The 1000 Genomes Project set out to provide a comprehensive description of common human genetic variation by applying whole-genome sequencing to a diverse set of individuals from multiple populations. Here we report completion of the project, having reconstructed the genomes of 2,504 individuals from 26 populations using a combination of low-coverage whole-genome sequencing, deep exome sequencing, and dense microarray genotyping. We characterized a broad spectrum of genetic variation, in total over 88 million variants (84.7 million single nucleotide polymorphisms (SNPs), 3.6 million short insertions/deletions (indels), and 60,000 structural variants), all phased onto high-quality haplotypes. This resource includes >99% of SNP variants with a frequency of >1% for a variety of ancestries. We describe the distribution of genetic variation across the global sample, and discuss the implications for common disease studies.

As a first step toward understanding how rare variants contribute to risk for complex diseases, we sequenced 15,585 human protein-coding genes to an average median depth of 111× in 2440 individuals of European (n = 1351) and African (n = 1088) ancestry. We identified over 500,000 single-nucleotide variants (SNVs), the majority of which were rare (86% with a minor allele frequency less than 0.5%), previously unknown (82%), and population-specific (82%). On average, 2.3% of the 13,595 SNVs each person carried were predicted to affect protein function of ∼313 genes per genome, and ∼95.7% of SNVs predicted to be functionally important were rare. This excess of rare functional variants is due to the combined effects of explosive, recent accelerated population growth and weak purifying selection. Furthermore, we show that large sample sizes will be required to associate rare variants with complex traits.

Tumour cells evade immune surveillance by upregulating the surface expression of programmed death-ligand 1 (PD-L1), which interacts with programmed death-1 (PD-1) receptor on T cells to elicit the immune checkpoint response. Anti-PD-1 antibodies have shown remarkable promise in treating tumours, including metastatic melanoma. However, the patient response rate is low. A better understanding of PD-L1-mediated immune evasion is needed to predict patient response and improve treatment efficacy. Here we report that metastatic melanomas release extracellular vesicles, mostly in the form of exosomes, that carry PD-L1 on their surface. Stimulation with interferon-γ (IFN-γ) increases the amount of PD-L1 on these vesicles, which suppresses the function of CD8 T cells and facilitates tumour growth. In patients with metastatic melanoma, the level of circulating exosomal PD-L1 positively correlates with that of IFN-γ, and varies during the course of anti-PD-1 therapy. The magnitudes of the increase in circulating exosomal PD-L1 during early stages of treatment, as an indicator of the adaptive response of the tumour cells to T cell reinvigoration, stratifies clinical responders from non-responders. Our study unveils a mechanism by which tumour cells systemically suppress the immune system, and provides a rationale for the application of exosomal PD-L1 as a predictor for anti-PD-1 therapy.

No abstract

Face alignment, which fits a face model to an image and extracts the semantic meanings of facial pixels, has been an important topic in CV community. However, most algorithms are designed for faces in small to medium poses (below 45• ), lacking the ability to align faces in large poses up to 90• . The challenges are three-fold: Firstly, the commonly used landmark-based face model assumes that all the landmarks are visible and is therefore not suitable for profile views. Secondly, the face appearance varies more dramatically across large poses, ranging from frontal view to profile view. Thirdly, labelling landmarks in large poses is extremely challenging since the invisible landmarks have to be guessed. In this paper, we propose a solution to the three problems in an new alignment framework, called 3D Dense Face Alignment (3DDFA), in which a dense 3D face model is fitted to the image via convolutional neutral network (CNN). We also propose a method to synthesize large-scale training samples in profile views to solve the third problem of data labelling. Experiments on the challenging AFLW database show that our approach achieves significant improvements over state-of-the-art methods.

Recently, very deep convolutional neural networks (CNNs) have been attracting considerable attention in image restoration. However, as the depth grows, the long-term dependency problem is rarely realized for these very deep models, which results in the prior states/layers having little influence on the subsequent ones. Motivated by the fact that human thoughts have persistency, we propose a very deep persistent memory network (MemNet) that introduces a memory block, consisting of a recursive unit and a gate unit, to explicitly mine persistent memory through an adaptive learning process. The recursive unit learns multi-level representations of the current state under different receptive fields. The representations and the outputs from the previous memory blocks are concatenated and sent to the gate unit, which adaptively controls how much of the previous states should be reserved, and decides how much of the current state should be stored. We apply MemNet to three image restoration tasks, i.e., image denosing, superresolution and JPEG deblocking. Comprehensive experiments demonstrate the necessity of the MemNet and its unanimous superiority on all three tasks over the state of the arts. Code is available at https://github.com/ tyshiwo/MemNet. the additive noise. With this mathematical model, extensive studies are conducted on many image restoration tasks, e.g., image denoising [2,5,9,37], single-image super-resolution (SISR) [15,38] and JPEG deblocking [18,26].As three classical image restoration tasks, image denoising aims to recover a clean image from a noisy observation, which commonly assumes additive white Gaussian noise with a standard deviation σ; single-image superresolution recovers a high-resolution (HR) image from a low-resolution (LR) image; and JPEG deblocking removes the blocking artifact caused by JPEG compression [7].Recently, due to the powerful learning ability, very deep convolutional neural network (CNN) is widely used to tackle the image restoration tasks. Kim et al. construct a 20-layer CNN structure named VDSR for SISR [20], and adopts residual learning to ease training difficulty. To control the parameter number of very deep models, the authors further introduce a recursive layer and propose a Deeply-Recursive Convolutional Network (DRCN) [21]. To mitegate training difficulty, Mao et al. [27] introduce symmetric skip connections into a 30-layer convolutional auto-encoder

Accurate deleteriousness prediction for nonsynonymous variants is crucial for distinguishing pathogenic mutations from background polymorphisms in whole exome sequencing (WES) studies. Although many deleteriousness prediction methods have been developed, their prediction results are sometimes inconsistent with each other and their relative merits are still unclear in practical applications. To address these issues, we comprehensively evaluated the predictive performance of 18 current deleteriousness-scoring methods, including 11 function prediction scores (PolyPhen-2, SIFT, MutationTaster, Mutation Assessor, FATHMM, LRT, PANTHER, PhD-SNP, SNAP, SNPs&GO and MutPred), 3 conservation scores (GERP++, SiPhy and PhyloP) and 4 ensemble scores (CADD, PON-P, KGGSeq and CONDEL). We found that FATHMM and KGGSeq had the highest discriminative power among independent scores and ensemble scores, respectively. Moreover, to ensure unbiased performance evaluation of these prediction scores, we manually collected three distinct testing datasets, on which no current prediction scores were tuned. In addition, we developed two new ensemble scores that integrate nine independent scores and allele frequency. Our scores achieved the highest discriminative power compared with all the deleteriousness prediction scores tested and showed low false-positive prediction rate for benign yet rare nonsynonymous variants, which demonstrated the value of combining information from multiple orthologous approaches. Finally, to facilitate variant prioritization in WES studies, we have pre-computed our ensemble scores for 87 347 044 possible variants in the whole-exome and made them publicly available through the ANNOVAR software and the dbNSFP database.

The purpose of the dbNSFP is to provide a one-stop resource for functional predictions and annotations for human non-synonymous single-nucleotide variants (nsSNVs) and splice site variants (ssSNVs), and to facilitate the steps of filtering and prioritizing SNVs from a large list of SNVs discovered in an exome-sequencing study. A list of all potential nsSNVs and ssSNVs based on the human reference sequence were created, functional predictions and annotations were curated and compiled for each SNV. Here we report a recent major update of the database to version 3.0. The SNV list has been rebuilt based on GENCODE 22 and currently the database includes 82,832,027 nsSNVs and ssSNVs. An attached database dbscSNV, which compiled all potential human SNVs within splicing consensus regions and their deleteriousness predictions, add another 15,030,459 potentially functional SNVs. Eleven prediction scores (MetaSVM, MetaLR, CADD, VEST3, PROVEAN, 4× fitCons, fathmm-MKL and DANN) and allele frequencies from the UK10K cohorts and the Exome Aggregation Consortium (ExAC), among others, have been added. The original seven prediction scores in v2.0 (SIFT, 2× Polyphen2, LRT, MutationTaster, MutationAssessor and FATHMM) as well as many SNV and gene functional annotations have been updated. dbNSFP v3.0 is freely available at http://sites.google.com/site/jpopgen/dbNSFP.