Bulletin of the American Physical Society
APS March Meeting 2023
Volume 68, Number 3
Las Vegas, Nevada (March 5-10)
Virtual (March 20-22); Time Zone: Pacific Time
Session T03: Machine Learning, Autonomous Experiments, and Big Data in Polymer Physics IIFocus
|
Hide Abstracts |
Sponsoring Units: DPOLY Chair: Tyler Martin, National Institute of Standards and Tech Room: Room 126 |
Thursday, March 9, 2023 11:30AM - 12:06PM |
T03.00001: Olexandr Isayev Invited Speaker: Olexandr Isayev
|
Thursday, March 9, 2023 12:06PM - 12:18PM |
T03.00002: Linking Rheological Properties with Molecular-Scale Features via Molecular Dynamics Simulations and Machine Learning Wenhui Li, JCS Kadupitiya, Vikram Jadhao Accurate molecular-scale understanding of rheological properties of small-molecular liquids and polymers is critical to optimizing their performance in practical applications such as lubrication and hydraulic fracking. Non-equilibrium molecular dynamics (NEMD) simulations can extract the flow behavior of materials sheared at very high strain rates that are difficult to reach in experiments. However, NEMD simulations produce a large amount of high-dimensional output data, which is often under-utilized in examining the link between molecular-scale features and rheological properties. We combine NEMD simulations with machine learning (ML) methods such as principal component analysis and t-distributed stochastic neighbor embedding to extract the correlation between molecular structure and shear thinning of small-molecular liquids such as squalane, polydecene trimer, and 9-octylheptadecane over a broad range of strain rates (105 - 1010 s-1), pressures (0.1 - 1000 MPa), and temperatures (293 - 373 K). ML reveals the competing contributions of end-to-end atom pairs and side branches toward shear thinning of fluids with low Newtonian viscosities. Data-driven insights into the molecular-scale origins of shear thinning of fluids with high Newtonian viscosity are also presented. |
Thursday, March 9, 2023 12:18PM - 12:30PM |
T03.00003: Active Learning Exploration of Thermally Conductive Strained Polymers Renzheng Zhang, Jiaxin Xu, Hanfeng Zhang, Tengfei Luo Thermal conductivity (TC), as an important transport property of polymers, can be improved when subject to strain since it can help align polymer chains. However, discovery of polymers that may have high TC after strain can be time-consuming and without guarantee of success. In this work, we employ an active learning scheme to speed up the discovery of high TC polymers. Polymers under strain were simulated using molecular dynamics (MD) and their TC are calculated. A Bayesian Neural Network (BNN) is then trained using these data. The BNN is used to screen the PoLyInfo database, and predicted mean TC and uncertainty are used towards an acquisition function to recommend polymers for MD labeling. The TC of these selected polymers is then calculated using MD simulations. The obtained data are then added to the training set to start another iteration in the active learning cycle. Through a few cycles, we were able to identify strained polymers with TC much better than the original dataset. |
Thursday, March 9, 2023 12:30PM - 12:42PM |
T03.00004: Connecting local structure and transport of small molecules through glassy polymers Samuel J Layding, Robert A Riggleman Understanding the relationship between structure and dynamics in disordered glassy materials is a key question in current scientific research. Several methods developed in the last decade have greatly improved understanding in this area, including the use of machine-learned softness fields to relate local structure with mobility and the propensity of particles to rearrange. For several model glass-formers it has been shown that rearrangement is an Arrhenius process in softness below the onset temperature of glassy dynamics with characteristic energetic and entropic scales. In this work we examine the diffusion of small particles through simple homopolymer melts below the onset of glassy dynamics and explore the relationship between softness and diffusion as the relative size and interaction potential of the diffusing particles changes. Additionally, we investigate how aging impacts transport inside the polymer glass by examining the time evolution of softness and the average energy required for particle rearrangement in the system. Characterizing this behavior provides an opportunity to better understand how softness may be used to predict or design for specific diffusion behavior, for example as applied to separations processes. |
Thursday, March 9, 2023 12:42PM - 12:54PM |
T03.00005: Deep Learning Approaches for Property Prediction and Inverse Design of Polymers JIHUN AHN, Su-Mi Hur, Yeojin Choe, Gabriella Pasya Irianti As an emerging paradigm, deep learning has attracted tremendous attention for unveiling molecular structure-property relationships and designing new materials. However, unlike highly crystalline inorganic materials, applying deep learning to polymeric systems is inherently challenged by the polymer’s multi-leveled features and complexity. Here, we present our efforts to apply deep learning models to polymeric systems by proposing an efficient and simplified polymer-to-string representation. We demonstrate that the proposed representation combined with RNN(Recurrent Neural Network) network provides satisfactory performance in predicting bulk properties of polymer melts. Furthermore, our representation is applied for an inverse design network that can generate polymer structures with targeted properties; the conciseness of the proposed representation guarantees the increased validity of output polymer structures. |
Thursday, March 9, 2023 12:54PM - 1:06PM |
T03.00006: Developing Universal Machine Learning Model for Predicting Polymer Properties Himanshu ., Tarak K Patra Machine learning models are gaining popularity and potency in predicting polymers' properties. These models can be built using pre-existing polymer structure-property data. They are very useful for rapid prediction of their properties for unknown structures and, thus, potentially accelerate their design. However, building an efficient, transferable, and universal machine learning model requires addressing several challenges. First, there are no general guidelines for selecting model architecture, hyperparameters, and polymer fingerprinting. Second, they need large volumes of homogeneous structure-property data, which is not readily available for many polymer design problems. Third, these models are interpolative in nature. Their extrapolation capability is poorly understood. To address these problems, here we propose a simple approach to build computationally cheap high-fidelity deep neural network models with optimal hyperparameters, introduce efficient polymer fingerprinting and data selection algorithm for universal model building. We implement these approaches for predicting several polymer properties, including single molecule radius of gyration, polymer compatibilizer, adhesion-free energy, and glass transition temperature, demonstrating the generality of these strategies. |
Thursday, March 9, 2023 1:06PM - 1:18PM |
T03.00007: Quantifying Pairwise Chemical Similarity of Polymers Jiale Shi, Nathan J Rebello, Dylan Walsh, Weizhong Zou, Michael E Deagen, Debra J Audus, Bradley D Olsen Computing the similarity between polymers is an important task in determining the diversity of different polymer libraries, developing search methods for polymer databases, and developing AI algorithms that can explore chemical space. In contrast to small molecules which have well-defined molecule structures and existing methods for similarity calculations, chemical similarity remains an outstanding problem due to the stochastic nature of polymers. Here, we design a similarity function built upon automata as a polymer graph representation. The graph representation is decomposed into three parts: repeat units, end groups, and topology. Similarity scores for each of the three parts are computed and linearly combined to yield a pairwise chemical similarity for polymers that are tunable based on the parameters used in the linear combination to reflect the needs of the user. We demonstrate the power of our method through a variety of case studies. This method gives a promising solution to quantitatively calculate the pairwise chemical similarity for polymers and presents an essential step towards making polymer data more structured and organized. |
Thursday, March 9, 2023 1:18PM - 1:30PM |
T03.00008: Canonicalizing BigSMILES for Polymer Informatics Using Chemical Intuition and State Machines Nathan J Rebello, Tzyy-Shyang Lin, Guang-He Lee, Melody A Morris, Bradley D Olsen Based on and fully compatible with the extremely popular SMILES line notation, BigSMILES is a user-friendly, human-readable text-based notation for encoding polymers as random graphs that will be transformative in machine learning applications and polymer informatics. However, a single polymer can have many BigSMILES representations, which makes tasks like searching polymers by string difficult. We introduce two algorithms to canonicalize BigSMILES into a single unique string representation. In the first algorithm, the user writes BigSMILES repeat units according to the monomers from which they are derived, and the output is a BigSMILES string that is human readable. The second algorithm does not depend on the choice of repeat units but rather the connectivity of the polymer: we propose that any linear polymer ensemble is a regular language as defined in computer science, and an abstract model of machines called finite automaton can describe that ensemble. Using algorithms in automata theory, an automaton with the fewest number of states can be derived, thus providing a means of canonicalizing a BigSMILES. These algorithms will be impactful in enabling chemists to search polymer structures in databases using strings. Moreover, using the conceptual advance of polymers as state machines, these stochastic graph representations can also be used in machine learning applications in polymer informatics, connecting structure to property. |
Thursday, March 9, 2023 1:30PM - 1:42PM |
T03.00009: CG-BigSMILES: a line notation for coarse-grained polymers Bruno S Leao, Weizhong Zou, Bradley D Olsen The non-deterministic character of polymers imposes a challenge for researchers to have a machine-readable molecular representation, especially for computationally originated coarse-grained materials. Although a previous work from our lab introduces BigSMILES, an atomistic line notation that represents polymers, the problem is still open for coarse-grained polymers. Here, we propose a layer based annotation syntax, called CG-BigSMILES, as complement to the non-coarse-grained BigSMILES. This new syntax contains specific information about the type of coarse-grained model, such as Kremer-Grest, Slip Link and DPD. The detailed mapping between the group of atoms and the coarse-grained “beads”, is also allowed to indicate the molecular topology. Moreover, the syntax can represent mixtures of various solvents, small molecules and polymers as a list of components. By associating the forcefield file with the string of CG-BigSMILES, the proposed syntax allows the representation of computational materials on our recently developed data platform (CRIPT), greatly contributing to data integration and searchability in both the simulation and experimental fields for polymer science. |
Thursday, March 9, 2023 1:42PM - 1:54PM |
T03.00010: Machine learning of phase diagram based on GPR for the DPD simulation of drug delivery to endothelial cells Saeed Akbari, Soumya Ray, Joao M Maia, Fei Zhou, Xiao Chen Despite huge effort over the years, the design of functionalized nanocarriers (NCs) for targeted drug delivery to endothelial cells is still to be completely unveiled. Dissipative Particle Dynamics (DPD) simulations, combined with an energy calculation method are used to find the phase diagram for adhesion of the designed NCs to the endothelial cell under the influence of series of parameters such as the shape, size, and ligand density of the NCs. However, preparing a phase diagram requires simulations of all possible NCs with the above-mentioned properties, which is not feasible. This challenge was addressed by applying a Gaussian Process Regression (GPR)-informed active learning strategy to the DPD results to drastically reduce the number of necessary simulations. We then validate the use of the ML-informed methodology by investigating computationally the morphology, dynamics, and inclusion free energy of NCs. |
Thursday, March 9, 2023 1:54PM - 2:06PM |
T03.00011: Deep Learning Boosted Langevin Field-Theoretic Simulation of Polymers Jaeup Kim, Daeseong Yong Physicists' community recently adopted machine learning for various research tasks, but performing polymer simulation using deep learning (DL) is still a relatively unexplored subject. For the widespread use of DL in a polymer simulation, accuracy of the DL prediction and the neural net training time are the two key issues that must be overcome. Field-theoretic simulations are promising tools in polymer field theory that can account for the compositional fluctuation effect, and among them, Langevin field-theoretic simulation (L-FTS) is known to be fast and free from the instability issue at low invariant polymerization index. However, it is still a computationally expensive tool, and it may take weeks to accurately calculate ensemble averages of thermodynamic quantities. In this presentation, we introduce a DNN that can be successively applied to determine the partial saddle point of the pressure field. Major deep learning (DL) models for semantic segmentation in computer vision are adopted to construct the optimal DNN architecture. Our model utilizing atrous convolutions in parallel is robust to the simulation parameter changes and can be reused after single training. Our DNN can achieve speedup of factor 6 or more compared to the conventional method without affecting accuracy. Open-source code for our deep Langevin FTS (DL-FTS) enables an easy and rapid Python scripting of SCFT and L-FTS incorporated with CPU or GPU parallelization and DL. |
Thursday, March 9, 2023 2:06PM - 2:18PM |
T03.00012: Machine learning strategies for the structure-property relationship of copolymers Ying Li, Lei Tao, Vikas Varshney, John Byrnes Establishing the structure-property relationship is extremely valuable for the molecular design of copolymers. However, machine learning (ML) models can incorporate both chemical composition and sequence distribution of monomers, and have the generalization ability to process various copolymer types (e.g., alternating, random, block, and gradient copolymers) with a unified approach are missing. To address this challenge, we formulate four different ML models for investigation, including a feedforward neural network (FFNN) model, a convolutional neural network (CNN) model, a recurrent neural network (RNN) model, and a combined FFNN/RNN (Fusion) model. We use various copolymer types to systematically validate the performance and generalizability of different models. We find that the RNN architecture that processes the monomer sequence information both forward and backward is a more suitable ML model for copolymers with better generalizability. As a supplement to polymer informatics, our proposed approach provides an efficient way for the evaluation of copolymers. |
Thursday, March 9, 2023 2:18PM - 2:30PM |
T03.00013: Data-driven identification and analysis of the glass transition in polymer melts Atreyee Banerjee, Hsiao-Ping Hsu, Kurt Kremer, Oleksandra Kukharenko On cooling, the dynamics of polymer melts slow down exponentially, leading to a glassy state without any drastic change in static structure. We employ data-driven approaches based on information about purely structural fluctuations of the chains to identify the glass transition from coarse-grained weakly semi-flexible polymer model simulations. More precisely, we used principal component analysis (PCA) to quantify conformational fluctuations and identify a sharp change in fluctuations around the glass transition temperature. The first PCA eigenvalue and the participation ratio show clear signatures of glass transition. The proposed method of glass transition temperature prediction is less ambiguous compared to the existing methods, as it requires minimal prior knowledge about the system and user inputs. |
Follow Us |
Engage
Become an APS Member |
My APS
Renew Membership |
Information for |
About APSThe American Physical Society (APS) is a non-profit membership organization working to advance the knowledge of physics. |
© 2024 American Physical Society
| All rights reserved | Terms of Use
| Contact Us
Headquarters
1 Physics Ellipse, College Park, MD 20740-3844
(301) 209-3200
Editorial Office
100 Motor Pkwy, Suite 110, Hauppauge, NY 11788
(631) 591-4000
Office of Public Affairs
529 14th St NW, Suite 1050, Washington, D.C. 20045-2001
(202) 662-8700