Abstract: The purported "black box" nature of neural networks is a barrier to adoption in applications where interpretability is essential. Here we present DeepLIFT (Deep Learning Important FeaTures), a method for decomposing the output prediction of a neural network on a specific input by backpropagating the contributions of all neurons in the network to every feature of the input. DeepLIFT compares the activation of each neuron to its 'reference activation' and assigns contribution scores according to the difference. By optionally giving separate consideration to positive and negative contributions, DeepLIFT can also reveal dependencies which are missed by other approaches. Scores can be computed efficiently in a single backward pass. We apply DeepLIFT to models trained on MNIST and simulated genomic data, and show significant advantages over gradient-based methods. Video tutorial: http://goo.gl/qKb7pL, code: http://goo.gl/RM8jvH.

Abstract: Accurate pathological diagnosis is crucial for optimal management of patients with cancer. For the approximately 100 known tumour types of the central nervous system, standardization of the diagnostic process has been shown to be particularly challenging-with substantial inter-observer variability in the histopathological diagnosis of many tumour types. Here we present a comprehensive approach for the DNA methylation-based classification of central nervous system tumours across all entities and age groups, and demonstrate its application in a routine diagnostic setting. We show that the availability of this method may have a substantial impact on diagnostic precision compared to standard methods, resulting in a change of diagnosis in up to 12% of prospective cases. For broader accessibility, we have designed a free online classifier tool, the use of which does not require any additional onsite data processing. Our results provide a blueprint for the generation of machine-learning-based tumour classifiers across other cancer entities, with the potential to fundamentally transform tumour pathology.

Abstract: We present the Natural Questions corpus, a question answering data set. Questions consist of real anonymized, aggregated queries issued to the Google search engine. An annotator is presented with a...

Abstract: Purpose Most patients with hepatocellular carcinoma (HCC) have associated chronic liver disease, the severity of which is currently assessed by the Child-Pugh (C-P) grade. In this international collaboration, we identify objective measures of liver function/dysfunction that independently influence survival in patients with HCC and then combine these into a model that could be compared with the conventional C-P grade. Patients and Methods We developed a simple model to assess liver function, based on 1,313 patients with HCC of all stages from Japan, that involved only serum bilirubin and albumin levels. We then tested the model using similar cohorts from other geographical regions (n = 5,097) and other clinical situations (patients undergoing resection [n = 525] or sorafenib treatment for advanced HCC [n = 1,132]). The specificity of the model for liver (dys)function was tested in patients with chronic liver disease but without HCC (n = 501). Results The model, the Albumin-Bilirubin (ALBI) grade, performed...

Abstract: This review provides a summary statement of recommended implementations of arterial spin labeling (ASL) for clinical applications. It is a consensus of the ISMRM Perfusion Study Group and the European ASL in Dementia consortium, both of whom met to reach this consensus in October 2012 in Amsterdam. Although ASL continues to undergo rapid technical development, we believe that current ASL methods are robust and ready to provide useful clinical information, and that a consensus statement on recommended implementations will help the clinical community to adopt a standardized approach. In this review, we describe the major considerations and trade-offs in implementing an ASL protocol and provide specific recommendations for a standard approach. Our conclusion is that as an optimal default implementation, we recommend pseudo-continuous labeling, background suppression, a segmented three-dimensional readout without vascular crushing gradients, and calculation and presentation of both label/control difference images and cerebral blood flow in absolute units using a simplified model.

Abstract: Recent studies show that the state-of-the-art deep neural networks (DNNs) are vulnerable to adversarial examples, resulting from small-magnitude perturbations added to the input. Given that that emerging physical systems are using DNNs in safety-critical situations, adversarial examples could mislead these systems and cause dangerous situations. Therefore, understanding adversarial examples in the physical world is an important step towards developing resilient learning algorithms. We propose a general attack algorithm, Robust Physical Perturbations (RP2), to generate robust visual adversarial perturbations under different physical conditions. Using the real-world case of road sign classification, we show that adversarial examples generated using RP2 achieve high targeted misclassification rates against standard-architecture road sign classifiers in the physical world under various environmental conditions, including viewpoints. Due to the current lack of a standardized testing method, we propose a two-stage evaluation methodology for robust physical adversarial examples consisting of lab and field tests. Using this methodology, we evaluate the efficacy of physical adversarial manipulations on real objects. With a perturbation in the form of only black and white stickers, we attack a real stop sign, causing targeted misclassification in 100% of the images obtained in lab settings, and in 84.8% of the captured video frames obtained on a moving vehicle (field test) for the target classifier.

Abstract: Cancer-associated fibroblasts (CAFs) are a key component of the tumour microenvironment with diverse functions, including matrix deposition and remodelling, extensive reciprocal signalling interactions with cancer cells and crosstalk with infiltrating leukocytes. As such, they are a potential target for optimizing therapeutic strategies against cancer. However, many challenges are present in ongoing attempts to modulate CAFs for therapeutic benefit. These include limitations in our understanding of the origin of CAFs and heterogeneity in CAF function, with it being desirable to retain some antitumorigenic functions. On the basis of a meeting of experts in the field of CAF biology, we summarize in this Consensus Statement our current knowledge and present a framework for advancing our understanding of this critical cell type within the tumour microenvironment.

Abstract: Robust and accurate visual tracking is one of the most challenging computer vision problems. Due to the inherent lack of training data, a robust approach for constructing a target appearance model is crucial. Recently, discriminatively learned correlation filters (DCF) have been successfully applied to address this problem for tracking. These methods utilize a periodic assumption of the training samples to efficiently learn a classifier on all patches in the target neighborhood. However, the periodic assumption also introduces unwanted boundary effects, which severely degrade the quality of the tracking model. We propose Spatially Regularized Discriminative Correlation Filters (SRDCF) for tracking. A spatial regularization component is introduced in the learning to penalize correlation filter coefficients depending on their spatial location. Our SRDCF formulation allows the correlation filters to be learned on a significantly larger set of negative training samples, without corrupting the positive samples. We further propose an optimization strategy, based on the iterative Gauss-Seidel method, for efficient online learning of our SRDCF. Experiments are performed on four benchmark datasets: OTB-2013, ALOV++, OTB-2015, and VOT2014. Our approach achieves state-of-the-art results on all four datasets. On OTB-2013 and OTB-2015, we obtain an absolute gain of 8.0% and 8.2% respectively, in mean overlap precision, compared to the best existing trackers.

Abstract: Importance Systematic reviews of diagnostic test accuracy synthesize data from primary diagnostic studies that have evaluated the accuracy of 1 or more index tests against a reference standard, provide estimates of test performance, allow comparisons of the accuracy of different tests, and facilitate the identification of sources of variability in test accuracy. Objective To develop the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) diagnostic test accuracy guideline as a stand-alone extension of the PRISMA statement. Modifications to the PRISMA statement reflect the specific requirements for reporting of systematic reviews and meta-analyses of diagnostic test accuracy studies and the abstracts for these reviews. Design Established standards from the Enhancing the Quality and Transparency of Health Research (EQUATOR) Network were followed for the development of the guideline. The original PRISMA statement was used as a framework on which to modify and add items. A group of 24 multidisciplinary experts used a systematic review of articles on existing reporting guidelines and methods, a 3-round Delphi process, a consensus meeting, pilot testing, and iterative refinement to develop the PRISMA diagnostic test accuracy guideline. The final version of the PRISMA diagnostic test accuracy guideline checklist was approved by the group. Findings The systematic review (produced 64 items) and the Delphi process (provided feedback on 7 proposed items; 1 item was later split into 2 items) identified 71 potentially relevant items for consideration. The Delphi process reduced these to 60 items that were discussed at the consensus meeting. Following the meeting, pilot testing and iterative feedback were used to generate the 27-item PRISMA diagnostic test accuracy checklist. To reflect specific or optimal contemporary systematic review methods for diagnostic test accuracy, 8 of the 27 original PRISMA items were left unchanged, 17 were modified, 2 were added, and 2 were omitted. Conclusions and Relevance The 27-item PRISMA diagnostic test accuracy checklist provides specific guidance for reporting of systematic reviews. The PRISMA diagnostic test accuracy guideline can facilitate the transparent reporting of reviews, and may assist in the evaluation of validity and applicability, enhance replicability of reviews, and make the results from systematic reviews of diagnostic test accuracy studies more useful.

Abstract: Weyl fermions possess exotic properties and can act like magnetic monopoles. Researchers show that TaAs is a Weyl semimetal, demonstrating for the first time that Weyl semimetals can be identified experimentally.

Abstract: The tree of life is one of the most important organizing principles in biology1. Gene surveys suggest the existence of an enormous number of branches2, but even an approximation of the full scale of the tree has remained elusive. Recent depictions of the tree of life have focused either on the nature of deep evolutionary relationships3–5 or on the known, well-classified diversity of life with an emphasis on eukaryotes6. These approaches overlook the dramatic change in our understanding of life's diversity resulting from genomic sampling of previously unexamined environments. New methods to generate genome sequences illuminate the identity of organisms and their metabolic capacities, placing them in community and ecosystem contexts7,8. Here, we use new genomic data from over 1,000 uncultivated and little known organisms, together with published sequences, to infer a dramatically expanded version of the tree of life, with Bacteria, Archaea and Eukarya included. The depiction is both a global overview and a snapshot of the diversity within each major lineage. The results reveal the dominance of bacterial diversification and underline the importance of organisms lacking isolated representatives, with substantial evolution concentrated in a major radiation of such organisms. This tree highlights major lineages currently underrepresented in biogeochemical models and identifies radiations that are probably important for future evolutionary analyses. An update to the ‘tree of life’ has revealed a dominance of bacterial diversity in many ecosystems and extensive evolution in some branches of the tree. It also highlights how few organisms we have been able to cultivate for further investigation.

Abstract: The problem of arbitrary object tracking has traditionally been tackled by learning a model of the object's appearance exclusively online, using as sole training data the video itself. Despite the success of these methods, their online-only approach inherently limits the richness of the model they can learn. Recently, several attempts have been made to exploit the expressive power of deep convolutional networks. However, when the object to track is not known beforehand, it is necessary to perform Stochastic Gradient Descent online to adapt the weights of the network, severely compromising the speed of the system. In this paper we equip a basic tracking algorithm with a novel fully-convolutional Siamese network trained end-to-end on the ILSVRC15 dataset for object detection in video. Our tracker operates at frame-rates beyond real-time and, despite its extreme simplicity, achieves state-of-the-art performance in multiple benchmarks.

Abstract: We propose an alternative generator architecture for generative adversarial networks, borrowing from style transfer literature. The new architecture leads to an automatically learned, unsupervised separation of high-level attributes (e.g., pose and identity when trained on human faces) and stochastic variation in the generated images (e.g., freckles, hair), and it enables intuitive, scale-specific control of the synthesis. The new generator improves the state-of-the-art in terms of traditional distribution quality metrics, leads to demonstrably better interpolation properties, and also better disentangles the latent factors of variation. To quantify interpolation quality and disentanglement, we propose two new, automated methods that are applicable to any generator architecture. Finally, we introduce a new, highly varied and high-quality dataset of human faces.

Abstract: We propose an efficient and unified framework, namely ThiNet, to simultaneously accelerate and compress CNN models in both training and inference stages. We focus on the filter level pruning, i.e., the whole filter would be discarded if it is less important. Our method does not change the original network structure, thus it can be perfectly supported by any off-the-shelf deep learning libraries. We formally establish filter pruning as an optimization problem, and reveal that we need to prune filters based on statistics information computed from its next layer, not the current layer, which differentiates ThiNet from existing methods. Experimental results demonstrate the effectiveness of this strategy, which has advanced the state-of-the-art. We also show the performance of ThiNet on ILSVRC-12 benchmark. ThiNet achieves 3.31 x FLOPs reduction and 16.63× compression on VGG-16, with only 0.52% top-5 accuracy drop. Similar experiments with ResNet-50 reveal that even for a compact network, ThiNet can also reduce more than half of the parameters and FLOPs, at the cost of roughly 1% top-5 accuracy drop. Moreover, the original VGG-16 model can be further pruned into a very small model with only 5.05MB model size, preserving AlexNet level accuracy but showing much stronger generalization ability.

Abstract: Data-independent acquisition (DIA) in liquid chromatography (LC) coupled to tandem mass spectrometry (MS/MS) provides comprehensive untargeted acquisition of molecular data. We provide an open-source software pipeline, which we call MS-DIAL, for DIA-based identification and quantification of small molecules by mass spectral deconvolution. For a reversed-phase LC-MS/MS analysis of nine algal strains, MS-DIAL using an enriched LipidBlast library identified 1,023 lipid compounds, highlighting the chemotaxonomic relationships between the algal strains.

Abstract: A recent article by Akira Takayama discusses an earlier paper by Harvey Averch and Leland L. Johnson on fair rate of return regulation of public utilities. Although Takayama (p. 255) agrees with Averch and Johnson's general conclusions "that a firm will tend to increase its investment with the introduction of an active constraint" on its rate of return, he criticizes the A-J argument as being "confusing, ambiguous, and in error." Takayama then attempts a clarification, and presents a new formulation which leads to the A-J result quoted above. This comment will discuss several of Takayama's criticisms in addition to showing that the so-called "A-J Effect" cannot be derived from the basic assumptions made by both Averch and Johnson and Takayama.1 We will show that the very assumptions used to prove the A-J Effect, by defining the region of X, require an assumption that the A-J Effect exists in the first place.

Abstract: The emergence of new sequencing technologies has facilitated the use of bacterial whole genome alignments for evolutionary studies and outbreak analyses. These datasets, of increasing size, often include examples of multiple different mechanisms of horizontal sequence transfer resulting in substantial alterations to prokaryotic chromosomes. The impact of these processes demands rapid and flexible approaches able to account for recombination when reconstructing isolates' recent diversification. Gubbins is an iterative algorithm that uses spatial scanning statistics to identify loci containing elevated densities of base substitutions suggestive of horizontal sequence transfer while concurrently constructing a maximum likelihood phylogeny based on the putative point mutations outside these regions of high sequence diversity. Simulations demonstrate the algorithm generates highly accurate reconstructions under realistically parameterized models of bacterial evolution, and achieves convergence in only a few hours on alignments of hundreds of bacterial genome sequences. Gubbins is appropriate for reconstructing the recent evolutionary history of a variety of haploid genotype alignments, as it makes no assumptions about the underlying mechanism of recombination. The software is freely available for download at github.com/sanger-pathogens/Gubbins, implemented in Python and C and supported on Linux and Mac OS X.

Abstract: Despite a century of research on complex traits in humans, the relative importance and specific nature of the influences of genes and environment on human traits remain controversial. We report a meta-analysis of twin correlations and reported variance components for 17,804 traits from 2,748 publications including 14,558,903 partly dependent twin pairs, virtually all published twin studies of complex traits. Estimates of heritability cluster strongly within functional domains, and across all traits the reported heritability is 49%. For a majority (69%) of traits, the observed twin correlations are consistent with a simple and parsimonious model where twin resemblance is solely due to additive genetic variation. The data are inconsistent with substantial influences from shared environment or non-additive genetic variation. This study provides the most comprehensive analysis of the causes of individual differences in human traits thus far and will guide future gene-mapping efforts. All the results can be visualized using the MaTCH webtool.

Abstract: Long-term scenarios play an important role in research on global environmental change. The climate change research community is developing new scenarios integrating future changes in climate and society to investigate climate impacts as well as options for mitigation and adaptation. One component of these new scenarios is a set of alternative futures of societal development known as the shared socioeconomic pathways (SSPs). The conceptual framework for the design and use of the SSPs calls for the development of global pathways describing the future evolution of key aspects of society that would together imply a range of challenges for mitigating and adapting to climate change. Here we present one component of these pathways: the SSP narratives, a set of five qualitative descriptions of future changes in demographics, human development, economy and lifestyle, policies and institutions, technology, and environment and natural resources. We describe the methods used to develop the narratives as well as how these pathways are hypothesized to produce particular combinations of challenges to mitigation and adaptation. Development of the narratives drew on expert opinion to (1) identify key determinants of these challenges that were essential to incorporate in the narratives and (2) combine these elements in the narratives in a manner consistent with scholarship on their inter-relationships. The narratives are intended as a description of plausible future conditions at the level of large world regions that can serve as a basis for integrated scenarios of emissions and land use, as well as climate impact, adaptation and vulnerability analyses.

Abstract: Existing deep learning based image inpainting methods use a standard convolutional network over the corrupted image, using convolutional filter responses conditioned on both valid pixels as well as the substitute values in the masked holes (typically the mean value). This often leads to artifacts such as color discrepancy and blurriness. Post-processing is usually used to reduce such artifacts, but are expensive and may fail. We propose the use of partial convolutions, where the convolution is masked and renormalized to be conditioned on only valid pixels. We further include a mechanism to automatically generate an updated mask for the next layer as part of the forward pass. Our model outperforms other methods for irregular masks. We show qualitative and quantitative comparisons with other methods to validate our approach.

Abstract: Coastal zones are exposed to a range of coastal hazards including sea-level rise with its related effects. At the same time, they are more densely populated than the hinterland and exhibit higher rates of population growth and urbanisation. As this trend is expected to continue into the future, we investigate how coastal populations will be affected by such impacts at global and regional scales by the years 2030 and 2060. Starting from baseline population estimates for the year 2000, we assess future population change in the low-elevation coastal zone and trends in exposure to 100-year coastal floods based on four different sea-level and socio-economic scenarios. Our method accounts for differential growth of coastal areas against the land-locked hinterland and for trends of urbanisation and expansive urban growth, as currently observed, but does not explicitly consider possible displacement or out-migration due to factors such as sea-level rise. We combine spatially explicit estimates of the baseline population with demographic data in order to derive scenario-driven projections of coastal population development. Our scenarios show that the number of people living in the low-elevation coastal zone, as well as the number of people exposed to flooding from 1-in-100 year storm surge events, is highest in Asia. China, India, Bangladesh, Indonesia and Viet Nam are estimated to have the highest total coastal population exposure in the baseline year and this ranking is expected to remain largely unchanged in the future. However, Africa is expected to experience the highest rates of population growth and urbanisation in the coastal zone, particularly in Egypt and sub-Saharan countries in Western and Eastern Africa. The results highlight countries and regions with a high degree of exposure to coastal flooding and help identifying regions where policies and adaptive planning for building resilient coastal communities are not only desirable but essential. Furthermore, we identify needs for further research and scope for improvement in this kind of scenario-based exposure analysis.

Abstract: Recent applications of Convolutional Neural Networks (ConvNets) for human action recognition in videos have proposed different solutions for incorporating the appearance and motion information. We study a number of ways of fusing ConvNet towers both spatially and temporally in order to best take advantage of this spatio-temporal information. We make the following findings: (i) that rather than fusing at the softmax layer, a spatial and temporal network can be fused at a convolution layer without loss of performance, but with a substantial saving in parameters; (ii) that it is better to fuse such networks spatially at the last convolutional layer than earlier, and that additionally fusing at the class prediction layer can boost accuracy; finally (iii) that pooling of abstract convolutional features over spatiotemporal neighbourhoods further boosts performance. Based on these studies we propose a new ConvNet architecture for spatiotemporal fusion of video snippets, and evaluate its performance on standard benchmarks where this architecture achieves state-of-the-art results.

Abstract: This paper aims to develop a method than can accurately estimate the crowd count from an individual image with arbitrary crowd density and arbitrary perspective. To this end, we have proposed a simple but effective Multi-column Convolutional Neural Network (MCNN) architecture to map the image to its crowd density map. The proposed MCNN allows the input image to be of arbitrary size or resolution. By utilizing filters with receptive fields of different sizes, the features learned by each column CNN are adaptive to variations in people/head size due to perspective effect or image resolution. Furthermore, the true density map is computed accurately based on geometry-adaptive kernels which do not need knowing the perspective map of the input image. Since exiting crowd counting datasets do not adequately cover all the challenging situations considered in our work, we have collected and labelled a large new dataset that includes 1198 images with about 330,000 heads annotated. On this challenging new dataset, as well as all existing datasets, we conduct extensive experiments to verify the effectiveness of the proposed model and method. In particular, with the proposed simple MCNN model, our method outperforms all existing methods. In addition, experiments show that our model, once trained on one dataset, can be readily transferred to a new dataset.

Abstract: Numerous important problems can be framed as learning from graph data. We propose a framework for learning convolutional neural networks for arbitrary graphs. These graphs may be undirected, directed, and with both discrete and continuous node and edge attributes. Analogous to image-based convolutional networks that operate on locally connected regions of the input, we present a general approach to extracting locally connected regions from graphs. Using established benchmark data sets, we demonstrate that the learned feature representations are competitive with state of the art graph kernels and that their computation is highly efficient.

Abstract: The SARS-CoV-2 coronavirus, the etiologic agent responsible for COVID-19 coronavirus disease, is a global threat. To better understand viral tropism, we assessed the RNA expression of the coronavirus receptor, ACE2, as well as the viral S protein priming protease TMPRSS2 thought to govern viral entry in single-cell RNA-sequencing (scRNA-seq) datasets from healthy individuals generated by the Human Cell Atlas consortium. We found that ACE2, as well as the protease TMPRSS2, are differentially expressed in respiratory and gut epithelial cells. In-depth analysis of epithelial cells in the respiratory tree reveals that nasal epithelial cells, specifically goblet/secretory cells and ciliated cells, display the highest ACE2 expression of all the epithelial cells analyzed. The skewed expression of viral receptors/entry-associated proteins towards the upper airway may be correlated with enhanced transmissivity. Finally, we showed that many of the top genes associated with ACE2 airway epithelial expression are innate immune-associated, antiviral genes, highly enriched in the nasal epithelial cells. This association with immune pathways might have clinical implications for the course of infection and viral pathology, and highlights the specific significance of nasal epithelia in viral infection. Our findings underscore the importance of the availability of the Human Cell Atlas as a reference dataset. In this instance, analysis of the compendium of data points to a particularly relevant role for nasal goblet and ciliated cells as early viral targets and potential reservoirs of SARS-CoV-2 infection. This, in turn, serves as a biological framework for dissecting viral transmission and developing clinical strategies for prevention and therapy.

Abstract: In the last years, Artificial Intelligence (AI) has achieved a notable momentum that may deliver the best of expectations over many application sectors across the field. For this to occur, the entire community stands in front of the barrier of explainability, an inherent problem of AI techniques brought by sub-symbolism (e.g. ensembles or Deep Neural Networks) that were not present in the last hype of AI. Paradigms underlying this problem fall within the so-called eXplainable AI (XAI) field, which is acknowledged as a crucial feature for the practical deployment of AI models. This overview examines the existing literature in the field of XAI, including a prospect toward what is yet to be reached. We summarize previous efforts to define explainability in Machine Learning, establishing a novel definition that covers prior conceptual propositions with a major focus on the audience for which explainability is sought. We then propose and discuss about a taxonomy of recent contributions related to the explainability of different Machine Learning models, including those aimed at Deep Learning methods for which a second taxonomy is built. This literature analysis serves as the background for a series of challenges faced by XAI, such as the crossroads between data fusion and explainability. Our prospects lead toward the concept of Responsible Artificial Intelligence, namely, a methodology for the large-scale implementation of AI methods in real organizations with fairness, model explainability and accountability at its core. Our ultimate goal is to provide newcomers to XAI with a reference material in order to stimulate future research advances, but also to encourage experts and professionals from other disciplines to embrace the benefits of AI in their activity sectors, without any prior bias for its lack of interpretability.

Abstract: Immunotherapy induces durable responses in a subset of patients with cancer. High tumor mutational burden (TMB) may be a response biomarker for PD-1/PD-L1 blockade in tumors such as melanoma and non-small cell lung cancer (NSCLC). Our aim was to examine the relationship between TMB and outcome in diverse cancers treated with various immunotherapies. We reviewed data on 1,638 patients who had undergone comprehensive genomic profiling and had TMB assessment. Immunotherapy-treated patients (N = 151) were analyzed for response rate (RR), progression-free survival (PFS), and overall survival (OS). Higher TMB was independently associated with better outcome parameters (multivariable analysis). The RR for patients with high (≥20 mutations/mb) versus low to intermediate TMB was 22/38 (58%) versus 23/113 (20%; P = 0.0001); median PFS, 12.8 months vs. 3.3 months (P ≤ 0.0001); median OS, not reached versus 16.3 months (P = 0.0036). Results were similar when anti-PD-1/PD-L1 monotherapy was analyzed (N = 102 patients), with a linear correlation between higher TMB and favorable outcome parameters; the median TMB for responders versus nonresponders treated with anti-PD-1/PD-L1 monotherapy was 18.0 versus 5.0 mutations/mb (P < 0.0001). Interestingly, anti-CTLA4/anti-PD-1/PD-L1 combinations versus anti-PD-1/PD-L1 monotherapy was selected as a factor independent of TMB for predicting better RR (77% vs. 21%; P = 0.004) and PFS (P = 0.024). Higher TMB predicts favorable outcome to PD-1/PD-L1 blockade across diverse tumors. Benefit from dual checkpoint blockade did not show a similarly strong dependence on TMB. Mol Cancer Ther; 16(11); 2598-608. ©2017 AACR.