Modular Ontology-Driven Data Capturing for Machine Learning-Assisted Supported Catalytically Active Metal Solutions Design

Modular Ontology-Driven Data Capturing for Machine LearningAssisted Supported Catalytically Active Metal Solutions Design Figure 1. Schematic representation of SCALMS in selective ethylene oligomerization to butene. Philipp Stangl1,*, Richard Lenz2, Marco Haumann1 1Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU), Lehrstuhl für Chemische Reaktionstechnik, Erlangen, Germany, Corresponding author: [email protected] 2Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU), Lehrstuhl für Datenmanagement, Erlangen, Germany Figure 2. Descriptors that characterize SCALMS catalysed reactions. Conclusions Modular ontology modelling facilitates understanding, modification, and adaptation by organizing it as an interconnected set of modules, each resonating with conceptualisations of domain experts. Reduced efforts for merging data sets by different researchers for a machine learning use case. Refrences [1] Søgaard et al., Catal. Sci. Technol., 2021, 11, 7535, [2] Taccardi et al., Nat. Chem., 2017, 9, 862, [3] Shimizu et al., Semantic Web, 2023, 14, 459, [4] Takahashi et al., ChemCatChem, 2019, 11, 1146 Funded by the European Research Council - Project number 786475 (SCALMS) Funded by the German Research Foundation (DFG) - Project number 441926934 (NFDI4Cat) Future Work Extending our work for the modelling of composite concepts. Modelling different reaction systems. Using our approach in an inter-institutional use case scenario, where each institution contributes data for a machine learning data set. Supported Catalytically Active Liquid Metal Solutions SCALMS consist of catalytically active metal (Ni) alloyed with a low melting metal (Ga) deposited on a porous support (SiO2)1 Metal alloy droplets become liquid under reaction conditions2 Reaction takes place at highly dynamic liquid metal/gas interface2 Study Objective Employing the modular ontology modelling approach3to capture experimental data from various researchers. The data is used for machine learning training for SCLAMS catalyst discovery.4 Capturing Experimental Data Integrate diverse raw data formats from experimental setups (text, spreadsheets, databases). Utilize ontology-based forms to systematically describe experimental designs, contexts, and methods. Output a local knowledge graph that documents and interlinks experimental metadata and results. Domain Modules Clearly define catalytic materials, their support structures, and related entities. Model relationships and properties specific to catalysis research (e.g., material support, identifiers). Link domain-specific concepts clearly and systematically using modular ontology patterns. Use Case Modules Support comparison of experimental data across different research studies and experiment setups. Define use case specific concepts. E.g., KPIs for SCALMS catalyst performance. Module can be extended by e.g., new performance indicators, by creating new subclasses. General Purpose Modules Define universally applicable concepts like identifiers and quantities. Employ standardized datatypes and encoding schemes to enhance data consistency and interoperability. Consistent representation and interoperability across different application domains. BFO:MaterialEntity Catalyst Support Identifier Material subclassOf hasSupport hasIdentifier ofMaterialType KeyPerformanceIndicator Activation Time Product Selectivity Quantity subclassOf … Average Selectivity subclassOf ofQuantity Identifier Identifier Encoding IdentifierType IdentifierSource IdentifierEncoding Scheme hasIdentifierEncoding hasIdentifierType xsd:string hasString Representation hasIdentifier EncodingScheme hasIdentifierSource Observation QuantityValue Voc4Cat:Unit hasQuantityValue hasUnit hasValue xsd:float Quantity Identifier Encoding Identifier Specifications modeled as Subclasses Other KPIs can be added as necessary Linking to data that is described in another module Datatype of captured values Using vocabulary from Voc4Cat Nested module within Identifier Catalyst Performance