Digital discovery最新文献

The rapid advancement of machine learning in computational chemistry has opened new doors for designing molecules, predicting molecular properties, and discovering novel materials. However, building scalable and robust models for molecular machine learning remains a significant challenge due to the vast size and complexity of chemical space. Recent advances in chemical foundation models hold considerable promise for addressing these challenges, but such models remain difficult to train and are often fully or partially proprietary. For this reason, we introduce ChemBERTa-3, an open source training and benchmarking framework designed to train and fine-tune large-scale chemical foundation models. ChemBERTa-3 provides: (i) unified, reproducible infrastructure for model pretraining and fine-tuning, (ii) systematic benchmarking tooling to evaluate proposed chemical foundation model architectures on tasks from the MoleculeNet suite, and (iii) fully open release of model weights, training configurations, and deployment workflows. Our experiments demonstrate that although both graph-based and transformer-based architectures perform well at small scale, transformer-based models are considerably easier to scale. We also discuss how to overcome the numerous challenges that arise when attempting to reproducibly construct large chemical foundation models, ranging from subtle benchmarking issues to training instabilities. We test ChemBERTa-3 infrastructure in both an AWS-based Ray deployment and in an on-premise high-performance computing cluster to verify the reproducibility of the framework and results. We anticipate that ChemBERTa-3 will serve as a foundational building block for next-generation chemical foundation models and for the broader project of creating open source LLMs for scientific applications. In support of reproducible and extensible science, we have open sourced all ChemBERTa3 models and our Ray cluster configurations.

Machine learning force fields (MLFFs) are rapidly evolving to provide molecular dynamics simulations of molecules and materials with an accuracy comparable to ab initio methods, while significantly reducing the need for computational resources. However, conventional MLFFs are generally system-specific; introducing new chemical components requires assembling a new training dataset and retraining the entire model from scratch. To address this, we present a density-based machine learning force field (DBMLFF). The key advantage of DBMLFF lies in its modular parametrization strategy: by modeling each molecular species independently, the resulting force fields achieve seamless transferability across diverse chemical environments and retaining high accuracy without the need for retraining. This significantly improves model portability and cross-system applicability. Unlike most of the statistically based MLFFs, DBMLFF is a physics-based force field with machine learning components in it. It computes intermolecular interactions directly from electron density, enabling accurate descriptions of complex non-bonded behavior. In terms of computational efficiency, DBMLFF is three orders of magnitude faster than ab initio molecular dynamics and exhibits linear scaling with system size, allowing efficient simulations of large-scale systems. These features make DBMLFF a robust tool for multi-component electrolyte MD simulations, ideal for practical electrochemical systems with variable compositions and large scales.

Solid-state electrolyte batteries are expected to replace liquid electrolyte lithium-ion batteries in the near future thanks to their higher theoretical energy density and improved safety. However, their adoption is currently hindered by imperfect electrode–electrolyte interfaces and a lower effective ionic conductivity, a quantity that governs charge and discharge rates. Identifying highly ion-conductive materials using conventional theoretical calculations and experimental validation is both time-consuming and resource-intensive. While machine learning holds the promise to expedite this process, relevant ionic conductivity and structural data is scarce. Here, we present OBELiX, a database of ∼600 synthesized solid electrolyte materials and their experimentally measured room temperature ionic conductivities gathered from literature and curated by domain experts. Each material is described by their measured composition, space group and lattice parameters. A full-crystal description in the form of a crystallographic information file (CIF) is provided for ∼320 structures for which atomic positions were available. We discuss various statistics and features of the dataset and provide training and testing splits carefully designed to avoid data leakage. Finally, we benchmark seven existing ML models on the task of predicting ionic conductivity and discuss their performance. The goal of this work is to facilitate the use of machine learning for solid-state electrolyte materials discovery.

Large Language Models (LLMs) with generative capabilities have garnered significant attention in various domains, including materials science. However, systematically evaluating their performance for structure generation tasks remains a major challenge. In this study, we fine-tune multiple LLMs on various density functional theory (DFT) datasets (including superconducting and semiconducting materials at different levels of DFT theory) and apply quantitative metrics to benchmark their effectiveness. Among the models evaluated, the Mistral 7 billion parameter model demonstrated excellent performance across several metrics. Leveraging this model, we generated candidate semiconductors and further screened them using a graph neural network property prediction model and validated them with DFT. Starting from nearly 100 000 generated candidates, we identified six semiconductor materials near the convex hull with DFT that were not present in any known datasets, one of which was found to be dynamically stable (Na₃S₂). This study demonstrates the effectiveness of a pipeline spanning fine-tuning, evaluation, generation, and validation for accelerating inverse design and discovery in computational materials science.

The semi-empirical Pariser–Parr–Pople (PPP) Hamiltonian is reviewed for its ability to provide a minimal model of the chemistry of conjugated π-electron systems, and its current applications and limitations are discussed. Since its inception, the PPP Hamiltonian has helped in the development of new computational approaches in instances where compute is constrained due to its inherent approximations that allow for an efficient representation and calculation of many systems of chemical and technological interest. The crucial influence of electron correlation on the validity of these approximations is discussed, and we review how PPP full configuration interaction-type calculations have enabled a deeper understanding of conjugated polymer systems. More recent usage of the PPP Hamiltonian includes its application in high-throughput screening activities to the inverse design problem, which we illustrate here for two specific fields of technological interest: singlet fission and singlet–triplet inverted energy gap molecules. Finally, we conjecture how utilizing the PPP model in quantum computing applications could be mutually beneficial.

In the study of non-adiabatic chemical processes such as photocatalysis and photosynthesis, non-adiabatic molecular dynamics (NAMD) is an indispensable theoretical tool, which requires precise potential energy surfaces (PESs) of ground and excited states. Quantum computing offers promising potential for calculating PESs that are intractable for classical computers. However, its realistic application poses significant challenges to the development of quantum algorithms that are sufficiently general to enable efficient and precise PES calculations across chemical systems with diverse properties and to seamlessly adapt existing NAMD theories to quantum computing. In this work, we introduce a quantum-adapted extension to the Landau–Zener-Surface-Hopping (LZSH) NAMD. This extension incorporates curvature-driven hopping corrections that protect the population evolution while maintaining the efficiency gained from avoiding the computation of non-adiabatic couplings (NACs) and preserving the trajectory independence that enables parallelization. Furthermore, to ensure the high-precision PESs required for surface hopping dynamics, we develop a sub-microhartree-accurate PES calculation protocol. This protocol supports active space selection, enables parallel acceleration either on quantum or classical clusters, and demonstrates adaptability to diverse chemical systems—including the charged H₃⁺ ion and the C₂H₄ molecule, a prototypical multi-reference benchmark. This work paves the way for practical application of quantum computing in NAMD, showcasing the potential of parallel simulation on quantum-classical heterogeneous clusters for ab initio computational chemistry.

The lack of flexible annotation tools has hindered the deployment of AI models in some scientific areas. Most existing image annotation software requires users to upload a precollected dataset, which limits support for on-demand pipelines and introduces unnecessary steps to acquire images. This constraint is particularly problematic in laboratory environments, where on-site data acquisition from instruments such as microscopes is increasingly common. In this work, we introduce LivePixel, a Python-based graphical user interface that integrates with imaging systems, such as webcams, microscopes, and others, to enable on-site image annotation. LivePyxel is designed to be easy to use through a simple interface that allows users to precisely delimit areas for annotation using tools commonly found in commercial graphics editing software. Of particular interest is the availability of Bézier splines and binary masks, and the software's capacity to work with non-destructive layers that enable high-performance editing. LivePyxel also integrates a wide compatibility across video devices, and it's optimized for object detection operations via the use of OpenCV in combination with high-performance libraries designed to handle matrix and linear algebra operations via Numpy effectively. LivePyxel facilitates seamless data collection and labeling, accelerating the development of AI models in experimental workflows. LivePyxel is freely available at https://github.com/UGarCil/LivePyxel.

Despite the rapidly growing applications of robots in industry, the use of robots to automate tasks in scientific laboratories is less prolific due to the lack of generalized methodologies and the high cost of hardware. This paper focuses on the automation of characterization tasks necessary for reducing cost while maintaining generalization and proposes a software architecture for building robotic systems in scientific laboratory environments. A dual-layer (Socket.IO and ROS) action server design is the basic building block, which facilitates the implementation of a web-based front end for user-friendly operation and the use of ROS Behavior Trees for convenient task planning and execution. A robotic platform for automating mineral and material sample characterization is built upon the architecture, with an open-source, low-cost three-axis computer numerical control gantry system serving as the main robot. A handheld laser induced breakdown spectroscopy (LIBS) analyzer is integrated with a 3D printed adapter, enabling (1) automated 2D chemical mapping and (2) autonomous sample measurement (with the support of an RGB-Depth camera). We demonstrate the utility of automated chemical mapping by scanning the surface of a spodumene-bearing pegmatite core sample with a 1071-point dense hyperspectral map acquired at a rate of 1520 bits per second. Furthermore, we showcase the autonomy of the platform in terms of perception, dynamic decision-making, and execution, through a case study of LIBS measurement of multiple mineral samples. The platform enables controlled and autonomous chemical quantification in the laboratory that complements field-based measurements acquired with the same handheld device, linking resource exploration and processing steps in the supply chain for lithium-based battery materials.

Real-time monitoring of laboratory experiments is essential for automating complex workflows and enhancing experimental efficiency. Accurate detection and classification of chemicals in varying forms and states support a range of techniques, including liquid–liquid extraction, distillation, and crystallization. However, challenges exist in the detection of chemical forms: some classes appear visually similar, and the classification of the forms is often context-dependent. In this study, we adapt the YOLO model into a multi-modal architecture that integrates scene images and task context for object detection. With the help of Large Language Models (LLM), the developed method facilitates reasoning about the experimental process and uses the reasoning result as the context guidance for the detection model. Experimental results show that by introducing context during training and inference, the performance of the proposed model, YOLO-text, has improved among all classes, and the model is able to make accurate predictions on visually similar areas. Compared to the baseline, our model increases 4.8% overall mAP without context given and 7% with context. The proposed framework can classify and localize substances with and without contextual suggestions, thereby enhancing the adaptability and flexibility of the detection process.

De novo 3D molecule generation is a pivotal task in drug discovery. However, many recent geometric generative models struggle to produce high-quality geometries, even if they able to generate valid molecular graphs. To tackle this issue and enhance the learning of effective molecular generation dynamics, we present Megalodon – a family of scalable transformer models. These models are enhanced with basic equivariant layers and trained using a joint continuous and discrete denoising co-design objective. We assess Megalodon's performance on established molecule generation benchmarks and introduce new 3D structure benchmarks that evaluate a model's capability to generate realistic molecular structures, particularly focusing on geometry precision. We show that Megalodon achieves state-of-the-art results in 3D molecule generation, conditional structure generation, and structure energy benchmarks using diffusion and flow matching. Furthermore, we demonstrate that scaling Megalodon produces up to 49× more valid molecules at large sizes and 2–10× lower energy compared to the prior best generative models. The code and the model are available at https://github.com/NVIDIA-Digital-Bio/megalodon.