Physics-based Methods Research Articles

Extrapolation of cavitation and hydrodynamic pressure in lubricated contacts: a physics-informed neural network approach

A comprehensive understanding of the dynamics of tribological interactions is essential for enhancing efficiency and durability in a multitude of technical domains. Conventional experimental techniques in tribology are frequently costly and time-consuming. In contrast, elastohydrodynamic lubrication (EHL) simulation models present a viable alternative for calculating frictional forces in sealing contacts. These calculations are based on the hydrodynamics within the sealing contact, as defined by the Reynolds equation, the deformation of the seal, and the contact mechanics. However, a significant drawback of these simulations is the time-consuming calculation process. To overcome these experimental and computational limitations, machine learning algorithms offer a promising solution. Physics-informed machine learning (PIML) improves on traditional data-driven models by incorporating physical principles. In particular, physics-informed neural networks (PINNs) are as effective hybrid solvers that combine data-driven and physics-based methods to solve the partial differential equations that drive EHL simulations. By integrating physical laws into the parameter optimization of the neural network (NN), PINNs provide accurate and fast solutions. Thus, unlike traditional NNs, PINNs have the potential to make accurate predictions beyond the limited training domain. The objective of this study is to demonstrate the feasibility of spatial and temporal extrapolation of the PINN and to analyze its reliability, both with and without consideration of cavitation. Two test cases are employed to examine the pressure and cavitation distribution within a sealing contact that extends beyond the spatial and temporal training range. The findings indicate that PINNs can surmount the typical constraints associated with NNs in the extrapolation of solution spaces, which represents a notable advancement in terms of computational efficiency and model flexibility.

Supraglacial Lake Depth Retrieval from ICESat-2 and Multispectral Imagery Datasets

Supraglacial lakes play an important role in the surface mass balance of ice sheets. With global warming, supraglacial lakes may become more extensive on ice sheet surfaces than they currently are. Therefore, accurate estimation of the volume of supraglacial lakes is important for characterizing their impact on ice sheets. In this study, we present a machine learning-based method for estimating the depth of supraglacial lakes through the combination of ICESat-2 ATL03 data with multispectral imagery. We tested this method via Landsat-8 and Sentinel-2 imagery and evaluated the accuracy of the algorithm on 7 test lakes on the Greenland Ice Sheet. Our results show that machine learning-based algorithms achieve better accuracy than traditional regression or physics-based methods do, especially for deeper lakes. The best accuracy was achieved when extreme gradient boosting was applied to a Sentinel-2 L1C image, with root mean square error, mean absolute error, and median absolute error values of 0.54 m, 0.43 m, and 0.36 m, respectively. Furthermore, we evaluated the effects of atmospheric corrections of multispectral imagery in the retrieval of supraglacial lake depth. On the basis of our results, we recommend the direct use of top-of-atmosphere reflectance products in mapping supraglacial lake bathymetry because of the low performance of atmospheric corrections for water and snow/ice in both the Landsat-8 and Sentinel-2 datasets. This study is expected to provide a more efficient method for estimating the depth of supraglacial lakes and laying the foundation for accurately quantifying meltwater volumes over large surface areas in subsequent studies.

The physics-AI dialogue in drug design.

A long path has led from the determination of the first protein structure in 1960 to the recent breakthroughs in protein science. Protein structure prediction and design methodologies based on machine learning (ML) have been recognized with the 2024 Nobel prize in Chemistry, but they would not have been possible without previous work and the input of many domain scientists. Challenges remain in the application of ML tools for the prediction of structural ensembles and their usage within the software pipelines for structure determination by crystallography or cryogenic electron microscopy. In the drug discovery workflow, ML techniques are being used in diverse areas such as scoring of docked poses, or the generation of molecular descriptors. As the ML techniques become more widespread, novel applications emerge which can profit from the large amounts of data available. Nevertheless, it is essential to balance the potential advantages against the environmental costs of ML deployment to decide if and when it is best to apply it. For hit to lead optimization ML tools can efficiently interpolate between compounds in large chemical series but free energy calculations by molecular dynamics simulations seem to be superior for designing novel derivatives. Importantly, the potential complementarity and/or synergism of physics-based methods (e.g., force field-based simulation models) and data-hungry ML techniques is growing strongly. Current ML methods have evolved from decades of research. It is now necessary for biologists, physicists, and computer scientists to fully understand advantages and limitations of ML techniques to ensure that the complementarity of physics-based methods and ML tools can be fully exploited for drug design.

PHYSICS BASED ALGORITHMS FOR FUZZY TRANSPORTATION PROBLEM

Transportation problems are computationally challenging due to their nondeterministic polynomial-time hard (NP-hard) nature, especially when uncertainties in costs, supply, and demand are involved. The fuzzy transportation problem addresses these uncertainties by representing them as fuzzy numbers, requiring robust methods for ranking and solving optimization models. Traditional exact methods become impractical for such problems, leading to the adoption of metaheuristic approaches. This study introduces three novel physics-based algorithms: one single-point-based and two population-based methods. These algorithms leverage principles from gravitational attraction, electromagnetism, and water flow dynamics, incorporating innovative neighborhood structures tailored to the fuzzy nature of transportation problems. The proposed methods were tested on diverse problem sizes and compared against established optimization techniques, such as genetic algorithms and commercial software, using Python implementations. An experimental design methodology was employed to optimize parameter tuning, enhancing computational efficiency and reducing experimentation time. Results demonstrate that the physics-based algorithms achieve competitive performance in terms of solution quality, computational efficiency, and adaptability to uncertainty. These findings underscore the potential of physics-based methods to advance optimization in fuzzy transportation and other domains with inherent uncertainties, paving the way for further research in hybrid and multi-objective approaches.

Recent Progress in Modeling and Simulation of Biomolecular Crowding and Condensation Inside Cells.

Macromolecular crowding in the cellular cytoplasm can potentially impact diffusion rates of proteins, their intrinsic structural stability, binding of proteins to their corresponding partners as well as biomolecular organization and phase separation. While such intracellular crowding can have a large impact on biomolecular structure and function, the molecular mechanisms and driving forces that determine the effect of crowding on dynamics and conformations of macromolecules are so far not well understood. At a molecular level, computational methods can provide a unique lens to investigate the effect of macromolecular crowding on biomolecular behavior, providing us with a resolution that is challenging to reach with experimental techniques alone. In this review, we focus on the various physics-based and data-driven computational methods developed in the past few years to investigate macromolecular crowding and intracellular protein condensation. We review recent progress in modeling and simulation of biomolecular systems of varying sizes, ranging from single protein molecules to the entire cellular cytoplasm. We further discuss the effects of macromolecular crowding on different phenomena, such as diffusion, protein-ligand binding, and mechanical and viscoelastic properties, such as surface tension of condensates. Finally, we discuss some of the outstanding challenges that we anticipate the community addressing in the next few years in order to investigate biological phenomena in model cellular environments by reproducing in vivo conditions as accurately as possible.

Integrated machine learning and physics-based methods assisted de novo design of Fatty Acyl-CoA synthase inhibitors

ABSTRACT Background Tuberculosis is an infectious disease that has become endemic worldwide. The causative bacteria Mycobacterium tuberculosis (Mtb) is targeted via several exciting drug targets. One newly discovered target is the Fatty Acyl-CoA synthase, which plays a significant role in activating the long-chain fatty acids. Research design & methods This study aims to generate novel compounds using Machine Learning (ML) algorithms to inhibit this synthase. Experimentally derived bioactive compounds were chosen from ChEMBL and used as inputs for effective molecule generation by Reinvent4. The library of new molecules generated was subjected to a two-tiered molecular docking protocol, and the results were further studied to obtain a binding free energy check. Results The ML-based de novo drug design (DNDD) approach successfully generated a diverse library of novel molecules targeting Fatty Acyl-CoA synthase. After rigorous molecular docking and binding free energy analysis, four new compounds were identified as potential lead candidates with promising inhibitory effects on Mtb lipid metabolism. Conclusions The study demonstrated the effectiveness of a machine-learning approach in generating novel drug candidates against Mtb. The identified hit compounds show potential as inhibitors of Fatty Acyl-CoA synthase, offering a new avenue for developing treatments for tuberculosis, particularly in combating drug-resistant strains.

Study of Cathode Materials for Na-Ion Batteries: Comparison Between Machine Learning Predictions and Density Functional Theory Calculations

Energy storage technologies have experienced significant advancements in recent decades, driven by the growing demand for efficient and sustainable energy solutions. The limitations associated with lithium’s supply chain, cost, and safety concerns have prompted the exploration of alternative battery chemistries. For this reason, research to replace widespread lithium batteries with sodium-ion batteries has received more and more attention. In the present work, we report cutting-edge research, where we explored a wide range of compositions of cathode materials for Na-ion batteries by first-principles calculations using workflow chains developed within the AiiDA framework. We trained crystal graph convolutional neural networks and geometric crystal graph neural networks, and we demonstrate the ability of the machine learning algorithms to predict the formation energy of the candidate materials as calculated by the density functional theory. This materials discovery approach is disruptive and significantly faster than traditional physics-based computational methods.

Conceptual framework for data-driven reservoir characterization: Integrating machine learning in petrophysical analysis

Reservoir characterization plays a critical role in optimizing oil and gas production by accurately determining subsurface properties such as porosity, permeability, and fluid saturation. Traditionally, this process has relied on the interpretation of well logs, seismic data, and core samples using manual or physics-based methods. However, these approaches often face limitations when dealing with complex datasets and heterogeneous reservoir properties. This review develops a conceptual framework for integrating machine learning (ML) into petrophysical analysis to revolutionize reservoir characterization. By utilizing advanced ML algorithms such as supervised learning, unsupervised learning, and deep learning, this model demonstrates the potential of data-driven methods to improve the accuracy and efficiency of reservoir analysis. The framework focuses on how ML techniques can automate the interpretation of well logs, enhance seismic data processing, and predict rock properties from core samples, leading to more accurate identification of reservoir boundaries and productive zones. This approach not only reduces interpretation errors but also accelerates decision-making in reservoir management. Moreover, machine learning can handle the vast amounts of data generated in reservoir studies, enabling real-time analysis and adaptive decision support. The study also addresses challenges such as data quality, model interpretability, and computational scalability, offering insights into future applications of ML in reservoir characterization. By combining petrophysical expertise with machine learning capabilities, this framework aims to significantly advance the field of reservoir characterization and contribute to more efficient resource management in the oil and gas industry.

Accurate de novo design of high-affinity protein binding macrocycles using deep learning.

The development of macrocyclic binders to therapeutic proteins typically relies on large-scale screening methods that are resource-intensive and provide little control over binding mode. Despite considerable progress in physics-based methods for peptide design and deep-learning methods for protein design, there are currently no robust approaches for de novo design of protein-binding macrocycles. Here, we introduce RFpeptides, a denoising diffusion-based pipeline for designing macrocyclic peptide binders against protein targets of interest. We test 20 or fewer designed macrocycles against each of four diverse proteins and obtain medium to high-affinity binders against all selected targets. Designs against MCL1 and MDM2 demonstrate KD between 1-10 μM, and the best anti-GABARAP macrocycle binds with a KD of 6 nM and a sub-nanomolar IC50 in vitro. For one of the targets, RbtA, we obtain a high-affinity binder with KD < 10 nM despite starting from the target sequence alone due to the lack of an experimentally determined target structure. X-ray structures determined for macrocycle-bound MCL1, GABARAP, and RbtA complexes match very closely with the computational design models, with three out of the four structures demonstrating Ca RMSD of less than 1.5 Å to the design models. In contrast to library screening approaches for which determining binding mode can be a major bottleneck, the binding modes of RFpeptides-generated macrocycles are known by design, which should greatly facilitate downstream optimization. RFpeptides thus provides a powerful framework for rapid and custom design of macrocyclic peptides for diagnostic and therapeutic applications.

A Nonconvex Approach with Structural Priors for Restoring Underwater Images

Underwater image restoration is a crucial task in various computer vision applications, including underwater target detection and recognition, autonomous underwater vehicles, underwater rescue, marine organism monitoring, and marine geological survey. Among other categories, the physics-based methods restore underwater images by improving the transmission map through optimization or regularization techniques. Conventional optimization-based methods often do not consider the effect of structural differences between guidance and transmission maps. To address this issue, in this paper, we present a regularization-based method for restoring underwater images that uses coherent structures between the guidance map and the transmission map. The proposed approach models the optimization of transmission maps through a nonconvex energy function comprising data and smoothness terms. The smoothness term includes static and dynamic structural priors, and the optimization problem is solved using a majorize-minimize algorithm. We evaluate the proposed method on benchmark datasets, and the results demonstrate the superiority of the proposed method over state-of-the-art techniques in terms of improving transmission maps and producing high-quality restored images.

Accurate prediction of protein–ligand interactions by combining physical energy functions and graph-neural networks

We introduce an advanced model for predicting protein–ligand interactions. Our approach combines the strengths of graph neural networks with physics-based scoring methods. Existing structure-based machine-learning models for protein–ligand binding prediction often fall short in practical virtual screening scenarios, hindered by the intricacies of binding poses, the chemical diversity of drug-like molecules, and the scarcity of crystallographic data for protein–ligand complexes. To overcome the limitations of existing machine learning-based prediction models, we propose a novel approach that fuses three independent neural network models. One classification model is designed to perform binary prediction of a given protein–ligand complex pose. The other two regression models are trained to predict the binding affinity and root-mean-square deviation of a ligand conformation from an input complex structure. We trained the model to account for both deviations in experimental and predicted binding affinities and pose prediction uncertainties. By effectively integrating the outputs of the triplet neural networks with a physics-based scoring function, our model showed a significantly improved performance in hit identification. The benchmark results with three independent decoy sets demonstrate that our model outperformed existing models in forward screening. Our model achieved top 1% enrichment factors of 32.7 and 23.1 with the CASF2016 and DUD-E benchmark sets, respectively. The benchmark results using the LIT-PCBA set further confirmed its higher average enrichment factors, emphasizing the model’s efficiency and generalizability. The model’s efficiency was further validated by identifying 23 active compounds from 63 candidates in experimental screening for autotaxin inhibitors, demonstrating its practical applicability in hit discovery.Scientific contributionOur work introduces a novel training strategy for a protein–ligand binding affinity prediction model by integrating the outputs of three independent sub-models and utilizing expertly crafted decoy sets. The model showcases exceptional performance across multiple benchmarks. The high enrichment factors in the LIT-PCBA benchmark demonstrate its potential to accelerate hit discovery.

Investigation of hybrid modeling and its transferability in building load prediction used for district heating systems

In the district heating systems, the historical operation data of the buildings in those areas would be partially or entirely missing. The traditional data-driven model is hard to predict the ground truth results because the historical data is not available for model training. However, utilizing the physics-based methods for load calculation takes a long time to process and encounters low accuracy issues. This paper investigates several hybrid models that integrate the data-driven model and the physics-based models with different fusion methods. The physics-based models calculate envelope load and infiltration load, based on Fourier's law and the grand canonical ensemble theory, respectively. After undergoing load processing, features fusion, and residual connection, the best advanced hybrid models generate 21.35%, 16.35%, and 12.73% better prediction results compared with the data-driven model. Moreover, the advanced hybride models also perform strong transferability across all the data quantity groups. In terms of practical application, the advanced hybrid models could be deployed with effective generalization in limited data scenarios and robust transfer capabilities. The selected best model constructed by hybrid modeling displays the highest performance and saves the total training costs with strong transferability.

Efficient generation of protein pockets with PocketGen

Designing protein-binding proteins is critical for drug discovery. However, artificial-intelligence-based design of such proteins is challenging due to the complexity of protein–ligand interactions, the flexibility of ligand molecules and amino acid side chains, and sequence–structure dependencies. We introduce PocketGen, a deep generative model that produces residue sequence and atomic structure of the protein regions in which ligand interactions occur. PocketGen promotes consistency between protein sequence and structure by using a graph transformer for structural encoding and a sequence refinement module based on a protein language model. The graph transformer captures interactions at multiple scales, including atom, residue and ligand levels. For sequence refinement, PocketGen integrates a structural adapter into the protein language model, ensuring that structure-based predictions align with sequence-based predictions. PocketGen can generate high-fidelity protein pockets with enhanced binding affinity and structural validity. It operates ten times faster than physics-based methods and achieves a 97% success rate, defined as the percentage of generated pockets with higher binding affinity than reference pockets. Additionally, it attains an amino acid recovery rate exceeding 63%.

A physics-informed neural network method for identifying parameters and predicting remaining life of fatigue crack growth

Predicting the remaining life of fatigue cracks is crucial for planning maintenance and repair strategies to prevent untoward incidents. This paper proposes a novel physics-informed neural network (PINN) method for identifying parameters and predicting remaining fatigue crack growth life (FCGL). Initially, the relationship between crack length and fatigue cycles is established through a neural network, and the gradient of fatigue cycles with respect to crack length is obtained by automatic differentiation. Subsequently, a composite loss function is designed to incorporate this gradient within the confines of physical knowledge, ensuring that the established relationship not only aligns with observed data but also adheres to physical knowledge. Furthermore, during the network training, the parameters in physical models are simultaneously updated to better conform to the individuality of the monitored subject. All predicted remaining FCGLs fall within the 1.5 times error band. Compared to purely data-driven or physics-based methods, the proposed method offers more robust and accurate predictions of remaining FCGLs.

Comparative assessment of physics-based in silico methods to calculate relative solubilities.

Relative solubilities, i.e. whether a given molecule is more soluble in one solvent compared to others, is a critical parameter for pharmaceutical and agricultural formulation development and chemical synthesis, material science, and environmental chemistry. In silico predictions of this crucial variable can help reducing experiments, waste of solvents and synthesis optimization. In this study, we evaluate the performance of different physics-based methods for predicting relative solubilities. Our assessment involves quantum mechanics-based COSMO-RS and molecular dynamics-based free energy methods using OPLS4, the open-source OpenFF Sage, and GAFF force fields, spanning over 200 solvent-solute combinations. Our investigation highlights the important role of compound multimerization, an effect which must be accounted for to obtain accurate relative solubility predictions. The performance landscape of these methods is varied, with significant differences in precision depending on both the method used and the solute considered, thereby offering an improved understanding of the predictive power of physics-based methods in chemical research.

Landslide Hazard and Rainfall Threshold Assessment: Incorporating Shallow and Deep-Seated Failure Mechanisms with Physics-Based Models

Landslides pose a significant threat worldwide, leading to numerous fatalities and severe economic losses. The city of Manizales, located in the Colombian Andes, is particularly vulnerable due to its steep topography and permeable volcanic ash-derived soils. This study aims to assess landslide hazards in Manizales by integrating shallow planar and deep-seated circular failure mechanisms using physics-based models (TRIGRS and Scoops3D). By combining hazard zonation maps with rainfall thresholds calibrated through historical data, we provide a refined approach for early warning systems (EWS) in the region. Our results underscore the significance of the landslide hazard maps, which combine shallow planar and deep-seated circular failure scenarios. By categorizing urban areas into high, medium, and low-risk zones, we offer a practical framework for urban planning. Moreover, we developed physics-based rainfall thresholds for early landslide warning, simplifying their application while aiming to enhance regional predictive accuracy. This comprehensive approach equips local authorities with essential tools to mitigate landslide risks, refine hazard zoning, and strengthen early warning systems, promoting safer urban development in the Andean region and beyond, as the physics-based methods used are well-established and implemented globally.

Different roads take me home: the nonlinear relationship between distance and flows during China’s Spring Festival

Human mobility modelling has attracted scholarly attention from physics-based methods and social science explanatory approaches. However, there is limited knowledge of the nonlinear relationship of flows and distance in intercity mobility and regional differences in the nonlinear relationship. Focusing on China’s long-distance and large-scale mobility during the Spring Festival, this paper develops a framework to explain the nonlinear relationship. Using the Gradient Boosting Decision Tree (GBDT) model and Tencent Big Data, we find that there are three types of nonlinear relationships, namely plateau (almost zero distance decay parameter), drop (decreasing distance decay parameter) and rebound (increasing distance decay parameter after decreasing). The provincial differences also reveal that the nonlinear relationships depend on the domestic relative location and the intra-provincial urban system. This result shows that the cities in the coastal province enjoy a more inclusive spatial structure, which supports the migration from the periphery of the province. In contrast, the inland cities are concerned with embracing the migrants and settling them down.

Alchemical Transformations and Beyond: Recent Advances and Real-World Applications of Free Energy Calculations in Drug Discovery.

Computational methods constitute efficient strategies for screening and optimizing potential drug molecules. A critical factor in this process is the binding affinity between candidate molecules and targets, quantified as binding free energy. Among various estimation methods, alchemical transformation methods stand out for their theoretical rigor. Despite challenges in force field accuracy and sampling efficiency, advancements in algorithms, software, and hardware have increased the application of free energy perturbation (FEP) calculations in the pharmaceutical industry. Here, we review the practical applications of FEP in drug discovery projects since 2018, covering both ligand-centric and residue-centric transformations. We show that relative binding free energy calculations have steadily achieved chemical accuracy in real-world applications. In addition, we discuss alternative physics-based simulation methods and the incorporation of deep learning into free energy calculations.

Exploring the potential of structure-based deep learning approaches for T cell receptor design.

Deep learning methods, trained on the increasing set of available protein 3D structures and sequences, have substantially impacted the protein modeling and design field. These advancements have facilitated the creation of novel proteins, or the optimization of existing ones designed for specific functions, such as binding a target protein. Despite the demonstrated potential of such approaches in designing general protein binders, their application in designing immunotherapeutics remains relatively underexplored. A relevant application is the design of T cell receptors (TCRs). Given the crucial role of T cells in mediating immune responses, redirecting these cells to tumor or infected target cells through the engineering of TCRs has shown promising results in treating diseases, especially cancer. However, the computational design of TCR interactions presents challenges for current physics-based methods, particularly due to the unique natural characteristics of these interfaces, such as low affinity and cross-reactivity. For this reason, in this study, we explored the potential of two structure-based deep learning protein design methods, ProteinMPNN and ESM-IF1, in designing fixed-backbone TCRs for binding target antigenic peptides presented by the MHC through different design scenarios. To evaluate TCR designs, we employed a comprehensive set of sequence- and structure-based metrics, highlighting the benefits of these methods in comparison to classical physics-based design methods and identifying deficiencies for improvement.

Physics-informed ensemble learning with residual modeling for enhanced building energy prediction

Accurate modeling of building energy use is important to support a diverse spectrum of its downstream applications, such as building energy efficiency assessment, resilience analysis, smart control, etc. As mainstream approaches of building energy modeling, physics-based modeling builds on different fidelities of physics rules yet are usually compromised in modeling accuracy due to insufficiency of physics rules to capture real-world dynamics and incomplete input information. Data-driven approaches are computationally efficient, but black box (uninterpretable) in nature. For improved modeling of building energy use, this work proposes a physics-informed ensemble learning approach in building energy prediction through residual modeling. Specifically, we first analyze the components of building energy use data. Evidence suggests that the building energy use data can be decomposed into physics-driven part, occupant-driven part, and white noise. Second, high-fidelity physics-based building models (EnergyPlus) and low-fidelity ones (RC models) are developed to capture the physics-driven part while time series methods are explored as the residual modeling approach to capture the occupant-driven discrepancies between physics-based simulation and measured building energy use (i.e., residuals). Finally, the physics-informed ensemble learning is proposed to integrate physics-based and data-driven models for enhanced accuracy and robustness of building energy modeling. Results demonstrate 40–90% increase of accuracy between modeling and field observations compared to traditional physics-based modeling methods. Moreover, when the training dataset size is small, the proposed ensemble model overperforms the pure data-driven models, demonstrating its higher robustness in extrapolation scenarios. This work makes fundamental contributions to the development of convergent modeling approaches in the building modeling field.