Source Code and Data for : From Physical Principles to ML-Ready Representations: A Unified Framework for Scanning Probe Microscope

This record contains the original data and source code accompanying the "From Physical Principles to ML-Ready Representations: A Unified Framework for Scanning Probe Microscope", by X. Pang, M. Xu, X. Yu, Richard Gubo, Q. Song, Y. Wang, Y-W. Li, P. Ren, X-D. Wen. All materials are provided to fully reproduce the reported results. This database includes crystal structures for DFT calculations, codes for data representation construction methods, and datasets for machine learning training. Folder structure and contents==============================The compressed file 'Source_Code_and_Data.zip' is organized into five numbered directories that mirror the research workflow: 1. All_Data Crystal_Data : Crystal database used to generate surface structure models. Training_Data : Training set containing input features and corresponding output target values for machine learning. 2. Data_Prepare Construct_Structure_Model : Scripts to construct all possible termination surface models for arbitrary Miller indices.. STM_Simulation : Code to visualize STM images from DFT-calculated PARCHG files. Vector_dh : Routines that encode surface environment sites and STM image into a machine‑learning‑ready representation. Vector_PCA_Optimized : Analysis and preprocessing of the input feature vectors. 3. Data_Analysis Analysis_Alldata : Statistical analysis and visualization of the full training dataset. DH_Values_Across_Different_Currents : Comparative analysis and visualization of data acquired under different tunneling currents. 4. Predict_STM_Images : Inference scripts that apply the trained model to predict STM images for any surface structure. 5. XGBoost_Model : User manual for retraining or fine‑tuning the XGBoost model. This Code or its derivative code will not be published or otherwise distributed.