Add precomputed ML mechanistic labels as parquet (ml_preds/)

Add Colab notebook for mechanistic label classifier

Fix disk space management in join_rasp_ddg: skip re-extraction, delete early

Replace vaex with h5py for RaSP conversion (no gcc required)

Add RaSP ΔΔG JOIN + proteome-wide download script

Bump CACHEBUST=8 + route esm1b/ parquets to separate data dir

Fix upload_large_folder: bake esm1b/ prefix into local dir structure

Switch to upload_large_folder + persistent output dir for resumable upload

Add ESM1b/VESM++ JOIN to backend get_db() + fix hf_index scoping

Fix join_vesm_scores: bulk download + local join + single upload_folder

Add VESM++ join script for proteome-wide ESM enrichment

Add ESM2 LLR pipeline for full proteome (20K proteins)

CACHEBUST=5 + fix download script to always overwrite parquets

refactor: serve parquet from local disk — baked in at build time