Graph neural networks (GNNs) have been widely used for predicting molecular properties, especially for single molecules. However, when treating multi-component systems, GNNs have mostly used simple data representations (concatenation, averaging,...
Surfactants are amphiphilic molecules that are widely used in consumer products, industrial processes, and biological applications. A critical property of a surfactant is the critical micelle concentration (CMC), which is the concentration at which surfactant molecules undergo cooperative self-assembly in solution. Notably, the primary method to obtain CMCs experimentally tensiometryis laborious and expensive. In this study, we show that graph convolutional neural networks (GCNs) can predict CMCs directly from the surfactant molecular structure. In particular, we developed a GCN architecture that encodes the surfactant structure in the form of a molecular graph and trained it using experimental CMC data. We found that the GCN can predict CMCs with higher accuracy on a more inclusive data set than previously proposed methods and that it can generalize to anionic, cationic, zwitterionic, and nonionic surfactants using a single model. Molecular saliency maps revealed how atom types and surfactant molecular substructures contribute to CMCs and found this behavior to be in agreement with physical rules that correlate constitutional and topological information to CMCs. Following such rules, we proposed a small set of new surfactants for which experimental CMCs are not available; for these molecules, CMCs predicted with our GCN exhibited similar trends to those obtained from molecular simulations. These results provide evidence that GCNs can enable high-throughput screening of surfactants with desired self-assembly characteristics.
A noninvasive, highly sensitive universal immunosensor platform for protein-based biomarker detection is described in this Article. A neutral charged sensing environment is constructed by an antibody with an oppositely charged amino acid as surface charge neutralizer. By adjusting the pH condition of the testing environment, this neutral charged immunosensor (NCI) directly utilizes the electrostatic charges of the analyte for quantification of circulating protein markers, achieving a wide dynamic range covering through the whole picomole level. Comparing with previous studies on electrostatic charges characterization, this NCI demonstrates its capability to analyze not only the negatively charged biomolecules but also positively charged analytes. We applied this NCI for the detection of HE4 antigen with a detection limit at 2.5 pM and Tau antigen with a detection limit at 0.968 pM, demonstrating the high-sensitivity property of this platform. Furthermore, this NCI possesses a simple fabrication method (less than 2 h) and a short testing turnaround time (less than 30 min), providing an excellent potential for further clinical point-of-care applications.
The industry is heading towards digitalization with production and quality assurance of processes and products. Novel practices such as machine learning or artificial intelligence make it possible for highly complex and nonlinear occurrences to be modeled and predicted with immense accuracy when real experiences are used to train the algorithm. For example, supervised learning in a closed‐loop system allows the user to analyze and predict outcomes and gives it the ability to adapt and add intelligence to the current system. This study focuses on the development of a neural network (NN) for surface defect prediction in injection molding of model polypropylene. Feature optimization allows us to conclude that rheological parameters such as the melt flow index and relaxation time (λ) can improve predictive accuracy. Furthermore, Bayesian optimization is implemented to optimize the NN structure. The optimization approach allowed for a cross‐validation (CV) accuracy of 90.2% ± 4.4% with only five input parameters, while the seven‐input parameter optimized structure arrived at a CV accuracy of 92.4% ± 11.4%. Although the full‐feature structure optimized with Bayesian optimization concluded with slightly higher accuracy, the error range dramatically increased, meaning that this structure tends to overfit.
Graph neural networks (GNNs) have been widely used for predicting molecular properties, especially for single molecules. However, when treating multi-component systems, GNNs have mostly used simple data representations (concatenation, averaging, or self-attention on features of individual components) that might fail to capture molecular interactions and potentially limit prediction accuracy. In this work, we propose a GNN architecture that captures molecular interactions in an explicit manner by combining atomic-level (local) graph convolution and molecular-level (global) message passing through a molecular interaction network. We tested the architecture (which we call SolvGNN) on a comprehensive phase equilibrium case study that aims to predict activity coefficients for a wide range of binary and ternary mixtures; we built this large dataset using the COnductor-like Screening MOdel for Real Solvation (COSMO-RS). We show that SolvGNN can predict composition-dependent activity coefficients with high accuracy and show that it outperforms a previously developed GNN used for predicting only infinite-dilution activity coefficients. We performed counterfactual analysis on the SolvGNN model that allowed us to explore the impact of functional groups and composition on equilibrium behavior. We also used the SolvGNN model for the development of a computational framework that automatically creates phase diagrams for a diverse set of complex mixtures. All scripts needed to reproduce the results are shared as open-source code.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.