MIST: An online tool automating mineral identification by stoichiometry

The identification of minerals is fundamental to the use and interpretation of earth and planetary materials. Minerals are defined by their chemistry and crystalline structure. A common way to identify minerals involves using instruments such as an Electron Probe Micro-Analyzer (EPMA) to measure the chemistry of a grain or crystal and compare element ratios to known minerals, i.e. stoichiometry, but this requires user expertise and often some prior knowledge of expected minerals. Here, we present MIST (Mineral Identification by SToichiometry), a mineral-stoichiometry-based model to identify geochemical observations with elemental ratios that match natural mineral compositions. MIST uses normalized oxide weight percentages and stoichiometric ratios between elements in a detailed hierarchical rules-based classification scheme based on validated mineral formulas and compositions to identify mineral phases. The model includes tolerances allowing the vacancies and elemental substitutions common in natural mineral analyses. MIST is focused on rock-forming mineral species containing oxygen and is tested against a standard dataset of validated mineral analyses. The current version of MIST, 3.0, can identify 246 mineral species or stoichiometrically indistinguishable sets of species, with the capability to expand the number of species recognized in future versions. MIST outputs precise mineral formulas, relevant mineral endmembers, and values used in intermediate calculations. As with other mineral identification methods, stoichiometric mineral identifications should be compared to other datasets, including oxide totals, textures, or structural information. We used MIST to filter over a million mineral chemistry analyses in the GEOROC database, resulting in over 875,000 natural mineral analyses with standardized labels, formulas, and mineral descriptors that can be used for machine learning models. MIST provides a rapid, accurate, standardized way to recognize minerals in high-resolution chemical datasets while minimizing required mineralogical expertise.