Tuning ORR Activity of N-Doped Biphenylene-Based Single-Atom Catalysts via DFT and Machine Learning Synergy

The oxygen reduction reaction (ORR) is critical for sustainable energy solutions, yet noble metal catalysts’ costs limit their scalability. This study investigates transition metal-doped biphenylene network (TM-BPNs) single-atom catalysts (SACs) with tailored nitrogen doping as affordable alternatives. Using density functional theory (DFT), we designed 460 TM-BPNs variants with 3d metals (Sc–Zn), evaluating their structures, electronic properties, and dual stability. Most TM-BPNs displayed quasi-metallic or semiconducting traits and robust thermodynamic and electrochemical stability, indicating synthetic viability. ORR assessments showed high potential, with V5/CCCC-Ni achieving an ultralow overpotential of 0.13 V. A novel approach combining the Extreme Gradient Boosting Regressor (XGBR) and Sure Independence Screening and Sparsifying Operator (SISSO) was developed to predict ORR performance. XGBR, with an R² of 0.96, identified key features such as the atomic number of TM (NA) and coordination environment influencing ΔG_*OH, validated by SHAP analysis. SISSO then derived a 3D descriptor (R² = 0.89) that elucidates physical properties governing catalysis, enhancing interpretability. This XGBR-SISSO synergy enables rapid screening and mechanistic insight, underscoring N-doping’s role in optimizing TM-BPNs. These findings provide a versatile framework for designing efficient, low-cost ORR electrocatalysts.