques that combine chemical analysis and bioassays to facilitate the identification of mixtures of chemicals that pose a combined risk.
Abstract: Chemicals have improved our quality of life, but the resulting environmental pollution has the potential to cause detrimental effects on humans and the environment. People and biota are chronically exposed to thousands of chemicals from various environmental sources through multiple pathways. Environmental chemists and toxicologists have moved beyond detecting and quantifying single chemicals to characterizing complex mixtures of chemicals in indoor and outdoor environments and biological matrices. We highlight analytical and bioanalytical approaches to isolating, characterizing, and tracking groups of chemicals of concern in complex matrices. Techniques that combine chemical analysis and bioassays have the potential to facilitate the identification of mixtures of chemicals that pose a combined risk.
••
01 Feb 2020
TL;DR: In this article, the impacts of dams on nutrient cycling and greenhouse production are discussed, emphasizing the need to consider biogeochemical cycling at all stages of dam lifespan, and regulating hydraulic residence time and environmental flows (or e-flows) can be used in planning and operation from dam conception to deconstruction.
Abstract: The increased use of hydropower is currently driving the greatest surge in global dam construction since the mid-20th century, meaning that most major rivers on Earth are now dammed. Dams impede the flow of essential nutrients, including carbon, phosphorus, nitrogen and silicon, along river networks, leading to enhanced nutrient transformation and elimination. Increased nutrient retention via sedimentation or gaseous elimination in dammed reservoirs influences downstream terrestrial and coastal environments. Reservoirs can also become hotspots for greenhouse gas emission, potentially impacting how ‘green’ hydropower is compared with fossil-fuel burning. In this Review, we discuss how damming changes nutrient biogeochemistry along river networks, as well as its broader environmental consequences. The influences of construction and management practices on nutrient elimination, the emission of greenhouse gases and potential remobilization of legacy nutrients are also examined. We further consider how regulating hydraulic residence time and environmental flows (or e-flows) can be used in planning and operation from dam conception to deconstruction. River damming can harness hydropower, control flooding and store water, but can also alter biogeochemistry in reservoirs and downstream environments. In this Review, the impacts of dams on nutrient cycling and greenhouse production are discussed, emphasizing the need to consider biogeochemical cycling at all stages of dam lifespan.
••
University of Marburg1, Goethe University Frankfurt2, University Medical Center Freiburg3, University of Münster4, Dresden University of Technology5, University of Hamburg6, Technische Universität München7, Augsburg College8, RWTH Aachen University9, Innsbruck Medical University10, University of Tübingen11, Ludwig Maximilian University of Munich12, University of Mainz13
TL;DR: Exploratory data show that patients with undetectable minimal residual disease (MRD) before HCT and those with detectable MRD after HCT derive the strongest benefit from sorafenib, and maintenance therapy reduces the risk of relapse and death after H CT for FLT3-ITD-positive AML.
Abstract: PURPOSEDespite undergoing allogeneic hematopoietic stem cell transplantation (HCT), patients with acute myeloid leukemia (AML) with internal tandem duplication mutation in the FMS-like tyrosine kin...
••
TL;DR: A quantitative atlas of the transcriptomes, proteomes and phosphoproteomes of 30 tissues of the model plant Arabidopsis thaliana provides a valuable resource for plant research.
Abstract: Plants are essential for life and are extremely diverse organisms with unique molecular capabilities1. Here we present a quantitative atlas of the transcriptomes, proteomes and phosphoproteomes of 30 tissues of the model plant Arabidopsis thaliana. Our analysis provides initial answers to how many genes exist as proteins (more than 18,000), where they are expressed, in which approximate quantities (a dynamic range of more than six orders of magnitude) and to what extent they are phosphorylated (over 43,000 sites). We present examples of how the data may be used, such as to discover proteins that are translated from short open-reading frames, to uncover sequence motifs that are involved in the regulation of protein production, and to identify tissue-specific protein complexes or phosphorylation-mediated signalling events. Interactive access to this resource for the plant community is provided by the ProteomicsDB and ATHENA databases, which include powerful bioinformatics tools to explore and characterize Arabidopsis proteins, their modifications and interactions.
••
Technical University of Denmark1, VU University Amsterdam2, Heidelberg University3, École Polytechnique Fédérale de Lausanne4, RWTH Aachen University5, University of California, San Diego6, University of Toronto7, National Autonomous University of Mexico8, Institute for Systems Biology9, University of Tübingen10, University of Queensland11, Argonne National Laboratory12, Leiden University13, Technical University of Madrid14, Spanish National Research Council15, Hanze University of Applied Sciences16, Norwegian University of Life Sciences17, Wellcome Trust18, KAIST19, Max Planck Society20, Humboldt University of Berlin21, Wageningen University and Research Centre22, Agency for Science, Technology and Research23, Sungkyunkwan University24, Royal Institute of Technology25, King's College London26, Chinese Academy of Sciences27, University of Virginia28, Chalmers University of Technology29, University of Arkansas for Medical Sciences30, Oxford Brookes University31, Nova Southeastern University32, University of Minho33, University of Düsseldorf34
TL;DR: A community effort to develop a test suite named MEMOTE (for metabolic model tests) to assess GEM quality, and advocate adoption of the latest version of the Systems Biology Markup Language level 3 flux balance constraints (SBML3FBC) package as the primary description and exchange format.
Abstract: We acknowledge D. Dannaher and A. Lopez for their supporting work on the Angular parts of MEMOTE; resources and support from the DTU Computing Center; J. Cardoso, S. Gudmundsson, K. Jensen and D. Lappa for their feedback on conceptual details; and P. D. Karp and I. Thiele for critically reviewing the manuscript. We thank J. Daniel, T. Kristjansdottir, J. Saez-Saez, S. Sulheim, and P. Tubergen for being early adopters of MEMOTE and for providing written testimonials. J.O.V. received the Research Council of Norway grants 244164 (GenoSysFat), 248792 (DigiSal) and 248810 (Digital Life Norway); M.Z. received the Research Council of Norway grant 244164 (GenoSysFat); C.L. received funding from the Innovation Fund Denmark (project “Environmentally Friendly Protein Production (EFPro2)”); C.L., A.K., N. S., M.B., M.A., D.M., P.M, B.J.S., P.V., K.R.P. and M.H. received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement 686070 (DD-DeCaF); B.G.O., F.T.B. and A.D. acknowledge funding from the US National Institutes of Health (NIH, grant number 2R01GM070923-13); A.D. was supported by infrastructural funding from the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation), Cluster of Excellence EXC 2124 Controlling Microbes to Fight Infections; N.E.L. received funding from NIGMS R35 GM119850, Novo Nordisk Foundation NNF10CC1016517 and the Keck Foundation; A.R. received a Lilly Innovation Fellowship Award; B.G.-J. and J. Nogales received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement no 686585 for the project LIAR, and the Spanish Ministry of Economy and Competitivity through the RobDcode grant (BIO2014-59528-JIN); L.M.B. has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement 633962 for project P4SB; R.F. received funding from the US Department of Energy, Offices of Advanced Scientific Computing Research and the Biological and Environmental Research as part of the Scientific Discovery Through Advanced Computing program, grant DE-SC0010429; A.M., C.Z., S.L. and J. Nielsen received funding from The Knut and Alice Wallenberg Foundation, Advanced Computing program, grant #DE-SC0010429; S.K.’s work was in part supported by the German Federal Ministry of Education and Research (de.NBI partner project “ModSim” (FKZ: 031L104B)); E.K. and J.A.H.W. were supported by the German Federal Ministry of Education and Research (project “SysToxChip”, FKZ 031A303A); M.K. is supported by the Federal Ministry of Education and Research (BMBF, Germany) within the research network Systems Medicine of the Liver (LiSyM, grant number 031L0054); J.A.P. and G.L.M. acknowledge funding from US National Institutes of Health (T32-LM012416, R01-AT010253, R01-GM108501) and the Wagner Foundation; G.L.M. acknowledges funding from a Grand Challenges Exploration Phase I grant (OPP1211869) from the Bill & Melinda Gates Foundation; H.H. and R.S.M.S. received funding from the Biotechnology and Biological Sciences Research Council MultiMod (BB/N019482/1); H.U.K. and S.Y.L. received funding from the Technology Development Program to Solve Climate Changes on Systems Metabolic Engineering for Biorefineries (grants NRF-2012M1A2A2026556 and NRF-2012M1A2A2026557) from the Ministry of Science and ICT through the National Research Foundation (NRF) of Korea; H.U.K. received funding from the Bio & Medical Technology Development Program of the NRF, the Ministry of Science and ICT (NRF-2018M3A9H3020459); P.B., B.J.S., Z.K., B.O.P., C.L., M.B., N.S., M.H. and A.F. received funding through Novo Nordisk Foundation through the Center for Biosustainability at the Technical University of Denmark (NNF10CC1016517); D.-Y.L. received funding from the Next-Generation BioGreen 21 Program (SSAC, PJ01334605), Rural Development Administration, Republic of Korea; G.F. was supported by the RobustYeast within ERA net project via SystemsX.ch; V.H. received funding from the ETH Domain and Swiss National Science Foundation; M.P. acknowledges Oxford Brookes University; J.C.X. received support via European Research Council (666053) to W.F. Martin; B.E.E. acknowledges funding through the CSIRO-UQ Synthetic Biology Alliance; C.D. is supported by a Washington Research Foundation Distinguished Investigator Award. I.N. received funding from National Institutes of Health (NIH)/National Institute of General Medical Sciences (NIGMS) (grant P20GM125503).
••
17 Jul 2020
TL;DR: The SemEval-2020 Task 12 on Multilingual Offensive Language Identification in Social Media (OffensEval 2020) as mentioned in this paper included three subtasks corresponding to the hierarchical taxonomy of the OLID schema, and was offered in five languages: Arabic, Danish, English, Greek, and Turkish.
Abstract: We present the results and the main findings of SemEval-2020 Task 12 on Multilingual Offensive Language Identification in Social Media (OffensEval-2020). The task included three subtasks corresponding to the hierarchical taxonomy of the OLID schema from OffensEval-2019, and it was offered in five languages: Arabic, Danish, English, Greek, and Turkish. OffensEval-2020 was one of the most popular tasks at SemEval-2020, attracting a large number of participants across all subtasks and languages: a total of 528 teams signed up to participate in the task, 145 teams submitted official runs on the test data, and 70 teams submitted system description papers.