ML4Tc

ML4Tc is an interpretable machine-learning framework to predict the superconducting critical temperature (Tc) of crystalline materials directly from their crystal structures and chemical compositions. We extract physics-informed descriptors — covering electronic structure, elemental composition, and crystallographic parameters — from CIF files and train an XGBoost regression model. SHAP (SHapley Additive exPlanations) is used to quantify each descriptor's contribution. A case study comparing Mg2IrH6 (superconducting) and Ca2IrH6 (non-superconducting) shows the model captures the contrasting behavior and the subtle descriptor differences driving it.

Highlights

• Physics-informed descriptor extraction from CIFs (electronic, elemental, crystallographic).
• XGBoost regression for accurate Tc prediction across diverse superconducting families.
• SHAP-based interpretability to link descriptor variations to physical outcomes.
• Case study: Mg2IrH6 vs Ca2IrH6 — model reproduces observed superconducting contrast.

Quick start

Create the conda environment and install dependencies:

conda create -n ml4tc python=3.10
conda activate ml4tc
pip install -r env.txt

Citation

If you use this code in published work, please cite the project and include a reference to the repository.