MOF–Metal Ion Adsorption Capacity Dataset (Aqueous Systems; N=209, 10 studies)
Description
This dataset contains n = 209 observations of equilibrium adsorption capacity for metal ions adsorbed by metal–organic frameworks (MOFs) in aqueous systems, obtained from 10 peer-reviewed studies. Each row corresponds to a unique experimental condition (MOF, metal ion, solution pH, initial concentration C₀, adsorbent dose). Input parameters span three descriptor groups used for machine-learning modeling and interpretability: MOF structural: BET (m² g⁻¹), PoreSize (nm), PoreVolume (cm³ g⁻¹), pHpzc Experimental: pH, C₀ (mg L⁻¹), AdsorbentDose(g L⁻¹) Ion properties: OxidationState, AtomicMass (g mol⁻¹), Electronegativity, IonizationEnergy (eV), AtomicRadius (Å) Intended use. Interpretable ML (e.g., XGBoost + SHAP) to predict metal ion adsorption capacity of MOFs and to identify governing factors across materials and operating conditions. Extrapolation beyond observed ranges is discouraged. Files in this deposit: dataset.xlsx (human-readable) and dataset.csv (UTF-8, machine-readable) README.md (context, usage) data_dictionary.md (definitions/units/ranges) references.docx and references.csv (StudyID ↔ full citation ↔ DOI mapping) code/train_ML_shap.py and requirements.txt mof-adsorption-dataset-v1.zip (all the files zipped together for one-click download)
Files
Steps to reproduce
1. Dataset was compiled from 10 peer-reviewed literature sources, each reporting the adsorption capacities of various MOF structures for different metal ions. For each entry, key physicochemical and structural characteristics of the MOFs were extracted [BET surface area (m²/g), pore size (nm), pore volume (cm³/g), and point of zero charge (pHpzc)]. 2. Articles identified in major scholarly databases using terms combining “MOF”, metal ions, “adsorption”, “aqueous media”, “adsorption capacity”. Records included when the study reported required descriptors needed for modeling. 3. Values were transcribed from tables; when only figures existed, points were digitized (Python-based digitization scripts) and rounded to the reported precision. 4. Units were harmonized. [Qₑ mg g⁻¹; C₀ mg L⁻¹; Adsorbent dose mg; BET m² g⁻¹; pore volume cm³ g⁻¹; pore size nm; pH and pHpzc unitless. Ion properties (oxidation state, atomic mass, electronegativity, ionization energy eV, atomic radius Å) retained as reported.] 5. Missing pore size values were predicted with the help of regression models trained on BET surface area and pore volume. 6. dataset.xlsx includes the curated master modeling dataset. 7. A Python 3.11 environmental was created, requirements (requirements.txt) were installed, and model training was run (python code/train_ML_shap.py).
Institutions
- Marmara Universitesi