Digital discovery最新文献

Machine learning (ML) has the potential to accelerate the discovery of high-performance materials by learning complex structure–property relationships and prioritizing candidates for costly experiments or simulations. However, ML efficiency is often offset by the need for large, high-quality training datasets, motivating strategies that intelligently select the most informative samples. Here, we formulate the search for top-performing functionalized nanoporous materials (metal–organic and covalent–organic frameworks) as a global optimization problem and apply Bayesian Optimization (BO) to identify regions of interest and rank candidates with minimal evaluations. We highlight the importance of a proper and efficient initialization scheme of the BO process, and we demonstrate how BO-acquired samples can also be used to train an XGBoost regression predictive model that can further enrich the efficient mapping of the region of high performing instances of the design space. Across multiple literature-derived adsorption and diffusion datasets containing thousands of structures, our BO framework identifies 2×- to 3×-more materials within a top-100 or top-10 ranking list, than random-sampling-based ML pipelines, and it achieves significantly higher ranking quality. Moreover, the surrogate enrichment strategy further boosts top-N recovery while maintaining high ranking fidelity. By shifting the evaluation focus from average predictive metrics (e.g., R², MSE) to task-specific criteria (e.g., recall@N and nDCG), our approach offers a practical, data-efficient, and computationally accessible route to guide experimental and computational campaigns toward the most promising materials.

The detection and classification of exfoliated two-dimensional (2D) material flakes from optical microscope images can be automated using computer vision algorithms. This has the potential to increase the accuracy and objectivity of classification and the efficiency of sample fabrication, and it allows for large-scale data collection. Existing algorithms often exhibit challenges in identifying low-contrast materials and typically require large amounts of training data. Here, we present a deep learning model, called MaskTerial, that uses an instance segmentation network to reliably identify 2D material flakes. The model is extensively pre-trained using a synthetic data generator that generates realistic microscopy images from unlabeled data. This results in a model that can quickly adapt to new materials with as little as 5 to 10 images. Furthermore, an uncertainty estimation model is used to finally classify the predictions based on optical contrast. We evaluate our method on eight different datasets comprising five different 2D materials and demonstrate significant improvements over existing techniques in the detection of low-contrast materials such as hexagonal boron nitride.

The growing demand for reproducible, high-throughput chemical experimentation calls for scalable digital infrastructures that support automation, traceability, and AI-readiness. A dedicated research data infrastructure (RDI) developed within Swiss Cat+ is presented, integrating automated synthesis, multi-stage analytics, and semantic modeling. It captures each experimental step in a structured, machine-interpretable format, forming a scalable, and interoperable data backbone. By systematically recording both successful and failed experiments, the RDI ensures data completeness, strengthens traceability, and enables the creation of bias-resilient datasets essential for robust AI model development. Built on Kubernetes and Argo Workflows and aligned with FAIR principles, the RDI transforms experimental metadata into validated Resource Description Framework (RDF) graphs using an ontology-driven semantic model. These graphs are accessible through a web interface and SPARQL endpoint, facilitating integration with downstream AI and analysis pipelines. Key features include a modular RDF converter and ‘Matryoshka files’, which encapsulate complete experiments with raw data and metadata in a portable, standardized ZIP format. This approach supports scalable querying and sets the stage for standardized data sharing and autonomous experimentation.

Correction for “Beyond training data: how elemental features enhance ML-based formation energy predictions” by Hamed Mahdavi et al., Digital Discovery, 2025, 4, 2972–2982, https://doi.org/10.1039/D5DD00182J.

Metal–organic frameworks (MOFs) exhibit immense structural diversity and hold promise for applications ranging from gas storage and separation to energy storage and conversion. However, structural flexibility makes accurate and scalable property prediction difficult. While machine learning potentials (MLPs) offer a compelling balance between accuracy and efficiency, most existing models are system-specific and lack transferability across different MOFs. In this work, we introduce FFLAME – Fragment-to-Framework Learning Approach for MOF Potentials, a fragment-centric strategy for training transferable MLPs. By decomposing MOFs into their constituent metal clusters and organic linkers, FFLAME enables efficient reuse of chemical environments and significantly reduces the need for full-framework training data. We demonstrate that fragment-informed training improves model generalizability, particularly in data-scarce regimes, and accelerates convergence during fine-tuning. FFLAME achieves near-target accuracy on unseen MOFs with minimal additional training. These results establish a robust and data-efficient pathway toward general-purpose MLPs for the simulation of diverse framework materials.

The integration of machine learning (ML) into materials science has the potential to accelerate material discovery and optimize properties. However, the reliability of ML models depends heavily on the consistency and reproducibility of experimental data. In this study, we present a methodology to combine automated, remotely-programmed synthesis protocols with ML to enable data-driven materials discovery. Experiments were programmed and conducted remotely through robotic syntheses at cloud laboratories, using multiple different liquid handlers and spectrometers across two independent facilities (Emerald Cloud Lab, Austin, TX and Carnegie Mellon University Automated Science Lab, Pittsburgh, PA). This multi-instrument approach ensured precise control over reaction parameters, eliminated both operator and instrument-specific variability, and enabled generation of high-quality datasets for ML training. From only 40 training samples, our approach predicts whether specific synthesis parameters will lead to successful formation of copper nanoclusters (CuNCs) with interpretable models providing mechanistic insights through SHAP analysis. Our workflow demonstrates how remotely accessed/cloud laboratory infrastructure coupled with ML can transform traditionally manual processes into autonomous, predictive systems. This multi-instrument validation demonstrates reproducibility critical for reliable ML-driven materials discovery and for advancing automated materials synthesis beyond single-laboratory demonstrations.

Predicting enzymatic reactions is crucial for applications in biocatalysis, metabolic engineering, and drug discovery, yet it remains a complex and resource-intensive task. Large Language Models (LLMs) have recently demonstrated remarkable success in various scientific domains, e.g., through their ability to generalize knowledge, reason over complex structures, and leverage in-context learning strategies. In this study, we systematically evaluate the capability of LLMs, particularly the Llama-3.1 family (8B and 70B), across three core biochemical tasks: enzyme commission number prediction, forward synthesis, and retrosynthesis. We compare single-task and multitask learning strategies, employing parameter-efficient fine-tuning via LoRA adapters. Additionally, we assess performance across different data regimes to explore their adaptability in low-data settings. Our results demonstrate that fine-tuned LLMs capture biochemical knowledge, with multitask learning enhancing forward- and retrosynthesis predictions by leveraging shared enzymatic information. We also identify key limitations, for example challenges in hierarchical EC classification schemes, highlighting areas for further improvement in LLM-driven biochemical modeling.

This study presents a novel integration of unsupervised learning and decision-making strategies for the advanced analysis of 4D-STEM datasets, with a focus on non-negative matrix factorization (NMF) as the primary clustering method. Our approach introduces a systematic framework to determine the optimal number of components (k) required for robust and interpretable orientation mapping. By leveraging the K-component loss method and Image Quality Assessment (IQA) metrics, we effectively balance reconstruction fidelity and model complexity. Additionally, we highlight the critical role of dataset preprocessing in improving clustering stability and accuracy. Furthermore, our spatial weight matrix analysis provides insights into overlapping regions within the dataset by employing threshold-based visualization, facilitating a detailed understanding of cluster interactions. The results demonstrate the potential of combining NMF with advanced IQA metrics and preprocessing techniques for reliable orientation mapping and structural analysis in 4D-STEM datasets, paving the way for future applications in multi-dimensional material characterization.

Accurate electronic structure simulations of strongly correlated metal oxides are crucial for the atomic level understanding of heterogeneous catalysts, batteries and photovoltaics, but remain challenging to perform in a computationally tractable manner. Hubbard corrected density functional theory (DFT+U) in a numerical atom-centred orbital framework has been shown to address this challenge but is susceptible to numerical instability when simulating common transition metal oxides (TMOs), e.g., TiO₂ and rare-earth metal oxides (REOs), e.g., CeO₂, necessitating the development of advanced DFT+U parameterisation strategies. In this work, the numerical instabilities of DFT+U are traced to the default atomic Hubbard projector, which we refine for Ti 3d orbitals in TiO₂ using Bayesian optimisation, with a cost function and constraints defined using symbolic regression (SR) and support vector machines, respectively. The optimised Ti 3d Hubbard projector enables the numerically stable simulation of electron polarons at intrinsic and extrinsic defects in both anatase and rutile TiO₂, with comparable accuracy to hybrid-DFT at several orders of magnitude lower computational cost. We extend the method by defining a general first-principles approach for optimising Hubbard projectors, based on reproducing orbital occupancies calculated using hybrid-DFT. Using a hierarchical SR-defined cost function that depends on DFT-predicted orbital occupancies, basis set parameters and atomic material descriptors, a generalised workflow for the one-shot computation of Hubbard U values and projectors is presented. The method transferability is shown for 10 prototypical TMOs and REOs, with demonstrable accuracy for unseen materials that extends to complex battery cathode materials like LiCo_1−xMg_xO_2−x. The work highlights the integration of advanced machine learning algorithms to develop cost-effective and transferable workflows for DFT+U parameterisation, enabling more accurate and efficient simulations of strongly correlated metal oxides.

Atomistic simulations driven by machine-learned interatomic potentials (MLIPs) are a cost-effective alternative to ab initio molecular dynamics (AIMD). Yet, their broad applicability in reaction modelling remains hindered, in part, by the need for large training datasets that adequately sample the relevant potential energy surface, including high-energy transition state (TS) regions. To optimise dataset generation and extend the use of MLIPs for reaction modelling, we present a data-efficient and fully automated workflow for MLIP training that requires only a small number (typically five to ten) of initial configurations and no prior knowledge of the TS. The approach combines automated active learning with well-tempered metadynamics to iteratively and selectively explore chemically relevant regions of configuration space. Using data-efficient architectures, such as the linear Atomic Cluster Expansion, we illustrate the performance of this strategy in various organic reactions where the environment is described at different levels, including the S_N2 reaction between fluoride and chloromethane in implicit water, the methyl shift of 2,2-dimethylisoindene in the gas phase, and a glycosylation reaction in explicit dichloromethane solution, where competitive pathways exist. The proposed training strategy yields accurate and stable MLIPs for all three cases, highlighting its versatility for modelling reactive processes.