Dense Variants Research Articles

DLAS: A Conceptual Model for Across-Stack Deep Learning Acceleration

Deep Neural Networks (DNNs) are very computationally demanding, which presents a significant barrier to their deployment, especially on resource-constrained devices. Significant work from both the machine learning and computing systems communities has attempted to accelerate DNNs. However, the number of techniques available and the required domain knowledge for their exploration continues to grow, making design space exploration (DSE) increasingly difficult. To unify the perspectives from these two communities, this paper introduces the Deep Learning Acceleration Stack (DLAS), a conceptual model for DNN deployment and acceleration. We adopt a six-layer representation that organizes and illustrates the key areas for DNN acceleration, from machine learning to software and computer architecture. We argue that the DLAS model balances simplicity and expressiveness, assisting practitioners from various domains in tackling co-design acceleration challenges. We demonstrate the interdependence of the DLAS layers, and thus the need for co-design, through an across-stack perturbation study, using a modified tensor compiler to generate experiments for combinations of a few parameters across the DLAS layers. Our perturbation study assesses the impact on inference time and accuracy when varying DLAS parameters across two datasets, seven popular DNN architectures, four compression techniques, three algorithmic primitives (with sparse and dense variants), untuned and auto-scheduled code generation, and four hardware platforms. The study observes significant changes in the relative performance of design choices with the introduction of new DLAS parameters (e.g., the fastest algorithmic primitive varies with the level of quantization). Given the strong evidence for the need for co-design, and the high costs of DSE, DLAS offers a valuable conceptual model for better exploring advanced co-designed accelerated deep learning solutions.

The Aryne-Abramov Reaction as a 1,2-Benzdiyne Platform for the Generation and Solvent-Dependent Trapping of 3-Phosphonyl Benzynes.

Synthetic methodology utilizing two aryne intermediates (i.e., a formal benzdiyne) enables the rapid generation of structurally complex molecules with diverse functionality. This report describes the sequential generation of two ortho-benzyne intermediates for the synthesis of 2,3-disubstituted aryl phosphonates. Aryl phosphonates have proven useful in medicinal chemistry and materials science, and the reported methodology provides a two-step route to functionally dense variants by way of 3-phosphonyl benzyne intermediates. The process begins with regioselective trapping of a 3-trifloxybenzyne intermediate by an O-silyl phosphite in an Abramov-like reaction to bond the strained Csp carbons with phosphorus and silicon. Standard aryne-generating conditions follow to convert the resulting 2-silylphenyl triflate into a 3-phosphonyl benzyne, which readily reacts with numerous aryne trapping reactants to form a variety of 2,3-difunctionalized aryl phosphonate products. DFT computational studies shed light on important mechanistic details and revealed that 3-phosphonyl benzynes are highly polarizable. Specifically, the distortion in the internal bond angles at each of the Csp atoms was strongly influenced by both the electronegativity of the phosphonate ester groups as well as the dielectric of the computational solvation model. These effects were verified experimentally as the regioselectivity of benzyl azide trapping increased with more electronegative esters and/or increasingly polar solvents. Conversely, replacing the conventional solvent, acetonitrile, with nonpolar alternatives provided attenuated or even inverted selectivities. Overall, these studies showcase new reactivity of benzyne intermediates and extend the aryne relay methodology to include organophosphonates. Furthermore, this work demonstrates that the regioselectivity of aryne trapping reactions could be tuned by simply changing the solvent.

Deep convolutional reconstruction for gradient-domain rendering

It has been shown that rendering in the gradient domain, i.e., estimating finite difference gradients of image intensity using correlated samples, and combining them with direct estimates of pixel intensities by solving a screened Poisson problem, often offers fundamental benefits over merely sampling pixel intensities. The reasons can be traced to the frequency content of the light transport integrand and its interplay with the gradient operator. However, while they often yield state of the art performance among algorithms that are based on Monte Carlo sampling alone, gradient-domain rendering algorithms have, until now, not generally been competitive with techniques that combine Monte Carlo sampling with post-hoc noise removal using sophisticated non-linear filtering. Drawing on the power of modern convolutional neural networks, we propose a novel reconstruction method for gradient-domain rendering. Our technique replaces the screened Poisson solver of previous gradient-domain techniques with a novel dense variant of the U-Net autoencoder, additionally taking auxiliary feature buffers as inputs. We optimize our network to minimize a perceptual image distance metric calibrated to the human visual system. Our results significantly improve the quality obtained from gradient-domain path tracing, allowing it to overtake state-of-the-art comparison techniques that denoise traditional Monte Carlo samplings. In particular, we observe that the correlated gradient samples --- that offer information about the smoothness of the integrand unavailable in standard Monte Carlo sampling --- notably improve image quality compared to an equally powerful neural model that does not make use of gradient samples.

Mapping-free variant calling using haplotype reconstruction from k-mer frequencies.

MotivationThe standard protocol for detecting variation in DNA is to map millions of short sequence reads to a known reference and find loci that differ. While this approach works well, it cannot be applied where the sample contains dense variants or is too distant from known references. De novo assembly or hybrid methods can recover genomic variation, but the cost of computation is often much higher. We developed a novel k-mer algorithm and software implementation, Kestrel, capable of characterizing densely packed SNPs and large indels without mapping, assembly or de Bruijn graphs.ResultsWhen applied to mosaic penicillin binding protein (PBP) genes in Streptococcus pneumoniae, we found near perfect concordance with assembled contigs at a fraction of the CPU time. Multilocus sequence typing (MLST) with this approach was able to bypass de novo assemblies. Kestrel has a very low false-positive rate when applied to the whole genome, and while Kestrel identified many variants missed by other methods, limitations of a purely k-mer based approach affect overall sensitivity.Availability and implementationSource code and documentation for a Java implementation of Kestrel can be found at https://github.com/paudano/kestrel. All test code for this publication is located at https://github.com/paudano/kescases.Supplementary information Supplementary data are available at Bioinformatics online.

Whole-genome characterization in pedigreed non-human primates using genotyping-by-sequencing (GBS) and imputation.

BackgroundRhesus macaques are widely used in biomedical research, but the application of genomic information in this species to better understand human disease is still in its infancy. Whole-genome sequence (WGS) data in large pedigreed macaque colonies could provide substantial experimental power for genetic discovery, but the collection of WGS data in large cohorts remains a formidable expense. Here, we describe a cost-effective approach that selects the most informative macaques in a pedigree for 30X WGS, followed by low-cost genotyping-by-sequencing (GBS) at 30X on the remaining macaques in order to generate sparse genotype data at high accuracy. Dense variants from the selected macaques with WGS data are then imputed into macaques having only sparse GBS data, resulting in dense genome-wide genotypes throughout the pedigree.ResultsWe developed GBS for the macaque genome using a digestion with PstI, followed by sequencing of size-selected fragments at 30X coverage. From GBS sequence data collected on all individuals in a 16-member pedigree, we characterized high-confidence genotypes at 22,455 single nucleotide variant (SNV) sites that were suitable for guiding imputation of dense sequence data from WGS. To characterize dense markers for imputation, we performed WGS at 30X coverage on nine of the 16 individuals, yielding 10,193,425 high-confidence SNVs. To validate the use of GBS data for facilitating imputation, we initially focused on chromosome 19 as a test case, using an optimized panel of 833 sparse, evenly-spaced markers from GBS and 5,010 dense markers from WGS. Using the method of “Genotype Imputation Given Inheritance” (GIGI), we evaluated the effects on imputation accuracy of 3 different strategies for selecting individuals for WGS, including 1) using “GIGI-Pick” to select the most informative individuals, 2) using the most recent generation, or 3) using founders only. We also evaluated the effects on imputation accuracy of using a range of from 1 to 9 WGS individuals for imputation. We found that the GIGI-Pick algorithm for selection of WGS individuals outperformed common heuristic approaches, and that genotype numbers and accuracy improved very little when using >5 WGS individuals for imputation. Informed by our findings, we used 4 macaques with WGS data to impute variants at up to 7,655,491 sites spanning all 20 autosomes in the 12 remaining macaques, based on their GBS genotypes at only 17,158 loci. Using a strict confidence threshold, we imputed an average of 3,680,238 variants per individual at >99 % accuracy, or an average 4,458,883 variants per individual at a more relaxed threshold, yielding >97 % accuracy.ConclusionsWe conclude that an optimal tradeoff between genotype accuracy, number of imputed genotypes, and overall cost exists at the ratio of one individual selected for WGS using the GIGI-Pick algorithm, per 3–5 relatives selected for GBS. This approach makes feasible the collection of accurate, dense genome-wide sequence data in large pedigreed macaque cohorts without the need for more expensive WGS data on all individuals.Electronic supplementary materialThe online version of this article (doi:10.1186/s12864-016-2966-x) contains supplementary material, which is available to authorized users.

Detection of Mendelian Consistent Genotyping Errors in Pedigrees

Detection of genotyping errors is a necessary step to minimize false results in genetic analysis. This is especially important when the rate of genotyping errors is high, as has been reported for high-throughput sequence data. To detect genotyping errors in pedigrees, Mendelian inconsistent (MI) error checks exist, as do multi-point methods that flag Mendelian consistent (MC) errors for sparse multi-allelic markers. However, few methods exist for detecting MC genotyping errors, particularly for dense variants on large pedigrees. Here, we introduce an efficient method to detect MC errors even for very dense variants (e.g., SNPs and sequencing data) on pedigrees that may be large. Our method first samples inheritance vectors (IVs) using a moderately sparse but informative set of markers using a Markov chain Monte Carlo-based sampler. Using sampled IVs, we considered two test statistics to detect MC genotyping errors: the percentage of IVs inconsistent with observed genotypes (A1) or the posterior probability of error configurations (A2). Using simulations, we show that this method, even with the simpler A1 statistic, is effective for detecting MC genotyping errors in dense variants, with sensitivity almost as high as the theoretical best sensitivity possible. We also evaluate the effectiveness of this method as a function of parameters, when including the observed pattern for genotype, density of framework markers, error rate, allele frequencies, and number of sampled inheritance vectors. Our approach provides a line of defense against false findings based on the use of dense variants in pedigrees.

X-Ray Fluorescence-Analyzed Mineral Micronutrient DensityVariation Among Kenyan Local Vegetables

<p>In Africa staple cereal foods are often eaten along with the African Leafy Vegetables (ALVs) rendering meals rich in micronutrients. To increase the mineral micronutrient value, it is imperative to intensify their cultivation given their under-utilized diversity, neglect until needed, slow rate of incorporation into crop value chains for lack clear micronutrient nutrition driven agenda. As a way of focusing on their ionomic nutraceutic attribute potential, the first objective was to investigate accession ionome differentials on the basis a soil mineral criterion.The second objective was to determine a method for variation assessment of ionomic micronutrient dense variants in key Kenyan local vegetables. Four ALVs species constituting 25 accessions were collected in short season rains of 2003 from north and souther stretch of western Kenyaand in early Long rains of 2004 from south eastern Kenya. A stratified sampling design organized had three collection points per farm; on three farms per site; and at three sites per phyto-region. ALV samples together with accompanying soils were Energy Dispersive X-ray fluorescence (XRF) analysed. The data were used, first, for ionomically differentiating populations by way of singling out respective element influenced accession differentials (SELIACDs). Second, elements were jointly resolved into multi-element influenced accession differentials (MELIACDs). Agro-edaphic effects on population ionome niching were assessed using analysis of variance and graphical aids. Geometric means were generated and awarded nutrametric merit scores to allow for nutrametric grading of accessions. Results suggest a great deal of ionomic phenotypic plasticity among the local vegetable accessions as a function at scale of farm, site and/or region soils. The SELIACD method was useful for piecemeal separation of accession on a single element basis but which method would require the development of a selection index with a certainty of a significant genetic gain at the onset. The joint MELIACD and the nutrametric grading methods are proposed as a promising basis for prebreeding tool prioritization given the XRF-analytic novelty and the emerging interest in ALV nutraceutical cropping.</p>

Genetic association of the gene encoding RPGRIP1L with susceptibility to vascular dementia

A previous genome-wide association study (GWAS) failed to discover any nucleotide sequence variant associated with susceptibility to vascular dementia (VaD) and remained a problem of false negatives produced by a low statistical power. The current study was conducted to identify such potential false negatives and to provide comprehensive evidence for the most plausible predisposing genetic factor using large-scale Korean cohorts. We identified the gene encoding retinitis pigmentosa GTPase regulator-interacting protein 1-like (RPGRIP1L) with multiple nucleotide variants associated with susceptibility to VaD by a modest significant threshold (P<10−4). Genetic associations were intensively examined with its sequence variants using 207 VaD patients and 207 age- and gender-matched control subjects. Genetic association analysis with dense variants in the region associated with VaD revealed 3 variants (P<0.0017) in strong linkage. Further analysis with VaD-related phenotypes using Korean Association REsource (KARE) cohort data showed that the region of the gene was associated with alanine aminotransferase (ALT), aspartate aminotransferase (AST), and blood pressure (BP) (P<7.6×10−4). The current study provided the first evidence of the association between RPGRIP1L gene and susceptibility of VaD. Functional studies are needed to understand underlying biological mechanism of the genetic association.

Self-Compensating Design for Reduction of Timing and Leakage Sensitivity to Systematic Pattern-Dependent Variation

Critical dimension (CD) variation caused by defocus is largely systematic with dense lines ldquosmilingrdquo through focus while isolated lines ldquofrown.rdquo In this paper, we propose a new design methodology that allows explicit compensation of focus-dependent CD variation, in particular, either within a cell (self-compensated cells) or across cells in a critical path (self-compensated design). By creating iso and dense variants for each library cell, we can achieve designs that are more robust to focus variation. Optimization with a mixture of dense and iso cell variants is possible, both for area and leakage power in timing constraints (critical delay), with the latter an interesting complement to existing leakage-reduction techniques, such as dual-Vth. We implement both a heuristic and mixed-integer linear-programming (MILP) solution methods to address this optimization and experimentally compare their results. Results indicate that designing with a self-compensated cell library incurs 12% area penalty and 6% leakage increase over a baseline library while compensating for focus-dependent CD variation (i.e., the design meets timing constraints across a large range of focus variation). We observe 27% area penalty and 7% leakage increase at the worst case defocus condition using only single-pitch cells. The area penalty of circuits after using both the heuristic and MILP optimization approaches is reduced to 3% while maintaining timing. We also apply the optimization to leakage, which traditionally shows very large variability due to its exponential relationship with gate CD. We conclude that a mixed iso/dense library that is combined with a sensitivity-based optimization approach yields much better area/timing/leakage tradeoffs than using a self-compensated cell library alone. Self-compensated designs show 25% less leakage power on average at the worst defocus condition compared to a design employing a conventional library for the benchmarks studied.

Anhydrous ammonioguanidinium(2+) and dihydrated bis[aminoguanidinium(1+)] hexafluorosilicates: new co-products of preparing ferroelectric ammonioguanidinium(2+) hexafluorozirconate.

Ammonioguanidinium hexafluorosilicate, CH8N4(2+).SiF6(2-), and bis(aminoguanidinium) hexafluorosilicate dihydrate, 2CH7N4+.SiF6(2-).2H2O, are new materials formed as by-products in course of preparing ferroelectric CH8N4ZrF6 in the presence of glassware. Their structures were determined for comparison with the corresponding hexafluorozirconates. All atoms including the eight H atoms in the CH8N4(2+) cation and the seven H atoms in the CH7N4+ cation have been located and refined with wR(F2) = 0.0653, R = 0.0255, S = 1.146 and wR(F2) = 0.0745, R = 0.0301, S = 1.065, respectively. The N2C-N-N backbone of the 2+ cation is close to planarity, while that of the 1+ cation does not differ significantly from planarity. The SiF6(2-) octahedron is nearly ideally regular in both materials, with < Si-F > = 1.684 (unbiassed estimator of standard uncertainty = 0.016) A in the anhydrous hexafluorosilicate and 1.6801 (unbiassed estimator of standard uncertainty = 0.0006) A in the dihydrate. The combination of coulombic and NH...F interactions in CH8N4SiF6 results in a relatively dense variant of the NaCl structure. In addition to similar forces, the dihydrate is also characterized by the role of the water molecule with its strong NH...O interactions; its packing efficiency is, however, appreciably less than that of the anhydrous hexafluorosilicate with an approximately 8% increase in void space. Cleaved crystals of the dihydrate are frequently twinned across the (001) composition plane, with a twofold rotation about the b axis as the twin operation.

Cheliped regeneration in the megalopa of the mud crab,Rhithropanopeus harrisii(Gould)

Summary The process of autotomy-induced cheliped regeneration in the megalopa of the mud crab, Rhithropanopeus harrisii has been studied by light microscopy and transmission electron microscopy. Following blastema eruption, four time intervals were selected arbitrarily to cover the entire period of limb regeneration. Epidermal proliferation preceded muscle and nerve regeneration. On the basis of morphologic observations, nerve tissue regenerated from existing pedal nerve. It is proposed that blood vessels arose from epidermal differentiation, while muscle differentiated from immigrant cells. Three types of haemocyte, resembling those described in adult crustaceans, were identified in the haemolymph of the larval limb bud: granular haemocytes, hyaline cells and lipoprotein cells. Two other immigrant cell types, which appear to be primitive in nature and may be variants of a single cell type, are also described. Morphological evidence is provided to suggest that the more electron dense variant differentiate...

Adult-onset Mesangiocapillary Glomerulonephritis: A Disease with a Poor Prognosis

A retrospective analysis of the records of 46 patients with adult-onset mesangiocapillary glomerulonephritis seen over a period of 15 years is reported. There were 28 males and 18 females with a mean age of 34 years. Mean observation period was 60 months. All patients had renal biopsies and characteristic changes, but no dense deposit variant was seen. HBsAg was positive in 20 per cent of the patients tested whereas in the general population it was about 10 per cent. Nephrotic syndrome was the commonest presentation. There was a mean delay of 8.5 months between the onset of symptoms and renal biopsy. Nineteen patients were given steroid/immunosuppressive therapy while 27 received no specific treatment. At the time of review, renal function was stable in 13 patients but progressively deteriorating in 33. In the latter, the slope of the graph of reciprocal of creatinine versus time could be accurately determined. Steroid/immunosuppressive therapy did not influence the rate of progression of renal failure but appeared to hasten and induce more partial remissions of the nephrotic state. There was no difference in cumulative renal survival between treated and untreated patients, and between HBsAg-positive and negative patients. Overall, 23 patients required dialysis/renal transplantation and an additional four had died of end-stage renal failure at the time of this review.

The distribution of mass in heteropolymer intermediate filaments assembled in vitro. Stem analysis of vimentin/desmin and bovine epidermal keratin.

We have studied the distribution of mass in several types of intermediate filaments (IF) assembled in vitro, by analyzing scanning transmission electron micrographs (STEM) of unstained specimens imaged in dark-field mode. Bovine epidermal keratin IF, which are obligate heteropolymers, were thus characterized and compared with facultative heteropolymers of vimentin and desmin and with earlier observations of homopolymer vimentin IF. The major components of each type of IF have linear densities of 37 kilodaltons/nm. Minor polymorphic variants are also present in each case, with linear densities of approximately 25 and 48 kilodaltons/nm, respectively. In view of the known subunit masses, these results are consistent with the proposition that at least three IF types (bovine epidermal keratin, vimentin, and desmin) are structurally homologous. For each type of IF studied, measurements taken near the ends of long IF or from short IF tend to have lower densities, characteristic of the least dense polymorphic variant. This form may represent a precursor "minimal core" structure in the sense of a minimal aggregate of protofilaments sufficiently stable to permit elongation. In the course of in vitro assembly, this form would be maturable to the normal IF structure at 37 kilodaltons/nm by the accretion of further protein. We also find the projected widths of these IF as measured from the STEM images to be significantly greater than the corresponding diameters determined in previous studies by conventional electron microscopy. Whereas the conventional techniques have routinely yielded diameters of 7-11 nm for IF contrasted by heavy metal staining, the STEM images of unstained IF provide estimates of 15 +/- 1 nm for the outer diameter of the major density class for each IF type studied and average values of 13.8 and 16.1 nm for the less dense and more dense variants, respectively.