SABER: processed single-cell proteomics benchmark data and precomputed protein language model embeddings (v1.0)

Processed inputs for the SABER manuscript (sequence-anchored bilinear expression reconstruction for imputing and expanding mass-spectrometry single-cell proteomes). The archive contains the materialized seven-dataset MS-SCP benchmark slice (AnnData matrices plus external-method reference annotations), the two case-study datasets (EasySCP liver zonation, PXD070185; Brunner cell cycle, PXD024043), and per-protein protein-language-model caches (ESM-2 650M for the main results; Ankh, ESM-2 150M, ESMC 600M, ProtBERT, ProtT5, SaProt, scGPT, and fused ESM-2||scGPT caches for the prior ablation). Together with the code repository, this archive supports full retraining of every figure in the manuscript without GPU-side feature extraction. File checksums are included (MANIFEST.sha256). Underlying mass-spectrometry data are available from PRIDE (PXD071205, PXD070185, PXD024043), the Slavov Lab data portal, and SPDB.