Data Generation for Machine Learning Interatomic Potentials and Beyond

TL;DR: Researchers develop a machine learning interatomic potential for niobium, accurately capturing non-equilibrium properties crucial for radiation damage simulations, outperforming existing potentials in accuracy and efficiency for large-scale MD simulations and collision cascade simulations.

TL;DR: Researchers introduce franken, a transfer learning framework that extracts atomic descriptors from pre-trained graph neural networks and transfers them to new systems using random Fourier features, achieving fast and accurate adaptation with minimal hyperparameter tuning.

Abstract: Abstract Molecular‐based electronic devices exploit the unique properties of single molecules or assemblies to surpass conventional solid‐state systems in miniaturization, efficiency, and functional diversity. Their performance hinges on controlling molecular orientation and packing at interfaces, which dictate charge transport and energy‐level alignment. Molecular orientation describes the directional alignment (e.g., face‐on, edge‐on) of organic semiconductors relative to substrates, while packing involves spatial arrangement, crystallinity, and interactions (π–π stacking, hydrogen bonding). Advanced techniques such as scanning probe microscopy (SPM), grazing‐incidence wide‐angle X‐ray scattering (GIWAXS), and Kelvin probe force microscopy (KPFM) elucidate these structural features, establishing correlations with optoelectronic properties like light absorption, exciton dynamics, and charge carrier mobility. The interplay between orientation and packing governs energy‐level alignment and charge transport via interfacial work function modulation. Emerging computational tools like machine learning (ML) and multiscale simulations enable predictive design of molecular configurations for targeted device functionalities. However, challenges remain in achieving uniform molecular alignment across practical device architectures and establishing robust structure‐property relationships under operational conditions. Addressing these requires integrating experimental characterization, computational modeling, and synthetic innovation. This review highlights the need for multidisciplinary approaches to advance molecular electronics toward practical, high‐performance applications, balancing fundamental insights with scalable fabrication.

Abstract: Ring Vault contains 201 546 cyclic molecules across 11 elements. AIMNet2 with 3D information outperformed 2D models in predicting the electronic properties of cyclic molecules.

TL;DR: For example, AlphaFold as mentioned in this paper predicts protein structures with an accuracy competitive with experimental structures in the majority of cases using a novel deep learning architecture. But the accuracy is limited by the fact that no homologous structure is available.

TL;DR: The relationship between transfer learning and other related machine learning techniques such as domain adaptation, multitask learning and sample selection bias, as well as covariate shift are discussed.

TL;DR: In this article, a modification of the nudged elastic band method for finding minimum energy paths is presented, where one of the images is made to climb up along the elastic band to converge rigorously on the highest saddle point.

TL;DR: A general Amber force field for organic molecules is described, designed to be compatible with existing Amber force fields for proteins and nucleic acids, and has parameters for most organic and pharmaceutical molecules that are composed of H, C, N, O, S, P, and halogens.