Probabilistic Model Research Articles

Probabilistic damage evaluation of isolated steel box-girder bridge excited by near-field earthquakes

Modern civilization regards bridges as lifelines, and their failure due to significant earthquakes in the past has raised concerns about their seismic vulnerability. The effect of near-fault earthquakes on the bridge vulnerability is a significant problem that has not been well explored. The present work utilizes fragility analysis to assess the probabilistic seismic risk evaluation of a continuous isolated steel box girder bridge exposed to near-field ground motions (NFGMs). Probabilistic seismic demand models (PSDMs) of the bridge are explored to reliably associate key demand and intensity measures to capture the seismic performance of the system. This study examines a steel box girder bridge in Guwahati, India, isolated with lead rubber (LRB) bearings and a friction pendulum system (FPS). To achieve a comparable seismic response, each bearing system is designed for a similar isolation period. The PGV/PGA ratio is a useful metric for classifying ground motion suites and estimating their frequency content by classifying the input excitations into low- and high-frequency motions. To perform probabilistic seismic risk assessments, two ensembles of forward directivity records with PGV: PGA >160 cm/s/g (high frequency) and PGV: PGA <160 cm/s/g (low frequency) are considered. Moreover, one ensembles of near-field incorporating the fling effect and far-field are also utilized. The failure probability under the seismic hazards for isolated bridges is assessed utilizing different damage measures (DM) such as maximum displacement ductility of pier (DDP), displacement of deck (MDD), drift ratio of pier (PDR), and rotational ductility (RDP). The near-field earthquake damage probability is significantly impacted by the PGV/PGA ratio, as a high PGV/PGA ratio results in a greater probability of damage than a low PGV/PGA ratio. In conclusion, guidelines are provided that will help the design engineers for designing bridges with LRB and FPS isolators for seismic isolation in the event of varying ground motion characteristics.

PostFocus: Automated Selective Post-acquisition High-throughput Focus Restoration using Diffusion model for Label-free Time-lapse Microscopy.

High-throughput time-lapse imaging is a fundamental tool for efficient living cell profiling at single-cell resolution. Label-free phase-contrast video microscopy enables non-invasive, non-toxic, and long-term imaging. The tradeoff between speed and throughput, however, implies that despite the state-of-the-art autofocusing algorithms, out-of-focus (OOF) cells are unavoidable due to the migratory nature of immune cells (velocities >10 μm/min). Here, we propose PostFocus to: (1) identifying OOF images within time-lapse sequences with a classifier, and (2) deploy a de-noising diffusion probabilistic model (DDPM) to yield reliable in-focus images. DDPM outperformed deep discriminative models with a superior performance on the whole image and around cell boundaries. In addition, PostFocus improves image analysis (cell and contact detection) accuracy and the yield of usable videos. Open-source code and sample data provided at: https://github.com/kwu14victor/PostFocus. Supplementary data are available at Bioinformatics online.

Application of Generative Artificial Intelligence in Predicting Membrane Partitioning of Drugs: Combining Denoising Diffusion Probabilistic Models and MD Simulations Reduces the Computational Cost to One-Third.

The optimal interaction of drugs with plasma membranes and membranes of subcellular organelles is a prerequisite for desirable pharmacology. Importantly, for drugs targeting the transmembrane lipid-facing sites of integral membrane proteins, the relative affinity of a drug to the bilayer lipids compared to the surrounding aqueous phase affects the partitioning, access, and binding of the drug to the target site. Molecular dynamics (MD) simulations, including enhanced sampling techniques such as steered MD, umbrella sampling (US), and metadynamics, offer valuable insights into the interactions of drugs with the membrane lipids and water in atomistic detail. However, these methods are computationally prohibitive for the high-throughput screening of drug candidates. This study shows that applying denoising diffusion probabilistic models (DDPMs), a generative AI method, to US simulation data reduces the computational cost significantly. Specifically, the models used only partial (one-third) data from the US simulations and reproduced the complete potential of mean force (PMF) profiles for three FDA-approved drugs (β2-adrenergic agonists) and ∼20 biologically relevant chemicals with known experimentally characterized bilayer locations. Intriguingly, the model can predict the solvation-free energies for partitioning and crossing the bilayer, preferred bilayer locations (low-energy well), and orientations of the ligands with high accuracy. The results indicate that DDPMs can be used to characterize the complete membrane partitioning profile of drug molecules using fewer umbrella sampling simulations at select positions along the bilayer normal (z-axis), irrespective of their amphiphilic-lipophilic-cephalophilic characteristics.

Quantifying the effect of pulse-like ground motions on the seismic design of SMA restrained isolation bridges using probabilistic repair cost method

The shape memory alloy (SMA) restrainer serves as an effective but expensive bridge restraining device. However, the impact of pulse effect on SMA design approach under pulse-like ground motions (PLGM) has not been fully quantified in previous studies, which hinders its application in near-fault regions. Therefore, based on the risk probability assessment method throughout the entire life cycle, by accounting for the comprehensive repair cost of each component damage, this paper introduces a parameter design method for SMA restrainer of near-fault bridges that considers structural parameters, pulse parameters, and economic indicators. Firstly, the repair cost ratio (RCR) of bridge system, which means the expenses for repairs expressed as a proportion of bridge replacement costs, was regarded as the life-time optimization goal and overall performance indicator. Secondly, by accounting for near-fault effects, the relationship between RCR and SMA design parameters was established by convolution algorithm of probabilistic seismic hazard analysis, demand analysis and capacity analysis. A novel probabilistic seismic demand model was utilized to quickly determine the RCR of the bridge system under PLGM. Finally, the influence of pulse period on the rational design parameter of SMA restrainer was comprehensively investigated by RCR-based method. A seismic isolation arch bridge was selected as the illustrated case in this article. The results indicate that the rational design parameters of SMA exhibit a pattern of initially rising, then falling as the pulse period increases, reaching the peak value when the pulse period approaches bridge fundamental period. Moreover, the price parameter of SMA significantly affect the optimal design parameters, and the effective range is also recommended.

DiffuseRT: predicting likely anatomical deformations of patients undergoing radiotherapy

Objective. Predicting potential deformations of patients can improve radiotherapy treatment planning. Here, we introduce new deep-learning models that predict likely anatomical changes during radiotherapy for head and neck cancer patients. Approach. Denoising diffusion probabilistic models (DDPMs) were developed to generate fraction-specific anatomical changes based on a reference cone-beam CT (CBCT), the fraction number and the dose distribution delivered. Three distinct DDPMs were developed: (1) the image model was trained to directly generate likely future CBCTs, (2) the deformable vector field (DVF) model was trained to generate DVFs that deform a reference CBCT and (3) the hybrid model was trained similarly to the DVF model, but without relying on an external deformable registration algorithm. The models were trained on 9 patients with longitudinal CBCT images (224 CBCTs) and evaluated on 5 patients (152 CBCTs). Results. The generated images mainly exhibited random positioning shifts and small anatomical changes for early fractions. For later fractions, all models predicted weight losses in accordance with the training data. The distributions of volume and position changes of the body, esophagus, and parotids generated with the image and hybrid models were more similar to the ground truth distribution than the DVF model, evident from the lower Wasserstein distance achieved with the image (0.33) and hybrid model (0.30) compared to the DVF model (0.36). Generating several images for the same fraction did not yield the expected variability since the ground truth anatomical changes were only in 76% of the fractions within the 95% bounds predicted with the best model. Using the generated images for robust optimization of simplified proton therapy plans improved the worst-case clinical target volume V95 with 7% compared to optimizing with 3 mm set-up robustness while maintaining a similar integral dose. Significance. The newly developed DDPMs generate distributions similar to the real anatomical changes and have the potential to be used for robust anatomical optimization.

Gaussian process regression driven rapid life-cycle based seismic fragility and risk assessment of laminated rubber bearings supported highway bridges subjected to multiple uncertainty sources

Structural attributes, corrosion of steel rebars in reinforced concrete (RC) columns, thermal oxidative aging of inherent rubber in laminated rubber bearings (LRBs), replacement of LRBs, and the time-dependent seismic hazard level at the bridge site are well recognized as the main factors that affect the accurate assessment of the seismic performance of RC bridges implemented with LRBs. However, the probabilistic seismic demand model (PSDM) and probabilistic structural capacity model (PSCM) of deteriorated bridge components may not exhibit homoscedastic log-linearity. And the expansion of seismic fragility surfaces and seismic risk curves in the temporal dimension requires a considerable computational cost on the nonlinear analysis, which regretfully has not been well addressed in current studies. To fill in this gap, this study proposes a life-cycle based seismic fragility and risk assessment framework for LRBs supported RC bridges considering deterioration mechanisms of bridge components during their life cycles demonstrated by a three-span LRBs supported highway bridge considering multiple uncertainties such as the replacement interval for LRBs. In this framework, Gaussian process regression (GPR) is employed to effectively construct the PSDM and PSCM of deteriorated bridge components, and then the time-dependent seismic fragility surfaces of bridge components and bridge system are established. Life-cycle based seismic risk analysis is conducted to quantify the coupled time-dependent effects of deterioration and site seismic hazard on a newly built bridge during its design service life and a deteriorated bridge within its remaining service life, respectively. Results indicate that GPR can successfully capture the nonlinearity and heteroscedasticity characteristics of PSDM and PSCM, leading to a significant reduced number of nonlinear analysis cases to generate the time-dependent seismic fragility surfaces and seismic risk curves. Consequently, the proposed framework using GPR can significantly improve the accuracy and efficiency of life-cycle based seismic performance assessment when uncertainties of structural attributes, deterioration progress and seismic hazard are taken into consideration.

A novel method addressing NGS-based mappability bias for sensitive detection of DNA alterations.

A turning point in cancer research is the introduction of massively parallel sequencing technology which greatly reduced the cost and time for genome sequencing. This enhanced the scope for detecting and analyzing the role of structural alterations in cancer. However, certain bias exists in NGS-based approaches, which badly affects the CNV identification process. Moreover, DNA repeats existing in CNV regions need special attention as they will degrade the performance of majority of the existing CNV detection tools, even after applying generalized bias correction method. This motivated this work, where a novel method has been designed to address the issue of DNA repeats and thereby mappability bias existing in regions of CNV.The method consists of three phases, where the first phase computes the alignment information of uniquely mapped DNA reads, considering the base quality and base mismatch parameters at nucleotide level precision. The second and the third phase use a novel approach to allocate the non-uniquely mapped reads to an optimal region of the DNA repeats based on a probabilistic membership model. The proposed method is capable of identifying CNVs present in coding, as well as non-coding region of the DNA, and is also capable of detecting CNVs existing in DNA repeat regions. The methodology achieves a sensitivity greater than [Formula: see text] during the performed simulations, and on real data, the detected variants are validated with the database of genomic variants, where the percentage overlap is also greater than 95%, and has achieved much better breakpoint prediction, as compared with other popular bias correction CNV detection methods.

Joint Modeling of Wind Speed and Power via a Nonparametric Approach

Power output from wind turbines is influenced by wind speed, but the traditional theoretical power curve approach introduces uncertainty into wind energy forecasting models. This is because it assumes a consistent power output for a given wind speed. To address this issue, a new nonparametric method has been proposed. It uses K-means clustering to estimate wind speed intervals, applies kernel density estimation (KDE) to establish the probability density function (PDF) for each interval and employs Monte Carlo simulation to predict power output based on the PDF. The method was tested using data from the MERRA-2 database, covering five wind farms in Brazil. The results showed that the new method outperformed the conventional estimation technique, improving estimates by an average of 47 to 49%. This study contributes by (i) proposing a new nonparametric method for modeling the relationship between wind speed and power; (ii) emphasizing the superiority of probabilistic modeling in capturing the natural variability in wind generation; (iii) demonstrating the benefits of temporally segregating data; (iv) highlighting how different wind farms within the same region can have distinct generation profiles due to environmental and technical factors; and (v) underscoring the significance and reliability of the data provided by the MERRA-2 database.

Memristive Crossbar Array-Based Probabilistic Graph Modeling.

Modern graph datasets with structural complexity and uncertainties due to incomplete information or data variability require advanced modeling techniques beyond conventional graph models. This study introduces a memristive crossbar array (CBA)-based probabilistic graph model (C-PGM) utilizing Cu0.3Te0.7/HfO2/Pt memristors, which exhibit probabilistic switching, self-rectifying, and memory characteristics. C-PGM addresses the complexities and uncertainties inherent in structural graph data across various domains, leveraging the probabilistic nature of memristors. C-PGM relies on the device-to-device variation across multiple memristive CBAs, overcoming the limitations of previous approaches that rely on sequential operations, which are slower and have a reliability concern due to repeated switching. This new approach enables the fast processing and massive implementation of probabilistic units at the expense of chip area. In this study, the hardware-based C-PGM feasibly expresses small-scale probabilistic graphs and shows minimal error in aggregate probability calculations. The probability calculation capabilities of C-PGM are applied to steady-state estimation and the PageRank algorithm, which is implemented on a simulated large-scale C-PGM. The C-PGM-based steady-state estimation and PageRank algorithm demonstrate comparable accuracy to conventional methods while significantly reducing computational costs.

Influence of Repeated Earthquake events on Structural Response of Multi‐storeyed RC buildings

Over the past decade, reinforced concrete (RC) multi-storeyed buildings have become more prevalent in urban environments. In addition, it has been well established in literature, that collapse of several RC buildings and resulting causalities, thereof are mainly attributed to the repeated nature of occurrence of earthquakes, more often leading to collapse. However, most existing design codes consider only a scenario earthquake characterized by the response spectrum specified by local codes of practice. This consideration do not address the repetitive nature of earthquakes, leading to inadequate estimation of design earthquake force. Hence, the present investigation focusses on proposing a probabilistic model to logically address the number of earthquake sequences to be considered at a chosen location. In addition, to ascertain attainment of collapse state of code designed 3D RC buildings, under a pattern of earthquake sequences (with simultaneous bi-directional ground motion data), a simple and computationally less intensive NLTHA (Nonlinear Time History Analysis) approach has been considered. Moreover, this approach captures the effect of sequential earthquakes and provides a quick rational decision on the structural behaviour, necessary for further analysis. Hence, eliminating the need to resort always for complex IDA (Incremental Dynamic Analysis) approaches to illustrate the structural behaviour. Moreover, the structural behaviour of RC buildings are represented in terms of base shear experienced and their corresponding response parameters (Horizontal displacement, Inter-story drift ratio (IDR), etc.,). The outcome of the results clearly contemplates the accumulation of damages in terms of response parameters, emphasizing the need for consideration of repeated earthquake sequences in seismic analysis.

Insights into canine rabies vaccination Disparities in Sierra Leone: A cross-sectional household study.

Annually, Sierra Leone records an estimated 301 human fatalities due to rabies. Canine vaccination is crucial for rabies prevention and control efforts. However, considerable variability exists in vaccination rates. Reasons for this variation remain unclear. We conducted a cross-sectional study across 2,558 dog-owning households (HHs) to provide insights into factors influencing canine vaccination for targeted prevention and control towards elimination by 2030. First, we described dog ownership practices, then built a probabilistic model to understand factors associated with dog vaccination, and finally used a spatial scan statistic to identify spatial clusters where vaccination rates were low. Our results indicated that only 14% (358/2,558) of participating HHs had fully vaccinated their dogs against rabies. The probability of dog vaccination increased when comparing civil servants to private workers/artisans, with an Odds Ratio(OR) of 1.14 (95% credible interval (Crl) of 0.82-1.56), residing in locations with a veterinary establishment vs. none (OR = 6.43, 95% Crl (4.97-8.35), providing care to dogs vs. allowing dogs to roam freely (OR = 2.38, 95% Crl(1.80-3.17) and owning a single dog vs multiple dogs (OR = 1.20, 95 Crl (0.92-1.56). Conversely, there was a decrease in the estimated probability of vaccination when comparing dog owners located in rural vs. urban areas (OR = 0.58, CrI 95% (0.43-0.78). Latent understanding, a measure of overall understanding of rabies virus, which we estimated using participant education levels and responses to questions about rabies epidemiology, was also an important predictor of vaccination probability (OR = 1.44, 95% Crl (1.04, 2.07). The spatial analysis identified high-risk clusters for low vaccination in the cities of Mayamba, with a radius of 40 km, a relative risk (RR) of 1.10, and Bo, with a radius of 19.9 km with RR of 1.11. These data do not support Sierra Leone reaching the 2030 goal of human rabies elimination caused by dogs. Our study highlights a critical need for public outreach and education, improved vaccination rates, increased accessibility to veterinary services, and targeted interventions in Bo and Mayamba to support rabies prevention and control efforts.

Complexity bound of a Levenberg–Marquardt algorithm based on probabilistic Jacobian models

Complexity bound of a Levenberg–Marquardt algorithm based on probabilistic Jacobian models

Temporal changes in cause of death among adolescents and adults in six countries in eastern and southern Africa in 1995–2019: a multi-country surveillance study of verbal autopsy data

The absence of high-quality comprehensive civil registration and vital statistics systems across many settings in Africa has led to little empirical data on causes of death in the region. We aimed to use verbal autopsy data to provide comparative, population-based estimates of cause-specific mortality among adolescents and adults in eastern and southern Africa. In this surveillance study, we harmonised verbal autopsy and residency data from nine health and demographic surveillance system (HDSS) sites in Kenya, Malawi, Tanzania, South Africa, Uganda, and Zimbabwe, each with variable coverage from Jan 1, 1995, to Dec 31, 2019. We included all deaths to adolescents and adults aged 12 or over that were residents of the study sites and had a verbal autopsy conducted. InSilicoVA, a probabilistic model, was used to assign cause of death on the basis of the signs and symptoms reported in the verbal autopsy. Levels and trends in all-cause and cause-specific mortality rates and cause-specific mortality fractions were calculated, stratified by HDSS site, sex, age, and calendar periods. 52 484 deaths and 5 157 802 person-years were reported among 1 071 913 individuals across the nine sites during the study period. 47 961 (91·4%) deaths had a verbal autopsy, of which 46 570 (97·1%) were assigned a cause of death. All-cause mortality generally decreased across the HDSS sites during this period, particularly for adults aged 20-59 years. In many of the HDSS sites, these decreases were driven by reductions in HIV and tuberculosis-related deaths. In 2010-14, the top causes of death were: road traffic accidents, HIV or tuberculosis, and meningitis or sepsis in adolescents (12-19 years); HIV or tuberculosis in adults aged 20-59 years; and neoplasms and cardiovascular disease in adults aged 60 years and older. There was greater between-HDSS and between-sex variation in causes of death for adolescents compared with adults. This study shows progress in reducing mortality across eastern and southern Africa but also highlights age, sex, within-HDSS, and between-HDSS differences in causes of adolescent and adult deaths. These findings highlight the importance of detailed local data to inform health needs to ensure continued improvements in survival. National Institute of Child Health and Human Development of the US National Institutes of Health.

Comparing components for seismic risk modelling using data from the 2019 Le Teil (France) earthquake

Abstract. Probabilistic seismic hazard and risk models are essential to improving our awareness of seismic risk, to its management, and to increasing our resilience against earthquake disasters. These models consist of a series of components, which may be evaluated and validated individually, although evaluating and validating these types of models as a whole is challenging due to the lack of recognized procedures. Estimations made with other models, as well as observations of damage from past earthquakes, lend themselves to evaluating the components used to estimate the severity of damage to buildings. Here, we are using a dataset based on emergency post-seismic assessments made after the Le Teil 2019 earthquake, third-party estimations of macroseismic intensity for this seismic event, shake maps, and scenario damage calculations to compare estimations under different modelling assumptions. First we select a rupture model using estimations of ground motion intensity measures and macroseismic intensity. Subsequently, we use scenario damage calculations based on different exposure models, including the aggregated exposure model in the 2020 European Seismic Risk Model (ESRM20), as well as different site models. Moreover, a building-by-building exposure model is used in scenario calculations, which individually models the buildings in the dataset. Lastly, we compare the results of a semi-empirical approach to the estimations made with the scenario calculations. The post-seismic assessments are converted to EMS-98 (Grünthal, 1998) damage grades and then used to estimate the damage for the entirety of the building stock in Le Teil. In general, the scenario calculations estimate lower probabilities for damage grades 3–4 than the estimations made using the emergency post-seismic assessments. An exposure and fragility model assembled herein leads to probabilities for damage grades 3–5 with small differences from the probabilities based on the ESRM20 exposure and fragility model, while the semi-empirical approach leads to lower probabilities. The comparisons in this paper also help us learn lessons on how to improve future testing. An improvement would be the use of damage observations collected directly on the EMS-98 scale or on the damage scale in ESRM20. Advances in testing may also be made by employing methods that inform us about the damage at the scale of a city, such as remote sensing or data-driven learning methods fed by a large number of low-cost seismological instruments spread over the building stock.

Modeling pediatric antibiotic use in an area of declining malaria prevalence

In malaria-endemic areas of Sub-Saharan Africa, overlap of clinical symptoms between malarial and non-malarial febrile illnesses can lead to empiric use of antibiotics among children. Our study aimed to illustrate the potential impact of decreasing malaria prevalence from malaria control efforts on antibiotic use. We constructed a probabilistic decision tree model representing antibiotic prescription in febrile children < 5 years. This model was used to predict change in absolute antibiotic use compared to baseline under levels of decreasing malaria prevalence. Model parameters were based on data from a hospital study in Ghana and validated via literature review. The baseline prevalence of malaria diagnoses was 52% among all hospitalized children. For our main results, we reported outcomes for a scenario representing a 50% decrease in malaria prevalence. Compared to baseline, absolute antibiotic prescription decreased from a baseline of 639 doses (95% CI 574–694) to 575 (95% CI 502–638). This reflected a 10% (95% CI 7%–13%) decrease in absolute antibiotic use. Our findings demonstrate that effective malaria control can reduce pediatric antibiotic use. However, until substantial progress is made in developing accurate diagnostics for non-malarial febrile illnesses, further reductions in antibiotic use will remain a challenge.

A novel probabilistic model: Simulation and modeling the time duration in musical education and engineering

Statistical analysis of the time-to-event scenarios commonly involves the use of the exponential and Rayleigh distributions. However, the exponential distribution is constrained by its constant failure rate, while the Rayleigh distribution exhibits a growing (or increasing) failure rate. The linear failure rate distribution provides a broader perspective by encompassing both of these distributions, including the possibility of a non-increasing hazard function. In this paper, a new distribution is introduced, which could be considered as one of the major modifications to the linear failure rate distribution, known as the weighted linear failure rate distribution. The model proposed illustrates both monotonic and a known non-monotonic failure rate. This paper extensively explores the characteristics of the proposed distribution. It presents the maximum likelihood estimates for the unknown parameters and includes a simulation study. The analysis focuses on three real data sets obtained from the fields of musical education and engineering. By employing three decisive tools, it is observed that the current distribution provides a superior fit in comparison to several other widely recognized distributions.

Probabilistic assessment of postfire debris-flow inundation in response to forecast rainfall

Abstract. Communities downstream of burned steep lands face increases in debris-flow hazards due to fire effects on soil and vegetation. Rapid postfire hazard assessments have traditionally focused on quantifying spatial variations in debris-flow likelihood and volume in response to design rainstorms. However, a methodology that provides estimates of debris-flow inundation downstream of burned areas based on forecast rainfall would provide decision-makers with information that directly addresses the potential for downstream impacts. We introduce a framework that integrates a 24 h lead-time ensemble precipitation forecast with debris-flow likelihood, volume, and runout models to produce probabilistic maps of debris-flow inundation. We applied this framework to simulate debris-flow inundation associated with the 9 January 2018 debris-flow event in Montecito, California, USA. When the observed debris-flow volumes were used to drive the probabilistic forecast model, analysis of the simulated inundation probabilities demonstrates that the model is both reliable and sharp. In the fully predictive model, however, in which debris-flow likelihood and volume were computed from the atmospheric model ensemble's predictions of peak 15 min rainfall intensity, I15, the model generally under-forecasted the inundation area. The observed peak I15 lies in the upper tail of the atmospheric model ensemble spread; thus a large fraction of ensemble members forecast lower I15 than observed. Using these I15 values as input to the inundation model resulted in lower-than-observed flow volumes which translated into under-forecasting of the inundation area. Even so, approximately 94 % of the observed inundated area was forecast to have an inundation probability greater than 1 %, demonstrating that the observed extent of inundation was generally captured within the range of outcomes predicted by the model. Sensitivity analyses indicate that debris-flow volume and two parameters associated with debris-flow mobility exert significant influence on inundation predictions, but reducing uncertainty in postfire debris-flow volume predictions will have the largest impact on reducing inundation outcome uncertainty. This study represents a first step toward a near-real-time hazard assessment product that includes probabilistic estimates of debris-flow inundation and provides guidance for future improvements to this and similar model frameworks by identifying key sources of uncertainty.

Physics-based natural gradient boosting probabilistic prediction model for seismic failure modes of reinforced concrete columns

A novel physics-based natural gradient boosting (PNGB) probabilistic prediction model for flexure failure (FF), flexure-shear failure (FSF) and shear failure (SF) of reinforced concrete (RC) columns under strong earthquakes has been developed to address the limitations of empirical and machine learning (ML) models, which often fail to learn physical laws, insufficiently consider uncertainties, and exhibit poor generalization performance. Firstly, an improved natural gradient boosting (NGB) probabilistic model of seismic failure modes was created by using natural gradient boosting and gradient boosting regressor algorithms. Then a new hyperparameter optimization method for the improved NGB probabilistic model was proposed based on the ML algorithms and the physical laws. Finally, the predictive performance and engineering practicability of the PNGB probabilistic model were verified compared to empirical and ML models. According to the analysis results, the PNGB model can effectively improve the predictive recall of FF, FSF and SF by about 19–65 %, 20–67 % and 20–56 % respectively compared with empirical models, and by about 1–7 %, 7–43 % and 6–32 % respectively compared with ML models. Moreover, the PNGB model exhibits satisfactory generalization performance and minimal dispersion, which can be used to calibrate the ML models based on the confidence intervals. Furthermore, the result predicted by the PNGB model satisfies the physical laws relating to the seismic failure modes and design parameters, which can further represent the competitive relationships of various seismic failure modes.

Assessing the Accuracy and Efficiency of Free Energy Differences Obtained from Reweighted Flow-Based Probabilistic Generative Models.

Computing free energy differences between metastable states characterized by nonoverlapping Boltzmann distributions is often a computationally intensive endeavor, usually requiring chains of intermediate states to connect them. Targeted free energy perturbation (TFEP) can significantly lower the computational cost of FEP calculations by choosing a set of invertible maps used to directly connect the distributions of interest, achieving the necessary statistically significant overlaps without sampling any intermediate states. Probabilistic generative models (PGMs) based on normalizing flow architectures can make it much easier via machine learning to train invertible maps needed for TFEP. However, the accuracy and applicability of approaches based on empirically learned maps depend crucially on the choice of reweighting method adopted to estimate the free energy differences. In this work, we assess the accuracy, rate of convergence, and data efficiency of different free energy estimators, including exponential averaging, Bennett acceptance ratio (BAR), and multistate Bennett acceptance ratio (MBAR), in reweighting PGMs trained by maximum likelihood on limited amounts of molecular dynamics data sampled only from end-states of interest. We carry out the comparisons on a set of simple but representative case studies, including conformational ensembles of alanine dipeptide and ibuprofen. Our results indicate that BAR and MBAR are both data efficient and robust, even in the presence of significant model overfitting in the generation of invertible maps. This analysis can serve as a stepping stone for the deployment of efficient and quantitatively accurate ML-based free energy calculation methods in complex systems.

Assessing the Accuracy and Efficiency of Free Energy Differences Obtained from Reweighted Flow-Based Probabilistic Generative Models.

Computing free energy differences between metastable states characterized by nonoverlapping Boltzmann distributions is often a computationally intensive endeavor, usually requiring chains of intermediate states to connect them. Targeted free energy perturbation (TFEP) can significantly lower the computational cost of FEP calculations by choosing a set of invertible maps used to directly connect the distributions of interest, achieving the necessary statistically significant overlaps without sampling any intermediate states. Probabilistic generative models (PGMs) based on normalizing flow architectures can make it much easier via machine learning to train invertible maps needed for TFEP. However, the accuracy and applicability of approaches based on empirically learned maps depend crucially on the choice of reweighting method adopted to estimate the free energy differences. In this work, we assess the accuracy, rate of convergence, and data efficiency of different free energy estimators, including exponential averaging, Bennett acceptance ratio (BAR), and multistate Bennett acceptance ratio (MBAR), in reweighting PGMs trained by maximum likelihood on limited amounts of molecular dynamics data sampled only from end-states of interest. We carry out the comparisons on a set of simple but representative case studies, including conformational ensembles of alanine dipeptide and ibuprofen. Our results indicate that BAR and MBAR are both data efficient and robust, even in the presence of significant model overfitting in the generation of invertible maps. This analysis can serve as a stepping stone for the deployment of efficient and quantitatively accurate ML-based free energy calculation methods in complex systems.