SEARCH · Search NASA

Results for “validation data sets”

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.

SourceResearch connectionFrom yearThrough year

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

At least 235 records · Page 13

All sources NASA NTRS DOE OSTI

DOE OSTI2026

Unified Description of Cuprate Superconductors by Fractionalized Electrons Emerging from Integrated Analyses of Photoemission Spectra and Quasiparticle Interference

Electronic structure of high-temperature superconducting cuprates is studied by analyzing experimental data independently obtained from two complementary spectroscopies: one, quasiparticle interference (QPI) measured by scanning-tunneling microscopy, and the other, angle-resolved photoemission spectroscopy (ARPES). We combine these two sets of data in a unified theoretical analysis. Through explicit calculations of experimentally measurable quantities, we show that a simple two-component fermion model (TCFM) representing electron fractionalization succeeds in reproducing various detailed features of these experimental data: ARPES and QPI data are concomitantly reproduced by the TCFM in full energy and momentum spaces. The measured QPI pattern reveals a signature characteristic of the TCFM, distinct from the conventional single-component prediction, supporting the validity of the electron fractionalization in the cuprates. The integrated analysis also solves the puzzles of ARPES and QPI data that are seemingly inconsistent with each other. The overall success of the TCFM offers a comprehensive understanding of the electronic structure of the cuprates, in particular, the unoccupied side of the spectra, of which momentum-resolved structure has long been unexplored experimentally. We further predict that a characteristic QPI pattern should appear in the unoccupied high-energy part if the fractionalization is at work. We propose that integrated-spectroscopy analyses offer a promising way to explore challenging issues of strongly correlated electron systems.

Sakai, Shiro [Sophia University; RIKEN Center for ↗

DOE OSTI2021

DEPRECATED AI-Batt-OS (Autonomous Identification of Battery Life Models - Open Source) [SWR 21-17]

DEPRECATED. This repository was archived by the owner on Jun 30, 2026. It is now read-only. Open source implementation of some of the methods utilized by AI-Batt, a battery lifetime modeling and analysis toolkit provided by the National Laboratory of the Rockies (NLR). This software demonstrates the use of bi-level optimization and symbolic regression techniques to semi-autonomously identify algebraic models predicting the capacity fade of lithium-ion batteries during calendar aging. Modeling the degradation of batteries is a complex task, due to the difficulty in separating the time-dependent and time-independent factors impacting cell level degradation, across multiple data series with different numbers of measurements and/or data quality. Bi-level optimization enables model parameters to be optimized to either the entire data set or to individual data series, allowing statistical disambiguation of global behaviors (data series independent) and local behaviors (data series dependent). Symbolic regression is used to automatically search for optimal low-dimesional models predicting the variation of locally optimized parameters versus time-independent experimental variables from millions of possible models, resulting in a more accurate and repeatable model identification process than is possible by a manual search. The provided tools also implement cross-validation and bootstrap resampling schemes, empowering statistical model comparison/selection and quantification of model uncertainties. An example script replicates the results from the manuscript "Challenging Practices of Algebraic Battery Life Models through Statistical Validation and Model Identification via Machine-Learning", submitted to ECS. All code is written in MATLAB. Requires the Statistics and Machine Learning Toolbox. Contact Dr. Paul Gasper at Paul.Gasper@nlr.gov for any questions.

Gasper, Paul [National Renewable Energy Lab. (NREL↗

DOE OSTI2025

Optimizing Facility Operations by Applying Machine Learning to the Army Reserve Enterprise Building Control System (Final Report)

Thousands of U.S. Department of Defense (DoD) buildings have building automation systems (BASs) and/or advanced meters. Although these systems have a wealth of data, performance optimization requires time and expertise to review and act on that information. Machine learning (ML) can provide automated and actionable insights to controls operators. This demonstration implemented proven ML methods on the Army Reserve Enterprise Building Control System. ML refers to algorithms that “learn” from data and improve their performance on a given task over time. In the buildings domain these tasks range from predicting future energy consumption, to identifying operational issues before faults occur, to optimizing control decisions. To learn, ML requires input data, which – for buildings – typically consists of instrument data such as energy consumption data and subsystem controls information such as set-point temperatures, and context data consisting of information such as the physical location of the building, the area of the building, and the weather. ML models use the relationships learned from the input data to make predictions with new, previously unseen, data. The team was able to investigate and successfully implement the following ML use cases: labeling consumption data as anomalous or non-anomalous; baseline whole-building load prediction (unknown fault status); fault detection (validation not possible); and site prioritization for energy-related projects. Due to the constraints of the project, interventions were not able to be implemented during the demonstration; therefore, assessments of operational cost savings and maintenance avoided could not be performed. The project has been presented at two leading national building conferences and two additional publications to peer-reviewed journals are currently in preparation.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

DOE OSTI2025

Evaluation of Seismic Artificial Intelligence with Uncertainty

Artificial intelligence has transformed the seismic community with deep learning models (DLMs) that are trained to complete specific tasks within workflows. However, there is still a lack of robust evaluation frameworks for evaluating and comparing DLMs. Here, we address this gap by designing an evaluation framework that jointly incorporates two crucial aspects: performance uncertainty and learning efficiency. To target these aspects, we meticulously construct the training, validation, and test splits using a clustering method tailored to seismic data and enact an expansive training design to segregate performance uncertainty arising from stochastic training processes and random data sampling. The framework’s ability to guard against misleading declarations of model superiority is demonstrated through the evaluation of PhaseNet (Zhu and Beroza, 2018), a popular seismic phase picking DLM, under three training approaches. Our framework helps practitioners choose the best model for their problem and set performance expectations by explicitly analyzing model performance with uncertainty at varying budgets of training data.

58 GEOSCIENCES↗

DOE OSTI2025

High-Temperature Gas Sensor Materials with Properties Predicted via First-Principles Calculations with Machine Learning Modeling and Experimental Corroboration

Understanding the temperature dependence of functional properties of sensing materials is vital for their applications in combustion environments. The electron-phonon coupling that derives the electronic structure change with temperatures is a key property of interest as it affects other sensing responses. Herein, we first assess the temperature dependence of band gap renormalization in sensing materials by employing Allen-Heine-Cardona (AHC) theory with density functional theory (DFT) simulations corroborated with experimental observation. As the AHC calculations are impractical for high-throughput screening of materials, we employ data-driven Gaussian process regression to predict the parameters employed in the O’Donnell empirical model from a set of physical features. To mitigate the reliability issues arising from the small size of the dataset, we apply a Bayesian technique to improve the generalizability of the data-driven models as well as to quantify the uncertainty associated with theoretical predictions. These models capture well the overall trend of the O’Donnell parameters with respect to a reduced feature set obtained by transforming the available physical features. Quantifying the associated uncertainty helps us understand the reliability of the predictions and, therefore, the variation of bandgap as a function of temperature for other novel materials. The predicted candidates from machine learning models are further validated by experiments and DFT calculations.

bandgap renormalization↗

DOE OSTI2025

DOE Repository Metadata Profile (DRMP): A Metadata Framework for Advancing Interoperability and AI Readiness Across Scientific Repositories

The Department of Energy (DOE) funds a diverse and distributed ecosystem of repositories that steward scientific data, publications, and software across its research programs, user facilities, and national laboratories. While significant progress has been made in standardizing dataset-level metadata, the metadata describing repositories themselves (their identity, governance, access interfaces, policies, and technical capabilities) remains inconsistent and fragmented across DOE-funded systems. This variability limits discoverability, interoperability, automated validation, and AI-driven analysis, all of which are increasingly essential for modern scientific workflows. To address this gap, the DOE Data Curation Working Group (DCWG) developed the DOE Repository Metadata Profile (DRMP). The DRMP is a practical, community-driven framework that defines how repositories can describe themselves in a consistent, machine-actionable, and scalable manner. The DRMP is not a new metadata schema. Instead, it is a mapping profile and structured element set capturing the essential characteristics of DOE repositories. It harmonizes repository-level metadata across six widely adopted community schemas: RE3Data; DCAT-US v3; Schema.org; Dublin Core; DataCite 4.6; and PREMIS 3.0. This harmonization eliminates reinvention and enables interoperability within DOE and across the broader scientific ecosystem. A core objective of the DRMP is to reduce burden on repositories by allowing them to reuse their existing metadata through a Rosetta-style crosswalk rather than redesigning local implementations. The profile introduces a three-level conformance model that supports incremental adoption: • Level 1 – Minimum Viable Record (MVR): foundational identification elements required for workflows, project registration, and basic repository presence. • Level 2 – Interoperable: structured metadata enabling alignment with national and international discovery systems. • Level 3 – AI-Ready: enhanced provenance, policy transparency, fixity, semantic context, and capabilities that support automated reasoning, model training governance, and machine-assisted curation. To support implementation, the DRMP includes JSON Schema definitions, OpenAPI patterns, and MCP templates that allow repositories to publish machine-readable metadata directly within existing platforms. These resources are modular and lightweight, enabling adoption without major architectural change. Adopting the DRMP enables repositories to: • Enhance discoverability and interoperability by aligning identifiers, classifications, and descriptive elements across widely used schema standards. • Support federated discovery and cross-registration across DOE systems, Data.gov, and international catalogs. • Enable AI agents and workflow orchestration systems to interpret repository-level metadata within the American Science Cloud (AmSC) through Model Context Protocol (MCP)-based context publication. • Demonstrate alignment with DOE’s open science, stewardship, and FAIR data priorities. This guidance represents a community-driven step forward. Through voluntary adoption and continued feedback, the DRMP advances a cohesive, machine-actionable description of DOE repositories that supports FAIR data practices, preparing the infrastructure for AI-enabled research, and strengthening the discoverability and reuse of DOE’s scientific outputs.

96 KNOWLEDGE MANAGEMENT AND PRESERVATION↗

DOE OSTI2026

Dataset of Generative AI Workload Power Profiles

This dataset provides a collection of high-resolution (5/10 Hz or every 0.2/0.1 seconds) power consumption profiles for generative artificial intelligence (GenAI) workloads executed on NLR's High Performance Computing (HPC) platform Kestrel. The dataset also includes examples of representative whole-facility power profiles generated using a bottom-up, event-driven, data center energy model . This dataset is designed to support research in energy modeling, infrastructure planning, energy system integration, and sustainability analysis for AI-driven computing systems. The dataset captures time-resolved electrical power measurements across a diverse set of configurations, including variations in job type (inference vs. training), workload (LLM vs. image generation), datasets, and number of compute nodes. Power traces are provided in a standardized format and include both raw/instantaneous and aggregated files. Each profile is accompanied by metadata describing workload parameters, enabling reproducibility and cross-study comparison. The dataset is intended for use in applications such as data center infrastructure planning, energy modeling, demand response and grid impact studies, and development and validation of system-level simulation tools. By making these workload-specific power profiles publicly available, this dataset aims to address the current lack of open, empirical energy data for generative AI systems and to facilitate transparent, reproducible research on the energy and environmental impacts of large-scale AI deployment. If you use this dataset, please cite the associated publication: Vercellino et al., “Measurement of Generative AI Workload Power Profiles for Whole-Facility Data Center Infrastructure Planning,” arXiv:2604.07345 (2026).

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2026

Object-Based Evaluation of Dynamical and Statistical Downscaled Precipitation Products over CONUS

High-resolution precipitation data, generated through dynamical downscaling (DD) or statistical downscaling (SD) of global climate model output, provide critical information for regional climate assessment and adaptation planning. Most downscaling development and validation have focused on accurate gridscale precipitation construction and ignored the spatial structure of precipitation across model grids and at the event scale. However, many applications, e.g., hydrologic modeling and the analysis using the downscaled precipitation, require a reasonable representation of the spatial structure of precipitation within watersheds. Therefore, a set of standard metrics to evaluate the representation of the spatial structure of individual storms across diverse downscaled precipitation products is desired. To address this need, we conducted an object-based evaluation of precipitation in decades-long DD and SD products over the contiguous United States (CONUS). Specifically, we evaluate their ability to reproduce various features of precipitation objects in the observations: total volume, precipitation area, peak intensity, and spatial structure. Multiple metrics (bias, Perkins score, and nonparametric statistical tests) are used to quantify model performance. Our evaluation reveals notable variations in performance among individual products across different climate zones and seasons, as well as between extreme and nonextreme events. In general, most DD products exhibit balanced performance across the four precipitation object features, while SD products vary more significantly in their performance across products. Based on this comprehensive evaluation, we provide guidance on choosing downscaled products for specific regions, seasons, and precipitation object features. These findings and recommendations can inform precipitation-relevant modeling and analysis over CONUS, guide future downscaling technique developments, and provide actionable information for climate impact assessment and adaptation.

Downscaling↗

DOE OSTI2026

Object-Based Evaluation of Dynamical and Statistical Downscaled Precipitation Products over CONUS

Environmental sciences↗

DOE OSTI2026

Direct Measurement of ICRF-Enhanced Plasma Potentials on WEST Using Reciprocating Emissive Probes

An extensive documentation of ICRF-enhanced plasma potentials has been conducted over two experimental campaigns on the WEST tokamak using reciprocating emissive probes magnetically connected to two ICRF antennas. The collected data spans a wide range of antenna electrical settings (coupled power, toroidal phasing, left–right power balance) and plasma parameters (density at the antenna limiter above and below the lower hybrid resonance, plasma current, minority fraction). By scanning the edge safety factor across multiple probe plunges, the magnetic connection between the probe and the antenna varied, enabling the construction of a 2D map of the plasma and floating potentials around an active ICRF antenna. This dataset will be used to validate RF simulation tools equipped with the sheath boundary condition and used to predict RF rectified potentials and ICRF-induced impurity sputtering in future machines. This paper presents the diagnostic and some initial measurements, while the rest will be reported elsewhere.

Diab, Raymond [Massachusetts Inst. of Technology (↗

DOE OSTI2025

Control Strategies and Validation in the Hybrid Optimization and Performance Platform (HOPP)

The Hybrid Optimization and Performance Platform (HOPP) is a tool that simulates hybrid power plants in various configurations, and also calculates the financial feasibility of these plants. This report outlines an overview of HOPP and the energy storage dispatch strategies available. It then presents three case studies which demonstrate different applications of HOPP. The first case looks at the profitability of hybrid power plants in different locations in the USA. The second case examines the availability of hybrid power plants to provide energy reliability services. The third case presents a plant that produces both hydrogen and electricity, and demonstrates a dispatch strategy that chooses the most profitable energy vector based on price signals. The next section shows the validation of HOPP on operational data, using data from both unit-scale and utility-scale power plants. This validation process demonstrated that HOPP can simulate the power output of both wind and solar PV plants at both scales with comparable fidelity to an existing commercial software tool. Finally, HOPP is applied in a field test which applies an optimal dispatch strategy to a physical battery in a unit-scale hybrid plant at NREL. HOPP's optimal dispatch strategy, applied in a real-world setting, improved this hybrid plant's ability to meet a load signal while minimizing operational costs.

14 SOLAR ENERGY↗

DOE OSTI2024

Baryon acoustic oscillation theory and modelling systematics for the DESI 2024 results

This paper provides a comprehensive overview of how fitting of baryon acoustic oscillations (BAO) is carried out within the upcoming Dark Energy Spectroscopic Instrument’s (DESI) 2024 results using its DR1 data set, and the associated systematic error budget from theory and modelling of the BAO. We derive new results showing how non-linearities in the clustering of galaxies can cause potential biases in measurements of the isotropic (⁠α iso ⁠) and anisotropic (⁠α ap ) BAO distance scales, and how these can be effectively removed with an appropriate choice of reconstruction algorithm. We then demonstrate how theory leads to a clear choice for how to model the BAO and develop, implement, and validate a new model for the remaining smooth-broad-band (i.e. without BAO) component of the galaxy clustering. Finally, we explore the impact of all remaining modelling choices on the BAO constraints from DESI using a suite of high-precision simulations, arriving at a set of best practices for DESI BAO fits, and an associated theory and modelling systematic error. Overall, our results demonstrate the remarkable robustness of the BAO to all our modelling choices and motivate a combined theory and modelling systematic error contribution to the post-reconstruction DESI BAO measurements of no more than 0.1 per cent (0.2 per cent) for its isotropic (anisotropic) distance measurements. We expect the theory and best practices laid out to here to be applicable to other BAO experiments in the era of DESI and beyond.

79 ASTRONOMY AND ASTROPHYSICS↗

DOE OSTI2025

AlloSHP: deconvoluting single homeologous polymorphism for phylogenetic analysis of allopolyploids

Background The genomic and evolutionary study of allopolyploid organisms involves multiple copies of homeologous chromosomes, making their assembly, annotation, and phylogenetic analysis challenging. Bioinformatics tools and protocols have been developed to study polyploid genomes, but sometimes require the assembly of their genomes, or at least the genes, limiting their use. Results We have developed AlloSHP, a command-line tool for detecting and extracting single homeologous polymorphisms (SHPs) from the subgenomes of allopolyploid species. This tool integrates three main algorithms, WGA, VCF2ALIGNMENT and VCF2SYNTENY, and allows the detection of SHPs for the study of diploid-polyploid complexes with available diploid progenitor genomes, without assembling and annotating the genomes of the allopolyploids under study. AlloSHP has been validated on three diploid-polyploid plant complexes, Brachypodium, Brassica, and Triticum-Aegilops, and a set of synthetic hybrid yeasts and their progenitors of the genus Saccharomyces. The results and congruent phylogenies obtained from the four datasets demonstrate the potential of AlloSHP for the evolutionary analysis of allopolyploids with a wide range of ploidy and genome sizes. Conclusions AlloSHP combines the strategies of simultaneous mapping against multiple reference genomes and syntenic alignment of these genomes to call SHPs, using as input data a single VCF file and the reference genomes of the known or closest extant diploid progenitor species. This novel approach provides a valuable tool for the evolutionary study of allopolyploid species, both at the interspecific and intraspecific levels, allowing the simultaneous analysis of a large number of accessions and avoiding the complex process of assembling polyploid genomes.

Allopolyploids↗

DOE OSTI2025

Multi-physics Topology OPtimization and Additive Manufacturing for High-temperature Heat Exchangers

This research significantly advances the understanding of high-temperature heat exchanger design through an integrated approach that combines topology optimization (TO), triply periodic minimal surface (TPMS) structures, additive manufacturing (AM) and thermohydraulic testing. Each of these components contributes uniquely to a unified, high-performance design, fabrication and testing workflow. Topology optimization serves as the foundation of the design methodology by providing a systematic way to determine the most effective material layout for separating hot and cold fluids while maximizing thermal performance. The researchers introduced a novel three-material optimization framework using two density fields to represent hot fluid, cold fluid, and solid domains. This approach enables automated discovery of optimal shapes and flow paths that cannot be intuitively designed, especially under constraints imposed by manufacturing technologies. Furthermore, constraints such as minimal wall thickness and overhang angles were embedded into the optimization process, ensuring that resulting designs are not only thermally efficient but also manufacturable using modern additive techniques. In parallel, the study delves into the use of Gyroid-based TPMS geometries for constructing the core of the heat exchanger. TPMS structures are known for their high surface area, excellent fluid mixing capabilities, and minimal pressure drop characteristics. The researchers applied a data-driven modeling framework using Heteroscedastic Sparse Gaussian Process Regression (HSGPR) combined with genetic algorithms. This allowed for the rapid evaluation and optimization of key geometric parameters such as frequency, iso-value, and phase shift. The result was a set of Gyroid structures tailored for high heat transfer and low flow resistance, demonstrating clear improvements over conventional straight-channel designs. After the designing process, additive manufacturing played a critical role by turning these highly complex, optimized geometries into physical components. Utilizing Laser Powder Bed Fusion (LPBF) with Haynes 282, the study demonstrated the feasibility of fabricating these heat exchangers at high precision. Post-processing methods, including dilation-erosion operations, were applied to ensure local features adhered to self-supporting constraints. The fabricated structures were then subjected to thermohydraulic testing under conditions representative of supercritical CO 2 Brayton cycles, validating the predicted performance and confirming the viability of the full design-to-fabrication pipeline. Finally, thermohydraulic testing across the above studies served as a crucial experimental validation of advanced heat exchanger. Under consistent high-temperature and high-pressure conditions using supercritical CO 2 , the testing demonstrated that both TO and Gyroid-based TPMS designs significantly outperformed conventional straight-channel HXs. The TO design achieved a 115% increase in UA and NTU and a 27.6% boost in gravimetric power density, while the data-driven optimized Gyroid design delivered a 166% increase in UA and NTU and improved effectiveness from 68.7% to 86.1%. These results validate the simulation models, confirm the manufacturability of complex geometries under AM constraints, and provide key insights into design-performance trade-offs, thereby advancing the development of high-efficiency, compact heat exchangers for extreme environments.

36 MATERIALS SCIENCE↗

DOE OSTI2025

Grey-Box System Identification of Grid-Forming Inverters

This paper demonstrates the use of grey-box system identification methods for simplifying and understanding the nonlinear power dynamics of grid-forming inverters (GFMs). The power and frequency outputs of complex high-order GFM models are fed into system identification software in order to fit them to a predetermined LTI system and learn system parameters such as (synthetic) inertia and droop constants. The same process is then run for a high-order synchronous generator model, and the outputs are fit to the same set of LTI equations. Simulation of a network of GFM inverters with diverse control architecture is also performed for the same process. The intent is threefold: first, to demonstrate the appropriateness of unified LTI models for describing the power and frequency dynamics of individual resources and connected networks, in order to facilitate analysis of larger heterogeneous networked systems; second, to discover the relationship between internal control parameters of GFMs and their externally observed values; and third, to validate that grey-box data-driven system identification techniques can be a valuable tool to discover the values of important parameters in the absence of explicit vendor models.

analytical models↗

DOE OSTI2024

JARVIS-Leaderboard: a large scale benchmark of materials design methods

Abstract Lack of rigorous reproducibility and validation are significant hurdles for scientific development across many fields. Materials science, in particular, encompasses a variety of experimental and theoretical approaches that require careful benchmarking. Leaderboard efforts have been developed previously to mitigate these issues. However, a comprehensive comparison and benchmarking on an integrated platform with multiple data modalities with perfect and defect materials data is still lacking. This work introduces JARVIS-Leaderboard, an open-source and community-driven platform that facilitates benchmarking and enhances reproducibility. The platform allows users to set up benchmarks with custom tasks and enables contributions in the form of dataset, code, and meta-data submissions. We cover the following materials design categories: Artificial Intelligence (AI), Electronic Structure (ES), Force-fields (FF), Quantum Computation (QC), and Experiments (EXP). For AI, we cover several types of input data, including atomic structures, atomistic images, spectra, and text. For ES, we consider multiple ES approaches, software packages, pseudopotentials, materials, and properties, comparing results to experiment. For FF, we compare multiple approaches for material property predictions. For QC, we benchmark Hamiltonian simulations using various quantum algorithms and circuits. Finally, for experiments, we use the inter-laboratory approach to establish benchmarks. There are 1281 contributions to 274 benchmarks using 152 methods with more than 8 million data points, and the leaderboard is continuously expanding. The JARVIS-Leaderboard is available at the website: https://pages.nist.gov/jarvis_leaderboard/

36 MATERIALS SCIENCE↗

DOE OSTI2025

Emulating the Lyman-Alpha forest 1D power spectrum from cosmological simulations: new models and constraints from the eBOSS measurement

We present the Lyssa suite of high-resolution cosmological simulations of the Lyman-α forest designed for cosmological analyses. These 18 simulations have been run using the Nyx code with 40963 hydrodynamical cells in a 120 Mpc (∼ 81 Mpc/h) comoving box and individually provide sub-percent level convergence of the Lyman-α forest 1d flux power spectrum. We build a Gaussian process emulator for the Lyssa simulations in the lym1d likelihood framework to interpolate the power spectrum at arbitrary parameter values. We validate this emulator based on leave-one-out tests and based on the parameter constraints for simulations outside of the training set. We also perform comparisons with a previous emulator, showing a percent level accuracy and a good recovery of the expected cosmological parameters. Using this emulator we derive constraints on the linear matter power spectrum amplitude and slope parameters A Lyα and n Lyα . While the best-fit Planck ΛCDM model has A Lyα = 8.79 and n Lyα = -2.363, from DR14 eBOSS data we find that A Lyα < 7.6 (95% CI) and n Lyα = -2.369 ± 0.008. The low value of A Lyα , in tension with Planck, is driven by the correlation of this parameter with the mean transmission of the Lyman-α forest. This tension disappears when imposing a well-motivated external prior on this mean transmission, in which case we find A Lyα = 9.8 ± 1.1 in accordance with Planck.

Walther, Michael↗

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↗