Derived ProteinGym Feature Matrices for TabICL and TabPFN3 Supervised Baselines

This dataset contains derived 960-dimensional float16 protein feature matrices and matching supervised ProteinGym CV CSV files used to reproduce TabICL and TabPFN3 supervised ProteinGym substitution benchmark predictions. Contents:- official_singles_features_all/<DMS_id>/X_float16.dat- official_singles_cv_all/<DMS_id>.csv- MANIFEST.csv- SHA256SUMS- README.md- FEATURE_PROVENANCE.md Each X_float16.dat file is a raw float16 memmap with shape (number_of_rows_in_matching_csv, 960). Row counts and paths are listed in MANIFEST.csv.