Curated Polyoxometalate Formula Dataset

Abstract

Reticular and cluster materials often feature complex formulas, making a comprehensive overview challenging due to the need to consult various resources. While datasets have been collected for metal-organic frameworks (MOFs), covalent organic frameworks (COFs), and zeolites, among others, there remains a gap in systematically organized information for polyoxometalates. This paper introduces a carefully curated dataset of 1984 polyoxometalate (POM) and related cluster metal oxide formula instances, currently connecting over 2500 POM material instances. These POM instances incorporate 75 different chemical elements, with compositions ranging from binary to octonary element clusters. This dataset not only enhances accessibility to polyoxometalate data but also aims to facilitate further research and development in the study of these complex inorganic compounds.

1. Summary

In material science, comprehensive datasets are indispensable for investigating complex materials such as covalent organic frameworks (COFs) [1], metal-organic frameworks (MOFs) [2], and zeolites [3,4]. These datasets facilitate the exploration of structural diversity and analysis of properties such as porosity, adsorption, and permeation. Moreover, they empower researchers to apply data intelligence in understanding trends in material properties and in predicting behaviors under various conditions, including applications in catalysis and sieving [5].

Polyoxometalates (POMs) are versatile metal-oxo clusters with diverse applications in catalysis [6], life sciences [7,8,9,10,11], and nanoelectronics [12,13]. Despite their structural complexity and promising potential in smart applications, there has been limited development of curated POM datasets crucial for advancing AI-driven technologies in exploring inorganic chemical spaces. The creation of these datasets is essential for accelerating POM chemistry through the understanding of POM speciation in solution via techniques such as nuclear magnetic resonance and mass spectrometry [14,15,16,17], utilizing POMs as building blocks in hybrid composite materials [18,19,20], and developing metallodrugs [11].

Over the recent century, the designation of POM formulas has posed significant challenges at the forefront of inorganic chemistry [21], as crystalline POM materials are among the most information-dense ones [22]. The structural complexity, intricacies of coordination bonding, and charge (de)localization make common chemical identifiers like InChI (IUPAC International Chemical Identifier) inadequate for fully capturing their nuances [23,24]. Although SciFinder has developed some collections of POM information, the proprietary nature of this database limits access to data [25]. Consequently, the development of open and curated datasets of POM formulations could significantly accelerate research and collaboration within the POM community.

In this work, we present the recently completed “Curated Polyoxometalate Formula Dataset”, now published on a Git platform under a Creative Commons License. This dataset addresses a critical data gap by linking POM formulas with their corresponding materials and provenances, specifically through Digital Object Identifiers (DOIs) [26]. It facilitates searches based on elements, charges, molecular mass, and POM formulas. Additionally, the structure and implementation of the dataset are designed for future reusability and expansion, enabling the development of new data-driven methods for exploring POM synthesis and other related research avenues.

2. Data Description

Polyoxometalates (POMs), as cluster materials, are featured as crystallographic motifs in a variety of POM-containing or POM-based materials. For instance, the $\alpha$-Wells–Dawson polyoxometalate composed of P, W, and O atoms with the formula ${\left[{\left({\mathrm{PO}}_4\right)}_2{\left({\mathrm{WO}}_3\right)}_{18}\right]}^{6-}$, as illustrated in Figure 1, exemplifies such motifs. This specific POM, reported in diverse materials like ${\mathrm{Li}}_6\left[{\left({\mathrm{PO}}_4\right)}_2{\left({\mathrm{WO}}_3\right)}_{18}\right]·28{\mathrm{H}}_2\mathrm{O}$ [27], showcases the potential for various formula descriptions within the same POM motif. The given formula, ${\left[{\left({\mathrm{PO}}_4\right)}_2{\left({\mathrm{WO}}_3\right)}_{18}\right]}^{6-}$, is a coordination formula easily interpreted by chemists, indicating an overall charge of 6⁻ and a composition that can be described as ${\mathrm{P}}_2{\mathrm{W}}_{18}{\mathrm{O}}_{62}$. This leads to possible empirical derivations like ${\mathrm{PW}}_9{\mathrm{O}}_{31}$, which can be explored further in synthetic studies.

Consider the connectivity outlined in Figure 2A, which depicts a data connection schema for polyoxometalates and polyoxometalate materials. Each material is linked to its formula and a Digital Object Identifier (DOI), encapsulating a POM as a distinct entity within the class, assigned a unique identifier. This POM entity connects to various attributes such as elemental composition, from which both the molecular and POM formulas—typically serving as coordination formulas—are derived. These features, coupled with specific charge and molecular mass data, enable domain chemists to label and classify the structure effectively.

In terms of implementation (see Figure 2B), our JSON data structure organizes information hierarchically about a specific POM entity, encapsulated within a top-level object. For example, we assign a unique four-letter key (e.g., “POM_GUKA”) as a POM identifier, referencing the phosphotungstate Wells–Dawson structure from Figure 1. This key points to a nested JSON object detailing the POM’s properties, such as formulas, elemental composition, and related materials. Keys like “POM Formula” and “Molecular Formula” store formulas as strings, “Contains Elements” maps to an object listing each element’s count, “Molecular Mass” is recorded as a string for precision, and “Charge” is stored as a floating-point number. An array under “Labels” collects descriptive tags, while “POM Material” includes nested objects with unique identifiers that detail each material’s formula and DOI, enhancing data retrieval for scientific analysis and database integration.

3. Methods

For the creation of the curated polyoxometalate formula dataset, we have meticulously compiled chemical data on POMs by studying and analyzing over 1300 distinct research papers, each identified by a unique Digital Object Identifier (DOI). Following the collection and structuring of POM formulas, we have carefully derived their more simplified molecular formula, which has enabled us to programmatically parse them into involved elements and overall formula charges. The simplified molecular formula also provided the quantity of each present element, which has enabled us to calculate the respective molecular weights. The resulting dataset currently includes 1984 unique entries, linking to more than 2500 POM-containing materials. Given the crucial role of nuclearity and charge in characterizing POMs, our data are presented in terms of overall charge plotted as a function of molecular mass, as illustrated in Figure 3A. This panel depicts the broad distribution of molecular weights and corresponding charges, highlighting the chemical diversity within polyoxometalates. Notably, extreme examples such as the polyoxomolybdate [Mo₃₆₈O₁₀₃₂H₁₆(H₂$O$)₂₄₀(SO₄)₄₈]⁴⁸⁻ [28], the polyoxotungstate [(P₈W₄₈O₁₈₄){(P₂W₁₄Mn₄O₆₀)(P₂W₁₅Mn₃O₅₈)₂}¹⁵²⁻ [29], the dense polyoxouranate [(U$O_2$)₁₂₀(C₂O₄)₉₀]¹⁸⁰⁻ [30], and the robust polyoxoniobate [Nb₂₈₈O₇₆₈(OH)₄₈(CO₃)₁₂]¹⁸⁰⁻ [31] are easily distinguishable, demonstrating the current extents of POMs’ molecular diversity.

A further qualitative analysis of the plot in Figure 3B reveals that as the charge increases, while the molecular mass decreases or remains constant, there are fewer or no POMs observed. This trend suggests an over-reduction of the addenda centers, which likely requires coordination changes not captured by this dataset. Nonetheless, this pattern can serve as empirical guidance to evaluate the feasibility of proposed POM formulas. Focusing on the region characterized by a charge of less than −75 and a molecular mass below 30,000 g/mol, we observe that this is where the majority of POMs are situated. By employing different colors for analysis, it becomes apparent that many polyoxomolybdates are located in the high-molecular-mass domain but have a relatively low overall charge. Conversely, polyoxoniobates appear as one of the most negatively charged species with lower molecular weights, highlighting the diverse charge and mass relationships within POMs.

On the left-bottom corners of our molecular mass versus overall formula charge diagrams in Figure 3, we do not observe any data points that could be related with synthesized and characterized POMs. As the molecular mass increases, the POMs tend to acquire specific charges. However, the section remains incomplete, reflecting an empirical expectation of overcharging, which is particularly relevant for battery research. The empirical boundaries defined by our analysis challenge claims such as those for supercharged Kegginoidal $\alpha$-[PMo₁₂₄₀]²⁷⁻, with Keplerate architectures [Mo₇₂V₃₀O₂₂₅]⁴⁹⁷⁻ being positioned outside the typical range observed for known POM structures [32,33]. This observation suggests that such extreme proposals might not conform to established POM chemistry, highlighting the necessity for rigorous scrutiny.

The current set of POM entries is primarily dominated by 1003 polyoxotungstate formulas. Careful analysis of these formulas shows that 82 of them are isopoly in character, that is, they contain a metal and oxygen ions or metal, oxygen, and hydrogen ions. Among the classical POMs, we also see a tendency for mixed-addenda species, namely vanado-molybdates (15), tungsto-niobates (42), vanado-niobates (9), vanado-tungstates (29), and molybdo-tungstates (9) instances. Among the polytungstates, phosphorous is the most common heteroelement, represented in over 400 POM formulas, which can be explained due to the frequent use of lacunary phosphotungstates in the development of many POM species. On the other hand, in polyoxomolybdates, polyoxovanadates, polyoxoniobates, and polyoxotantalates, carbon appears as the most common heteroelement, which can be related to the development of hybrid organic–inorganic POM species. The prevalence of tungstates is likely due to tungsten’s ability to form strong, non-labile bonds in high oxidation states and to generate lacunary POM species that facilitate further chemical reactions, enhancing the diversity of tungstate-based POMs. Following tungstates are polyoxomolybdates with 465 entries, polyoxovanadates with 209 entries, and polyoxoniobates with 100 entries, as well as an additional 207 entries based on other addenda elements (see Figure 4A), including some positively charged metal-oxo clusters. The dataset showcases chemical diversity with 75 unique elements across all entries, illustrating a broad range of structural combinations. Specifically, the distribution of unique elements per POM varies from two to eight, with the majority of structures, 587, containing five elements (see Figure 4B), often reflecting the integration of heteroatoms and hybrid species involving carbon, which underscores the complex compositional variety within this dataset. Inorganic functionalization with 3d and 4f centers is mainly prevalent for polyoxotungstates, being mainly driven by the modular use of the lacunary species. On the other hand, 3d functionalization is less prevalent among polyoxovanadates, while 4f functionalization is less prevalent in polyoxoniobates (see Figure 4C,D).

4. Conclusions

This paper presents a comprehensive dataset of nearly 2000 POMs, currently featuring details such as molecular formulas, charges, and DOI identifiers, with an aim for a future expansion to include topological, synthetic, and functional properties. The dataset is designed to evolve, incorporating charge distributions and mass projections to enhance material data projects through machine learning and cheminformatics. Upcoming enhancements will also cover polyoxometalates’ speciation in various environments to assess protonation levels, isomerism, and charge distribution for automated computational analyses, as well as integrated components for retrosynthetic strategies, significantly advancing both the fundamental and applied research in polyoxometalate chemistry.

The curated POM dataset will enhance the ongoing development of the domain-specific OntoPOM ontology, encompassing all aspects of POM chemistry to render POM data machine-actionable. Furthermore, the expanded semantic description of POMs and the training of large language models based on these data will improve discoverability in the field for both experts and students, similar to the recent advancements in zeolitic crystalline materials and metal-organic polyhedra [34,35]. This detailed semantic framework will facilitate the functional compartmentalization of POM geometries in digital spaces, supporting the advanced, data-driven exploration of POM chemistry [36].