Disordered solids lack structural order and exhibit complex atomic environments that challenge conventional theoretical and simulation frameworks. This structural complexity makes it difficult to establish quantitative structure–property relationships and to rationally design new amorphous materials. Recent advances in machine learning (ML) offer powerful tools to address these challenges by enabling high‐dimensional representations of disorder, uncovering hidden correlations, and efficiently exploring expansive compositional and configurational spaces. In this review, we survey the rapidly growing interface between ML and the study of disordered solids. We first examine structural representations that connect atomic disorder with ML models, including composition/property descriptors, atomic‐density encodings, graph representations, and topology‐based methods. We then provide an overview of traditional supervised learning, modern deep‐learning architectures, unsupervised approaches, and generative models for accelerating inverse discovery. Explainable ML techniques are highlighted for their ability to relate data‐driven predictions to underlying physical mechanisms. Building on these foundations, we review emerging applications of ML in predicting global and local structure–property relationships and in enabling inverse design of disordered solids with targeted performance. We conclude by outlining key challenges and opportunities, including the need for large multimodal datasets, interpretable and transferable models, and generative frameworks capable of producing physically meaningful atomic configurations. Collectively, these developments illustrate how ML is becoming an essential tool for understanding and designing disordered solids.
more »
« less
A review of mathematical representations of biomolecular data
Recently, machine learning (ML) has established itself in various worldwide benchmarking competitions in computational biology, including Critical Assessment of Structure Prediction (CASP) and Drug Design Data Resource (D3R) Grand Challenges. However, the intricate structural complexity and high ML dimensionality of biomolecular datasets obstruct the efficient application of ML algorithms in the field. In addition to data and algorithm, an efficient ML machinery for biomolecular predictions must include structural representation as an indispensable component. Mathematical representations that simplify the biomolecular structural complexity and reduce ML dimensionality have emerged as a prime winner in D3R Grand Challenges. This review is devoted to the recent advances in developing low-dimensional and scalable mathematical representations of biomolecules in our laboratory. We discuss three classes of mathematical approaches, including algebraic topology, differential geometry, and graph theory. We elucidate how the physical and biological challenges have guided the evolution and development of these mathematical apparatuses for massive and diverse biomolecular data. We focus the performance analysis on protein–ligand binding predictions in this review although these methods have had tremendous success in many other applications, such as protein classification, virtual screening, and the predictions of solubility, solvation free energies, toxicity, partition coefficients, protein folding stability changes upon mutation, etc.
more »
« less
- PAR ID:
- 10170687
- Date Published:
- Journal Name:
- Physical Chemistry Chemical Physics
- Volume:
- 22
- Issue:
- 8
- ISSN:
- 1463-9076
- Page Range / eLocation ID:
- 4343 to 4367
- Subject(s) / Keyword(s):
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Peptide misfolding and aberrant assembly in membranous micro-environments have been associated with numerous neurodegenerative diseases. The biomolecular mechanisms and biophysical implications of these amyloid membrane interactions have been under extensive research and can assist in understanding disease pathogenesis and potential development of rational therapeutics. But, the complex nature and diversity of biomolecular interactions, structural transitions, and dependence on local environmental conditions have made accurate microscopic characterization challenging. In this review, using cases of Alzheimer's disease (amyloid-beta peptide), Parkinson's disease (alpha-synuclein peptide) and Huntington's disease (huntingtin protein), we illustrate existing challenges in experimental investigations and summarize recent relevant numerical simulation studies into amyloidogenic peptide–membrane interactions. In addition we project directions for future in silico studies and discuss shortcomings of current computational approaches.more » « less
-
Polymeric membranes have become essential for energy-efficient gas separations such as natural gas sweetening, hydrogen separation, and carbon dioxide capture. Polymeric membranes face challenges like permeability-selectivity tradeoffs, plasticization, and physical aging, limiting their broader applicability. Machine learning (ML) techniques are increasingly used to address these challenges. This review covers current ML applications in polymeric gas separation membrane design, focusing on three key components: polymer data, representation methods, and ML algorithms. Exploring diverse polymer datasets related to gas separation, encompassing experimental, computational, and synthetic data, forms the foundation of ML applications. Various polymer representation methods are discussed, ranging from traditional descriptors and fingerprints to deep learning-based embeddings. Furthermore, we examine diverse ML algorithms applied to gas separation polymers. It provides insights into fundamental concepts such as supervised and unsupervised learning, emphasizing their applications in the context of polymer membranes. The review also extends to advanced ML techniques, including data-centric and model-centric methods, aimed at addressing challenges unique to polymer membranes, focusing on accurate screening and inverse design.more » « less
-
While recent years have witnessed a fast growth in mathematical artificial intelligence (AI). One of the most successful mathematical AI approaches is topological data analysis via persistent homology (PH) that provides explainable AI by extracting multiscale structural features from complex datasets. Interpretability is crucial for world models, the new frontier in AI that can understand and simulate reality. This article investigates the interpretability and representability of three foundational mathematical AI methods, PH, persistent Laplacians (PL) derived from topological spectral theory, and persistent commutative algebra (PCA) rooted in Stanley–Reisner theory. We apply these methods to a set of data, including geometric shapes, synthetic complexes, fullerene structures, and biomolecular systems to examine their geometric, topological, and algebraic properties. PH captures topological invariants such as connected components, loops, and voids through persistence barcodes. PL extends PH by incorporating spectral information, quantifying topological invariants, geometric stiffness, and connectivity via harmonic and nonharmonic spectra. PCA introduces algebraic invariants such as graded Betti numbers, facet persistence, and ‐vectors, offering combinatorial, topological, geometric, and algebraic perspectives on data over scales. Comparative analysis reveals that while PH offers computational efficiency and intuitive visualization, PL provides enhanced geometric sensitivity, and PCA delivers rich algebraic interpretability. Together, these methods form a hierarchy of mathematical representations, enabling explainable and generalizable AI for real‐world data.more » « less
-
Structure-based drug design (SBDD) leverages the three-dimensional geometry of proteins to identify potential drug candidates. Traditional approaches, rooted in physicochemical modeling and domain expertise, are often resource-intensive. Recent advancements in geometric deep learning, which effectively integrate and process 3D geometric data, alongside breakthroughs in accurate protein structure predictions from tools like AlphaFold, have significantly propelled the field forward. This article systematically reviews the state-of-the-art in geometric deep learning for SBDD. We begin by outlining foundational tasks in SBDD, discussing prevalent 3D protein representations, and highlighting representative predictive and generative models. Next, we provide an in-depth review of key tasks, including binding site prediction, binding pose generation, de novo molecule generation, linker design, protein pocket generation, and binding affinity prediction. For each task, we present formal problem definitions, key methods, datasets, evaluation metrics, and performance benchmarks. Lastly, we explore current challenges and future opportunities in SBDD. Challenges include oversimplified problem formulations, limited out-of-distribution generalization, biosecurity concerns related to the misuse of structural data, insufficient evaluation metrics and large-scale benchmarks, and the need for experimental validation and enhanced model interpretability. Opportunities lie in integrating biomedical AI agents, leveraging multimodal datasets, developing comprehensive benchmarks, establishing criteria aligned with clinical outcomes, and designing foundation models to expand the scope of design tasks. We also curatehttps://github.com/zaixizhang/Awesome-SBDD, reflecting ongoing contributions and new datasets in SBDD.more » « less
An official website of the United States government

