Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “large scale network”

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 1,117 records · Page 62
All sourcesNASA NTRSDOE OSTI
DOE OSTI2025

High-frequency Data Integration for Landscape Model Calibration of Carbon Fluxes Across Diverse Tidal Marshes

Terrestrial Aquatic Interfaces (TAIs), and tidal wetlands in particular, store large amounts of carbon yet are not well represented in Earth System Models (ESMs). Predictions of carbon cycling and greenhouse gas (GHG) emissions in tidal wetlands are highly uncertain. Eddy covariance (EC) towers provide ecosystem-scale GHG flux data at a temporal resolution (every 30min) that is helpful for parameterizing and improving mechanistic realism in ESMs. We propose to use a network of eddy covariance towers and standardized ancillary data streams, along with mesocosm experiments and statistical analyses, across diverse tidal wetlands of North America to develop and improve biogeochemical modeling at the TAI. Our overarching objective is to improve understanding and process-based modeling of gross primary productivity (GPP) and CH4 emission responses, both non-linear and asynchronous, to stressors including plant inundation, disturbance, salinity and nitrogen loading.

54 ENVIRONMENTAL SCIENCES↗
NASA NTRS1986

Optical fiber characteristics and standards; Proceedings of the Meeting, Cannes, France, November 25-27, 1985

Papers are presented on outside vapor deposition, the plasma activated CVD process for large scale production of telecommunication fibers, axial lateral plasma deposition technology from plastic clad silica, coatings for optical fibers, primary coating characterization, and radiation-induced time dependent attenuation in a fiber. Topics discussed include fibers with high tensile strength, the characteristics and specifications of airborne fiber optic components, the baseband frequency response of multimode fibers, and fibers for local and broadband networks. Consideration is given to industrial measurements for single mode and multimode fibers, the characterization of source power distribution in a multimode fiber by a splice offset technique, the measurement of chromatic dispersion in a single mode optical, and the effect of temperature on the refracted near-field optical fiber profiling technique.

Bouillie, Remy↗
NASA NTRS1993

Cloud fields derived from satellite and surface data during FIRE cirrus phase 2

The interpretation of surface and aircraft measurements of cloud properties taken during field programs must take into account the large-scale cloud and meteorological conditions. Cloud properties are also required at scales beyond the point and line data taken from ground and aircraft platforms. Satellite data can provide a quantitative description of these large-scale cloud properties. When derived from geostationary satellite data, the cloud fields constitute a unique source for evaluating the development and demise of a cloud system. Satellites, however, can only see the tops of clouds, so that cloud layers below the uppermost cloud deck may remain undetected resulting in a incomplete depiction of the cloud system. Some multilayer clouds are amenable to detection from satellites. Many, especially in midlatitude cyclonic systems, can only be observed from the surface. A combination of surface and satellite cloud observations should be the most complete quantification of large-scale cloudiness if there are sufficient surface measurements. During the First International Satellite Cloud Climatology Project (ISCCP) Regional Experiment Phase 2 (FIRE-2) Cirrus Intensive Field Observation (IFO) period (November 13 - December 7, 1991) conducted at Coffeyville, Kansas, cirrus observations were taken in a variety of conditions. The IFO area was selected for a variety of reasons including the relatively dense network of surface weather stations and special surface instrumentation sites. Thus, the FIRE-2 IFO presents an excellent opportunity to combine cloud observations from surface and satellite observations. This paper presents an analysis of cloud properties on a mesoscale grid using satellite cloud property retrievals, surface observer data, and rawinsonde temperature and humidity profiles.

Minnis, Patrick↗
NASA NTRS2024

Automatic Detection and Classification of Aurora in THEMIS All‐Sky Images

We report a novel machine-learning algorithm for automatically detecting and classifying aurora in all–sky images (ASI) that is largely trained without requiring ground–truth labels. By including a small number of labeled images, we are able to automatically label all of the approximately 700 million images in the Time History of Events and Macroscale Interactions during Substorms (THEMIS) ASI data set from 2008 to 2022. We use a two–stage approach. In the first stage, we adapt the Simple framework for Contrastive Learning of Representations (SimCLR) algorithm to learn latent representations of THEMIS all–sky images. We then finetune a classifier network on the latent representations our model learns of the manually labeled Oslo aurora THEMIS (OATH) data set. We demonstrate that this two–stage approach achieves excellent classification results on data for which there is no current ML classification benchmark. The outcome of this work will facilitate efficient information retrieval for researchers interested in specific categories of aurora and will enable large scale statistical studies and machine learning analyses of THEMIS all–sky images that have not previously been possible. To demonstrate possible ways to utilize this database, we performed a statistical analysis of the occurrence rates of auroral labels with respect to solar wind parameters, interplanetary magnetic field vector, and geomagnetic indices. We further investigate the occurrence rates of auroral phenomena in the annotated data set and their geoeffectiveness by utilizing the co–located THEMIS ground magnetometer data set.

Jeremiah W Johnson↗
DOE OSTI2024

Multiscale Flow for robust and optimal cosmological analysis

We propose Multiscale Flow, a generative Normalizing Flow that creates samples and models the field-level likelihood of two-dimensional cosmological data such as weak lensing. Multiscale Flow uses hierarchical decomposition of cosmological fields via a wavelet basis and then models different wavelet components separately as Normalizing Flows. The log-likelihood of the original cosmological field can be recovered by summing over the log-likelihood of each wavelet term. This decomposition allows us to separate the information from different scales and identify distribution shifts in the data such as unknown scale-dependent systematics. The resulting likelihood analysis can not only identify these types of systematics, but can also be made optimal, in the sense that the Multiscale Flow can learn the full likelihood at the field without any dimensionality reduction. We apply Multiscale Flow to weak lensing mock datasets for cosmological inference and show that it significantly outperforms traditional summary statistics such as power spectrum and peak counts, as well as machine learning–based summary statistics such as scattering transform and convolutional neural networks. We further show that Multiscale Flow is able to identify distribution shifts not in the training data such as baryonic effects. Finally, we demonstrate that Multiscale Flow can be used to generate realistic samples of weak lensing data.

79 ASTRONOMY AND ASTROPHYSICS↗
NASA NTRS2002

The M2 Proton Channel of Influenza Virus: How Does It Work?

The transport of protons across membranes is an essential process for both bioenergetics of modem cells and the origins of cellular life. All living systems make use of proton gradients across cell walls to convert environmental energy into a high-energy chemical compound, adenosine triphosphate (ATR), synthesized from adenosine diphosphate. ATR, in turn, is used as a source of energy to drive many cellular reactions. The ubiquity of this process in biology suggests that even the earliest cellular systems were relying on proton gradient for harvesting environmental energy needed to support their survival and growth. In contemporary cells, proton transfer is assisted by large, complex proteins embedded in membranes. The issue addressed in this study was: how the same process can be accomplished with the aid of similar, but much simpler molecules that could have existed in the protobiological milieu? The model system used in the study contained a bilayer membrane made of phospholipid, dimyristoylphosphatidylcholine (DMPC), which is a good model of the biological membranes focusing cellular boundaries. Both sides of the bilayer were surrounded by water which simulated the environment inside and outside the cell. Embedded in the membrane was a fragment of the Influenza-A M2 protein and enough sodium counterions to maintain system neutrality. This protein has been shown to exhibit remarkably high rates of proton transport and, therefore, is an excellent model to study the formation of proton gradients across membranes. The Influenza M2 protein is 97 amino acids in length, but a fragment 25 amino acids long, which contains a transmembrane domain of 19 amino acids flanked by 3 amino acids on each side, is sufficient to transport protons. Four identical protein fragments, each folded into a helix, aggregate to form small channels spanning the membrane. Protons are conducted through a narrow pore in the middle of the channel in response to applied voltage. This channel is large enough to contain water molecules, and is normally filled with water. In analogy to the mechanism of proton transfer in some other channels, it has been postulated that protons are translocated along the network of water molecules filling the pore of the channel. This mechanism, however, must involve an additional, important step because the channel contains four histidine amino acid residues, one from each of the helices, which are sufficiently large to occlude the pore and interrupt the water network. The histidine residues ensure channel selectivity by blocking transport of small such as sodium or potassium. They have been also implicated in gating protons due to the ability of each histidine to become positively charged by accepting an additional proton. Two mechanisms of gating have been proposed. In one mechanism, all four histidines acquire an additional proton and, due to repulsion between their positive charges, move away from one another, thus opening the channel. The alternative mechanism relies of the ability of protons to move between different atoms in a molecule (tautomerization). Thus, a proton is captured on one side of the gate while another proton is released from the opposite side, and the molecule returns to the initial state through tautomerization. The simulations were designed to test these two mechanisms. Large-scale, atomic-level molecular dynamics simulations of the channel, in which the histidine residues were in different protonation states revealed that all intermediate states of the system involved in the tautomerization mechanism are structurally stable and the arrangement of water molecules in the channel is conducive to the proton transport. In contrast, in the four-protonated state, postulated to exist in the gate-opening mechanism, the electrostatic repulsion between the histidine residues appears to be so large that the channel looses its structural integrity and one helix moves away from the remaining three. This result indicates that such a mechanism of proton transport is unlikely. The simulations revealed that translocation along a network of water molecules in the channel and tautomerization of the histidine residues in the M2 proteins in the most likely mechanism of proton transport. The results not only explain how a remarkably simple protein system can efficiently aid in the formation of proton gradients across cell walls, but also suggest how this system can be genetically re-engineered to become a directional, reversible proton pump. Such a pump can provide energy to laboratory-built models of simple cellular systems. If they were successfully constructed it would greatly advance our understanding of the beginnings of life and find important applications in medicine and pharmacology.

Pohorille, Andrew↗
DOE OSTI2024

Detectability of Varied Hybridization Scenarios Using Genome-Scale Hybrid Detection Methods

Hybridization events complicate the accurate reconstruction of phylogenies, as they lead to patterns of genetic heritability that are unexpected under traditional, bifurcating models of species trees. This phenomenon has led to the development of methods to infer these varied hybridization events, both methods that reconstruct networks directly, as well as summary methods that predict individual hybridization events from a subset of taxa. However, a lack of empirical comparisons between methods – especially those pertaining to large networks with varied hybridization scenarios – hinders their practical use. Here, we provide a comprehensive review of popular summary methods: TICR, MSCquartets, HyDe, Patterson’s D-Statistic (ABBA-BABA), D3, and Dp. TICR and MSCquartets are based on quartet concordance factors gathered from gene tree topologies and HyDe, Patterson’s D-Statistic, D3, and Dp use site pattern frequencies to identify hybridization events between sets of three taxa. We then use simulated data to address questions of method accuracy and ideal use scenarios by testing methods against complex networks which depict gene flow events that differ in depth (timing), quantity (single vs. multiple, overlapping hybridizations), and rate of gene flow (γ). We find that deeper or multiple hybridization events may introduce noise and weaken the signal of hybridization, leading to higher relative false negative rates across all methods. Despite some forms of hybridization eluding quartet-based detection methods, MSCquartets displays high precision in most scenarios. While HyDe results in high false negative rates when tested on hybridizations involving extinct or unsampled ghost lineages, HyDe is the only method able to identify the direction of hybridization, distinguishing the source parental lineages from recipient hybrid lineages. Lastly, we test the methods on a dataset of ultraconserved elements from the bee subfamily Nomiinae, finding possible hybridization events between clades which correspond to regions of poor support in the species tree estimated in a previous study.

Bjorner, Marianne B.↗
DOE OSTI2025

Generalizable machine learning potentials for quantum-accurate predictions of non-equilibrium behavior in 2D materials

Machine learning interatomic potentials (ML-IAPs) are emerging as transformative tools in materials modeling, promising quantum-level accuracy at a fraction of the computational cost. However, their ability to generalize beyond equilibrium configurations and to reliably capture defect- and temperature-driven behavior remains underexplored. Here, we develop and benchmark two state-of-the-art ML-IAPs, Spectral Neighbor Analysis Potential (SNAP) and Allegro, on a comprehensive dataset for monolayer MoSe₂. Using density functional theory (DFT) as the reference, we evaluate their performance in capturing stress–strain behavior, phase transition energetics, defect evolution, edge stability, and fracture toughness. Allegro, a deep equivariant neural network potential, surpasses both SNAP and the classical Tersoff potential in accuracy, efficiency, and transferability. Importantly, both ML potentials accurately reproduce experimental fracture measurements and ab initio predictions of inversion domain formation—phenomena well beyond their training sets. Our findings establish ML-IAPs as viable replacements for traditional force fields in the study of non-equilibrium mechanical phenomena, enabling large-scale, high-fidelity simulations in 2D materials and beyond. In conclusion, this work provides a broadly applicable framework for data-driven modeling of structural and functional transformations under extreme conditions.

2D materials↗
DOE OSTI2022

Using soil library hyperspectral reflectance and machine learning to predict soil organic carbon: Assessing potential of airborne and spaceborne optical soil sensing

Soil organic carbon (SOC) is a key variable to determine soil functioning, ecosystem services, and global carbon cycles. Spectroscopy, particularly optical hyperspectral reflectance coupled with machine learning, can provide rapid, efficient, and cost-effective quantification of SOC. However, how to exploit soil hyperspectral reflectance to predict SOC concentration, and the potential performance of airborne and satellite data for predicting surface SOC at large scales remain relatively underknown. Here, this study utilized a continental-scale soil laboratory spectral library (37,540 full-pedon 350–2500 nm reflectance spectra with SOC concentration of 0–780 g·kg –1 across the US) to thoroughly evaluate seven machine learning algorithms including Partial-Least Squares Regression (PLSR), Random Forest (RF), K-Nearest Neighbors (KNN), Ridge, Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), and Long Short-Term Memory (LSTM) along with four preprocessed spectra, i.e. original, vector normalization, continuum removal, and first-order derivative, to quantify SOC concentration. Furthermore, by using the coupled soil-vegetation-atmosphere radiative transfer model, we simulated twelve airborne and spaceborne hyper/multi-spectral remote sensing data from surface bare soil laboratory spectra to evaluate their potential for estimating SOC concentration of surface bare soils. Results show that LSTM achieved best predictive performance of quantifying SOC concentration for the whole data sets (R 2 = 0.96, RMSE = 30.81 g·kg –1 ), mineral soils (SOC ≤ 120 g·kg –1 , R 2 = 0.71, RMSE = 10.60 g·kg –1 ), and organic soils (SOC > 120 g·kg –1 , R 2 = 0.78, RMSE = 62.31 g·kg –1 ). Spectral data preprocessing, particularly the first-order derivative, improved the performance of PLSR, RF, Ridge, KNN, and ANN, but not LSTM or CNN. We found that the SOC models of mineral and organic soils should be distinguished given their distinct spectral signatures. Finally, we identified that the shortwave infrared is vital for airborne and spaceborne hyperspectral sensors to monitor surface SOC. This study highlights the high accuracy of LSTM with hyperspectral/multispectral data to mitigate a certain level of noise (soil moisture <0.4 m 3 ·m –3 , green leaf area < 0.3 m 2 ·m –2 , plant residue <0.4 m 2 ·m –2 ) for quantifying surface SOC concentration. Forthcoming satellite hyperspectral missions like Surface Biology and Geology (SBG) have a high potential for future global soil carbon monitoring, while high-resolution satellite multispectral fusion data can be an alternative.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2021

Improved dryland carbon flux predictions with explicit consideration of water-carbon coupling

Dryland ecosystems are dominant influences on both the trend and interannual variability of the terrestrial carbon sink. Despite their importance, dryland carbon dynamics are not well-characterized by current models. Here, we present DryFlux, an upscaled product built on a dense network of eddy covariance sites in the North American Southwest. To estimate dryland gross primary productivity, we fuse in situ fluxes with remote sensing and meteorological observations using machine learning. DryFlux explicitly accounts for intra-annual variation in water availability, and accurately predicts interannual and seasonal variability in carbon uptake. Applying DryFlux globally indicates existing products may underestimate impacts of large-scale climate patterns on the interannual variability of dryland carbon uptake. We anticipate DryFlux will be an improved benchmark for earth system models in drylands, and prompt a more sensitive accounting of water limitation on the carbon cycle.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2021

Emergent Recursive Multiscale Interaction in Complex Systems

This project studied the potential for multiscale group dynamics in complex social systems, including emergent recursive interaction. Current social theory on group formation and interaction focuses on a single scale (individuals forming groups) and is largely qualitative in its explanation of mechanisms. We combined theory, modeling, and data analysis to find evidence that these multiscale phenomena exist, and to investigate their potential consequences and develop predictive capabilities. In this report, we discuss the results of data analysis showing that some group dynamics theory holds at multiple scales. We introduce a new theory on communicative vibration that uses social network dynamics to predict group life cycle events. We discuss a model of behavioral responses to the COVID-19 pandemic that incorporates influence and social pressures. Finally, we discuss a set of modeling techniques that can be used to simulate multiscale group phenomena.

42 ENGINEERING↗
NASA NTRS1993

A synoptic-scale overview of the TOGA COARE intensive observing period November 1992 to February 1993 based on analyses from US operational global data assimilation systems

The operational global analyses from the two major U.S. numerical weather prediction centers, the Navy's Fleet Numerical Oceanography Center and the National Meteorological Center, are used to describe the synoptic-scale features of the 1 Nov. 1992 to 28 Feb. 1993 TOGA COARE intensive observing period (IOP). TOGA COARE is an international field experiment in which a large number of research scientists from the Goddard Laboratory for Atmospheres (Code 910) and the Laboratory for Hydrospheres (Code 970) participated. Two high-amplitude intraseasonal (30-60 day) oscillations passed through the TOGA COARE observational network located in the equatorial western Pacific. Associated with the oscillations were two 6-10 day periods of persistent westerly surface winds at the equator or 'westerly wind bursts.' These events are depicted through time series and time-longitude cross sections of divergence/velocity potential, surface winds, precipitation, ocean mixed-layer depth, and sea surface temperature. The high and low frequency components of the flow in which the intraseasonal oscillations were embedded are shown using seasonal, monthly, and 5-day averages of the surface, 850 and 200 mb winds, precipitation, and sea-level pressure, and a time-longitude cross section of tropical cyclone activity. Independent verification of precipitation comes from near real-time satellite estimates, and a reference climatology is given based on 9 years of ECMWF analyses. Daily 00 UTC analyses of surface winds and sea-level pressure for the entire western Pacific and Indian Ocean are provided to trace the evolution of individual synoptic events.

Fiorino, M.↗
NASA NTRS2003

Experiences with Testing the Largest Ground System NASA Has Ever Built

In the 1980s, the National Aeronautics and Space Administration (NASA) embarked upon a major Earth-focused program called Mission to Planet Earth. The Goddard Space Flight Center (GSFC) was selected to manage and develop a key component - the Earth Observing System (EOS). The EOS consisted of four major missions designed to monitor the Earth. The missions included 4 spacecraft. Terra (launched December 1999), Aqua (launched May 2002), ICESat (Ice, Cloud, and Land Elevation Satellite, launched January 2003), and Aura (scheduled for launch January 2004). The purpose of these missions was to provide support for NASA s long-term research effort for determining how human-induced and natural changes affect our global environment. The EOS Data and Information System (EOSDIS), a globally distributed, large-scale scientific system, was built to support EOS. Its primary function is to capture, collect, process, and distribute the most voluminous set of remotely sensed scientific data to date estimated to be 350 Gbytes per day. The EOSDIS is composed of a diverse set of elements with functional capabilities that require the implementation of a complex set of computers, high-speed networks, mission-unique equipment, and associated Information Technology (IT) software along with mission-specific software. All missions are constrained by schedule, budget, and staffing resources, and rigorous testing has been shown to be critical to the success of each mission. This paper addresses the challenges associated with the planning, test definition. resource scheduling, execution, and discrepancy reporting involved in the mission readiness testing of a ground system on the scale of EOSDIS. The size and complexity of the mission systems supporting the Aqua flight operations, for example, combined with the limited resources available, prompted the project to challenge the prevailing testing culture. The resulting success of the Aqua Mission Readiness Testing (MRT) program was due in no small measure to re-structuring the traditional programmatic and technical approach to a more efficient and robust program. Programmatically, it meant gaining the endorsement, commitment, and cooperation of the numerous subsystem element managers and other stakeholder organizations. Technically, it required an MRT program that was agile, could rapidly adapt to requirements changes, and was flexible in its overall approach. Furthermore, this paper addresses the following questions: 1. What are the key ingredients (e.g., test tools, organization) needed to conduct a successful MRT program? 2. What distinguishes EOS MRT from the traditional system testing approach? 3. Where should the focus of testing be since it is infeasible to test every element or subsystem? 4. How can MRT be applied effectively to other systems or missions? To provide answers to these questions, this paper relies heavily on real-life, hands-on experiences ("lessons learned") gained during mission readiness testing of the Terra ground system and, most recently, the Aqua and ICESat missions. Moreover, this paper explores how lessons learned were turned into lessons applied for the upcoming Aura mission. Although derived from the EOS missions, MRT techniques and strategies can be applied to enhance the testing of other missions.

Lehtonen, Ken↗
NASA NTRS2020

Multi-Agent Motion Planning using Deep Learning for Space Applications

State-of-the-art motion planners cannot scale to a large number of systems. Motion planning for multiple agents is an NP (non-deterministic polynomial-time) hard problem, so the computation time increases exponentially with each addition of agents. This computational demand is a major stumbling block to the motion planner's application to future NASA missions involving the swarm of space vehicles. We applied a deep neural network to transform computationally demanding mathematical motion planning problems into deep learning-based numerical problems. We showed optimal motion trajectories can be accurately replicated using deep learning-based numerical models in several 2D and 3D systems with multiple agents. The deep learning-based numerical model demonstrates superior computational efficiency with plans generated 1000 times faster than the mathematical model counterpart.

Madani, Ramtin↗
DOE OSTI2022

Attention-based convolutional capsules for evapotranspiration estimation at scale

Evapotranspiration (ET) measures the amount of water lost from the Earth's surface to the atmosphere and is an integral metric for both agricultural and environmental sciences. Understanding and quantifying ET is critical for achieving effective management of freshwater and irrigation systems. However, current ET estimation models suffer from a trade-off between accuracy and spatial coverage. In this study, we introduce our model Quench, a neural network architecture that achieves highly-accurate ET estimates over large continuous spatial extents. Quench uses our novel Attention-Based Convolutional Capsule for its neural network layers to identify areas of focus and efficiently extract ET information from satellite imagery. Benchmarks that profile our model's performance show substantive improvements in accuracy, with up to 128% increase in accuracy compared to traditional convolutional-based and process-based models. Finally, Quench also demonstrates consistent model performance over high geospatial variability and a diverse array of regions, seasons, climates, and vegetations.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2026

Comparing Delay-, Distance-, and Cordon-Based Congestion Pricing Strategies Via Large-Scale Simulation

This study compares the impacts of delay-, distance-, and cordon-based congestion pricing strategies for Austin, Texas, using the POLARIS agent-based activity-based travel demand simulation model. This approach enables agent-level heterogeneity and realistic choice options (including destination, mode, and activity scheduling) for dynamic traffic assignment and congestion feedbacks across a major metro region, which are features lacking in past work. To ensure comparability, distance-based tolls were set to generate the same revenue as delay-based tolling of $3.5 M/day, averaging $1.17/resident/day or $0.42/vehicle-trip. Delay-based pricing delivers 44% lower network delay and 13% lower VHT compared to the no-toll baseline, levels unmatched by other pricing strategies. At the height of the AM peak, drivers pay up to $0.13/mile on average, though most links in the network remain untolled. Distance-based pricing is the most effective at reducing VMT (by 4%), but VHT reductions (of 6%) primarily stem from drivers selecting closer destinations, achieving only one-fourth the delay reduction of delay-based pricing. Across various implementations of delay- and distance-based pricing, the results suggest that spatial variations of tolls are far more important than temporal variations. Cordon tolls produce minimal impacts at the network-wide level, but offer substantial delay reductions inside the cordon. Other major findings include: 1) delay-based pricing increases trip-making during the PM peak period due to backward shifts in discretionary-activity start times by higher-income residents; and 2) tolls’ spatial impacts, including changes in network flows and tolls paid by residents, vary substantially between delay- and distance-based pricing strategies.

Agent-based modeling↗
DOE OSTI2024

Harnessing large language models’ zero-shot and few-shot learning capabilities for regulatory research

Abstract Large language models (LLMs) are sophisticated AI-driven models trained on vast sources of natural language data. They are adept at generating responses that closely mimic human conversational patterns. One of the most notable examples is OpenAI's ChatGPT, which has been extensively used across diverse sectors. Despite their flexibility, a significant challenge arises as most users must transmit their data to the servers of companies operating these models. Utilizing ChatGPT or similar models online may inadvertently expose sensitive information to the risk of data breaches. Therefore, implementing LLMs that are open source and smaller in scale within a secure local network becomes a crucial step for organizations where ensuring data privacy and protection has the highest priority, such as regulatory agencies. As a feasibility evaluation, we implemented a series of open-source LLMs within a regulatory agency’s local network and assessed their performance on specific tasks involving extracting relevant clinical pharmacology information from regulatory drug labels. Our research shows that some models work well in the context of few- or zero-shot learning, achieving performance comparable, or even better than, neural network models that needed thousands of training samples. One of the models was selected to address a real-world issue of finding intrinsic factors that affect drugs' clinical exposure without any training or fine-tuning. In a dataset of over 700 000 sentences, the model showed a 78.5% accuracy rate. Our work pointed to the possibility of implementing open-source LLMs within a secure local network and using these models to perform various natural language processing tasks when large numbers of training examples are unavailable.

Biochemistry & Molecular Biology↗
DOE OSTI2021

Deep learning for Alzheimer's disease: Mapping large-scale histological tau protein for neuroimaging biomarker validation

Abnormal tau inclusions are hallmarks of Alzheimer's disease and predictors of clinical decline. Several tau PET tracers are available for neurodegenerative disease research, opening avenues for molecular diagnosis in vivo. However, few have been approved for clinical use. Understanding the neurobiological basis of PET signal validation remains problematic because it requires a large-scale, voxel-to-voxel correlation between PET and (immuno) histological signals. Large dimensionality of whole human brains, tissue deformation impacting co-registration, and computing requirements to process terabytes of information preclude proper validation. We developed a computational pipeline to identify and segment particles of interest in billion-pixel digital pathology images to generate quantitative, 3D density maps. The proposed convolutional neural network for immunohistochemistry samples, IHCNet, is at the pipeline's core. We have successfully processed and immunostained over 500 slides from two whole human brains with three phospho-tau antibodies (AT100, AT8, and MC1), spanning several terabytes of images. Our artificial neural network estimated tau inclusion from brain images, which performs with ROC AUC of 0.87, 0.85, and 0.91 for AT100, AT8, and MC1, respectively. Introspection studies further assessed the ability of our trained model to learn tau-related features. We present an end-to-end pipeline to create terabytes-large 3D tau inclusion density maps co-registered to MRI as a means to facilitate validation of PET tracers.

60 APPLIED LIFE SCIENCES↗