The LESS classifier is proposed, utilizing bond orientational order (BOO) parameters as lightweight features for crystal structure classification, enabling efficient recognition of mono/binary/amorphous and other crystal structures with over 98.8% accuracy and low retraining cost for new phases.
Abstract
Accelerating data-driven materials design is critical for advancing sustainable energy, optoelectronics, and catalysis, yet traditional approaches suffer from computational inefficiency, poor generalization, or inadequate capture of complex structure-property relationships, highlighting the urgent need for systematic and effective feature engineering. Herein, we present a series of our recent work to address this challenge. We proposed the LESS classifier, utilizing bond orientational order (BOO) parameters as lightweight features for crystal structure classification, enabling efficient recognition of mono/binary/amorphous and other crystal structures with over 98.8% accuracy and low retraining cost for new phases. Building on the LESS framework, we further developed luMOD, a 24-dimensional universal descriptor integrating convoluted BOO parameters and neighbor type encoding, delivering superior performance in multispecies systems (e.g., perovskites, olivines) with minimal computational overhead. Harnessing the robust encoding capacity of large language models (LLMs), we introduced EvoMD-LLM, which abstracts MD trajectories into macro-symbolic sequences to model species-level reaction dynamics, bridging static linguistic knowledge with dynamic temporal evolution of chemical systems. Additionally, we proposed a transferable bandgap prediction framework for perovskites, integrating ChemGPT-derived atomic embeddings with global attention graph networks to capture organic-inorganic synergies and enable reliable extrapolation to unseen compositions. Collectively, these studies advance feature engineering across static/dynamic, single/multispecies, and equilibrium/reactive systems, unifying interpretability, efficiency, and transferability. They aim to nhance the reliability and scalability of data-driven materials discovery, propelling autonomous functional materials design.
Materials with negative thermal expansion (NTE) are essential for applications requiring precise control of thermal expansion. Owing to their exceptional chemical tunability, flexible architectures, and low-energy lattice vibrations, metal-organic frameworks (MOFs) represent a rich platform for exploring NTE. However, uncovering the structural motifs that govern NTE across the enormous MOF design space remains experimentally challenging, and large-scale first-principles phonon calculations are computationally prohibitive. Here, we comprehensively evaluate the factors influencing NTE in MOFs by utilizing a high-throughput workflow based on MACE-MP-MOF0, a machine learning interatomic potential fine-tuned for MOFs with near-ab initio accuracy, to construct PhononMOFdb, a database of phonons, inelastic neutron scattering spectra, bulk moduli, and heat capacities for over 12,000 MOFs. High-throughput screening of this database reveals that highly porous cubic topology frameworks with heavier, lower-valent metal nodes favor strong NTE, while linker functionalization provides a practical handle for tuning NTE magnitude and sign without compromising mechanical stability. Experimental validation via high-resolution temperature-dependent synchrotron powder X-ray diffraction on the Ce-UiO-66 MOF and its brominated variants confirms the design recipe and yields volumetric NTE coefficients surpassing current records. This work establishes a data-driven strategy for engineering NTE in MOFs, showing how machine learning-accelerated discovery and targeted experimental validation together unlock predictive materials design.
Prathami Divakar Kamath, F. Tavani, A. Elena et al.· 1 citation
High-entropy materials (HEMs) have emerged as promising electrocatalyst platforms because compositional diversity enables tunable electronic structures and abundant active sites. With rising demands in energy conversion and environmental applications, theory now plays a central role in HEM design, in particular density functional theory (DFT), molecular dynamics (MD), and machine learning (ML). Recent advances in computational strategies for elucidating mechanisms and selecting optimal compositions are synthesized. The shift from experiment-driven studies to proactive, theory- and data-driven discovery is highlighted, enabled by the integration of high-throughput calculations with ML for efficient screening. Method limits and key challenges are assessed, including black-box interpretability, modeling of disorder and coverage, and gaps between theoretical models and experimental conditions, and directions in workflow standardization, open databases, and reproducible benchmarks are outlined. Building on these elements, a compact DFT-ML-MD closed-loop framework is proposed that links atomic-scale energetics to device-level metrics, with the aim of guiding the accelerated discovery and deployment of high-activity, durable high-entropy electrocatalysts for sustainable energy.
Fangshi Fan, Weiwei Cai, Zhen Huang et al.· Advances in Materials· 0 citations
Magnetic functional alloys (MFAs) constitute a vital category of smart materials, underpinning transformative technologies by the virtue of their emergent multifunctional properties. Nevertheless, the rational design of MFAs continues to pose significant challenges, stemming from expansive compositional space, complex interplay between magnetic behavior and crystal structure, and inherent limitations in conventional experimental and theoretical techniques. In the context of big data, machine learning (ML) has arisen as a revolutionary approach, facilitating the high-throughput discovery of new MFAs with tailored functionalities while simultaneously offering insights into their fundamental physical mechanisms. This review begins by outlining a general framework of ML-driven materials design, covering key algorithms and workflow strategies. Recent breakthroughs in the development of MFAs aided by ML are then highlighted, providing a comprehensive survey of data-driven advances in their design and application. Finally, we discuss persistent challenges and prospects arising from the integration of data-driven methodologies with physics-based models to propel this rapidly evolving field forward.
This work introduces a verified and interpretable pathway for high‐throughput screening of orthorhombic perovskites and provides fundamental insight into the descriptor‐property connections governing different perovskite polymorphs.
Q. Fatima, A. A. Haidry, Usaid Ahmad et al.· Advanced Theory and Simulati...· 0 citations
The rapid integration of machine learning (ML) into organic light-emitting diode (OLED) materials design is increasingly constrained by a critical yet under-recognized limitation: the prioritisation of large datasets and complex models at the expense of chemical relevance, physical fidelity, and interpretability. This perspective argues that meaningful progress requires a shift toward a “quality over quantity” framework, prioritising domain-specific molecular representations, physics-informed descriptors, and interpretable models over generic, data-intensive black-box approaches. Emerging evidence substantiates this view. For example, the organic electronic fingerprint (OEFP), designed for conjugated systems, reduces prediction errors by approaching 50% in out-of-distribution tasks compared with conventional descriptors. Similarly, incorporating physically grounded priors, such as triplet excitation energies, bond dissociation energies, and transition dipole moments, improves predictive accuracy while reducing data requirements. In this context, interpretability must be treated as a core design principle rather than a post hoc tool, enabling mechanistic insights that inform rational molecular design. Complementary data-centric approaches, including active learning, transfer learning, and multi-fidelity modelling, further maximize the value of limited high-quality data. By shifting toward chemically informed intelligence and integrating molecular understanding with device and manufacturing constraints, OLED research can develop ML frameworks that are not only predictive but also interpretable and practically transformative for next-generation materials discovery.
Qi Zhu, Jing-Lin Xu, Qiao-Jun Li· Journal of engineering and a...· 0 citations
Metal-organic frameworks (MOFs) and MOF-like porous materials exhibit vast structural diversity and support critical applications in gas storage, separations, and catalysis. Predictive modeling remains difficult because their structure-property relationships are multiscale and cage-like, governed by both local chemical environments and global pore-network topology. These challenges, together with sparse and unevenly distributed labeled data, hinder generalization across material families. We develop an interaction topology theory and propose the interaction topological transformer (ITT), a data-efficient framework that captures materials information across multiple scales and levels, including structural, elemental, atomic, and pairwise-elemental organization. ITT extracts scale-aware features reflecting both compositional and relational structures in complex porous frameworks and integrates them through a transformer architecture for joint reasoning across scales. Using self-supervised pretraining on more than 0.6 million unlabeled structures followed by supervised fine-tuning, ITT achieves accurate, transferable, state-of-the-art predictions for adsorption, transport, and stability properties across 17 tasks, providing a principled and scalable strategy for learning-guided discovery in diverse MOF-like materials.
