Data-driven Computations Research Articles

Data-driven risk/benefit estimator for multiple sclerosis therapies.

Multiple sclerosis (MS) disease-modifying treatments (DMTs) are tested in patients pre-selected for favorable risk/benefits ratios but prescribed broadly in clinical practice. We aimed to establish data-driven computations of individualized risk/benefit ratios to optimize MS care. We derived determinants of DMTs efficacy on disability progression from re-analysis and integration of 61 randomized, blinded Phase 2b/3 clinical trials that studied 46,611 patients for 91,787 patient-years. From each arm we extracted 80 and computed 30 features to identify and adjust for biases, and to use in multiple regression models. DMTs mortality risks were estimated from age mortality tables modified by published hazard ratios. Baseline characteristics of the recruited patients determine disability progression rates and DMTs efficacies with high effect sizes. DMTs efficacies increase with MS lesional activity (LA) measured by relapses or contrast-enhancing lesions and decrease with increasing age, disease duration and disability. Unexpectedly, as placebo arms' relapse rate rapidly declines with trial duration, efficacy of MS DMTs likewise decreases quickly with treatment duration. Conversely, DMTs morbidity/mortality risks increase with age, advanced disability, and comorbidities. We integrated these results into an interactive personalized web based DMTs risk/benefit estimator. Results predict that prescribing DMTs to patients traditionally excluded from MS clinical trials causes more harm than benefit. Treatment with high efficacy drugs at MS onset followed by de-escalation to DMTs that do not increase infectious risks would optimize risk/benefit. DMTs targeting mechanisms of progression independent of LA are greatly needed as current DMTs inhibit disability caused by LA only.

Image Synthesis for Solar Flare Prediction

Solar flare prediction is a topic of interest to many researchers owing to the potential of solar flares to affect various technological systems, both terrestrial and in orbit. In recent years, the forecasting task has become progressively more reliant on data-driven computations and machine-learning algorithms. Although these efforts have improved solar flare predictions, they still falter in doing so for large solar flares, in particular under operational conditions, since large-flare data are very scarce and labeled data are heavily imbalanced. In this work, we seek to address this fundamental issue and present a scheme for generating synthetic magnetograms to reduce the imbalance in the data. Our method consists of (1) synthetic oversampling of line-of-sight magnetograms using Gaussian mixture model representation, followed by (2) a global optimization technique to ensure consistency of both physical features and flare precursors, and (3) the mapping of the generated representations to realistic magnetogram images using deep generative models. We show that these synthetically generated data indeed improve the capacity of solar flare prediction models and that, when tested on such a state-of-the-art model, it significantly enhances its forecasting performance, achieving an F1-score as high as 0.43 ± 0.08 and a true skill statistic of 0.64 ± 0.10 for X-class flares in the 24 hr operational solar flare data split.

Tau data-driven evaluation of the Hadronic Vacuum Polarization [formula omitted

Windows in Euclidean time have become a standard tool for comparing lattice QCD and data-driven computations of the hadronic vacuum polarization (HVP) contribution to the muon g−2. Here we review our results, obtained using isospin-rotated τ−→π−π0ντ data instead of e+e−→π+π− measurements, and compare them to other approaches. Consistency of the tau-based and lattice results hints to underestimated uncertainties in the e+e− data. If that is the case, the theory prediction of the muon g−2 would only lie at ≲2.5σ from its measured value.

JAX-FEM: A differentiable GPU-accelerated 3D finite element solver for automatic inverse design and mechanistic data science

This paper introduces JAX-FEM, an open-source differentiable finite element method (FEM) library. Constructed on top of Google JAX, a rising machine learning library focusing on high-performance numerical computing, JAX-FEM is implemented with pure Python while scalable to efficiently solve problems with moderate to large sizes. For example, in a 3D tensile loading problem with 7.7 million degrees of freedom, JAX-FEM with GPU achieves around 10× acceleration compared to a commercial FEM code depending on platform. Beyond efficiently solving forward problems, JAX-FEM employs the automatic differentiation technique so that inverse problems are solved in a fully automatic manner without the need to manually derive sensitivities. Examples of 3D topology optimization of nonlinear materials are shown to achieve optimal compliance. Finally, JAX-FEM is an integrated platform for machine learning-aided computational mechanics. We show an example of data-driven multi-scale computations of a composite material where JAX-FEM provides an all-in-one solution from microscopic data generation and model training to macroscopic FE computations. The source code of the library and these examples are shared with the community to facilitate computational mechanics research. Program summaryProgram Title: JAX-FEMCPC Library link to program files:https://doi.org/10.17632/hgwshjbcw6.1Developer's repository link:https://github.com/tianjuxue/jax-am/tree/main/jax_am/femLicensing provisions: GPLv3Programming language: PythonNature of problem: Implementation of the finite element method (FEM) with several appealing features that classic FEM software typically does not have: realized with pure Python; running on CPU/GPU; differentiable for solving PDE-constrained optimization problems; seamless integration with machine learning.Solution method: Our framework JAX-FEM is based on Google JAX, a high-performance numerical computing library with automatic differentiation features and supporting both CPU/GPU. Unlike classic FEM software written in Fortran or C/C++, JAX-FEM is implemented with pure Python and can easily be installed as a Python package. We demonstrate our software by solving problems including forward PDE prediction, inverse design/optimization, and data-driven analysis.

Performance effectiveness of vital parameter combinations for early warning of sepsis-an exhaustive study using machine learning.

ObjectiveTo carry out exhaustive data-driven computations for the performance of noninvasive vital signs heart rate (HR), respiratory rate (RR), peripheral oxygen saturation (SpO2), and temperature (Temp), considered both independently and in all possible combinations, for early detection of sepsis.Materials and methodsBy extracting features interpretable by clinicians, we applied Gradient Boosted Decision Tree machine learning on a dataset of 2630 patients to build 240 models. Validation was performed on a geographically distinct dataset. Relative to onset, predictions were clocked as per 16 pairs of monitoring intervals and prediction times, and the outcomes were ranked.ResultsThe combination of HR and Temp was found to be a minimal feature set yielding maximal predictability with area under receiver operating curve 0.94, sensitivity of 0.85, and specificity of 0.90. Whereas HR and RR each directly enhance prediction, the effects of SpO2 and Temp are significant only when combined with HR or RR. In benchmarking relative to standard methods Systemic Inflammatory Response Syndrome (SIRS), National Early Warning Score (NEWS), and quick-Sequential Organ Failure Assessment (qSOFA), Vital-SEP outperformed all 3 of them.ConclusionIt can be concluded that using intensive care unit data even 2 vital signs are adequate to predict sepsis upto 6 h in advance with promising accuracy comparable to standard scoring methods and other sepsis predictive tools reported in literature. Vital-SEP can be used for fast-track prediction especially in limited resource hospital settings where laboratory based hematologic or biochemical assays may be unavailable, inaccurate, or entail clinically inordinate delays. A prospective study is essential to determine the clinical impact of the proposed sepsis prediction model and evaluate other outcomes such as mortality and duration of hospital stay.

Data-driven finite element computation of open-cell foam structures

This paper presents a model-free data-driven strategy for linear and non-linear finite element computations of open-cell foam. Employing sets of material data, the data-driven problem is formulated as the minimization of a distance function subjected to the physical constraints from the kinematics and the balance laws of the mechanical problem.The material data sets of the foam are deduced here from representative microscopic material volumes. These volume elements capture the stochastic characteristics of the open-cell microstructure and the properties of the polyurethane material. Their computation provides the required stress–strain data used in the macroscopic finite element computations without postulating a specific constitutive model. The paper shows how to derive suitable material data sets for the different cases of (non-)linear and (an-)isotropic material behavior efficiently.Exemplarily, we compare data-driven finite element computations with linearized and finite deformations and show that in a data-driven computation the linear kinematic is sufficiently accurate to capture the material’s non-linearity up to 50% of straining. The numerical example of a rubber sealing illustrates possible areas of application, the expenditure, and the proposed strategy’s versatility.

Applying Data Structure Succinctness to Graph Numbering For Efficient Graph Analysis

Graph algorithms have inherent characteristics, including data-driven computations and poor locality. These characteristics expose graph algorithms to several challenges, because most well studied (parallel) abstractions and implementation are not suitable for them. In our previous work [21, 22, 24], we show how to use some complex-network properties, including community structure and heterogeneity of node degree, to improve performance, by a proper memory management (Cn-order) and an appropriate thread scheduling (comm-deg-scheduling). In recent work [23], Besta et al. proposed log(graph), a graph representation that outperforms existing graph compression algorithms. In this paper, we show that our graph numbering heuristic and our scheduling heuristics can be improved when they are combined with log(graph) data structure. Experiments were made on multi-core machines. For example, on one node of a multi-core machine (Troll from Grid'5000), we showed that when combining our previously proposed heuristics with graph compression, with Pagerank being executing on Live Journal dataset, we can reduce with cn-order: cache-references from 29.94% (without compression) to 39.56% (with compression), cache-misses from 37.87% to 51.90% and hence time from 18.93% to 28.66%.

A multi-level method for data-driven finite element computations

In the data-driven finite element analysis proposed by Kirchdoerfer and Ortiz (2016) the material data are a direct part of the optimization problem. The mechanical problem is rewritten as a minimization problem of a distance function subject to the conservation laws. For every material point of the finite element geometry nearest neighbor searches need to be performed to find the closest data points describing equilibrium. With increasing dimensions and data density the numerical cost increases significantly. The fact that detailed data are required for good results and that three dimensional examples cannot always be simplified leads to a high computation time and memory demand.Here a multi-level method is proposed to reduce the computational cost of such analyses. Starting point is a coarser initial subset of the data, a first solution is then approximated and successively improved by adaptively refined data sets. Repeated simulations lead to an adequate solution using just a fraction of the total data set.While introducing the method on a simple truss structure example it is shown that this multi-level method significantly reduces computational costs and enables complex data-driven finite element computations by means of an engineering example.

In Situ Heating of the Nightside Martian Upper Atmosphere and Ionosphere: The Role of Solar Wind Electron Precipitation

Abstract In the absence of solar radiation, precipitating electrons from the solar wind (SW) are generally thought to be the dominant source of energy deposition in the nightside Martian upper atmosphere, creating a patchy ionosphere and possibly also affecting the nightside thermal budget of various neutral and ionized species. Previous model calculations have not taken into account in situ heating via SW electron impact. In the present study, we utilize extensive measurements made by several instruments on board the Mars Atmosphere and Volatile Evolution spacecraft, in order to perform data-driven computations of the nightside neutral, ion, and electron heating rates. Considering the large range of energetic electron intensity observed on the nightside of Mars, we divide the entire data set into two subsamples, either with or without energetic electron depletion, a notable feature of the nightside Martian ionosphere. Our calculations indicate that in situ nightside neutral heating is dominated by exothermic chemistry and Maxwell interaction with thermal ions for regions with depletion, and by direct SW impact for regions without. Collisional quenching of excited state species produced from a variety of channels, such as electron impact excitation, dissociation, and ionization, as well as dissociative recombination, makes a substantial contribution to neutral heating, except during depletion. For comparison, nightside ion heating is mainly driven by energetic ion production under all circumstances, which occurs mainly via ion-neutral reaction O+ + CO2 and CO2 + predissociation.

CINEMAS: Comprehensively INtegrated Environment for advanced MAterials Simulations

Incessantly increasing capabilities of Density Functional Theory(DFT) to interpret and predict materials properties have established it as an important method in the domain of materials simulations. This naturally attracts several users beyond conventional computational theorists. In this work, we introduce Comprehensively INtregrated Environment for advanced MAterials Simulations (CINEMAS) [ www.ikst.res.in/cinemas]. CINEMAS is a completely graphical user interface(GUI) based simulation platform to facilitate DFT calculations and more, for both non-expert and expert users. At the same time, different types of GUI enabled automation provides a platform to execute a large number of data-driven computations. The main workhorse of CINEMAS is ‘workflows’ which is a graphical representation of a single or a sequence of dependent and independent calculations. These workflows lead to complete preparation and execution of simulations from the beginning to end, providing handles for multiple analysis, and finally, prepare graphics meeting publication standards. External Python-based packages can be used with these workflows in CINEMAS. This workflow representation can trace back the calculations along a sequence, gives an overview of the whole project from a single point, and makes data sharing among collaborators very easy. We discuss the concept and features of CINEMAS, and how several tools required at different stages of a calculation are seamlessly integrated through its rich and easy GUI. We discuss how CINEMAS is self-sufficient to carry out calculations using a DFT code and more without requiring any other software application to prepare, execute, and analyze the simulations.

Data-Driven Photoluminescence Tuning in Eu2+-Doped Phosphors.

Discovery of rare earth phosphors has generally relied on the chemical intuition and time-intensive trial-and-error synthesis; therefore, finding new materials assisted by data-driven computations is urgent. Herein, we utilize a regression model to predict the emission wavelengths of Eu2+-doped phosphors by revealing the relationships between the crystal structure and luminescence property. The emission wavelengths of [Rb(1-x)K(x)]3LuSi2O7:Eu2+ (0 ≤ x ≤ 1) phosphors, as examples for the data-driven photoluminescence tuning, are successfully predicted on the basis of the existing data of only eight systems, also consistent with the experimental results. These phosphors can be excited by blue light and exhibit broad-band red and near-infrared emission ranging from 619 to 737 nm. These findings in Eu2+-doped silicate phosphors indicate that data-driven computations through the regression mode would have bright application in discovering novel phosphors with a target emission wavelengths.

Critical Review—Data-Driven Discovery of Novel Phosphors

Innovative phosphor materials are continuously pursued with advances in solid-state lighting and display technologies. Traditional methods of searching for new phosphors are usually based on a time-consuming and laborious trail-and-error process. Newly emerging approaches, such as high-throughput calculations and machine learning, provide great potentials to significantly accelerate the design and discovery of new materials. Here, we overview the recent development on searching for phosphor materials assisted by data-driven computations, aiming to (i) clarify the composition-structure-property relationships; (ii) construct materials descriptors by computation methods; and (iii) design novel phosphors by combining computation with experiments. Finally, future prospects for data-driven discovery of novel phosphors are highlighted.

Non-parametric material state field extraction from full field measurements

Data-driven computations propose a completely new paradigm to the computational mechanics research community and to experimentalists. Classically, admissible material states can only be obtained experimentally for homogeneous stress/strain configurations or using a parametric optimization of material laws based on heterogeneous tests. Data-driven algorithms aim at circumventing these limitations. However, data-driven algorithms require a large database of admissible material states, otherwise extrapolation is required and some limitations of the classical constitutive equation based approach remain. In this paper, an inverse data-driven approach based on full field measurements is presented. The main idea is to extract, with no assumption on the constitutive equations, rich (i.e. heterogeneous and multiaxial) material state fields from displacement fields and external load measurements. The capability of the proposed method to extract databases of admissible material states and to evaluate stress fields without parametric constitutive equations is illustrated through three examples dedicated to non-linear elasticity, plasticity and dynamics.

Processing of big heterogeneous genomic datasets for tertiary analysis of Next Generation Sequencing data.

We previously proposed a paradigm shift in genomic data management, based on the Genomic Data Model (GDM) for mediating existing data formats and on the GenoMetric Query Language (GMQL) for supporting, at a high level of abstraction, data extraction and the most common data-driven computations required by tertiary data analysis of Next Generation Sequencing datasets. Here, we present a new GMQL-based system with enhanced accessibility, portability, scalability and performance. The new system has a well-designed modular architecture featuring: (i) an intermediate representation supporting many different implementations (including Spark, Flink and SciDB); (ii) a high-level technology-independent repository abstraction, supporting different repository technologies (e.g., local file system, Hadoop File System, database or others); (iii) several system interfaces, including a user-friendly Web-based interface, a Web Service interface, and a programmatic interface for Python language. Biological use case examples, using public ENCODE, Roadmap Epigenomics and TCGA datasets, demonstrate the relevance of our work. The GMQL system is freely available for non-commercial use as open source project at: http://www.bioinformatics.deib.polimi.it/GMQLsystem/. Supplementary data are available at Bioinformatics online.

Modeling and interoperability of heterogeneous genomic big data for integrative processing and querying

While a huge amount of (epi)genomic data of multiple types is becoming available by using Next Generation Sequencing (NGS) technologies, the most important emerging problem is the so-called tertiary analysis, concerned with sense making, e.g., discovering how different (epi)genomic regions and their products interact and cooperate with each other. We propose a paradigm shift in tertiary analysis, based on the use of the Genomic Data Model (GDM), a simple data model which links genomic feature data to their associated experimental, biological and clinical metadata. GDM encompasses all the data formats which have been produced for feature extraction from (epi)genomic datasets. We specifically describe the mapping to GDM of SAM (Sequence Alignment/Map), VCF (Variant Call Format), NARROWPEAK (for called peaks produced by NGS ChIP-seq or DNase-seq methods), and BED (Browser Extensible Data) formats, but GDM supports as well all the formats describing experimental datasets (e.g., including copy number variations, DNA somatic mutations, or gene expressions) and annotations (e.g., regarding transcription start sites, genes, enhancers or CpG islands). We downloaded and integrated samples of all the above-mentioned data types and formats from multiple sources. The GDM is able to homogeneously describe semantically heterogeneous data and makes the ground for providing data interoperability, e.g., achieved through the GenoMetric Query Language (GMQL), a high-level, declarative query language for genomic big data. The combined use of the data model and the query language allows comprehensive processing of multiple heterogeneous data, and supports the development of domain-specific data-driven computations and bio-molecular knowledge discovery.

Dynamic Data Driven Applications System Concept for Information Fusion

We present a framework of Information Fusion (IF) using the Dynamic Data Driven Applications Systems (DDDAS) concept. Existing literature at the intersection of these two topics supports environmental modeling (e.g., terrain understanding) for context enhanced applications. Taking advantage of sensor models, statistical methods, and situation- specific spatio-temporal fusion products derived from wide area sensor networks, DDDAS demonstrates robust multi-scale and multi-resolution geographical terrain computations. We highlight the complementary nature of these seemingly parallel approaches and propose a more integrated analytical framework in the context of a cooperative multimodal sensing application. In particular, we use a Wide-Area Motion Imagery (WAMI) application to draw parallels and contrasts between IF and DDDAS systems that warrants an integrated perspective. This elementary work is aimed at triggering a sequence of deeper insightful research towards exploiting sparsely sampled piecewise dense WAMI measurements – an application where the challenges of big-data with regards to mathematical fusion relationships and high-performance computations remain significant and will persist. Dynamic data-driven adaptive computations are required to effectively handle the challenges with exponentially increasing data volume for advanced information fusion systems solutions such as simultaneous target tracking and identification.

Identification of candidate SNPs for drug induced toxicity from differentially expressed genes in associated tissues

The growing collection of publicly available high-throughput data provides an invaluable resource for generating preliminary in silico data in support of novel hypotheses. In this study we used a cross-dataset meta-analysis strategy to identify novel candidate genes and genetic variations relevant to paclitaxel/carboplatin-induced myelosuppression and neuropathy. We identified genes affected by drug exposure and present in tissues associated with toxicity. From ten top-ranked genes 42 non-synonymous single nucleotide polymorphisms (SNPs) were identified in silico and genotyped in 94 cancer patients treated with carboplatin/paclitaxel. We observed variations in 11 SNPs, of which seven were present in a sufficient frequency for statistical evaluation. Of these seven SNPs, three were present in ABCA1 and ATM, and showed significant or borderline significant association with either myelosuppression or neuropathy. The strikingly high number of associations between genotype and clinically observed toxicity provides support for our data-driven computations strategy to identify biomarkers for drug toxicity.

Models, methods and middleware for grid-enabled multiphysics oil reservoir management

Meeting the demands for energy entails a better understanding and characterization of the fundamental processes of reservoirs and of how human made objects affect these systems. The need to perform extensive reservoir studies for either uncertainty assessment or optimal exploitation plans brings up demands of computing power and data management in a more pervasive way. This work focuses on high performance numerical methods, tools and grid-enabled middleware systems for scalable and data-driven computations for multiphysics simulation and decision-making processes in integrated multiphase flow applications. The proposed suite of tools and systems consists of (1) a scalable reservoir simulator, (2) novel stochastic optimization algorithms, (3) decentralized autonomic grid middleware tools, and (4) middleware systems for large-scale data storage, querying, and retrieval. The aforementioned components offer enormous potential for performing data-driven studies and efficient execution of complex, large-scale reservoir models in a collaborative environment.