114333 research outputs found
Sort by
Unusual activity of class II diterpene cyclases
No full textTerpenoids comprise a profoundly diverse and evolutionarily ancient superfamily of natural products that exhibit broad applications from ecology to therapeutics. These structurally and stereochemically complex metabolites are difficult to produce with traditional synthetic chemistry alone. Therefore, elucidation of the enzymatic biosynthesis pathways is of great interest, both to expand accessibility and to support broader commercial applications. Terpenoids are produced from simple five-carbon isoprene subunits sequentially concatenated to form terpene chains of varying lengths. Terpene subfamilies are characterized based on their carbon numeration. Diterpenoids are the twenty-carbon containing subclass, many of which have industrial applications including flavorings, fragrances, biofuels, cosmetics, and because of anti-inflammatory, antimicrobial, and antitumor properties, they are highly sought after in the pharmaceutical and agricultural sectors as well. Labdane-related diterpenoids (LRDs) define over 15,000 of these compounds and account for the majority of hydrocarbon backbone diversity. Class II diterpene cyclases (DTCs) are the enzymatic gatekeepers to labdane-related diterpenoid biosynthesis committing the universal substrate, geranylgeranyl diphosphate (GGPP), to diterpenoid production. DTCs catalyze the bicyclization of GGPP via an acid-base mechanism by which multiple highly reactive carbocation intermediates rearrange the carbon scaffold to yield a diterpene product. The DTC active site influences substrate entry, rearrangement, and product outcome. For this reason, investigations into the broadly conserved motifs, and unusual variations thereof, are worthwhile to gain insights on how the regio- and stereo- specificity is controlled. Here, a singularity was identified in the fungus Clitopilus passeckerianus by which its one-of-a-kind variation in the DTC catalytic acid motif was discovered to be essential in the formation of the unusual multidienyl diphosphate product. When reverted to the highly conserved motif, a novel DTC product was achieved. Interestingly, of the nearly 80 obviously possible DTC products, only 20 have been produced experimentally. Natural product chemists regularly identify novel diterpenoids, some of which contain carbon backbones of the not-yet-achieved DTC products. Consequently, genomic analysis on these unique diterpenoid-producing species to identify novel DTCs should be pursued to unveil new chemistries associated with these exceptional natural products. Dodonaea viscosa is an example of such a species by which genome sequencing and assembly revealed two novel DTCs, discovered and elaborated upon in this work. Findings reported henceforth contribute to the breadth of knowledge underlying DTC catalysis. Such a comprehensive understanding can be applied to better predict and engineer these enzymes for industrial biosynthesis or genetic engineering for desired phenotypes
Quantifying resilience for distribution system customers with SALEDI
No full textThe impact of routine smaller outages on distribution system customers in terms of customer minutes interrupted can be tracked using conventional reliability indices. However, the customer minutes interrupted in large blackout events are extremely variable, and this makes it difficult to quantify the customer impact of these extreme events with resilience metrics. We solve this problem with the System Average Large Event Duration Index SALEDI that logarithmically transforms the customer minutes interrupted. We explain how this new resilience metric works, compare it with alternatives, quantify its statistical accuracy, and illustrate its practical use with standard outage data from five utilities.This is a preprint from Ahmad, Arslan, and Ian Dobson. "Quantifying resilience for distribution system customers with SALEDI." arXiv preprint arXiv:2602.07684 (2026).
doi: https://doi.org/10.48550/arXiv.2602.07684.Support from USA NSF grants 2153163 and 2429602, Argonne National Laboratory, Iowa State University Electric Power Research Center, and PSerc project S110 is gratefully acknowledged
Quantification of per- and polyfluoroalkyl substances in plasma and follicular fluid of patients undergoing in vitro fertilization in Iowa: A pilot study
No full textPer- and polyfluoroalkyl substances (PFAS) are a class of persistent synthetic chemicals that impart oil, water, and stain repellency to products. The use of PFAS has expanded in industrial and consumer products, including cleaning, personal care, and food items. The PFAS are absorbed via the gastrointestinal tract, respiratory system, and skin, after which they accumulate in the liver through interactions with fatty acid-binding proteins. They subsequently are distributed systemically via noncovalent binding to serum albumin, facilitating their presence in tissues, such as plasma and follicular fluid. Unlike many xenobiotics, the PFAS investigated in this study are not metabolized; rather, they are excreted largely unchanged. Legacy PFAS are long-chain compounds, including perfluorooctane sulfonic acid (PFOS) and perfluorooctanoic acid (PFOA), historically used in industrial and consumer products that have been phased out or restricted because of their persistence, bioaccumulation, and toxicity. Replacement PFAS were introduced as alternatives and include short-chain analogs and polyfluorinated ether derivatives. However, replacement PFAS are recognized increasingly for their environmental persistence and potential health risks. Growing evidence associates PFAS to adverse health effects related to endocrine and fertility function, which could affect reproductive outcomes in patients undergoing in vitro fertilization (IVF) (1). Our objective was to quantify PFAS in plasma and follicular fluid of patients undergoing infertility treatment in Iowa, United States.This article is published as Singh-Herren, Prapti, Samantha Good, Karen M. Summers, Joseph A. Charbonnet, Amy E. Sparks, and Aileen F. Keating. "Quantification of per-and polyfluoroalkyl substances in plasma and follicular fluid of patients undergoing in vitro fertilization in Iowa: A pilot study." F&S Reports 7, no. 1 (2026): 90-93. doi: https://doi.org/10.1016/j.xfre.2025.11.009.Supported by institutional departmental funding at the University of Iowa and from the Presidential Interdisciplinary Research Seed Grant from Iowa State University and by Iowa State University Data Driven Discovery Program Traineeship (National Science Foundation Grant No. 2152117; to S.G.)
Simulation and analysis of gradient coding protocols- Leveraging partial stragglers within gradient coding
No full textDistributed machine learning has become essential for training modern deep learning models on large-scale datasets. However, heterogeneous worker performance in cloud computing environments that are characterized by variable computation speeds, network congestion, and hardware variability create significant bottlenecks in synchronous training. Workers that are slower than expected, known as \emph{stragglers}, force the parameter server to wait indefinitely for gradient computations, severely degrading training efficiency.
Gradient coding, a coding-theoretic approach to distributed learning, addresses this problem by introducing redundancy in the assignment of data chunks to workers. By carefully encoding gradients using linear combinations, the parameter server can reconstruct the full gradient even when some workers fail to respond. However, existing gradient coding protocols treat workers as either fully operational or completely failed, ignoring the valuable partial work completed by slow-but-not-failed workers. Furthermore, communication-efficient variants that reduce transmitted gradient dimensionality suffer from severe numerical instability due to their reliance on polynomial interpolation over ill-conditioned Vandermonde matrices.
This thesis presents a practical implementation and empirical evaluation of a novel partial straggler gradient coding protocol that addresses these limitations. Our protocol introduces minimal interactive communication between workers and the parameter server, enabling dynamic coordination that exploits intermediate gradient computations from slow workers. Workers periodically report their progress, and when sufficient redundancy is achieved across the cluster, the parameter server broadcasts an encode-and-transmit signal. Each worker then independently computes encoding coefficients by solving a minimum -norm least-squares problem, ensuring numerical stability through the use of random Gaussian matrices rather than polynomial bases.
We implement the complete protocol using the Message Passing Interface (MPI) framework and deploy it on distributed clusters with 10 to 40 workers. Extensive experiments on the MNIST and CIFAR-10 datasets with neural networks containing 100,000 to 600,000 parameters demonstrate substantial performance improvements. Under heterogeneous conditions with simulated stragglers, our protocol achieves -- speedup over the original gradient coding approach across different cluster sizes and model architectures. The speedup factors increase superlinearly with cluster size, indicating that benefits compound at scale. Critically, our approach maintains excellent numerical stability, with gradient reconstruction errors bounded by machine precision (), and introduces only -- encoding overhead under ideal conditions.
We further demonstrate that our protocol integrates seamlessly with modern distributed training paradigms, including ZeRO memory optimization, pipeline parallelism, and hybrid 3D parallelism. The protocol operates at the data-parallel dimension, providing straggler tolerance while maintaining the memory efficiency of ZeRO and the computational benefits of model parallelism. Theoretical analysis confirms that exact gradient recovery preserves convergence guarantees for modern optimizers such as Adam and LAMB.
Our results validate that leveraging partial work from slow workers through numerically stable encoding provides significant practical benefits for distributed machine learning. The protocol is particularly well-suited for cloud deployments where straggler rates are high and cost efficiency is paramount, achieving training time reductions of -- and corresponding cost savings. This work establishes the viability of partial straggler exploitation for production distributed training systems and provides a foundation for future research in adaptive redundancy, heterogeneity-aware scheduling, and communication-efficient approximate gradient coding
Mechanics of incompatible asymmetric grain boundary migration
No full textGrain boundary (GB) migration governs microstructure evolution and can mediate plastic deformation through sliding or shear coupling. Numerous experimental and numerical studies have reported a wide range of behaviors associated with boundary migration, such as defect emission or mode switching. Notably, recent studies have reported directionally asymmetric migration rates under symmetric loading, attributing this behavior to intrinsically asymmetric mobility; however, a mechanistic mesoscale explanation for this behavior remains lacking. In this work, we introduce a constitutive flow rule for grain-boundary eigendeformation within a multiphase-field framework, in which interfacial shear evolves in response to its mechanically conjugate driving force through the phase field Allen-Cahn equations. The formulation systematically employs regularized grain boundary shear kinematics informed by crystallography, and enables elastic compatibility to modulate boundary motion. Migration thresholds, residual back-stress, and apparent directional asymmetry appear naturally as emergent mechanical behavior. Simulations of symmetric and asymmetric tilt grain boundaries under mechanical, synthetic, and curvature-driven loading reveal persistent defect-like residuals following incompatible migration, transitions from planar motion to lamination at large inclinations, and even "ratcheting" behavior. These results provide a mechanically transparent explanation for behaviors such as effective mobility asymmetry and establish elastic compatibility as a constitutive mechanism in mesoscale models of boundary-mediated plasticity.This is a preprint from Runnels, Brandon. "Mechanics of incompatible asymmetric grain boundary migration." arXiv preprint arXiv:2602.02387 (2026). doi: https://doi.org/10.48550/arXiv.2602.02387.This work was funded by the National Science Foundation through the CAREER program, grant # 2341922
Improved antioxidant and anti-melanogenic effects of ovalbumin derived from egg white after enzymatic hydrolysis
No full textConsumption of oxidized foods or exposure of skin to ultraviolet (UV) can produce reactive oxygen species (ROS) that increase oxidative stresses that can lead to skin aging and melanogenesis. Hydrolyzing proteins like ovalbumin (OA) can significantly increase their bioactivities. In this study, OA was hydrolyzed using Alcalase 2.4 L FG, Multifect PR 14 L, and Papain T100 MG, and the antioxidant and melanogenesis inhibitory activities of the hydrolysates were determined using B16F10 melanoma cells. All OA hydrolysates (OAH) showed higher antioxidant activities than the natural OA. The OAH prepared with Multifect PR 14 L (OAMF) exhibited the highest antioxidant activities and significantly inhibited melanin production and tyrosinase activity in the stimulated B16F10 cells using α-melanocyte-stimulating hormone. The DOPA staining confirmed the melanogenesis inhibitory (melanin formation and dendrite development) effects of OAMF. The OAMF (100, 200, and 400 µg/mL) treatment downregulated the mRNA expression of microphthalmia-associated transcription factor (MITF) and melanogenic enzymes [tyrosinase, tyrosinase-related protein 1 (TRP-1), and tyrosinase-related protein 2 (TRP-2)] dose-dependently. Therefore, OAMF can help reduce oxidative stress in humans when consumed and can be used as an anti-aging and skin-lightening agent in the cosmetic industry.This article is published as Cho, H.Y., Lee, JE., Lee, J.H. et al. Improved antioxidant and anti-melanogenic effects of ovalbumin derived from egg white after enzymatic hydrolysis. Food Sci Anim Resour 46, 15 (2026). https://doi.org/10.1007/s44463-025-00024-
Mitigation of Ammonia and Hydrogen Sulfide Emissions from Biochar-Treated Swine Manure After Application to Cropland Soil
No full textAir quality management is essential for sustainable livestock production. Manure generates odorous and greenhouse gas emissions during storage and cropland application. Ammonia (NH3) emissions during manure application to soil represent a loss of nitrogen and decrease the value of manure. Biochar (BC) can mitigate emissions during manure storage. In this research, we hypothesized that BC-treated manure generates less emissions after post-storage application to cropland soil. This study tested the effect of BC thickness (dose) and manure treatment timing on gaseous emissions with controlled lab-scale experiments. Both NH3 and hydrogen sulfide (H2S) emissions from cropland soil amended with BC treated manure were reduced. Overall, average percent reductions in the first 6 h / 24 h was 89.9% / 89.7% and 59.6% / 60.8% for NH3 and H2S emissions, respectively, when manure was treated with BC immediately prior to application to the soil. When manure was treated with BC during storage, emissions of NH3 and H2S were reduced 20.7% / 21.8% and 12.2% / 13.5%, respectively. The percent reductions in NH3 and H2S for 13- and 6.5-mm BC application thicknesses were not significantly different, but the 2.5-mm application thickness treatment was less effective at reducing gas emissions. The effects of BC application on methane), carbon dioxide, and odor were inconclusive. Observational data indicated a possible mitigation trend in odor, especially for the BC treatment added immediately prior to application to soil, but further trials are needed. Scaling up the technoeconomic analysis showed that less than 6 m3 of BC would be needed to treat more than 900 m2 surface of stored manure in a typical 1,200-head swine barn with a uniform 6.5 mm thick BC layer. Subsequent cropland application of such BC-treated manure would result in 0.1 m3 ha-1 BC addition to soil in a typical corn-soybean crop rotation system. This research shows that BC has the potential to mitigate gaseous emissions while promoting nutrient cycling and the sustainability of livestock waste as a fertilizer.This article is published as O'Brien, Samuel C., Jacek A. Koziel, Brett C. Ramirez, and Anna Ortiz. "Mitigation of Ammonia and Hydrogen Sulfide Emissions from Biochar-Treated Swine Manure After Application to Cropland Soil." Applied Engineering in Agriculture (2026): 0. doi: https://doi.org/10.13031/aea.16394.This project (2022-2 Sustainability of Manure) was funded by the Leopold Center for Sustainable Agriculture (Ames, Iowa, USA). JK (while primarily affiliated with Iowa State University) and BR’s participation was partially supported by the Iowa Agriculture and Home Economics Experiment Station, Ames, Iowa. Project No. IOW05556
Influence of resuspension on the fate and transport of antibiotic-resistant genes in rivers
No full textSediment resuspension can elevate the transport of antibiotic resistance genes in rivers. We identify the conditions under which resuspension is important and quantify its influence on transported mass over one reach travel time by comparing simulations with and without resuspension under the same hydraulics and upstream concentrations. The model represents intracellular and extracellular DNA in water and a uniform, active bed with exchange, and it uses first order fate. Resuspension is isolated by ignoring settling and longitudinal dispersion. Resuspension is important when the bed stores large antibiotic resistant gene (ARG) concentrations, shear stress exceeds the critical erosion threshold, and sediments are fine and cohesive, because cohesion raises the threshold but events that exceed it release ARG rich fines that remain in the water column. An application to six rivers showed that transport increased by about 65% and 58% in the strongest systems, by about 28% and 22% in intermediate systems, and by about 8% and 1% in weak systems. Regime maps built from a resuspension number that compares bed to water exchange to advection and from the ARG mass per bed area in the active layer reproduced these outcomes. Uncertainty analysis showed that 95% intervals, from first order propagation of independent input ranges, preserved the river order and identified marginal cases where plausible resuspension rates and bed concentrations yield little or uncertain impact. Sensitivity and variance results ranked the resuspension rate and bed concentrations as the largest positive controls, while water concentrations and mean velocity reduced the effect of resuspension through dilution and shorter exposure; biological rate constants were positive and smaller on this time scale. We conclude that this approach tells when resuspension matters for river ARG transport and which bulk measurements to prioritize; when the screening indicates little influence, a model without resuspension suffices
LSTM sequence-to-sequence based System Identification and feedforward-feedback control of piezoelectric actuators
No full textPiezo-electric actuators (PEA) have wide applications in nano/micro-positioning systems, such as
Atomic Force Microscopy (AFM), micro-forming/machining, adaptive optics, and more. The popularity of
PEA applications come from its advantages, including high resolution, high stiffness, and fast response,
which align with the performance demands of nano/micro-positioning systems. However, PEA also
presents unwanted properties, such as hysteresis, creep, and drift, which introduce nonlinearity and
uncertainty, posing challenges to control.
To address these issues, we propose a novel sequence-to-sequence LSTM-based algorithm to achieve
system identification and 2 Degree-of-Freedom (2DOF) feedforward-feedback inversion control for a
commercial PEA.
In the first study (presented in Ch. 2), an LSTM sequence-to-sequence (LSTMseq2seq or LSTMs2s)
algorithm was used to model the dynamics of the PEA. The training set for the input signals was generated
using both sinusoidal and triangular waveforms with different frequency-amplitude (f-A) combinations. To
ensure homogeneous coverage of the working range of frequency and amplitude, the f-A combinations
were randomized within the range and selected using the k-means clustering method. Since LSTMseq2seq
enables parallel training with faster speeds, the training set can be significantly larger than that of
conventional RNNs, containing thousands of signals with different frequencies, amplitudes, and
waveforms. In addition to the superior training performance, LSTMseq2seq also offers the capability to
learn the pattern of system behavior under a specific frequency and amplitude, considering the entire signal
as one sample for system input and output. To evaluate the accuracy of the model, the predicted output was
compared to the actual system outputs. The results demonstrated the superiority of the proposed modeling
approach over RNNs in improving system identification accuracy, particularly with a broader working
frequency bandwidth.
In the second study (presented in Ch. 3), the LSTMs2s algorithm developed as an inversion model was
applied to a 2DOF feedforward-feedback controller for the PEA. The inverse LSTM sequence-to-sequence
(invLSTMs2s) model and a linear MPC were deployed as the feedforward and feedback controllers,
respectively. In contrast to the first study, the LSTMs2s were trained with an inverted training set, which
flipped the input and output signals of the commercial PEA. The resulting invLSTMs2s model could
predict the corresponding input for the reference output. By cascading the accurate invLSTMs2s model to
the commercial PEA, the nonlinearity of the system behavior was compensated, significantly reducing the
burden on the feedback controller. To achieve better accuracy across a wider bandwidth, a
frequency-dependent gain was added to both the feedforward and feedback controllers. With this design,
the MPC was focused on controlling the PEA in the low-frequency range, where the system is more linear,
while the invLSTMs2s focused on high-frequency control, effectively compensating for system
nonlinearity. After testing the trajectory tracking performance of the proposed controller with multiple
signal forms, the results were compared to PID and MPC controllers, demonstrating improvements in
tracking accuracy, working bandwidth, and robustness to different waveforms. Furthermore, we applied the
proposed controller to control an AFM nano-positioning system, evaluating its potential for practical
applications. The performance in this application aligned with the results of the major tests.
Altogether, this work presents a new LSTM sequence-to-sequence model for system identification and
a corresponding inversion model combined with a 2DOF feedforward-feedback controller for controlling
the PEA. Both models demonstrated improvements and superiority compared to existing popular methods