State-Constrained SHAP Distills Polymer Property Models into Transferable Structural Rules for Fixed-Interval Screening

Abstract

This repository accompanies the manuscript “State-Constrained SHAP Distills Polymer Property Models into Transferable Structural Rules for Fixed-Interval Screening” submitted to the Journal of Chemical Information and Modeling. The package contains SC-SHAP source code, plotting scripts, environment files, synthetic demonstration data, manuscript figure/table source data, and a PoLyInfo reconstruction protocol. The software implements state-constrained SHAP rule extraction, fixed-interval screening, Morgan–t-SNE interpolation/extrapolation splitting, Tanimoto-distance validation, AR-WB support-floor sensitivity, and XGB-DC-K selection-budget-matched comparison. The synthetic demonstration data are provided only for software execution tests and do not reproduce the numerical results reported in the manuscript. The experimental polymer records used in the manuscript were obtained from PoLyInfo, the Polymer Database in MatNavi operated by the National Institute for Materials Science (NIMS). Because PoLyInfo/MatNavi terms of use restrict redistribution of DATA and Processed DATA, this public repository does not redistribute PoLyInfo-derived SMILES strings, property values, fingerprint input tables, or interpolation/extrapolation split input files. Instead, it provides property-specific retrieval criteria, processing metadata, and reconstruction scripts for users with authorized PoLyInfo access. For peer review only, an exact PoLyInfo polymer-ID manifest without SMILES strings or property values is provided separately through the ACS submission system as Supporting Information for Review Only.