Computational Modelling — Discovery of Dual Ion-Electron Conductivity of Metal-Organic Frameworks via Machine Learning-Guided Experimentation

3 measurement groups · 26 results

Reported values remain attached to the sample, method, conditions, extraction quality and source location that produced them.

ML classification with RF, SVM, NN and LR ensemble voting implemented in scikit-learn

CoREMOF computational screening set · Model

Models trained on ca. 52,300 OQMD inorganic compounds; CoREMOF candidates labelled conductive if two or more classifiers voted conductive; both 44- and 70-feature models used with intersection as final result.

Context: model_system
Measurement source: 1144 · 2.1 Machine Learning; 2.2.3 Ensemble Voting Analysis · Figure 1

Property	Reported value	Normalised value	Uncertainty	Origin and quality	Source
Experimentally validated dual-conductive MOFs	only two MOFs	—	—	Text Exact Reported	1146 · 3.1 Synthesis of MOFs
LR conducting precision	0.86	—	—	Table Exact Reported	1147 · 4.1 ML Predictions · Table 1
NN conducting precisionMarked as a best value within this paper	0.91	—	—	Table Exact Reported	1147 · 4.1 ML Predictions · Table 1
ML-predicted conductive MOFs	60 conductive MOFs	—	—	Text Exact Reported	1144 · Introduction
RF conducting precision	0.90	—	—	Table Exact Reported	1147 · 4.1 ML Predictions · Table 1
CoREMOF structures screened	14,000 MOF structures	—	—	Text Rounded Reported	1144 · Introduction
SVM conducting precision	0.89	—	—	Table Exact Reported	1147 · 4.1 ML Predictions · Table 1
OQMD training compounds	approximately 52,300 inorganic compounds	—	—	Text Approximate	1144 · Introduction
Cu occurrence in predicted conductive MOFsMarked as a best value within this paper	70%	0.7 fraction	—	Table Exact Reported	1147 · 4.1 ML Predictions · Table 2
Cyanide linker occurrence in predicted conductive MOFsMarked as a best value within this paper	38%	0.38 fraction	—	Table Exact Reported	1147 · 4.1 ML Predictions · Table 2
Random split Neural Network 70 features weighted precision	0.8696	—	conducting precision 0.9112; non-conducting precision 0.7798	SI Table Exact Reported	S14 · Machine learning - Model performances · Table S5
Random split Random Forest 44 features weighted precision	0.8759	—	conducting precision 0.9075; non-conducting precision 0.8077	SI Table Exact Reported	S14 · Machine learning - Model performances · Table S5
Similarity split Random Forest 44 features weighted precision	0.7584	—	conducting precision 0.7346; non-conducting precision 0.7864	SI Table Exact Reported	S14 · Machine learning - Model performances · Table S5
Similarity split SVM 70 features weighted precision	0.7421	—	conducting precision 0.6973; non-conducting precision 0.7944	SI Table Exact Reported	S14 · Machine learning - Model performances · Table S5

Five-layer fully connected neural-network regression model for predicted band gaps

CoREMOF computational screening set · Model

Regression trained on OQMD after removing metals; used to rank CoREMOF candidates rather than as an absolute band-gap predictor.

Context: model_system
Measurement source: S16 · Regression results · Figure S16

Property	Reported value	Normalised value	Uncertainty	Origin and quality	Source
Regression model R2 for band-gap prediction	R2 score of 0.65	—	—	Text Exact Reported	1147 · 4.1 ML Predictions · Figure 3
Regression model Spearman correlation	Spearman correlation of 0.78	—	—	Text Exact Reported	1147 · 4.1 ML Predictions · Figure 3
OQMD/CoREMOF feature-space silhouette score	0.1463	—	—	Text Exact Reported	1145 · 2.2.1 Features · Figure 2
MOF 1 predicted Eg from regression table	1.3328198 eV	—	—	SI Table Exact Reported	S2 · Table S1 · Table S1
MOF 2 predicted Eg from regression table	1.3328198 eV	—	—	SI Table Exact Reported	S2 · Table S1 · Table S1
Predicted band gap from regression analysis for validated MOFs	1.33 eV	—	—	Text Exact Reported	1148 · 4.2.2 Optical and Electrical Properties · Table S1
Random split 44-feature silhouette score	0.0008	—	—	SI Table Exact Reported	S15-S16 · Silhouette similarities · Figures S14-S16
Similarity split 44-feature silhouette score	0.1342	—	—	SI Table Exact Reported	S15-S16 · Silhouette similarities · Figures S14-S16
OQMD vs CoREMOF 44-feature silhouette score	0.0917	—	—	SI Table Exact Reported	S15-S16 · Silhouette similarities · Figures S14-S16

Spin-polarised DFT PDOS using VASP, r2SCAN+rVV10, PAW/PBE pseudopotentials

Density functional theory

MOF 1 as-synthesised crystals/pellet · Pellet

272-atom model of 1; 500 eV plane-wave cutoff; Gamma-centred 2 x 2 x 2 k-point sampling; convergence 1e-5 eV and force criterion 0.01 eV/A.

Geometry: periodic computational model
Context: model of pristine framework
Measurement source: S8 · Electronic band and projected density of states (PDOS) · Figure S3

Property	Reported value	Normalised value	Uncertainty	Origin and quality	Source
MOF 1 PDOS band gap	0.43 eV	—	—	Text Exact Reported	S8 · Electronic band and projected density of states (PDOS) · Figure S3
MOF 2 model relaxed lattice parameter a	21.018 A	—	—	Text Exact Reported	S8 · Electronic band and projected density of states (PDOS) · Figure S3
MOF 2 experimental lattice parameter a	21.334 A	—	—	Text Exact Reported	S8 · Electronic band and projected density of states (PDOS) · Figure S3