Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “matrix factorization”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 91 records · Page 5
All sourcesNASA NTRSDOE OSTI
DOE OSTI2026

NMF-Based Anomaly Detection in CMS 2D Tracking Occupancy Histograms

The CMS experiment relies on Data Quality Monitoring (DQM) to ensure that recorded collision data are suitable for physics analysis. During LHC Run 3, each run contains many lumisections and tracking monitoring elements, making offline inspection challenging, especially for localized detector effects that may appear only for short periods of time. This poster presents an unsupervised machine-learning approach to identify anomalous lumisections in CMS tracking occupancy histograms using Non-Negative Matrix Factorization (NMF). The workflow uses offline CMS DQMIO tracking histograms retrieved with the CMS DIALS API and organized as two-dimensional occupancy maps for each lumisection. After selecting stable lumisections, the occupancy maps are normalized and arranged into a non-negative data matrix. The NMF model learns a compact set of basis patterns describing normal tracking occupancy. Each lumisection is then reconstructed from these learned components, and the reconstruction error is used as an anomaly score. Large residuals indicate occupancy patterns that deviate from normal detector behavior and are flagged for further inspection. This NMF-based approach provides a fast and interpretable way to flag lumisections whose tracking occupancy patterns differ from normal detector behavior. Preliminary studies show sensitivity to known tracking anomalies, and ongoing work is focused on validating the method across additional Run 3 Pixel and Strip detector issues.

Rodríguez Ramos, Iliomar [Puerto Rico U., Mayaguez↗
DOE OSTI2022

General-Purpose Unsupervised Cyber Anomaly Detection via Non-Negative Tensor Factorization

Distinguishing malicious anomalous activities from unusual but benign activities is a fundamental challenge for cyber defenders. Prior studies have shown that statistical user behavior analysis yields accurate detections by learning behavior profiles from observed user activity. These unsupervised models are able to generalize to unseen types of attacks by detecting deviations from normal behavior, without knowledge of specific attack signatures. However, approaches proposed to date based on probabilistic matrix factorization are limited by the information conveyed in a two-dimensional space. Non-negative tensor factorization, on the other hand, is a powerful unsupervised machine learning method that naturally models multi-dimensional data, capturing complex and multi-faceted details of behavior profiles. Herein, our new unsupervised statistical anomaly detection methodology matches or surpasses state-of-the-art supervised learning baselines across several challenging and diverse cyber application areas, including detection of compromised user credentials, botnets, spam e-mails, and fraudulent credit card transactions.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS2016

Emissions Relationships Among Western Forest Fire Plumes: II. Plume Typing for Sources of Ozone and Aerosol Absorption

Previous studies of emission factors from biomass burning are prone to largeerrors since they ignore the interplay of mixing and varying pre-fire backgroundCO2 levels. Such complications severely affected our studies of 446 forest fireplume samples measured in the Western US by the science teams of NASAs SEAC4RS and ARCTAS airborne missions. Consequently we propose a MixedEffects Regression Emission Technique (MERET) to check techniques like theNormalized Emission Ratio Method (NERM), where use of sequentialobservations cannot disentangle emissions and mixing. We also evaluate asimpler consensus technique. All techniques relate emissions to fuel burnedusing Cburn Ctot added to the fire plume, where Ctot (CO2 + CO). Mixed-effectsregression can estimate pre-fire background values of Ctot (indexed byobservation j) simultaneously with emissions factors indexed by individualspecies i, xi (Cburn )i,j., MERET and consensus require more than twoemissions indicators. Our studies excluded samples where exogenous CO orCH4 might have been fed into a fire plume, mimicking emission.We sought to let the data on 13 gases and particulate properties suggest clustersof variables and plume types, using non-negative matrix factorization (NMF).While samples were mixtures, the NMF unmixing suggested purer burn types.Particulate properties (bscat, babs, SSA, AE) and gas-phase emissions were interrelated.Finally, we sought a simple categorization useful for modeling ozone productionin plumes. Two kinds of fires produced high ozone: those with large fuel nitrogenas evidenced by remnant CH3CN in the plumes, and also those from veryintense large burns. Fire types with optimal ratios of delta-NOydelta-HCHO associate with the highest additional ozone per unit Cburn, Perhaps theseplumes exhibit limited NOx binding to reactive organics. Perhaps these plumesexhibit limited NOx binding to reactive organics.

ozone↗
NASA NTRS2017

Emissions Relationships Among Western Forest Fire Plumes: I. Emission Factors Free from Mixing Errors

Previous studies of emission factors from biomass burning are prone to largeerrors since they ignore the interplay of mixing and varying pre-fire backgroundCO2 levels. Such complications severely affected our studies of 446 forest fireplume samples measured in the Western US by the science teams of NASAsSEAC4RS and ARCTAS airborne missions. Consequently we propose a MixedEffects Regression Emission Technique (MERET) to check techniques like theNormalized Emission Ratio Method (NERM), where use of sequentialobservations cannot disentangle emissions and mixing. We also evaluate asimpler consensus technique. All techniques relate emissions to fuel burnedusing C burn = Δ Ctot added to the fire plume, where ≈ Ctot (CO2 + CO). Mixed-effectsregression can estimate pre-fire background values of Ctot (indexed byobservation j) simultaneously with emissions factors indexed by individual species i, δέλτα-xi (Cburn )i,j., MERET and consensus require more than twoemissions indicators. Our studies excluded samples where exogenous CO orCH4 might have been fed into a fire plume, mimicking emission.We sought to let the data on 13 gases and particulate properties suggest clustersof variables and plume types, using non-negative matrix factorization (NMF).While samples were mixtures, the NMF unmixing suggested purer burn types.Particulate properties (bscat, babs, SSA, AÅE) and gas-phase emissions were interrelated.Finally, we sought a simple categorization useful for modeling ozone productionin plumes. Two kinds of fires produced high ozone: those with large fuel nitrogenas evidenced by remnant CH3CN in the plumes, and also those from veryintense large burns. Fire types with optimal ratios of delta-NOydelta-HCHO associate with the highest additional ozone per unit Cburn, Perhaps theseplumes exhibit limited NOx binding to reactive organics. Perhaps these plumesexhibit limited NOx binding to reactive organics.

plumes↗
DOE OSTI2021

Exploring Transport Behavior in Hybrid Perovskites Solar Cells via Machine Learning Analysis of Environmental-Dependent Impedance Spectroscopy

Hybrid organic–inorganic perovskites are one of the promising candidates for the next-generation semiconductors due to their superlative optoelectronic properties. However, one of the limiting factors for potential applications is their chemical and structural instability in different environments. Herein, the stability of (FAPbI3)0.85(MAPbBr3)0.15 perovskite solar cell is explored in different atmospheres using impedance spectroscopy. An equivalent circuit model and distribution of relaxation times (DRTs) are used to effectively analyze impedance spectra. DRT is further analyzed via machine learning workflow based on the non-negative matrix factorization of reconstructed relaxation time spectra. This exploration provides the interplay of charge transport dynamics and recombination processes under environment stimuli and illumination. The results reveal that in the dark, oxygen atmosphere induces an increased hole concentration with less ionic character while ionic motion is dominant under ambient air. Under 1 Sun illumination, the environment-dependent impedance responses show a more striking effect compared with dark conditions. In this case, the increased transport resistance observed under oxygen atmosphere in equivalent circuit analysis arises due to interruption of photogenerated hole carriers. The results not only shed light on elucidating transport mechanisms of perovskite solar cells in different environments but also offer an effective interpretation of impedance responses.

36 MATERIALS SCIENCE↗
DOE OSTI2021

Machine learning to identify geologic factors associated with production in geothermal fields: A casestudy using 3D geologic data, Brady geothermal field, Nevada

In this paper, we present an analysis using unsupervised machine learning (ML) to identify the key geologic factors that contribute to the geothermal production in the Brady geothermal field. Brady is a hydrothermal system in northwestern Nevada that supports both electricity production and direct use of hydrothermal fluids. Transmissive fluid flow pathways are relatively rare in the subsurface but are critical components of hydrothermal systems like Brady and many other types of fluid flow systems in fractured rock. The ML method, non-negative matrix factorization with k-means clustering (NMFk), is applied to a library of fourteen 3D geologic characteristics hypothesized to control hydrothermal circulation in the Brady geothermal field. Our results indicate the macro-scale faults and a local step-over in the fault system preferentially occur along with production wells when compared to injection wells and non-productive wells. We infer that these are the key geologic characteristics that control the through-going hydrothermal transmission pathways at Brady. Our results demonstrate 1) the specific geologic controls on the Brady hydrothermal system and 2) the efficacy of pairing ML techniques with 3D geologic characterization to enhance the understanding of subsurface processes.

15 GEOTHERMAL ENERGY↗
NASA NTRS2015

Techniques for Estimating Emissions Factors from Forest Burning: ARCTAS and SEAC4RS Airborne Measurements Indicate Which Fires Produce Ozone

Previous studies of emission factors from biomass burning are prone to large errors since they ignore the interplay of mixing and varying pre-fire background CO2 levels. Such complications severely affected our studies of 446 forest fire plume samples measured in the Western US by the science teams of NASA's SEAC4RS and ARCTAS airborne missions. Consequently we propose a Mixed Effects Regression Emission Technique (MERET) to check techniques like the Normalized Emission Ratio Method (NERM), where use of sequential observations cannot disentangle emissions and mixing. We also evaluate a simpler "consensus" technique. All techniques relate emissions to fuel burned using C(sub burn) = delta C(sub tot) added to the fire plume, where C(sub tot) approximately equals (CO2 + CO). Mixed-effects regression can estimate pre-fire background values of Ctot (indexed by observation j) simultaneously with emissions factors indexed by individual species i, delta epsilon lambda tau alpha−x(sub i)/(C(sub burn))i,j., MERET and "consensus" require more than two emissions indicators. Our studies excluded samples where exogenous CO or CH4 might have been fed into a fire plume, mimicking emission. We sought to let the data on 13 gases and particulate properties suggest clusters of variables and plume types, using non-negative matrix factorization (NMF). While samples were mixtures, the NMF unmixing suggested purer burn types. Particulate properties (bscat, babs, SSA, AAE) and gas-phase emissions were interrelated. Finally, we sought a simple categorization useful for modeling ozone production in plumes. Two kinds of fires produced high ozone: those with large fuel nitrogen as evidenced by remnant CH3CN in the plumes, and also those from very intense large burns. Fire types with optimal ratios of delta-NOy/delta- HCHO associate with the highest additional ozone per unit Cburn, Perhaps these plumes exhibit limited NOx binding to reactive organics. Perhaps these plumes exhibit limited NOx binding to reactive organics.

fires↗
NASA NTRS2016

Techniques for Estimating Emissions Factors from Forest Burning: ARCTAS and SEAC4RS Airborne Measurements Indicate which Fires Produce Ozone

Previous studies of emission factors from biomass burning are prone to large errors since they ignore the interplay of mixing and varying pre-fire background CO2 levels. Such complications severely affected our studies of 446 forest fire plume samples measured in the Western US by the science teams of NASA's SEAC4RS and ARCTAS airborne missions. Consequently we propose a Mixed Effects Regression Emission Technique (MERET) to check techniques like the Normalized Emission Ratio Method (NERM), where use of sequential observations cannot disentangle emissions and mixing. We also evaluate a simpler "consensus" technique. All techniques relate emissions to fuel burned using C(burn) = delta C(tot) added to the fire plume, where C(tot) approximately equals (CO2 = CO). Mixed-effects regression can estimate pre-fire background values of C(tot) (indexed by observation j) simultaneously with emissions factors indexed by individual species i, delta, epsilon lambda tau alpha-x(sub I)/C(sub burn))I,j. MERET and "consensus" require more than emissions indicators. Our studies excluded samples where exogenous CO or CH4 might have been fed into a fire plume, mimicking emission. We sought to let the data on 13 gases and particulate properties suggest clusters of variables and plume types, using non-negative matrix factorization (NMF). While samples were mixtures, the NMF unmixing suggested purer burn types. Particulate properties (b scant, b abs, SSA, AAE) and gas-phase emissions were interrelated. Finally, we sought a simple categorization useful for modeling ozone production in plumes. Two kinds of fires produced high ozone: those with large fuel nitrogen as evidenced by remnant CH3CN in the plumes, and also those from very intense large burns. Fire types with optimal ratios of delta-NOy/delta- HCHO associate with the highest additional ozone per unit Cburn, Perhaps these plumes exhibit limited NOx binding to reactive organics. Perhaps these plumes exhibit limited NOx binding to reactive organics

ozone↗
DOE OSTI2021

Machine learning to identify geologic factors associated with production in geothermal fields: a case-study using 3D geologic data, Brady geothermal field, Nevada

Abstract In this paper, we present an analysis using unsupervised machine learning (ML) to identify the key geologic factors that contribute to the geothermal production in Brady geothermal field. Brady is a hydrothermal system in northwestern Nevada that supports both electricity production and direct use of hydrothermal fluids. Transmissive fluid-flow pathways are relatively rare in the subsurface, but are critical components of hydrothermal systems like Brady and many other types of fluid-flow systems in fractured rock. Here, we analyze geologic data with ML methods to unravel the local geologic controls on these pathways. The ML method, non-negative matrix factorization with k -means clustering (NMF k ), is applied to a library of 14 3D geologic characteristics hypothesized to control hydrothermal circulation in the Brady geothermal field. Our results indicate that macro-scale faults and a local step-over in the fault system preferentially occur along production wells when compared to injection wells and non-productive wells. We infer that these are the key geologic characteristics that control the through-going hydrothermal transmission pathways at Brady. Our results demonstrate: (1) the specific geologic controls on the Brady hydrothermal system and (2) the efficacy of pairing ML techniques with 3D geologic characterization to enhance the understanding of subsurface processes.

58 GEOSCIENCES↗
DOE OSTI2024

Investigating New Particle Formation and Growth Over an Urban Location in the Eastern Mediterranean

Abstract This study investigates the new particle formation (NPF) events at an urban location in the Eastern Mediterranean. Particle size distribution, particulate chemical composition, and gaseous pollutants were monitored in Rehovot, Israel (31°53″N 34°48″E) during two campaigns: from April 29 to 3 May 2021 (Campaign 1) and from May 3 to 11 May 2023 (Campaign 2), coinciding with an intensive bonfire burning festival. The organic aerosols (OA) source apportionment identified two major factors—Hydrocarbon‐like OA and Biomass‐burning OA—as well as two secondary factors—MO‐OOA (more oxidized‐oxygenated OA) and LO‐OOA (low oxidized oxygenated OA). NPF events were frequently observed during the day (mostly well‐defined nucleation events) and at night (burst of ultrafine mode particles without any discernible growth). A condensation sink value of (9.4 ± 4.0) × 10 −3 s −1 during Campaign 1 and (14.2 ± 6.0) × 10 −3 s −1 during Campaign 2 was obtained. The daytime events were associated with enhanced sulfuric acid proxy concentrations of (2–12) × 10 6 molecules cm −3 , suggesting the role of gas‐phase photochemistry in promoting NPF. A novel approach of hybrid positive matrix factorization analysis was used to deconvolve the chemical species responsible for the observed events. The results suggest the involvement of multiple components, including ammonium sulfate and MO‐OOA, in the nucleation; Nitrate, HOA and LO‐OOA participate in the subsequent particle growth for the daytime events. Nighttime events involve only semi‐volatile species (LO‐OOA, HOA and nitrate) along with ammonium sulfate.

Meteorology & Atmospheric Sciences↗
DOE OSTI2022

Submicrometer spectromicroscopy of UO 2 aged under high humidity conditions

We report the oxidation of uranium dioxide is a complicated process, depending on factors including humidity, temperature, and microstructure. To further determine the characteristics of this process, UO 2 particles were allowed to age and agglomerate under 98% relative humidity at room temperature for 378 days. A focused ion beam (FIB) section of this agglomeration was then measured at the O K-edge, U N 5 -edge, and C K-edge using the scanning transmission x-ray microscope (STXM) at the Advanced Light Source. O K-edge and U N 5 -edge x-ray absorption measurements allowed for the elemental and chemical species mapping of the agglomerates and indicated the formation of schoepite at the submicrometer scale in specific locations. Non-negative matrix factorization was employed to elucidate the main components at the O K-edge, which were uranyl (schoepite) formed primarily at the interface of the sample with controlled atmosphere, a UO 2 -like bulk component present in the majority of the sample, and an oxygen species present at the surface of the FIB section, which is likely adsorbed water. STXM spectromicroscopy measurements at the U N 5 -edge measurements also confirmed the location of oxidized uranium. This analysis is a valuable insight into the formation of schoepite on UO 2 and shows the sensitivity to and utility of STXM spectromicroscopy for uranium speciation.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2023

Compression of tokamak boundary plasma simulation data using a maximum volume algorithm for matrix skeleton decomposition

This report demonstrates satisfactory data compression of SOLPS-ITER simulation output ranging from 2D fields, 1D profiles, and 0D scalar variables with a novel matrix decomposition approach. The singular value decomposition (SVD) scales poorly for large matrix sizes and is unsuited to the application on high dimensional data common to fusion plasma physics simulation. In this work, we employ the columns-submatrix-rows (CUR) matrix factorization technique in order to compute a low-rank approximation up to two orders of magnitude faster than the SVD, but within a nominal L2-norm relative error of ε = 10 –2 . In addition, the CUR approach maintains the original format of the data, in its extracted columns and rows, allowing for interpretable data storage at the original resolution of the simulation. We utilize an iterative algorithm to compute the CUR decomposition of simulation output by maximizing the volume, or linearly independent information content, of a low-rank submatrix contained within the data. Experiments over $\textit{n} × \textit{n}$ randomized test matrices with embedded rank-deficient features show that this maximum volume implementation of CUR matrix approximation has reduced asymptotic computational complexity on the order of n compared to the SVD, which scales approximately as $n^3$. These results show that the CUR technique can be used to effectively select time step snapshots (columns) of over 140 SOLPS-ITER output variables and the associated discretized coordinate timeseries (rows) allowing for reconstruction of the complete simulation dynamics.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
NASA NTRS2011

Graph Theory Roots of Spatial Operators for Kinematics and Dynamics

Spatial operators have been used to analyze the dynamics of robotic multibody systems and to develop novel computational dynamics algorithms. Mass matrix factorization, inversion, diagonalization, and linearization are among several new insights obtained using such operators. While initially developed for serial rigid body manipulators, the spatial operators and the related mathematical analysis have been shown to extend very broadly including to tree and closed topology systems, to systems with flexible joints, links, etc. This work uses concepts from graph theory to explore the mathematical foundations of spatial operators. The goal is to study and characterize the properties of the spatial operators at an abstract level so that they can be applied to a broader range of dynamics problems. The rich mathematical properties of the kinematics and dynamics of robotic multibody systems has been an area of strong research interest for several decades. These properties are important to understand the inherent physical behavior of systems, for stability and control analysis, for the development of computational algorithms, and for model development of faithful models. Recurring patterns in spatial operators leads one to ask the more abstract question about the properties and characteristics of spatial operators that make them so broadly applicable. The idea is to step back from the specific application systems, and understand more deeply the generic requirements and properties of spatial operators, so that the insights and techniques are readily available across different kinematics and dynamics problems. In this work, techniques from graph theory were used to explore the abstract basis for the spatial operators. The close relationship between the mathematical properties of adjacency matrices for graphs and those of spatial operators and their kernels were established. The connections hold across very basic requirements on the system topology, the nature of the component bodies, the indexing schemes, etc. The relationship of the underlying structure is intimately connected with efficient, recursive computational algorithms. The results provide the foundational groundwork for a much broader look at the key problems in kinematics and dynamics. The properties of general graphs and trees of nodes and edge were examined, as well as the properties of adjacency matrices that are used to describe graph connectivity. The nilpotency property of such matrices for directed trees was reviewed, and the adjacency matrices were generalized to the notion of block weighted adjacency matrices that support block matrix elements. This leads us to the development of the notion of Spatial Kernel Operator SKO kernels. These kernels provide the basis for the development of SKO resolvent operators.

Jain, Abhinandan↗
DOE OSTI2020

Exploration of Electrochemical Reactions at Organic–Inorganic Halide Perovskite Interfaces via Machine Learning in In Situ Time-of-Flight Secondary Ion Mass Spectrometry

The instability of hybrid organic–inorganic perovskite (HOIP) devices is one of the significant challenges preventing commercialization. Exploring these phenomena is severely limited by the complexity of the intrinsic electrochemistry of HOIPs, the presence of multiple volatile and mobile ionic species, and the possible role of environmentally induced reactions at surfaces and triple-phase junctions. Here, in situ studies of the electrochemistry of methylammonium lead bromide perovskite with the Au electrode interface are reported via light- and voltage-dependent time-of-flight secondary ion mass spectrometry (ToF-SIMS) imaging of lateral perovskite heterostructures. While ToF-SIMS allows for the visualization of the chemical composition along the surface and its evolution with light and electrical bias, the interpretation of the multidimensional data obtained is often limited due to strong correlations between chemical signatures and the need to track multiple peaks at once. Here, a machine learning workflow combining the Hough transform and non-negative matrix factorization and no-negative tensor decomposition is developed to avoid this limitation and extract salient features of associated chemical changes and to separate the light- and voltage-dependent dynamics. Finally, combining these in situ characterizations and the machine learning workflow provides comprehensive information on the chemical nature of moving species, ion accumulation, and interfacial electrochemical reactions in HOIP devices.

36 MATERIALS SCIENCE↗
DOE OSTI2022

Latent Mechanisms of Polarization Switching from In Situ Electron Microscopy Observations

In situ scanning transmission electron microscopy enables observation of the domain dynamics in ferroelectric materials as a function of externally applied bias and temperature. The resultant data sets contain a wealth of information on polarization switching and phase transition mechanisms. However, identification of these mechanisms from observational data sets has remained a problem due to a large variety of possible configurations, many of which are degenerate. Here, an approach based on a combination of deep learning-based semantic segmentation, rotationally invariant variational autoencoder (VAE), and non-negative matrix factorization to enable learning of a latent space representation of the data with multiple real-space rotationally equivalent variants mapped to the same latent space descriptors is introduced. By varying the size of training sub-images in the VAE, the degree of complexity in the structural descriptors is tuned from simple domain wall detection to the identification of switching pathways. Importantly, this yields a powerful tool for the exploration of the dynamic data in mesoscopic electron, scanning probe, optical, and chemical imaging. Moreover, this work adds to the growing body of knowledge of incorporating physical constraints into the machine and deep-learning methods to improve learned descriptors of physical phenomena.

36 MATERIALS SCIENCE↗
DOE OSTI2022

Developing Reaction Chemistry Models from Reactive Molecular Dynamics: TATB

Reactive Molecular Dynamic (RMD) are used to simulate the cook-off chemistry of TATB at a variety of fixed density and fixed temperature conditions. The chemical transformations are monitored using a Coordination Geometry Analysis (CGA) approach which tracks which atom types are bonded to each specific atom. This particularly identifies oxidation state changes that occur during the transformations. Correlations between these different chemical changes are identified using a Non-negative Matrix Factorization (NMF) approach. These identify reduced order chemistry models for the TATB system which contains six components whose concentration profiles are a function of both the temperature and density/pressure. The time histories of these transformations appear to show exponential growth/decay properties that could be fit with Arrhenius rates. These components should form the basis of deflagration rate models for these materials which could then be used in mesoscale simulations to analyze accidental initiation, shock-to-detonation and detonation propagation.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2021

Compact representations of structured BFGS matrices

For general large-scale optimization problems compact representations exist in which recursive quasi-Newton update formulas are represented as compact matrix factorizations. For problems in which the objective function contains additional structure, recent structured quasi-Newton methods exploit available second-derivative information and approximate unavailable second derivatives. Here, this article develops the compact representations of two structured Broyden-Fletcher-Goldfarb-Shanno update formulas. The compact representations enable efficient limited memory and initialization strategies. Two limited memory line search algorithms are described for which extensive numerical results demonstrate the efficacy of the algorithms, including comparisons to IPOPT on large machine learning problems, and to L-BFGS on a real world large scale ptychographic imaging application.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

DOME: Directional medical embedding vectors from Electronic Health Records

Motivation: The increasing availability of Electronic Health Record (EHR) systems has created enormous potential for translational research. Recent developments in representation learning techniques have led to effective large-scale representations of EHR concepts along with knowledge graphs that empower downstream EHR studies. However, most existing methods require training with patient-level data, limiting their abilities to expand the training with multi-institutional EHR data. On the other hand, scalable approaches that only require summary-level data do not incorporate temporal dependencies between concepts. Methods: We introduce a DirectiOnal Medical Embedding (DOME) algorithm to encode temporally directional relationships between medical concepts, using summary-level EHR data. Specifically, DOME first aggregates patient-level EHR data into an asymmetric co-occurrence matrix. Then it computes two Positive Pointwise Mutual Information (PPMI) matrices to correspondingly encode the pairwise prior and posterior dependencies between medical concepts. Following that, a joint matrix factorization is performed on the two PPMI matrices, which results in three vectors for each concept: a semantic embedding and two directional context embeddings. They collectively provide a comprehensive depiction of the temporal relationship between EHR concepts. Results: We highlight the advantages and translational potential of DOME through three sets of validation studies. First, DOME consistently improves existing direction-agnostic embedding vectors for disease risk prediction in several diseases, for example achieving a relative gain of 5.5% in the area under the receiver operating characteristic (AUROC) for lung cancer. Second, DOME excels in directional drug-disease relationship inference by successfully differentiating between drug side effects and indications, correspondingly achieving relative AUROC gain over the state-of-the-art methods by 10.8% and 6.6%. Finally, DOME effectively constructs directional knowledge graphs, which distinguish disease risk factors from comorbidities, thereby revealing disease progression trajectories. The source codes are provided at https://github.com/celehs/Directional-EHRembedding.

60 APPLIED LIFE SCIENCES↗