Search NASA⌕ Search

SEARCH · Search NASA

Results for “Linear Regression”

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 379 records · Page 21

Stream Temperature Predictions for River Basin Management in the Pacific Northwest and Mid-Atlantic Regions Using Machine Learning

Stream temperature (Ts) is an important water quality parameter that affects ecosystem health and human water use for beneficial purposes. Accurate Ts predictions at different spatial and temporal scales can inform water management decisions that account for the effects of changing climate and extreme events. In particular, widespread predictions of Ts in unmonitored stream reaches can enable decision makers to be responsive to changes caused by unforeseen disturbances. In this study, we demonstrate the use of classical machine learning (ML) models, support vector regression and gradient boosted trees (XGBoost), for monthly Ts predictions in 78 pristine and human-impacted catchments of the Mid-Atlantic and Pacific Northwest hydrologic regions spanning different geologies, climate, and land use. The ML models were trained using long-term monitoring data from 1980–2020 for three scenarios: (1) temporal predictions at a single site, (2) temporal predictions for multiple sites within a region, and (3) spatiotemporal predictions in unmonitored basins (PUB). In the first two scenarios, the ML models predicted Ts with median root mean squared errors (RMSE) of 0.69–0.84 °C and 0.92–1.02 °C across different model types for the temporal predictions at single and multiple sites respectively. For the PUB scenario, we used a bootstrap aggregation approach using models trained with different subsets of data, for which an ensemble XGBoost implementation outperformed all other modeling configurations (median RMSE 0.62 °C).The ML models improved median monthly Ts estimates compared to baseline statistical multi-linear regression models by 15–48% depending on the site and scenario. Air temperature was found to be the primary driver of monthly Ts for all sites, with secondary influence of month of the year (seasonality) and solar radiation, while discharge was a significant predictor at only 10 sites. The predictive performance of the ML models was robust to configuration changes in model setup and inputs, but was influenced by the distance to the nearest dam with RMSE <1 °C at sites situated greater than 16 and 44 km from a dam for the temporal single site and regional scenarios, and over 1.4 km from a dam for the PUB scenario. Our results show that classical ML models with solely meteorological inputs can be used for spatial and temporal predictions of monthly Ts in pristine and managed basins with reasonable (<1 °C) accuracy for most locations.

54 ENVIRONMENTAL SCIENCES↗

Neural network-based classification and regression of magnetohydrodynamic modes in tokamaks

We present a machine learning-based magnetohydrodynamic (MHD) classifier and regressor that utilizes real or complex-valued 3D magnetic sensor array data to determine neoclassical tearing mode (NTM) onset times in tokamaks with millisecond accuracy. The input dataset consists of poloidal profiles of complex Fourier amplitudes with an n = 1 toroidal mode number from 144 human-labeled ITER Baseline Scenario discharges in the DIII-D tokamak, spanning both tearing-dominated and sawtooth-dominated regimes. Since m, n = 2,1 NTMs frequently emerge alongside sawteeth at the same frequency in this scenario, the focus is on isolating the m = 1 and m = 2 components of the n = 1 MHD mode near the tearing onset. To improve model regularization and prediction stability, singular value decomposition was applied to balance the sawtooth and tearing datasets. The enriched datasets facilitated training neural networks that learn the key distinguishing features of sawtooth and tearing modes in the poloidal profiles of their magnetic amplitude and phase. When the modes occur independently, the networks achieve perfect classification due to the modes’ distinct characteristics and low measurement noise. In the more experimentally relevant case where both modes coexist, the networks maintain exceptional performance across key metrics. Tests on synthetic data with known ground truth demonstrate the superior accuracy of the neural network trained on complex-valued input compared to models using real amplitude, phase, or pseudo-complex data, achieving both a mean time delay and standard deviation below 1 ms. Notably, standard linear regression methods fitting the dominant singular modes to the data closely match the neural network’s performance. Applying these methods across a broad range of H-mode scenarios will enable future studies to systematically identify dominant NTM triggers as scenario-specific variables, paving the way for more effective tearing mode avoidance strategies in future fusion reactor designs.

machine learning↗

Prediction of muscle performance during dynamic repetitive movement

BACKGROUND: During long-duration spaceflight, astronauts experience progressive muscle atrophy and often perform strenuous extravehicular activities. Post-flight, there is a lengthy recovery period with an increased risk for injury. Currently, there is a critical need for an enabling tool to optimize muscle performance and to minimize the risk of injury to astronauts while on-orbit and during post-flight recovery. Consequently, these studies were performed to develop a method to address this need. METHODS: Eight test subjects performed a repetitive dynamic exercise to failure at 65% of their upper torso weight using a Lordex spinal machine. Surface electromyography (SEMG) data was collected from the erector spinae back muscle. The SEMG data was evaluated using a 5th order autoregressive (AR) model and linear regression analysis. RESULTS: The best predictor found was an AR parameter, the mean average magnitude of AR poles, with r = 0.75 and p = 0.03. This parameter can predict performance to failure as early as the second repetition of the exercise. CONCLUSION: A method for predicting human muscle performance early during dynamic repetitive exercise was developed. The capability to predict performance to failure has many potential applications to the space program including evaluating countermeasure effectiveness on-orbit, optimizing post-flight recovery, and potential future real-time monitoring capability during extravehicular activity.

Physical Endurance/physiology↗

Different binding modes of Cu and Pb vs. Cd, Ni, and Zn with the trihydroxamate siderophore desferrioxamine B at seawater ionic strength

The solution speciation in seawater of divalent trace metals (Cd, Cu, Ni, Pb, Zn) is dominated by strong, ostensibly metal-specific organic ligands that may play important roles in microbial metal acquisition and/or detoxification processes. We compare the effective stabilities of these metal-organic complexes to the stabilities of their complexes with a model siderophore, desferrioxamine B (DFOB). While metal-DFOB complexation has been studied in various dilute but often moderately coordinating media, for the purpose of this investigation we measured the stability constants in a non-coordinating background electrolyte at seawater ionic strength (0.7 M NaClO4). Potentiometric titrations of single metals (M) were performed in the presence of ligand (L) at different M:L molar ratios, whereupon the stability constants of multiple complexes were simultaneously determined by non-linear regression of the titration curves with FITEQL, using the optimal binding mode for each metal. Cadmium, Ni, and Zn, like trivalent Fe, sequentially form a bi-, tetra-, and hexadentate complex with DFOB as pH increases, consistent with their coordination number of 6 and regular octahedral geometry. Copper has a Jahn-Teller-distorted square-bipyramidal geometry whereas the geometry of Pb is cryptic, involving a range of bond lengths. Supported by a thermodynamic argument, our data suggest that this impedes binding of the third hydroxamate group and that the hexadentate Cu-DFOB and Pb-DFOB complex identified in earlier reports may instead be a deprotonated tetradentate complex. Absence of the hexadentate complex promotes the formation of a dinuclear (bidentate-tetradentate) complex, M2HL2+, albeit not for Pb in 0.7 M NaCl, evidently due to extensive complexation with chloride. Stabilities of the hexadentate Ni-DFOB, Zn-DFOB, and the tetradentate Pb-DFOB complex are nearly equal, yet about 2 orders of magnitude higher and 4 orders of magnitude lower than those of the hexadentate Cd-DFOB and tetradentate Cu-DFOB complex, respectively. Linear free-energy relations defined by the rare earth elements are able to predict stabilities of the Cd, Zn, and one of the Pb complexes, but underestimate those of the Ni and Cu complexes. The comparison with metal-specific organic ligands detected in seawater yields fair agreement for three of the five metals, implying that they could be siderophore-like. The Cd- and Ni-specific ligands are much stronger and may contain quite different functional groups. Calculations with MINEQL incorporating our new stability constants indicate that very high DFOB concentrations would be required to match the extent of metal-organic complexation observed in seawater, however DFOB may well represent a much broader class of structurally related ligands.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Interindividual variability of arterial impulse response to intravenous injection of nonionic contrast agent (Iohexol) in DCE-CT study

Purpose: It is known that the arterial input function (AIF) in dynamic contrast-enhanced (DCE)-CT differs among patients even for fixed contrast injection protocols. Therefore, a study has been performed to investigate the interindividual variability of the AIF with respect to patient factors (such as weight, height, and age). In addition, it has been demonstrated that the relations from the interindividual variability investigation can be further used for the estimation of AIF for a patient without the requirement of measurement. Methods: DCE-CT data for a cohort of 34 patients with cervical carcinoma were used for the investigation of interindividual variability of the AIF. To dissociate the effect of different durations of contrast injection, the arterial impulse response (AIR) to intravenous contrast injection was calculated and examined for its correlations with these patient factors. An empirical functional form was proposed to model the AIR with temporal intensity of a first pass of contrast agent followed by recirculation and quasiequilibrium state of contrast concentration. Specific features (onset time, peak time, and amplitudes) of the AIR were tested for correlations with the patient factors. Linear regression was applied to cases that show significant strong correlation between the AIR amplitudes and patient factors. The results were then used to predict the AIR for any given patient based on the patient factors. It was shown that using the predicted AIR, the AIF of the patient can be estimated without the requirement of measurement given the injection protocol is known. The method of AIF estimation was tested in DCE-CT data from another group of 14 patients. The efficacy of individually estimated AIF on pharmacokinetic analysis was assessed against the use of measured AIF and population-averaged AIF as the latter is another possible strategy for AIF generation if AIF measurement is not available. Results: It was found that the amplitudes of AIR postonset time were significantly correlated with patient bodyweight at most time points (r<-0.44, P<0.01) except 8-17 s (P>0.01). When the adiabatic tissue homogeneity (ATH) and the Kety models were applied to the pharmacokinetic analysis, the mean percentage errors in kinetic parameter estimates induced by using the estimated AIF instead of the measured one for both models ranged from -4.98% to 11.19%. Conclusions: It was found that there are strong linear relationships between AIR and patient weight; thereby the AIR of a patient can be estimated with patient weight. The proposed method of AIF estimation is potentially applicable in cases where AIF measurement is not possible.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Accuracy of Robotic Radiosurgical Liver Treatment Throughout the Respiratory Cycle

Purpose: To quantify random uncertainties in robotic radiosurgical treatment of liver lesions with real-time respiratory motion management. Methods and Materials: We conducted a retrospective analysis of 27 liver cancer patients treated with robotic radiosurgery over 118 fractions. The robotic radiosurgical system uses orthogonal x-ray images to determine internal target position and correlates this position with an external surrogate to provide robotic corrections of linear accelerator positioning. Verification and update of this internal–external correlation model was achieved using periodic x-ray images collected throughout treatment. To quantify random uncertainties in targeting, we analyzed logged tracking information and isolated x-ray images collected immediately before beam delivery. For translational correlation errors, we quantified the difference between correlation model–estimated target position and actual position determined by periodic x-ray imaging. To quantify prediction errors, we computed the mean absolute difference between the predicted coordinates and actual modeled position calculated 115 milliseconds later. We estimated overall random uncertainty by quadratically summing correlation, prediction, and end-to-end targeting errors. We also investigated relationships between tracking errors and motion amplitude using linear regression. Results: The 95th percentile absolute correlation errors in each direction were 2.1 mm left–right, 1.8 mm anterior–posterior, 3.3 mm cranio–caudal, and 3.9 mm 3-dimensional radial, whereas 95th percentile absolute radial prediction errors were 0.5 mm. Overall 95th percentile random uncertainty was 4 mm in the radial direction. Prediction errors were strongly correlated with modeled target amplitude (r=0.53-0.66, P<.001), whereas only weak correlations existed for correlation errors. Conclusions: Study results demonstrate that model correlation errors are the primary random source of uncertainty in Cyberknife liver treatment and, unlike prediction errors, are not strongly correlated with target motion amplitude. Aggregate 3-dimensional radial position errors presented here suggest the target will be within 4 mm of the target volume for 95% of the beam delivery.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Decadal-Scale Responses in Middle and Upper Stratospheric Ozone From SAGE II Version 7 Data

Stratospheric Aerosol and Gas Experiment (SAGE II) version 7 (v7) ozone profiles are analyzed for their decadal-scale responses in the middle and upper stratosphere for 1991 and 1992-2005 and compared with those from its previous version 6.2 (v6.2). Multiple linear regression (MLR) analysis is applied to time series of its ozone number density vs. altitude data for a range of latitudes and altitudes. The MLR models that are fit to the time series data include a periodic 11 yr term, and it is in-phase with that of the 11 yr, solar UV (Ultraviolet)-flux throughout most of the latitude/ altitude domain of the middle and upper stratosphere. Several regions that have a response that is not quite in-phase are interpreted as being affected by decadal-scale, dynamical forcings. The maximum minus minimum, solar cycle (SClike) responses for the ozone at the low latitudes are similar from the two SAGE II data versions and vary from about 5 to 2.5% from 35 to 50 km, although they are resolved better with v7. SAGE II v7 ozone is also analyzed for 1984-1998, in order to mitigate effects of end-point anomalies that bias its ozone in 1991 and the analyzed results for 1991-2005 or following the Pinatubo eruption. Its SC-like ozone response in the upper stratosphere is of the order of 4%for 1984-1998 vs. 2.5 to 3%for 1991-2005. The SAGE II v7 results are also recompared with the responses in ozone from the Halogen Occultation Experiment (HALOE) that are in terms of mixing ratio vs. pressure for 1991-2005 and then for late 1992- 2005 to avoid any effects following Pinatubo. Shapes of their respective response profiles agree very well for 1992-2005. The associated linear trends of the ozone are not as negative in 1992-2005 as in 1984-1998, in accord with a leveling off of the effects of reactive chlorine on ozone. It is concluded that the SAGE II v7 ozone yields SC-like ozone responses and trends that are of better quality than those from v6.2.

Remsberg, E. E.↗

The tonotopic map in the embryonic chicken cochlea

The purpose of the present study was to determine the tonotopic map in the chicken cochlea at 19 days of incubation (E19) by obtaining characteristic frequencies (CFs) for primary afferents, labeling the characterized neurons, and documenting their projections to the papilla. The lowest and highest CFs recorded were 188 and 1623 Hz respectively. The embryonic tonotopic map coincided with maps reported for post-hatch chicks. There were no evidence that neurons selective to low frequencies project inappropriately to more basal locations of the embryonic papilla. Linear regression was used to estimate the frequency gradient (b = 0.037 +/- 0.012 In Hz/% [b +/- SEb]) and intercept (In C, where C = 111 Hz) of the semilog plot of frequency versus cochlear position (in % distance from apex). From these estimates the octave distribution was calculated to be 18.7%/octave or 0.58 mm/octave. These quantities were not significantly different from those found in post hatch chickens. We conclude that the tonotopic map of the avian cochlea for CFs between 100 and 1700 Hz is stable and relatively mature from age E19 to post-hatch day 21 (P21). The most striking sign of immaturity in the E19 embryo is the limited range of high CFs. We offer the hypothesis that, between the ages of E19 and P21, improvements in middle ear admittance alone or in combination with functional maturation of the cochlear base may be the principal factors responsible for the appearance of adult-like high CF limits and not an apically shifting tonotopic map.

NASA Discipline Neuroscience↗

THE SPATIALLY RESOLVED STAR FORMATION LAW FROM INTEGRAL FIELD SPECTROSCOPY: VIRUS-P OBSERVATIONS OF NGC 5194

We investigate the relation between the star formation rate (SFR) surface density (SIGMA{sub SFR}) and the mass surface density of gas (SIGMA{sub gas}) in NGC 5194 (a.k.a. M51a, Whirlpool Galaxy). Visible Integral field Replicable Unit Spectrograph Prototype (VIRUS-P) integral field spectroscopy of the central 4.1 x 4.1 kpc{sup 2} of the galaxy is used to measure Halpha, Hbeta, [O III]lambda5007, [N II]lambdalambda6548,6584, and [S II]lambdalambda6717,6731 emission line fluxes for 735 regions approx170 pc in diameter. We use the Balmer decrement to calculate nebular dust extinctions, and correct the observed fluxes in order to accurately measure SIGMA{sub SFR} in each region. Archival H I 21 cm and CO maps with spatial resolution similar to that of VIRUS-P are used to measure the atomic and molecular gas surface density for each region. We present a new method for fitting the star formation law (SFL), which includes the intrinsic scatter in the relation as a free parameter, allows the inclusion of non-detections in both SIGMA{sub gas} and SIGMA{sub SFR}, and is free of the systematics involved in performing linear regressions over incomplete data in logarithmic space. After rejecting regions whose nebular spectrum is affected by the central active galactic nucleus in NGC 5194, we use the [S II]/Halpha ratio to separate spectroscopically the contribution from the diffuse ionized gas (DIG) in the galaxy, which has a different temperature and ionization state from those of H II regions in the disk. The DIG only accounts for 11% of the total Halpha luminosity integrated over the whole central region, but on local scales it can account for up to a 100% of the Halpha emission, especially in the inter-arm regions. After removing the DIG contribution from the Halpha fluxes, we measure a slope N = 0.82 +- 0.05, and an intrinsic scatter epsilon = 0.43 +- 0.02 dex for the molecular gas SFL. We also measure a typical depletion timescale tau=SIGMA{sub H{sub i+H{sub 2}}}/SIGMA{sub SFR}approx2 Gyr, in good agreement with recent measurements by Bigiel et al. The atomic gas density shows no correlation with the SFR, and the total gas SFL in the sampled density range closely follows the molecular gas SFL. Integral field spectroscopy allows a much cleaner measurement of Halpha emission line fluxes than narrow-band imaging, since it is free of the systematics introduced by continuum subtraction, underlying photospheric absorption, and contamination by the [N II] doublet. We assess the validity of different corrections usually applied in narrow-band measurements to overcome these issues and find that while systematics are introduced by these corrections, they are only dominant in the low surface brightness regime. The disagreement with the previous measurement of a super-linear molecular SFL by Kennicutt et al. is most likely due to differences in the fitting method. Our results support the recent evidence for a low, and close to constant, star formation efficiency (SFE =tau{sup -1}) in the molecular component of the interstellar medium. The data show an excellent agreement with the recently proposed model of the SFL by Krumholz et al. The large intrinsic scatter observed may imply the existence of other parameters, beyond the availability of gas, which are important in setting the SFR.

79 ASTRONOMY AND ASTROPHYSICS↗

HydraGNN_Predictive_GFM_2026 - Ensemble of predictive graph foundation models for atomistic materials modeling

This release contains data and parameters of HydraGNN-based graph foundation models trained as a result of the work published in the pre-print "Exascale Multi-Task Graph Foundation Models for Imbalanced, Multi-Fidelity Atomistic Data" by M. Lupo Pasini et al. (https://arxiv.org/abs/2604.15380). We jointly train on 16 open first-principles datasets (544+ million structures covering 85+ elements) using a multi-task architecture with per-dataset heads and a scalable ADIOS2/DDStore data pipeline. On Frontier, we execute six large-scale DeepHyper hyperparameter optimization campaigns in FP64 and promote the top-performing message-passing models to sustained 2,048-node training, yielding a PaiNN-based lead model. The version of HydraGNN used to generate the outputs provided in this release is HydraGNN v5.0 (https://github.com/ORNL/HydraGNN/releases/tag/v5.0) The list of datasets used for the training of the graph foundation model is the following: 1) Alexandria [1] 2) ANI1x [2] 3) MPTrj [3] 4) Open Catalyst 2020 (OC20) [4] 5) Open Catalyst 2022 (OC22) [5] 6) Open Catalyst 2025 (OC25) [6] 7) Open Direct ir Capture 2023 (ODAC23) [7] 8) Open Materials 2024 (OMat24) [8] 9) Open Molecules 2025 (OMol25) [9] 10) OMol25-neutral (subset of OMol25 that contains only molecules with zero total charge) 11) OMol25-non-neutral (subset of OMol25 that contains only molecules with non-zero total charge) 12) Open Polymers 2026 (OPoly2026) [10] 13) Nabla2DFT [11] 14) QCML [12] 15) QM7X [reference 13] 16) transition1x [14] Dataset references: [1] J. Schmidt et al., “A dataset of 175k stable and metastable materials calculated with the PBEsol and SCAN functionals,” Scientific Data, vol. 9, p. 64, 2022. [2] J. S. Smith et al., “The ANI-1ccx and ANI-1x data sets, coupled-cluster and density functional theory properties for molecules,” Scientific Data, vol. 7, p. 134, 2020. [Online]. Available: https: //www.nature.com/articles/s41597-020-0473-z [3] A. Jain et al., “Commentary: The Materials Project: A materials genome approach to accelerating materials innovation,” APL Materials, vol. 1, no. 1, p. 011002, 07 2013. [Online]. Available: https://doi.org/10.1063/1.4812323 [4] L. Chanussot et al., “Open catalyst 2020 (oc20) dataset and community challenges,” ACS Catalysis, vol. 11, no. 10, pp. 6059–6072, 2021. [Online]. Available: https://doi.org/10.1021/acscatal.0c04525 [5] K. Tran et al., “Open catalyst 2022 (oc22) dataset and challenges for oxidation electrocatalysts,” ACS Catalysis, vol. 13, no. 5, pp. 3066–3084, 2023. [Online]. Available: https://doi.org/10.1021/acscatal.2c05426 [6] S. J. Sahoo et al., “The open catalyst 2025 (oc25) dataset and models for solid-liquid interfaces,” arXiv preprint arXiv:2509.17862, 2025. [Online]. Available: https://arxiv.org/abs/2509.17862 [7] A. Sriram et al., “The open DAC 2023 dataset and challenges for sorbent discovery in direct air capture,” ACS Central Science, vol. 10, no. 5, pp. 923–941, 2024. [8] L. Barroso-Luque et al., “Open materials 2024 (omat24) inorganic materials dataset and models,” 2024. [Online]. Available: https://arxiv.org/abs/2410.12771 [9] D. S. Levine et al., “The open molecules 2025 (OMol25) dataset, evaluations, and models,” 2025. [Online]. Available: https://arxiv.org/abs/2505.08762 [10] D. S. Levine et al., The open polymers 2026 (OPoly26) dataset and evaluations,” arXiv preprint arXiv:2512.23117, 2025. [Online]. Available: https://arxiv.org/abs/2512.23117 [11] K. Khrabrov et al., “Nabla2dft: A universal quantum chemistry dataset of drug-like molecules and a benchmark for neural network potentials,” in NeurIPS 2024 Datasets and Benchmarks Track, 2024. [Online]. Available: https://openreview.net/forum?id=ElUrNM9U8c [12] S. Ganscha et al., “The QCML dataset, quantum chemistry reference data from 33.5M DFT and 14.7B semi-empirical calculations,” Scientific Data, vol. 12, p. 406, 2025. [13] J. Hoja et al., “QM7-X, a comprehensive dataset of quantum-mechanical properties spanning the chemical space of small organic molecules,” Scientific Data, vol. 8, p. 43, 2021. [Online]. Available: https://www.nature.com/articles/s41597-021-00812-2 [14] M. Schreiner et al., “Transition1x - a dataset for building generalizable reactive machine learning potentials,” Scientific Data, vol. 9, p. 779, 2022. The folder "datasets_ADIOS2_format" contains the set of pre-processed datasets in Adaptable I/O System (ADIOS) format (https://www.exascaleproject.org/research-project/adios/) that have been used for the development and training of GFMs in this work. The "datasets_ADIOS2_format" directory contains 2 sub-directories, one for the version "v1" of the datasets and one for the version "v2" of the datasets. The version "v1" of the datasets provides values of the total energy as they are extracted from the original data as it was released by the respective institutions. The version "v2" of the datasets provides values of the energy that have been realigned. The realignment was performed by training a linear regression model that predicts the total energy as a function of the chemical composition of the atomistic structure, and then subtract such prediction from the original value of the total energy. Both folders "v1" and "v2" contain 16 sub-directories, each corresponding to an ADIOS2-formatted dataset The folder "DeepHyper-results" contains the configurational files and model's parameters for all the 186 HPO trials that were successfully completed by the scalable hyperparameter optimization (HPO) runs on Frontier. The content of the folder "DeepHyper-results" I structured as follows: 1) task-list.txt: list of mpnn name, jobid, and deephyper task id 2) gfm_${MPNN}_${JOBID}_0.${TASKID}: run directory with checkpoint files 3) gfm_${MPNN}: deephyper summary directory (*.csv) for each specific MPNN type 4) deephyper-experiment-${JOBID}: output and error logs for each job The file "deephyper-sorted.csv" contains the details of each HydraGNN model built and tested by HPO, obtained by merging the (*.csv) filed from each HPO run executed. Out of all the HPO trials, we selected 10 to continue the training of the respective HydraGNN models. Due to limited computational budget available in the LRN070 allocation we could not complete the training till convergence for all these 10 selected models. The folder "models" contains multiple sub-folders, one per each HydraGNN model trained. Each model sub-folder contains the parameters of each HydraGNN model, with multiple checkpoint-restarts. The list of sub-folders are as follows: 1) multidataset_hpo-BEST1-fp64 2) multidataset_hpo-BEST2-fp64 3) multidataset_hpo-BEST3-fp64 4) multidataset_hpo-BEST4-fp64 5) multidataset_hpo-BEST5-fp64 6) multidataset_hpo-BEST6-fp64 7) multidataset_hpo-BEST7-fp64 8) multidataset_hpo-BEST8-fp64 9) multidataset_hpo-BEST9-fp64 10) multidataset_hpo-BEST10-fp64 Within each one of these folders, additional auxiliary log files are provided with descriptions about how the training proceeded. The lead PaiNN-model is contained inside "multidataset_hpo-BEST6-fp64". The file "mlp_branch_weights" contains the parameters of the multi-layer perceptron (MLP) used to reconcile the predictions of the 16 output decoding heads of the HydragNN architectures. The MLP takes in input the chemical composition of the atomistic structure and predicts averaging weights to linearly mix the predictions of each output decoding head toward consolidating them into a single one. The folder "1.1billion-structure-inference" contains 1.1 billion atomistic structures randomly generated. Each structures is associated with energy and forces predicted with the lead-PaiNN model combined with the MLP model for reconciliation of the multi-branch predictions generated by the 16 output decoding heads. The folder "1.1billion-structure-inference" contains 9,300 (*.tar.gz) subdirectories, one per Frontier compute node used to execute the inference at exascale. Once uncompressed, each (*.tar.gz) subdirectory contains an ADIOS2 (*.bp) file container, where each atomistic structure is stored as a PyTorch-Geometric Data object. The file "export_dataset_environment_variables.sh" contains the environment variables that need to be set before running the HydraGNN code to reproduce the results provided in this dataset release. The code that can be used to load the ADIOS2 files, load HydraGNN models, and run inference is available at: https://github.com/ORNL/HydraGNN/releases/tag/v5.0

36 MATERIALS SCIENCE↗

The Evolution of Randomized Clinical Trial Designs to Assess Therapeutics in Alzheimer Disease

Importance The success of recent randomized clinical trials (RCTs) for Alzheimer disease (AD), particularly those focusing on anti-amyloid therapies, has been discussed at length. However, the evolution of RCT design features for AD that preceded this success remain underexplored. Objective To describe temporal changes in the features of RCT design for interventions in AD. Evidence Review PubMed, Scopus, and Web of Science databases were searched in January 2025 for phase 2 and 3 AD RCTs published between January 1992 and December 2024. RCTs that investigated an intervention for AD, with a placebo or standard-of-care control group, were included. Four assessors independently reviewed full-text articles to capture study characteristics. Main Outcomes and Measures The number of participants and the duration of RCTs as well as the target population, outcomes, and funding were extracted from published reports. These features were analyzed with respect to time using linear regression and χ 2 analyses. Results The study included 203 RCTs with 79 589 participants testing interventions in AD. From 1992 to 2024, the mean sample size increased by 464% for phase 2 RCTs (from 42 to 237), and 50% for phase 3 RCTs (from 632 to 951), while the mean trial duration increased by 188% (from 16 to 46 weeks) for phase 2, and 256% (from 20 to 71 weeks) for phase 3 RCTs. This longer duration of RCTs may be partially attributed by a greater share of disease-modifying rather than symptomatic treatments. Similarly, more recent trials required AD biomarker evidence for enrollment (from 1 of 36 [2.7%] before 2006 to 40 of 76 [52.6%] since 2019). A substantial difference in the type of therapeutics researched was observed, with anti-amyloid and anti-tau RCTs being more likely to be funded by the pharmaceutical industry compared with neurotransmitter or other RCTs (anti-amyloid or anti-tau, 68 of 71 [95.8%]; neurotransmitter, 52 of 69 [77.6%]; other, 33 of 52 [63.5%]). RCT transparency improved, with more frequent data accessibility statements, registered reports, and better reporting on race and ethnicity. Conclusions and Relevance This methodology research of AD RCTs highlights substantial changes in key features of AD clinical trials from 1992 to 2024. AD RCTs have become larger and longer, such that they are powered to detect smaller clinical differences. The increased sample sizes and duration should enable the detection of smaller and more slowly occurring outcomes, which may lead to successful RCTs of therapies with slower and more subtle efficacy.

General & Internal Medicine↗

Investigation of the marine boundary layer cloud and CCN properties under coupled and decoupled conditions over the Azores

Here, six coupled and decoupled marine boundary layer (MBL) clouds were chosen from the 19 month Atmospheric Radiation Measurement Mobile Facility data set over the Azores. Thresholds of liquid water potential temperature difference Δθ L < 0.5 K (>0.5 K) and total water mixing ratio difference Δq t < 0.5 g/kg (>0.5 g/kg) below the cloud base were used for selecting the coupled (decoupled) cases. A schematic diagram was given to demonstrate the coupled and decoupled MBL vertical structures and how they associate with nondrizzle, virga, and rain drizzle events. Out of a total of 2676 5 min samples, 34.5% were classified as coupled and 65.5% as decoupled, 36.2% as nondrizzle and 63.8% as drizzle (47.7% as virga and 16.1% as rain), and 33.4% as daytime and 66.6% as nighttime. The decoupled cloud layer is deeper (0.406 km) than coupled cloud layer (0.304 km), and its liquid water path and cloud droplet effective radius (r e ) values (122.1 gm -2 and 13.0 µm) are higher than coupled ones (83.7 gm -2 and 10.4 µm). Conversely, decoupled stratocumuli have lower cloud droplet number concentration (N d ) and surface cloud condensation nucleus (CCN) concentration (N CCN ) (74.5 cm -3 and 150.9 cm -3 ) than coupled stratocumuli (111.7 cm -3 and 216.4 cm -3 ). The linear regressions between r e and N d with N CCN have demonstrated that coupled r e and N d strongly depend on N CCN and have higher correlations (-0.56 and 0.59) with N CCN than decoupled results (-0.14 and 0.25). The MBL cloud properties under nondrizzle and virga drizzle conditions are similar to each other but significantly different to those of rain drizzle.

54 ENVIRONMENTAL SCIENCES↗

Substantial soil organic carbon retention along floodplains of mountain streams

Small, snowmelt–dominated mountain streams have the potential to store substantial organic carbon in floodplain sediment because of high inputs of particulate organic matter, relatively lower temperatures compared with lowland regions, and potential for increased moisture conditions. This work (i) quantifies mean soil organic carbon (OC) content along 24 study reaches in the Colorado Rocky Mountains using 660 soil samples, (ii) identifies potential controls of OC content based on soil properties and spatial position with respect to the channel, and (iii) and examines soil properties and OC across various floodplain geomorphic features in the study area. Stepwise multiple linear regression (adjusted r 2 = 0.48, p < 0.001) indicates that percentage of silt and clay, sample depth, percent sand, distance from the channel, and relative elevation from the channel are significant predictors of OC content in the study area. Principle component analysis indicates limited separation between geomorphic floodplain features based on predictors of OC content. A lack of significant differences among floodplain features suggests that the systematic random sampling employed in this study can capture the variability of OC across floodplains in the study area. Mean floodplain OC (6.3 ± 0.3%) is more variable but on average greater than values in uplands (1.5 ± 0.08% to 2.2 ± 0.14%) of the Colorado Front Range and higher than published values from floodplains in other regions, particularly those of larger rivers.

58 GEOSCIENCES↗

Partitioning evapotranspiration using long‐term carbon dioxide and water vapor fluxes

Abstract The separate components of evapotranspiration (ET) elucidate the pathways and time scales over which water is returned to the atmosphere, but ecosystem‐scale measurements of transpiration ( T ) and evaporation ( E ) remain elusive. We propose a novel determination of E and T using multiyear eddy covariance estimates of ET and gross ecosystem photosynthesis (GEP). The method is applicable at water‐limited sites over time periods during which a linear regression between GEP (abscissa) and ET (ordinate) yields a positive ET axis intercept, an estimate of E . At four summer‐rainfall semiarid sites, T /ET increases to a peak coincident with maximum GEP and remains elevated as the growing season progresses, consistent with previous, direct measurements. The seasonal course of T /ET is related to increasing leaf area index and declining frequency of rainy days—an index of the wet surface conditions that promote E —suggesting both surface and climatic controls on ET partitioning.

Scott, Russell L.↗

Temporal Coupling of Subsurface and Surface Soil CO 2 Fluxes: Insights From a Nonsteady State Model and Cross-Wavelet Coherence Analysis

Inferences about subsurface CO 2 fluxes often rely on surface soil respiration (R soil ) estimates because directly measuring subsurface microbial and root respiration (collectively, CO 2 production, S Total ), is difficult. To evaluate how well R soil serves as a proxy for S Total , we applied the nonsteady state DEconvolution of Temporally varying Ecosystem Carbon componenTs model (0.01-m vertical resolution), using 6-hourly data from a Wyoming grassland, in six simulations that cross three soil types (clay, sandy loam, and sandy) with two depth distributions of subsurface biota. We used cross-wavelet coherence analysis to examine temporal coherence (localized linear correlation) and offsets (lags) between S Total and R soil and fluxes and drivers (e.g., soil temperature and moisture). Cross-wavelet coherence revealed higher coherence between fluxes and drivers than linear regressions between concurrent variables. Soil texture and moisture exerted the strongest controls over coherence between CO 2 fluxes. Coherence between CO 2 fluxes in all soil types was strong at short (~1 day) and long periods (>8 days), but soil type controlled lags, and rainfall events decoupled the fluxes at periods of 1–8 days for several days in sandy soil, up to 1 week in sandy loam, and for a month or more in clay soil. Concentrating root and microbial biomass nearer the surface decreased lags in all soil types and increased coherence up to 10% in clay soil. The assumption of high temporal coherence between R soil and S Total is likely valid in dry, sandy soil, but may lead to underestimates of short-term STotal in semiarid grasslands with fine-grained and/or wet soil.

54 ENVIRONMENTAL SCIENCES↗

Battery Charge Curve Prediction via Feature Extraction and Supervised Machine Learning

Real-time onboard state monitoring and estimation of a battery over its lifetime is indispensable for the safe and durable operation of battery-powered devices. In this study, a methodology to predict the entire constant-current cycling curve with limited input information that can be collected in a short period of time is developed. A total of 10 066 charge curves of LiNiO 2 -based batteries at a constant C-rate are collected. With the combination of a feature extraction step and a multiple linear regression step, the method can accurately predict an entire battery charge curve with an error of < 2% using only 10% of the charge curve as the input information. The method is further validated across other battery chemistries (LiCoO 2 -based) using open-access datasets. The prediction error of the charge curves for the LiCoO 2 -based battery is around 2% with only 5% of the charge curve as the input information, indicating the generalization of the developed methodology for predicting battery cycling curves. The developed method paves the way for fast onboard health status monitoring and estimation for batteries during practical applications.

25 ENERGY STORAGE↗

Factorization Machine‐Based Active Learning for Functional Materials Design with Optimal Initial Data

The optimization of functional materials is important to enhance their properties, but their complex geometries pose great challenges to optimization. Data-driven algorithms efficiently navigate such complex design spaces by learning relationships between material structures and performance metrics to discover high-performance functional materials. Surrogate-based active learning, continually improving its surrogate model by iteratively including high-quality data points, has emerged as a cost-effective data-driven approach. Furthermore, it can be coupled with quantum computing to enhance optimization processes, especially when paired with a special form of surrogate model (i.e., quadratic unconstrained binary optimization), formulated by factorization machine (FM). However, current practices often overlook the variability in design space sizes when determining the initial data size for optimization. In this work, we investigate the optimal initial data sizes required for efficient convergence across various design space sizes. By employing averaged piecewise linear regression, we identify initiation points where convergence begins, highlighting the crucial role of employing adequate initial data in achieving efficient optimization. These results contribute to the efficient optimization of functional materials by ensuring faster convergence and reducing computational costs in FM-based active learning.

active learning↗